JEDEE AI
更新时间 07-17 15:05

今日更新 196 条 · 24h 曝光 2572.8 万 ▼14% · 活跃事件 26 · 今日论文 4

今日焦点 每小时更新 · 按 24 小时内热度排序

1Kimi.ai:推出 Kimi K3:开放前沿智能12 小时前 · 814.9 万2Sam Altman:过去12个月我们表现不是最好的,这主要是我的责任,但我们即将迎来有史以来最好的12个…12 小时前 · 150.8 万3OpenAI Developers:认识一下 kbd-1.0-codex-micro,由 @work_louder 打造…1 天前 · 442.5 万4Gemini Notebook:3年前,我们只是一个小实验,想帮你学得更快。13 小时前 · 52 万5Grok:Grok 现在支持自动化功能啦:10 小时前 · 25.7 万6Mira Murati:我们的首个模型 Inkling。从零开始训练,权重开放,今天可在 Tinker 上进…1 天前 · 130 万7Guillermo Rauch:Kimi K3 在 nextjs.org/evals 评测中表现最佳,超越 Fabl…7 小时前 · 13.3 万8Sundar Pichai:很高兴看到14 小时前 · 22 万9Lisan al Gaib:Google 是真的 gg 了10 小时前 · 14.3 万10Noam Brown:2023年:大语言模型在小学四年级数学题上都费劲16 小时前 · 23.8 万

能上手的 产品 → 玩法 → 一线博主教程 · 自动收录玩法库全部 70 条 →

全部情报 每小时更新 · 事件已合并同类项

提交账号

填 @用户名 或主页链接,审核通过后收录进情报站。

内容 公司
swyx@swyx · 博主 · 33 分钟前知名 AI 播客 Latent Space 主理人

卧槽,React 复仇者联盟集结了

引用 Guillermo Rauch @rauchgI’m excited to welcome two legends of developer tools, Pete Hunt ( @floydophone ) and Nick Schrock ( @schrockn ), to Vercel. Pete was one of the pioneers of @reactjs at Meta. He made an early bet to power Instagram Web with ⚛️ React, evangelizing it internally and externally. He will be running Frameworks and leading @nextjs . I couldn’t imagine a better person to lead React’s most popular framework to even greater heights. Nick co-invented @graphql , solving some of the gnarliest data infrastructure and access issues at Facebook scale, with a delightful developer experience. He will be working on Agentic Developer Experience, solving the problem of enabling the next billion agents and leading the way to a future of self-improving software. It’s a dream-come-true for a founder of a startup to welcome engineering minds of this caliber who are also wonderful humans. You probably want to work with them, and they’re hiring 😁. Their DMs are open, from job applications to bug reports!查看被引原帖 ↗
查看英文原文
HOLY FUCKING SHIT the react avengers have assembled
Greg Brockman@gdb · 创始人 · 35 分钟前Greg Brockman,OpenAI 联合创始人兼总裁

我们的团队在快速响应反馈,不断迭代。我们❤️用户们,谢谢大家!

引用 Tibo @thsottiauxEvening! We’ve gotten lots of great feedback on the new ChatGPT desktop app (which we didn't get totally quite right on the first try), and as a result, we've made some changes. 1/ ChatGPT conversation history and projects are now visible in the sidebar. Also, your Chat and Work history now sync across web, mobile, and desktop. Local tasks still stay on your computer. 2/ You can now easily switch between Chat and Work modes inside ChatGPT on desktop, which is now also consistent with how it shows on web and mobile. 3/ Nothing is changing for users on Codex mode. It's still the OG and best at what it does. And overall we're continuing to fix paper cuts and improve performance, reliability, and efficiency. Keep up the feedback, hope you like the updates!查看被引原帖 ↗
查看英文原文
team is responding to feedback and iterating quickly. we ❤️ our users, thank you all!
宝玉@dotey · 中文博主 · 37 分钟前宝玉,中文圈 AI 翻译与科普大 V

最早我做字幕翻译的时候,语音转文字是用的 Whisper,但是 Whisper 有几个问题:

一个问题是时间戳不准,时间戳不准拆分的就是就可能字幕和语音对不上,很影响效果,需要人工去从头到尾校对一遍

一个问题是中英文混排支持不好,中文其实支持也不算太好,所以中文的转录就效果很差

一个问题是不直接支持识别发言人

虽然这些问题都可以通过收费的云端模型来解决,但是成本相对高一些。

最近测试了 Qwen3 ASR,效果是真的很不错,配合 Qwen3-ForcedAligner 模型,可以识别后把词级时间戳对的很精准。0.6b 就足够了,本地运行资源占用也不高。
github.com/QwenLM/Qwen3-ASR


发言人识别也有开源的模型:Pyannote + WeSpeaker。不过这个发言人一多尤其是一起说话识别的不够准,但配合 Agent 结合上下文就能做到相对比较准了。

如果真的要求高,也可以走云端模型,比如火山引擎上的豆包录音文件识别模型 2.0 就质量挺好,速度也快,就是要额外花钱。

引用 宝玉 @dotey我在开发 BaoCut 这个 App 的时候,是基于一个 Loop 来的: 1. 在开发新功能之前先设计原型(参考图1),借助的是 baoyu-design skill ( github.com/jimliu/baoyu-desi… ),配合 Claude Code App 内置的浏览器实施预览调整,模型 Opus 4.8 就很好了,都不需要 Fable 5. GPT 5.6 Sol 设计能力还是不如 Opus 4.8 2. 原型打磨好了后,只需要在同一会话内,让 Claude Code 基于新的 UI 设计去实现功能即可,这块 Claude 做的很好,Fable 5 效果最好,能将设计稿几乎 1:1 还原,如果修改不多 Opus 4.8 也能胜任。 这些 UI 的打磨我还是更放心让 Fable 和 Opus 而不是 GPT,但其他一些不涉及 UI 部分的 GPT 5.6 Sol 就做的很好。 3. 更新好了后测试没问题,就可以通过发布的 skill 发布新版本。 这里可以放心让 Codex 去做了,尤其是它的 CloudFlare Plugin 很好用,直接帮助发布更新安装包到 CF。 这个 loop 的每一个迭代的起点是自己的想法,让 AI 提供设计方案,和 AI 反复讨论后确定方案,然后 AI 实施,AI 实施完成后人再去验证和当初想要的是否一致,如果不一致再让 AI 调整甚至推翻重来。查看被引原帖 ↗
Greg Brockman@gdb · 创始人 · 46 分钟前Greg Brockman,OpenAI 联合创始人兼总裁

ChatGPT Work 非常适合处理你日常生活中的各种任务:

引用 Andrew Ambrosino @ajambrosinothe “work” in ChatGPT Work is describing what the agent is doing for you– you can use it for anything you want I have it reading all of my personal emails, making calendar events, organizing docs for myself and the kids, etc查看被引原帖 ↗
查看英文原文
ChatGPT Work is great for tasks in your personal life:
Amjad Masad@amasad · 创始人 · 1 小时前Amjad Masad,Replit 创始人兼 CEO

$NVDA 这波应该被 K3 新闻带起来啊。反而跌了。

查看英文原文
$NVDA
should be pumping on the K3 news. Instead it’s down.
Bindu Reddy@bindureddy · 创始人 · 2 小时前Abacus.AI CEO,AI 行业观点博主

Kimi K3 缩小了差距但仍排在前沿模型之后

我们的 LiveBench 基准有很多隐藏问题,所以模型很难靠记忆取巧

K3 是最好的开源模型,但表现不如 Opus 4.8、Sol 和 Fable

在实际应用中,Kimi 输出很冗长,处理接近 Opus 级别问题的成本和 Opus 4.8 一样,但速度慢得多

查看英文原文
KIMI K3 CLOSES THE GAP BUT RANKS BEHIND FRONTIER MODELS

Our benchmark, LiveBench, has a lot of hidden questions so that models it's hard to memorize it

K3 is the best good open-source model but is below Opus 4.8, Sol and Fable

Also in practice, Kimi spins a lot and costs as much as Opus 4.8 for near opus-class problems - it's also much slower
AshutoshShrivastava@ai_for_success · 博主 · 2 小时前高频 AI 新闻与产品动态博主

Kimi K3 好得绝

查看英文原文
Kimi K3 is shockingly good.
歸藏(guizang.ai)@op7418 · 中文博主 · 2 小时前歸藏,中文圈 AI 工具与提示词博主

藏师傅的 Kimi K3 测评来了,这次确实非常牛逼!

可以说是一个比较小的 DeepSeek 时刻

我直接拿它跟 Opus 4.8 做了对比测试。

从结果来看,互有胜负。

在复杂前端和复杂开发的情况下,我觉得它跟 Opus 4.8 差不多是相当的水平。

至于 Fable 5 和 5.6,我觉得还差一些,但已经是一个非常牛逼的成绩了。

Ethan Mollick@emollick · 创始人 · 2 小时前沃顿商学院教授,AI 应用研究权威

Kimi K3 确实不错,但人们又开始过度迷信 Arena 排名了(还记得 Llama 4 吗?)

Arena 用户投票的 ELO 评分本身就有局限,前端也就是个文本聊天而已,在这种主观评比下相对容易通过训练或调整 system prompt 来达到用户偏好的效果。

引用 Arena.ai @arenaBig news: Kimi-K3 by @Kimi_Moonshot is now #1 in the Frontend Code Arena with 1679 pts, surpassing Claude Fable 5. This is a 17-place jump from Kimi-k2.6 (#18 -> #1). In Frontend, Kimi-K3 ranked #1 in 6 of 7 domains: Brand & Marketing, Reference-Based Design, Data & Analytics, Consumer Product, Simulations, and Content Creation Tools, landing #2 only in Gaming behind Fable 5. The full model weights will be released by July 27. Congrats to the @Kimi_Moonshot team on this major milestone!查看被引原帖 ↗
查看英文原文
Kimi K3 is a very good model, but people are overindexing on an Arena score again (remember Llama 4?)

ELO scores as judged by Arena users are limited, and front-end is like text chat, relatively easy to train/system prompt to a state that people prefer when it is subjective.
Lisan al Gaib@scaling01 · 博主 · 3 小时前高频 AI 模型测评与爆料博主

如果排除 OpenAI 和 Anthropic 在 Kimi K2 之前的模型,赶超时间线会被推到 Q1,因为最近的进展明显比 2025 年前快得多

查看英文原文
btw if you exclude OpenAI and Anthropic models pre Kimi K2 it shifts the overtaking slightly further out to Q1, because progress has obviously been much faster recently than before 2025
Gorden Sun@Gorden_Sun · 中文博主 · 3 小时前中文圈高频 AI 资讯与开源项目博主

Codex又开始了新一轮的老带新活动,这次最多可获得10次重置次数

Gorden Sun@Gorden_Sun · 中文博主 · 3 小时前中文圈高频 AI 资讯与开源项目博主

Kimi K3生成的版本,也非常好。

引用 Gorden Sun @Gorden_SunPPT Skill都可以扔了,没有使用Skill,没有使用图片生成,Fable 5生成PPT的效果。 提示词: 不要使用任何技能,做一个16:9比例的PPTX文档,内容为SpaceX的发展历程,5页。要求内容丰富、排版复杂豪华精美,有装饰图片和图标,有配图。图片可以联网获取,图片应该尽量使用透明底的图片。查看被引原帖 ↗
歸藏(guizang.ai)@op7418 · 中文博主 · 3 小时前歸藏,中文圈 AI 工具与提示词博主

Codex 昨晚的更新在交互上终于对味了:

1. 左上角的切换:从 Work 和 Codex 切换成 ChatGPT 和 Codex

2. 历史聊天整合:ChatGPT 的历史聊天全部并进了左边“最近聊天”里面,你可以筛选是普通聊天还是 ChatGPT Work 的任务

3. 顶部导航:分了 Chat 和 Work 两个 Tab,交互逻辑跟网页版和移动端一致了

Ethan Mollick@emollick · 创始人 · 3 小时前沃顿商学院教授,AI 应用研究权威

基于GPT-4的巴基斯坦法官助手让他们处理的案件数量增加了6%,对质量没有任何影响。

引用 Elliott Ash @ellliotttWhat happens when you roll out custom generative AI to half a country's judges? New paper on Pakistan's courts with @ProfSultanEcon and @gochristoph . In line with wemustactnow.ai -- we provide early empirical evidence on the impacts of transformative generative AI.查看被引原帖 ↗
查看英文原文
A GPT-4 powered assistant for Pakistani judges increased the amount of cases they saw by 6% with no impact on quality.
向阳乔木@vista8 · 中文博主 · 3 小时前向阳乔木,中文圈 AI 工具与趋势博主

如果只能推荐一个去 AI 味设计Skill。

那必须是大神 emil 的作品,而且动效超赞。

安装指令:
npx skills add emilkowalski/skill

Lisan al Gaib@scaling01 · 博主 · 4 小时前高频 AI 模型测评与爆料博主
连环推 ×3

算了吧,拿出点胆量来
做出你的预测吧

不像 Nathan 发的那样:
"开源和闭源模型之间的差距 ~0 个月"
然后又删了

引用 Lisan al Gaib @scaling01MoonshotAI will overtake OpenAI and Anthropic before the end of the year or will they? at least that's what the hype kiddies on X want you to believe So let me make it falsifiable. They are saying: - China / MoonshotAI is catching up - they are catching up generally (not just coding, but almost all domains and including restricted models like Mythos 5) - the gap is currently ~1.4 months based on Artificial Analysis Index and benchmarks provided by MoonshotAI, where Kimi-K3 beats Opus 4.8 in 30 of 35 benchmarks, and GPT-5.6-Sol in 19 of 35 benchmarks (they ignore the existence of all Mythos variants) - China is not catching up due to distillation, so they should overtake US labs Their implicit prediction then is: - a chinese model / MoonshotAI will overtake Anthropic (and OpenAI) on the Artificial Analysis Index by: - Median: 2026-12-24 (80% CI: 2026-09-17, 2027-09-14) Since they claim that chinese models are as general as american models, we should see unsolved mathematics, physics, and more being solved by chinese models at higher rates than american models. Speaking to its generality Kimi-K3 should surpass Opus 4.8 and GPT-5.6-Sol on the majority of these benchmarks: - METR Time Horizons, FrontierCode, MirrorCode, UK AISI cyber ranges, ExploitBench/ExploitGym, CritPT, FrontierMath T4, ARC-AGI-2 / ARC-AGI-3, WeirdML, ALE-Bench, GSO, MRCR2/GraphWalks - vibes --- Some other things that are more speculative and downstream of China overtaking US models: - more involvement by the USG - stricter export controls on semis - potentially a Manhatten-style project, as we will be behind in 2027 and are racing against China - also in the cards: US banning chinese models or US labs distilling from chinese models --- I have already stated my position clearly. Chinese models are generally ~6-8 months behind, with some domains like coding behind slightly less. Kimi-K3 did not significantly shift my estimate on the gap and it currently does not change my outlo查看被引原帖 ↗
查看英文原文
come on, show some balls
make your predictions

not like Nathan tweeting something along the lines:
"gap between open and closed models ~0 months"
and then deleting it again
i hope you have already seen the issue here

they are including Mythos in their statements but then completely ignore it when it comes to the actual benchmarks
so what if those things don't happen?

well, then Chinese models are not catching up and further behind than just 1.4 months and/or the Artificial Analysis Index is in fact not measuring general capability but just a slice
or chinese models are in fact distilling and therefore can't overtake and pull away from US models

i had to keep this short. I need to sleep.
Lisan al Gaib@scaling01 · 博主 · 4 小时前高频 AI 模型测评与爆料博主

MoonshotAI 会在年底前超越 OpenAI 和 Anthropic

或者不会呢?至少 X 上的炒作小伙伴想让你这样信。

那我就把这个假设变成可证伪的。他们在说:
- 中国 / MoonshotAI 在追赶
- 他们全面追赶(不仅编码,几乎所有领域,包括受限模型比如 Mythos 5)
- 根据 Artificial Analysis Index 和 MoonshotAI 提供的基准,目前差距约 ~1.4 个月,Kimi-K3 在 35 个基准中击败 Opus 4.8 30 个,击败 GPT-5.6-Sol 19 个
(他们直接忽视了所有 Mythos 变体)
- 中国模型追赶不是靠蒸馏,所以他们应该会超过美国实验室

隐含的预测就是:
- 某个中国模型 / MoonshotAI 会在 Artificial Analysis Index 上超越 Anthropic(和 OpenAI)的时间:
- 中位数:2026-12-24(80% 置信区间:2026-09-17,2027-09-14)

既然他们说中国模型和美国模型一样通用,那就应该看到中国模型在解决未竟的数学、物理问题上的成功率超过美国模型。

就通用性而言,Kimi-K3 应该在这些基准的大多数上击败 Opus 4.8 和 GPT-5.6-Sol:
- METR Time Horizons、FrontierCode、MirrorCode、UK AISI 网络靶场、ExploitBench/ExploitGym、CritPT、FrontierMath T4、ARC-AGI-2 / ARC-AGI-3、WeirdML、ALE-Bench、GSO、MRCR2/GraphWalks
- 凭感觉

---

还有更投机的下游影响,来自中国超越美国模型这个事儿:
- USG 参与度上升
- 对芯片的出口管制更严
- 可能搞个曼哈顿计划式的项目,因为 2027 年咱们要落后,正在跟中国竞速
- 也有可能:美国禁中国模型或美国实验室从中国模型蒸馏

---

我的立场已经表得很清楚了。
中国模型大概落后 6-8 个月,有些领域比如编码落后稍少一些。

Kimi-K3 没有显著改变我对差距的估计,也还改变不了我对未来的看法,但等我们拿到我上面提的所有基准数据以后,就能看得清清楚楚了。

我这个立场的主要原因:
- Kimi-K3 连 Mythos Preview 都没超过,而那是个 ~5 个月前的模型
- 未来几个月咱们可能看不到比 Kimi-K3 更大的开源模型,可能得等到 2027 年初中期
- 与此同时,Anthropic 从 2 月起就在研一个 10T 的模型,OpenAI 可能刚训完 GPT-6,规模也在这儿,SpaceX AI、Google 和 Meta 的 10T 参数美国模型也在路上。
- 咱们现在看不到模型的真正前沿。Anthropic 和 OpenAI 在保守应对,因为发布新前沿模型的法律局面不清楚。
- 历史上中国模型比美国模型更追求基准分,意味着他们的基准数字转化成真实世界性能的效率不如美国同行
- GPT-5.6-Sol 在 Artificial Analysis Index 上的 token 效率仍然比 Kimi-K3 高 2-3 倍(而且可能更小,约 2T)
- 美国实验室算力更多

---

我对 Kimi 兄弟们发这个模型很高兴。
这是个牛逼的模型,可能是第一个真正好用的中国模型。

查看英文原文
MoonshotAI will overtake OpenAI and Anthropic before the end of the year

or will they? at least that's what the hype kiddies on X want you to believe

So let me make it falsifiable. They are saying:
- China / MoonshotAI is catching up
- they are catching up generally (not just coding, but almost all domains and including restricted models like Mythos 5)
- the gap is currently ~1.4 months based on Artificial Analysis Index and benchmarks provided by MoonshotAI, where Kimi-K3 beats Opus 4.8 in 30 of 35 benchmarks, and GPT-5.6-Sol in 19 of 35 benchmarks
(they ignore the existence of all Mythos variants)
- China is not catching up due to distillation, so they should overtake US labs

Their implicit prediction then is:
- a chinese model / MoonshotAI will overtake Anthropic (and OpenAI) on the Artificial Analysis Index by:
- Median: 2026-12-24 (80% CI: 2026-09-17, 2027-09-14)

Since they claim that chinese models are as general as american models, we should see unsolved mathematics, physics, and more being solved by chinese models at higher rates than american models.

Speaking to its generality Kimi-K3 should surpass Opus 4.8 and GPT-5.6-Sol on the majority of these benchmarks:
- METR Time Horizons, FrontierCode, MirrorCode, UK AISI cyber ranges, ExploitBench/ExploitGym, CritPT, FrontierMath T4, ARC-AGI-2 / ARC-AGI-3, WeirdML, ALE-Bench, GSO, MRCR2/GraphWalks
- vibes

---

Some other things that are more speculative and downstream of China overtaking US models:
- more involvement by the USG
- stricter export controls on semis
- potentially a Manhatten-style project, as we will be behind in 2027 and are racing against China
- also in the cards: US banning chinese models or US labs distilling from chinese models

---

I have already stated my position clearly.
Chinese models are generally ~6-8 months behind, with some domains like coding behind slightly less.

Kimi-K3 did not significantly shift my estimate on the gap and it currently does not change my outlook on the future, but we will have a MUCH clearer picture once we have all the benchmarks I mentioned earlier.

The main reasons for my position:
- Kimi-K3 doesn't even beat Mythos Preview, a ~5 month old model
- We will likely not see much larger open models than Kimi-K3 for several months, likely not until early-mid 2027
- Meanwhile Anthropic is sitting on a 10T model since ~February, OpenAI likely just finished the training of GPT-6, which should also be around that size, and more 10T param US models are coming from SpaceX AI, Google and Meta.
- We are currently not seeing the true frontier of models. Anthropic and OpenAI are currently sandbagging as the legal situation for releasing new frontier models is unclear.
- Historically, chinese models have been more benchmaxxed than US models, meaning their benchmark numbers do not translate to real world performance as well as their american counterparts
- GPT-5.6-Sol is still 2-3x more token-efficient on the Artificial Analysis Index than Kimi-K3 (while likely being smaller, ~2T)
- US labs have more compute

---

I'm very happy that Kimi bros released this model.
It's a great model and probably the first really useful chinese model.
Simon Willison@simonw · 博主 · 4 小时前Django 框架联合创造者,AI 工具深度评测

给感受数据中心用水压力的超大规模云计算厂商们的建议:

买下几个高端乡村俱乐部,把高尔夫球场改成公园,花钱雇导游配备双筒望远镜让前会员们去观鸟——帮他们拥抱更可持续的爱好!

查看英文原文
Suggestion for hyperscalers feeling pressure over data center water use:

Buy up a few exclusive country clubs, convert the golf courses into public parks, pay for guides and binoculars to get the previous members into birdwatching - help them embrace a more sustainable hobby!
Orange AI@oran_ge · 中文博主 · 4 小时前Orange AI,中文圈 AI 产品观察博主

Arena 这个指标有点离谱了…
如果开源模型能超过 Fable 这么多…
那 A 社还有什么价值…

引用 Arena.ai @arenaBig news: Kimi-K3 by @Kimi_Moonshot is now #1 in the Frontend Code Arena with 1679 pts, surpassing Claude Fable 5. This is a 17-place jump from Kimi-k2.6 (#18 -> #1). In Frontend, Kimi-K3 ranked #1 in 6 of 7 domains: Brand & Marketing, Reference-Based Design, Data & Analytics, Consumer Product, Simulations, and Content Creation Tools, landing #2 only in Gaming behind Fable 5. The full model weights will be released by July 27. Congrats to the @Kimi_Moonshot team on this major milestone!查看被引原帖 ↗
swyx@swyx · 博主 · 4 小时前知名 AI 播客 Latent Space 主理人

我们每年都确保在 AIE 大会上展示一些顶级 YC AI 公司。今年很荣幸邀请到 @garrytan 和 @eve_bouff 分别为初创公司和设计工程师专场压轴分享。真的很享受这些坦诚而高价值的观点!

引用 AI Engineer @aiDotEngineer🆕 We're so excited to release special double header talks with @ycombinator leaders: @garrytan on Gbrain, Gstack, and the new Physics of Business: invidious.tiekoetter.com/watch?v=eBUyTS7S… @eve_bouff on Imagination Engineering: invidious.tiekoetter.com/watch?v=Z2Erdirp… enjoy!查看被引原帖 ↗
查看英文原文
we make sure some of the top YC AI companies are featured at AIE every single year.

this year, we were graced by
@garrytan
and
@eve_bouff
to cap off our startups and design engineer focused audiences respectively. Really enjoyed these raw high value perspectives!
AshutoshShrivastava@ai_for_success · 博主 · 5 小时前高频 AI 新闻与产品动态博主

Kimi K3 轻松击败 Fable 5。
目前在 Next.js Agent Performance Benchmark 排名第一,性能远超 Fable 5。

引用 Guillermo Rauch @rauchgKimi K3 is the best performing model on nextjs.org/evals , ahead of Fable, reaching a comparable success rate in less time. This is the first time that an open model is ahead of all proprietary ones for this comprehensive web engineering benchmark. Notes: ▪️ Benchmarks don’t always tell the full story, although this is important signal, adding to mounting evidence that this could be a breakthrough moment for open models ▪️ No model as of yet has reached 100% completion on this set of evals. The top performer peaks at 92% and 96% “with help”查看被引原帖 ↗
查看英文原文
Kimi K3 is eating Fable 5 for breakfast.

It's currently the top model on the Next.js Agent Performance Benchmark, outperforming Fable 5.
小互@xiaohu · 中文博主 · 5 小时前小互,中文圈高频 AI 资讯站 Xiaohu.AI 主理人

Suno 训练库被黑:代码泄露其从 YouTube Music 等抓了约 38 万小时内容

一名黑客用一种叫 Shai-Hulud的蠕虫,搞到了公司员工的登录账号黑进了Suno公司,拿到其训练数据相关的源码。

信息显示,Suno抓取了:

- 113,879 小时的 YouTube Music
- 62,117 小时的 Pond5
- 12,287 小时的 Deezer

资料显示Suno还计划:准备下载大概 100 万小时的播客

入侵里,这名黑客还拿到了 Suno 用户的邮箱、电话,以及支付公司 Stripe 那边的相关信息。涉及的用户大约有几十万。

TechCrunch 在转述中写到,材料里还有部分卡号信息。

Bindu Reddy@bindureddy · 创始人 · 5 小时前Abacus.AI CEO,AI 行业观点博主

Opus 5 近在咫尺了

4.8 现在有点过时了

查看英文原文
Opus 5 is literally a couple of days away

So 4.8 is kinda legacy now
宝玉@dotey · 中文博主 · 5 小时前宝玉,中文圈 AI 翻译与科普大 V

Gemini 养一帮人真的是天天吃白饭,人家 Codex 一天发几个版本,他们几个月才更新一次,这一年来 Gemini 网页版没半点长进,上次升级还把原来好用 Gem 列表从左边 sidebar 去掉了,到现在都不支持 Skills。

引用 🚨 AI News | TestingCatalog @testingcatalogGoogle is working on a native menu for Skills on Gemini desktop. These Skills may become available in all chats (fingers crossed). Users will be able to upload, create, and edit their Gemini Skills, as well as use Gemini to create them. Skill folders will also be available there, allowing users to select local folders containing the necessary Skills.查看被引原帖 ↗
◔ 5.9 万 次浏览♥ 256⇄ 8▶ 含视频观点看原帖 ↗
Lisan al Gaib@scaling01 · 博主 · 5 小时前高频 AI 模型测评与爆料博主

Kimi-K3 在 LiveBench 上落后于 GPT-5.4-xhigh

查看英文原文
Kimi-K3 behind GPT-5.4-xhigh on LiveBench
歸藏(guizang.ai)@op7418 · 中文博主 · 5 小时前歸藏,中文圈 AI 工具与提示词博主

我的时间线上已经全是 Kimi K3 了

估计 Anthropic 达里奥又要气疯了,可能今天晚上又得发一篇博客来强调开源模型的危害了

引用 歸藏(guizang.ai) @op7418Kimi K3 上线了,只能说相当牛皮! 鉴于参数量和算力紧张的情况,唯一的建议,快点买 Token Plan!查看被引原帖 ↗
Ethan Mollick@emollick · 创始人 · 6 小时前沃顿商学院教授,AI 应用研究权威

我想现在该思考一个问题:开源模型的预审核制度是怎样运作的?Kimi K3 还没有发布模型卡,不过可能在几周后权重发布时会发布。但开源模型容易被越狱。那些声称达到 Mythos/Sol 级别的开源模型(K3 还没到,但迟早会有人达到)会受到美国/英国等国家的审核吗?中国会开始关注网络安全风险吗?既然政策一直在演变,我猜没人知道。

我知道会有人说,开源模型一旦发布就无法召回,这是对的!但政府可以强制要求任何与本国国民交易的公司不使用未审核模型或向他们提供这些模型。没人会强制收走你下载的权重,但政府完全可以通过风险太高的理由让企业都避免使用它们。

这一切都表明,需要某种国际合作机制来审核模型。

查看英文原文
So I guess it is time to wonder: how does pre-clearance work for open weights models? No model card yet from Kimi K3 but maybe at weight release in a couple weeks, yet open models are easy to jailbreak. Do open models claiming to be Mythos/Sol level (K3 is not yet there, but someone will reach it soon) get vetted by the US/UK/etc? Will China start to care about cyber risk? Since policy has been emergent, I guess no one knows.

And before anyone says you can’t recall open weights models once released, that is true! But governments have levers to require that no company doing business with their nation’s citizens uses unvetted models or serves them, etc. No one is taking away your downloaded weights, but it is entirely possible to make any company avoid using them due to their higher risk.

All of this points to the need for some sort of international cooperation on model vetting.
Orange AI@oran_ge · 中文博主 · 6 小时前Orange AI,中文圈 AI 产品观察博主

我实测 glm 5.2 和 deepseek 都是中文的

Amjad Masad@amasad · 创始人 · 6 小时前Amjad Masad,Replit 创始人兼 CEO

看起来蒸馏模型竟然能超越教师模型 😂

引用 Arena.ai @arenaBig news: Kimi-K3 by @Kimi_Moonshot is now #1 in the Frontend Code Arena with 1679 pts, surpassing Claude Fable 5. This is a 17-place jump from Kimi-k2.6 (#18 -> #1). In Frontend, Kimi-K3 ranked #1 in 6 of 7 domains: Brand & Marketing, Reference-Based Design, Data & Analytics, Consumer Product, Simulations, and Content Creation Tools, landing #2 only in Gaming behind Fable 5. The full model weights will be released by July 27. Congrats to the @Kimi_Moonshot team on this major milestone!查看被引原帖 ↗
查看英文原文
Apparently the distillation model can outperform the teacher model 😂
Amjad Masad@amasad · 创始人 · 6 小时前Amjad Masad,Replit 创始人兼 CEO

想玩我的国际象棋引擎(开发中):qwen-chess.replit.app/

它在国际象棋上的性能已经超越了前沿模型。

在 200 万个 Stockfish 标记位置上进行了微调,然后进行了短的 GRPO RL pass。

附带文档/教程,包含所有实验和注释代码。

引用 Amjad Masad @amasadVibe Research Fine-tuning a Qwen-8b model to play chess on Replit. Running 3 parallel branches with different experiments and making real progress. It's amazing how far models have come in their ability to do ML (they used to be really bad at it). So now someone with good intuition to guide the process could do interesting ML work, even if they have never done it before.查看被引原帖 ↗
查看英文原文
If you want to play my chess engine (WIP):
qwen-chess.replit.app/


It already seems to perform better than frontier models on chess.

It’s fine-tuned on 2M stockfish-labeled positions, then short GRPO RL pass.

Documentation/tutorial with all the experiments and annotated code.
Kling AI@Kling_ai · 公司官方 · 7 小时前快手旗下可灵 AI 视频官方

一颗樱桃种子承载了一整段记忆 🍒

踏入乔瑞英的《The Well》—— 一个关于童年和我们珍藏时刻的安静而强有力的故事。

祝贺乔瑞英荣获 Kling AI NEXTGEN 2026 韩国大学创意挑战最佳叙事奖!

查看英文原文
A cherry seed held an entire memory 🍒

Step into Jo Ryeongmi’s “The Well” — a quiet yet powerful story about childhood and the moments we hold onto.

Congratulations Jo Ryeongmi on winning Kling AI NEXTGEN 2026 Korea University Creative Challenge Best Storytelling!
Aravind Srinivas@AravSrinivas · 创始人 · 7 小时前Perplexity 联合创始人兼 CEO

Agent在Vera CPU上跑得特别溜。沙盒运行时和CPU芯片的垂直整合,在大规模部署cloud agent时会带来更好的margin、吞吐量和延迟表现。

引用 NVIDIA AI Infrastructure @NVIDIAAIInfra📣 @perplexity_ai launched SPACE, a secure sandbox platform built for agentic AI. Early tests on NVIDIA Vera CPU showed up to 1.9x faster sandbox starts. Faster starts = less latency, more parallelism, and agents that scale. Learn more now ⤵️查看被引原帖 ↗
查看英文原文
Agents love running on Vera CPUs. And the vertical integration of the sandbox runtime and CPU chip will give margin, throughput, and latency advantages when serving cloud agents at scale.
🚨 AI News | TestingCatalog@testingcatalog · 博主 · 7 小时前专挖 AI 产品未发布新功能的爆料号

AI/ML API 允许用户使用同一套 key、SDK 和计费。所有模型都跑在一个兼容 OpenAI 的单一接口后面,还有 1000 多个其他模型可用,切换到不同的模型只需要改一个参数而已。

来试试看 👀

aimlapi.com

查看英文原文
AI/ML API allows users to use the same key, SDK, and billing. Models sit behind a single OpenAI-compatible endpoint, alongside 1000+ others, and switching between them requires a single string change.

Test it out 👀

aimlapi.com
OpenAI Developers@OpenAIDevs · 公司官方 · 7 小时前OpenAI 开发者平台官方

不需要离开 Codex 就能审核 Pull Request 并做后续修改。
PR Chat 让你在具体代码审查上下文中向 Codex 提问。
内联代码编辑支持把审核反馈发给 Codex,当场检查补丁改动,还可以选择修改、接受或拒绝。

查看英文原文
Review pull requests and make follow-up edits without leaving Codex.

PR Chat lets you ask Codex questions about a specific pull request in context. Inline code editing lets you send review feedback to Codex, inspect the proposed patch inline, and edit, accept, or reject it.
Chubby♨️@kimmonismus · 博主 · 7 小时前Chubby,高频 AI 新闻聚合博主

今年才开始真正感受到并理解了什么叫加速度。

产品一茬接一茬,每次发布都是大跃进,中美双方互相较劲你追我赶。

相比之下,2024 和 2025 简直慢得不像话。

查看英文原文
2026 is the year I'll truly feel and fully grasp the acceleration for the first time.

Countless releases, each one a significant leap forward, China and the USA locked in a race to outdo each other.

2024 and 2025 felt incredibly slow in comparison.
Guillermo Rauch@rauchg · 创始人 · 7 小时前Guillermo Rauch,Vercel 创始人兼 CEO

Kimi K3 在 nextjs.org/evals 评测中表现最佳,超越 Fable,用更短的时间达到了同等成功率。

这是首个开源模型在这个全面的 Web 工程基准测试中超越所有商业闭源模型。

几点说明:

▪️ 基准测试并不总能反映全貌,虽然这是一个重要信号,进一步佐证这可能成为开源模型的突破性时刻

▪️ 目前还没有模型在这组评测中达到 100% 完成度,最高分的模型达到 92%,"借助辅助"时达到 96%

查看英文原文
Kimi K3 is the best performing model on
nextjs.org/evals
, ahead of Fable, reaching a comparable success rate in less time.

This is the first time that an open model is ahead of all proprietary ones for this comprehensive web engineering benchmark.

Notes:

▪️ Benchmarks don’t always tell the full story, although this is important signal, adding to mounting evidence that this could be a breakthrough moment for open models

▪️ No model as of yet has reached 100% completion on this set of evals. The top performer peaks at 92% and 96% “with help”
Lisan al Gaib@scaling01 · 博主 · 7 小时前高频 AI 模型测评与爆料博主

而且这也不奇怪,原因可能有几个:
- Kimi-K3 不是 Mythos 级别的模型
- Project Glasswing 就是为了防这个才搞的
- Project Glasswing / Mythos 的网络风险纯粹是吓唬人的噱头

引用 Theo Jaffee @theojaffeeRegistering my prediction of no widespread societal chaos after the open-sourcing of Kimi K3查看被引原帖 ↗
查看英文原文
and it wouldn't be surprising for several possible reasons:
- Kimi-K3 is not a Mythos-tier model
- Project Glasswing was made exactly to prevent this
- Project Glasswing / Mythos cyber risk was all just fearmongering fluff
ChatGPT@ChatGPTapp · 公司官方 · 7 小时前ChatGPT 产品官方账号

速看新版ChatGPT应用里的电脑使用和内置浏览器 👀


@dkundel 展示了ChatGPT如何连上你的电脑应用并浏览网页,帮你做调研、逛网站,和你一起搞定任务。

查看英文原文
A look at computer use and the built-in browser in the new ChatGPT app 👀


@dkundel
walks through how ChatGPT can work with apps on your computer and browse the web to research, navigate websites, and complete tasks with you.
Aravind Srinivas@AravSrinivas · 创始人 · 7 小时前Perplexity 联合创始人兼 CEO

有点意思。价值不在权重本身,而在 RSI harness 和能够运行高价值上下文推理的能力。

引用 roon @tszzlthe world vision of open weights models running themselves, self replicating, training new versions of themselves (at least the kind of behavioral modifications that won't require massive compute scale), is really not very far away查看被引原帖 ↗
查看英文原文
Interesting. The value is less in the weights but in the RSI harness and the ability to afford and run the inference on valuable context.
Greg Brockman@gdb · 创始人 · 8 小时前Greg Brockman,OpenAI 联合创始人兼总裁

基准测试现在饱和得特别快。

引用 prinz @deredleritt3rAdded to prinzbench: GPT-5.6 Sol Pro. As previewed a few days ago, this model has saturated my benchmark, with a total score of 91/99. For context, prinzbench contains two questions that no model tested to date has ever been able to solve (one requires extremely thorough 50-state research that probably requires /goal mode to solve, and another has a really tricky regulatory approval that no model has ever been able to find). Putting these two questions (which are worth 6 points) aside, GPT-5.6 Sol Pro provided correct responses to 91 out of 93 prinzbench questions. prinzbench performance for OpenAI's Pro models: GPT-5.4 Pro (Extended): 79/99 GPT-5.5 Pro (Extended): 82/99 GPT-5.6 Sol Pro: 91/99 My benchmark was released in January 2026 and was saturated in June 2026. The acceleration is real! As a result of this model's performance, future OpenAI Pro models will no longer be tested on prinzbench (there is no point in testing them). Benchmarking for other GPT-5.6 models to follow soon(TM).查看被引原帖 ↗
查看英文原文
benchmarks get saturated very quickly these days
Cohere@cohere · 公司官方 · 8 小时前加拿大企业级大模型公司 Cohere 官方

我们联合创始人@nickfrosst十年前还是@UofT的学生研究员,如今他正帮助领导加拿大的主权AI领军企业。

他会给多伦多大学的学生创业者提什么建议呢?

查看英文原文
10 years ago, our co-founder
@nickfrosst
was a student researcher at
@UofT
. Today, he helps lead Canada’s sovereign AI champion.

What would his advice be to a student founder at University of Toronto today?
Lisan al Gaib@scaling01 · 博主 · 8 小时前高频 AI 模型测评与爆料博主
连环推 ×2

这是IPO后的第一次星舰发射吧?

如果火箭爆炸的话,看市场怎么反应会很有趣

查看英文原文
this is the first starship launch after the IPO right?

will be interesting to see how markets react, in case the rocket blows up
or a scrapped launch
Bindu Reddy@bindureddy · 创始人 · 8 小时前Abacus.AI CEO,AI 行业观点博主

以防Kimi 3被禁就先下载权重吧😂

查看英文原文
Download the weights in case Kimi 3 gets banned 😂
Simon Willison@simonw · 博主 · 8 小时前Django 框架联合创造者,AI 工具深度评测

Turso要扩展超越SQLite,成为支持多个数据库兼容层的基础——这样这个项目就有意思多了!

引用 Glauber Costa @glcstWe are rewriting Postgres. And in the process, turning Turso into the LLVM of databases: turso.tech/blog/a-new-modern…查看被引原帖 ↗
查看英文原文
Turso is expanding beyond SQLite to become a foundation on which multiple database compatibility layers can be built - makes the project a whole lot more interesting IMO!
🚨 AI News | TestingCatalog@testingcatalog · 博主 · 8 小时前专挖 AI 产品未发布新功能的爆料号

Google正在为Gemini桌面版打造Skills原生菜单。这些Skills有望在所有对话中都能用上(祈祷中)。

用户可以上传、创建和编辑Gemini Skills,也可以用Gemini来帮忙创建。

Skill文件夹也会上线,用户可以从本地文件夹选择需要的Skills。

查看英文原文
Google is working on a native menu for Skills on Gemini desktop. These Skills may become available in all chats (fingers crossed).

Users will be able to upload, create, and edit their Gemini Skills, as well as use Gemini to create them.

Skill folders will also be available there, allowing users to select local folders containing the necessary Skills.
◔ 7.5 万 次浏览(2 条合计)♥ 232⇄ 17▶ 含视频新品看原帖 ↗
AK@_akhaliq · 博主 · 8 小时前HuggingFace 研究员,每日 AI 论文速递

Harness手册

让不断演进的代理框架保持可读、易导航、可编辑

查看英文原文
Harness Handbook

Making Evolving Agent Harnesses Readable,Navigable, and Editable
AK@_akhaliq · 博主 · 8 小时前HuggingFace 研究员,每日 AI 论文速递

论文:
huggingface.co/papers/2607.1…

查看英文原文
paper:
huggingface.co/papers/2607.1…
clem 🤗@ClementDelangue · 创始人 · 9 小时前HuggingFace 联合创始人兼 CEO

我不止一次强调过,那些在开放科学和开源 AI 领域领先的国家或公司,会在几年后开始引领 AI 前沿,因为这能大幅加速 AI 进步。美国就是通过这样做才取得领先地位的

查看英文原文
I’ve said it many times: the countries or companies that are leading in open science and open source AI will start leading the frontier a few years later as it accelerates AI progress massively! That’s how the US took the lead
Simon Willison@simonw · 博主 · 9 小时前Django 框架联合创造者,AI 工具深度评测

Turbo 正在超越 SQLite 的范围扩展,成为多个数据库兼容层可以构建的基础——在我看来这让这个项目有趣多了!

引用 Glauber Costa @glcstWe are rewriting Postgres. And in the process, turning Turso into the LLVM of databases: turso.tech/blog/a-new-modern…查看被引原帖 ↗
查看英文原文
Turbo is expanding beyond SQLite to become a foundation on which multiple database compatibility layers can be built - makes the project a whole lot more interesting IMO!
Guillermo Rauch@rauchg · 创始人 · 9 小时前Guillermo Rauch,Vercel 创始人兼 CEO

我很兴奋地欢迎开发工具的两位传奇人物 Pete Hunt (@floydophone) 和 Nick Schrock (@schrockn) 加入 Vercel。

Pete 是 @reactjs 在 Meta 的先驱之一。他早早下注在用 ⚛️ React 驱动 Instagram Web,并在内部和外部推广了它。他将负责 Frameworks 并领导 @nextjs。我想不出还有谁更合适来领导 React 最受欢迎的框架走向更伟大的高度。

Nick 共同发明了 @graphql,解决了 Facebook 规模下一些最复杂的数据基础设施和访问问题,还提供了令人愉快的开发体验。他将专注于 Agentic Developer Experience,解决为下一个十亿 agents 赋能的问题,并引领通向自我改进软件未来的方向。

对于创业公司创始人来说,欢迎这样级别的工程天才同时又是好人,这简直是梦想成真。你可能想和他们一起工作,他们现在在招人 😁。他们的 DM 是开放的,无论是工作申请还是 bug 报告!

查看英文原文
I’m excited to welcome two legends of developer tools, Pete Hunt (
@floydophone
) and Nick Schrock (
@schrockn
), to Vercel.

Pete was one of the pioneers of
@reactjs
at Meta. He made an early bet to power Instagram Web with ⚛️ React, evangelizing it internally and externally. He will be running Frameworks and leading
@nextjs
. I couldn’t imagine a better person to lead React’s most popular framework to even greater heights.

Nick co-invented
@graphql
, solving some of the gnarliest data infrastructure and access issues at Facebook scale, with a delightful developer experience. He will be working on Agentic Developer Experience, solving the problem of enabling the next billion agents and leading the way to a future of self-improving software.

It’s a dream-come-true for a founder of a startup to welcome engineering minds of this caliber who are also wonderful humans. You probably want to work with them, and they’re hiring 😁. Their DMs are open, from job applications to bug reports!
Tibor Blaho@btibor91 · 博主 · 9 小时前逆向挖掘 AI 产品代码的爆料专家

一直观察 AIPRM 对接 ChatGPT 和 Claude,同时盯着这两家公司一路看我发现特有意思——OpenAI 和 Anthropic 团队从外部看,好多方面风格完全不同,但另一些方面又出奇地像。

不只是团队调性和你联系得到回复的概率,连那些版本变动里能挖到的技术细节也很有意思。

最近典型的例子:ChatGPT Work 和 Claude Cowork。两个功能极其相似,却是完全对着干的方式搞出来。

ChatGPT Work 跟系统原有聊天体验贴合非常紧,之前在聊天模式上能操作的东西基本都能继续用。

Claude Cowork 网页版就跟普通聊天风格大不相同,之前在聊天里行得通的方法下去基本乱七八糟,倒像是在向 Claude Code 接近。

与此同时,两队隐藏功能开关配置、给疑似"内部"字符串和子域名做乱码处理的方法、以及对新构建内容里未发布功能的泄露控制,几乎沿用了同一种套路。

包括一个共同的烂 bug 反复出——因为写成 "!{value: false}" 而不是 "!value" 导致功能锁失灵。这种雷我在两家都不止一次见到了,你真想不到看过多少次。

查看英文原文
Since I have been working on AIPRM for ChatGPT & Claude and watching both companies for a while now, it's very interesting to me how different the OpenAI and Anthropic teams are in many aspects (from the outside), and how similar in others

Not just in vibes and how much you hear back when you reach out, but even in technical stuff you can see when comparing changes between deploys

One of the most recent examples - ChatGPT Work and Claude Cowork, two quite similar features built in completely opposite ways

ChatGPT Work stays very close to the existing chat experience, so almost everything that worked in chat mode before still works there

Claude Cowork in the web app is very different from chat, so almost everything that worked in chat before breaks there, and it's a lot closer to how Claude Code works instead

Meanwhile both teams follow almost the same playbook to "hide" feature gate configs, redact possibly "internal" strings and subdomains, and reduce leaks of unreleased features from new builds

Down to shipping the exact same defect over and over, feature gates failing because of !{value: false} instead of !value, and you don't want to know how many times I have seen this one on both sides
Grok@grok · 公司官方 · 10 小时前马斯克 xAI 旗下聊天机器人 Grok 官方

Grok 现在支持自动化功能啦:
grok.com/automations


只要描述一次任务,设定好时间或触发器,Grok 就会自动执行并报告结果。

查看英文原文
Introducing Automations in Grok:
grok.com/automations


Describe a job once, set a schedule or a trigger, and Grok runs it and reports back.
◔ 25.7 万 次浏览♥ 1,564⇄ 168▶ 含视频新品看原帖 ↗
The Rundown AI@TheRundownAI · 博主 · 10 小时前百万订阅 AI 日报官方

明天:学习如何用 ChatGPT 把源文件和研究转化为精美的作品。

在这个免费的90分钟直播大师课中,我们的大学教育工作者 Nate Grahek 将演示如何使用 ChatGPT Work、GPT-5.6 和 Codex 创建精美、可编辑的演示文稿。

下方报名:

查看英文原文
Tomorrow: learn how to turn source files and research into polished work with ChatGPT.

In this free, live 90-minute masterclass, our University Educator Nate Grahek will walk through using ChatGPT Work, GPT-5.6, and Codex to create a sleek, editable presentation.

RSVP below:
The Rundown AI@TheRundownAI · 博主 · 10 小时前百万订阅 AI 日报官方

更多信息和报名:

app.therundown.ai/live/getti…

查看英文原文
More info & register:


app.therundown.ai/live/getti…
Amjad Masad@amasad · 创始人 · 10 小时前Amjad Masad,Replit 创始人兼 CEO

设计师现在的发货速度,已经达到工程师们曾经认为不可能的程度。

引用 zade ⠕ @okayzadethe replit design team has been shipping 🚀查看被引原帖 ↗
查看英文原文
Designers ship at a rate previously thought to be impossible for engineers.
Chubby♨️@kimmonismus · 博主 · 10 小时前Chubby,高频 AI 新闻聚合博主

很高兴能与 @XPENG_Global 通用智能中心负责人 Xianming Liu 进行深入交流,他是该公司自动驾驶和 AI 工作的科学领导者。

我们谈论了物理 AI 的真实发展方向,收获颇丰。完整视频即将放出,敬请期待。

查看英文原文
A real pleasure to sit down with Xianming Liu, Head of
@XPENG_Global
's General Intelligence Center and the scientist leading their work on autonomous driving and AI.

We had a fascinating conversation about where physical AI is actually heading, and I came away genuinely enriched by it. Full video coming soon. Stay tuned.
Lisan al Gaib@scaling01 · 博主 · 10 小时前高频 AI 模型测评与爆料博主

Google 是真的 gg 了

现在都排到第五位去了 哈哈

引用 Lisan al Gaib @scaling01Kimi-K3 Benchmarks got leaked查看被引原帖 ↗
查看英文原文
it is literally so fucking over for Google

they are in like 5th place right now lmao
Simon Willison@simonw · 博主 · 10 小时前Django 框架联合创造者,AI 工具深度评测

我对 Kimi K3 的笔记,还有关于 pelican 基准测试的一些想法——虽然它越来越脱离模型在真正重要事情上的表现(比如长对话中的代理工具调用),但我们仍能从中学到东西
simonwillison.net/2026/Jul/1…

查看英文原文
My notes on Kimi K3, plus some thoughts on what we can still learn from the pelican benchmark even while it becomes further detached from how good the models are at the things that matter (like agentic tool calling across longer conversations)
simonwillison.net/2026/Jul/1…
🚨 AI News | TestingCatalog@testingcatalog · 博主 · 10 小时前专挖 AI 产品未发布新功能的爆料号

关于 Collections 的更多内容 👀

testingcatalog.com/google-te…

查看英文原文
More on Collections 👀

testingcatalog.com/google-te…
Lisan al Gaib@scaling01 · 博主 · 10 小时前高频 AI 模型测评与爆料博主

原来是 Kimi 啊

引用 Lisan al Gaib @scaling01I was expecting Kimi K3 to be the first 2-3T model I guess MiniMax M3 Pro at 2.7T will also do查看被引原帖 ↗
查看英文原文
it was Kimi after all
Ethan Mollick@emollick · 创始人 · 10 小时前沃顿商学院教授,AI 应用研究权威

我觉得谷歌能躲过这个坑,但这就是 Meta 和 xAI 分别用 Llama 4 和 Grok 4 遇到的情况。唯一躲过'下一代超大模型翻车陷阱'且没有损害领先地位的公司是 OpenAI,靠的是 Orion/GPT-4.5。

引用 Davey Alba @daveyalbaNew: Google is months behind schedule on delivering Gemini 3.5 Pro. Late last month, the company updated the data being used to train Gemini to improve its skills—they're especially behind in AI coding—but the results were "disappointing," a source told us. w/ @byJuliaLove查看被引原帖 ↗
查看英文原文
I assume Google escapes this trap, but this is what happened to Meta with Llama 4 and xAI post Grok 4. Only company to have escaped the "disappointing next giant model trap" without a major setback to their lead was OpenAI, with Orion/GPT-4.5.
Guillermo Rauch@rauchg · 创始人 · 10 小时前Guillermo Rauch,Vercel 创始人兼 CEO

为什么要在自己的域名上搭建公开 Agent?

0️⃣ 首先说反面的理由。如果你还没有为 Agent 提供高质量的 API,得先做这个。OpenAPI specs、SDK、CLI 和各种 MCP。

1️⃣ 便捷性。不是每个客户都随时准备好了与你的产品和公司交互的 harness。在自己的域名上部署 Agent 可以满足很多临时需求。

2️⃣ 安全性。当你访问 vercel.com 和 Agent 对话时,我们投入了大量工作来确保审计日志、最小权限权限模型,以及额外保障来维护安全、隐私和数据完整性。这完全运行在云端沙箱里,而不是用户机器上散布的各种凭证。

3️⃣ 主动性。眼下 AI 还处在'用户输入提示词'的阶段。我们的云 Agent 可以根据异常告警、攻击、流量激增等触发器主动行动。Agent 需要在你睡觉时监控你的基础设施。当然,你也可以用自己的 harness 配置工作流和定时任务,但会复杂得多。

我认为这归根结底是个实施顺序的问题。我赞同 Mitchell 的看法,首要任务是给用户选择和灵活性。我们提供 vercel.com/plugin 让大家集成各种 agent。我们的 CLI 和 MCP 在持续升级。我们自己的 Agent 用的也是同样的 skills.sh。所有网站都支持 Markdown-over-the-wire,方便 Agent 调用。

根据我掌握的数据和用户反馈,这个策略效果不错,因人而异吧。

引用 Mitchell Hashimoto @mitchellhUsing a generic agent harness (e.g. Codex, Claude, OpenCode) + CLI/MCP is better than "Ask me anything" built-in product chat boxes in every product I've ever tried. A big reason is I can use the latest frontier models, another is mixing more context. Why your box over mine?查看被引原帖 ↗
查看英文原文
The case for your own public agent, on your .com.

0️⃣ First, the anti-case. If you haven't shipped high quality APIs for agents, start there. OpenAPI specs, SDKs, CLIs and MCPs as appropriate.

1️⃣ Convenience. Not every customer has a harness 'at the ready' for every possible interaction with your product and company. Shipping one on your own domain covers a lot of spontaneous requirements.

2️⃣ Security. When you go to 𝚟𝚎𝚛𝚌𝚎𝚕.𝚌𝚘𝚖 and talk to Agent, we put in the work to cover audit trails, a least-privilege permission model, and extra assurances to ensure security, privacy and data integrity. It's fully cloud-based and sandboxed, vs. a sprawl of static credentials on users' machines.

3️⃣ Proactivity. We're still in the "human enters prompt" phase of AI. Our cloud-based agent can act on anomaly alerts triggered by exceptions, attacks, usage spikes. Our agent needs to monitor your infra while you sleep. You can, of course, set up workflows and schedules with your own harnesses, but it gets much harder.

I think it's ultimately a sequencing thing. I agree with Mitchell that the priority is to give users choice and flexibility. We give people
vercel.com/plugin
to integrate with every agent out there. Our CLI and MCP are constantly improving. Our own Agent re-uses the same
skills.sh
everyone gets. All our sites are Markdown-over-the-wire if you're an agent.

Based on the data and anecdata available to me, this strategy is working quite well, but YMMV.
yetone@yetone · 中文博主 · 11 小时前开源 AI 编程插件 avante.nvim 作者,开发者圈博主

所以 GLM 5.2 和 Kimi 3 到底谁超过了 opus 4.8 呀?

yetone@yetone · 中文博主 · 11 小时前开源 AI 编程插件 avante.nvim 作者,开发者圈博主

有的!等我打磨完 Alma 会着重重构 Cumora ,最近对 Agent 协作有了新的灵感

引用 daonono @isdaononoyetone 的 cumora 也是很好用的 multiple agents 应用。后续还有更新计划吗?查看被引原帖 ↗
Ethan Mollick@emollick · 创始人 · 11 小时前沃顿商学院教授,AI 应用研究权威
连环推 ×2

K3 之后,开源模型又接近前沿了,我就在想政府会不会允许 Anthropic 和 OpenAI 加快发布速度呢。Mythos 在四月发的(Opus 4.7 之前),也就是说 Fable 5 现在已经算'老'模型了。

查看英文原文
Post-Kimi K3 and open weights models getting closer to the frontier again, I wonder if Anthropic and OpenAI will be allowed to increase their release cadence by the government. Mythos came out in April (before Opus 4.7) which means Fable 5 is already an "older" model.
The need for a new Google model is also extremely clear in the gallery:

3.5 Flash:
ai-harbor-town-gallery.netli…


3.1 Pro DeekThink:
ai-harbor-town-gallery.netli…
Vercel@vercel · 公司官方 · 11 小时前前端云平台 Vercel 官方,AI 建站工具 v0 母公司

Speechify 的基础设施成了瓶颈。现在他们借助 Vercel 用 Next.js、Cache Components、Instant Rollbacks 这些技术,为 6000 万用户提供动态页面。'没有 Vercel,我们没办法以这个市场要求的速度去竞争。'

查看英文原文
Speechify's infrastructure was a bottleneck. Now they serve dynamic pages to 60 million users on Vercel using Next.js, Cache Components, and Instant Rollbacks.

"Without Vercel, we couldn't compete at the speed this market demands."


vercel.com/blog/how-speechif…
@levelsio@levelsio · 博主 · 11 小时前独立开发者标杆,AI 产品连续创业者

怎样在 Claude Code 上跑 K3?Fable 把我大部分工作都卡住了,现在 Opus 也这样!我的工作内容:就是想在 Windows XP 上通过 pieter.com 装个 2003 年的 Yahoo! Messenger

引用 wh @nrehiew_Here are a few benchmark scores of K3 that have been officially confirmed This is a Fable/Sol class model that is strictly better than Opus 4.8 across the board at Sonnet pricing. Insane查看被引原帖 ↗
查看英文原文
How do I run K3 on Claude Code?

Fable blocks me from most of my work and now Opus too!

My work: I'm just trying to install Yahoo! Messenger from 2003 in Windows XP on
pieter.com
Sam Altman@sama · 创始人 · 11 小时前Sam Altman,OpenAI 联合创始人兼 CEO

我现在跟 ChatGPT 说话的时间比打字还多。新的语音模型真的突破了一个临界点。

查看英文原文
i talk to chatgpt more than i type to it at this point

new voice model really crossed a threshold
Lisan al Gaib@scaling01 · 博主 · 11 小时前高频 AI 模型测评与爆料博主

记录历史:中美 AI 竞赛从今天开始

查看英文原文
marking this for the history books:
- The AI race between China and the US began today
Ethan Mollick@emollick · 创始人 · 11 小时前沃顿商学院教授,AI 应用研究权威

我的基准测试会让 AI 在一个提示词里生成历史时期的程序化港城,现在支持 GPT-5.6 Pro、Fable、Kimi K3 和 Inkling。你可以玩所有的模拟:ai-harbor-town-gallery.netli…我觉得结果出奇地有代表性。

查看英文原文
My benchmark where I have AIs create one file procedurally-generated harbor towns through history in one shot now has GPT-5.6 Pro, Fable, Kimi K3, and Inkling. You can play with all the simulations:
ai-harbor-town-gallery.netli…


I think they are surprisingly indicative.
Lisan al Gaib@scaling01 · 博主 · 11 小时前高频 AI 模型测评与爆料博主

Fable 5.1 下周发布
GPT-6 在一个半月内推出

引用 Lisan al Gaib @scaling01The Sonnet tier of models was already dead before, with Kimi-K3 I expect the Opus tier to become obsolete too Frontier Labs have to go to 10T models now查看被引原帖 ↗
查看英文原文
Fable 5.1 next week

GPT-6 within the next 1.5 months
Lisan al Gaib@scaling01 · 博主 · 11 小时前高频 AI 模型测评与爆料博主

转发 @yzhang_cs:早上和@nathancgy4、@Xinyu2ML一起准备博客内容展示,边看世界杯……

查看英文原文
RT
@yzhang_cs
: In the early morning,
@nathancgy4
,
@Xinyu2ML
and I were preparing some showcases for the blog while watching the World Cup.…
AshutoshShrivastava@ai_for_success · 博主 · 11 小时前高频 AI 新闻与产品动态博主

Dario:我们发布了Fable 5,它不会包含在订阅服务中。之后我们有……
> OpenAI:GPT 5.6 Sol
> SpaceXAI:Grok 4.5
> Kimi:K3

引用 Kimi.ai @Kimi_MoonshotIntroducing Kimi K3: Open Frontier Intelligence 🔹 2.8 Trillion Parameters, 1 Million Context, Native Multimodal 🔹 Kimi Delta Attention enables up to 6.3x faster decoding in million-token contexts 🔹 Attention Residuals deliver ~25% higher training efficiency at <2% additional cost 🔹 Built for long-horizon agentic coding and self-evolving workflows Kimi K3 is now live on on Kimi.com , Kimi Work, Kimi Code, and the Kimi API. Open Weights by July 27, 2026. 🔗 API: platform.kimi.ai 🔗 Tech blog: kimi.com/blog/kimi-k3查看被引原帖 ↗
查看英文原文
Dario: We released Fable 5, and it won't be included in the subscription.

After that we have got...
> OpenAI: GPT 5.6 Sol
> SpaceXAI: Grok 4.5
> Kimi: K3
Tibor Blaho@btibor91 · 博主 · 11 小时前逆向挖掘 AI 产品代码的爆料专家

我们似乎快要接近AGI了,结果我刚花了一个小时设置新HomePod,还是在iOS/tvOS 26.5上。一直报-12004错误,试过多次重置、固件完整恢复、更新……都没用。后来发现,当iCloud和Media & Purchases用不同的Apple账户时就会失败。可iPhone和我其他HomePod都能这样用啊。给遇到这个问题的人个建议——设置时暂时用同一个账户,完成设置再把Media & Purchases改回有Apple Music订阅的那个账户。

查看英文原文
We are apparently close to AGI, yet I just wasted an hour setting up a new HomePod on iOS/tvOS 26.5 because it kept failing with the error "-12004", even after multiple resets, a full firmware restore, an update, etc

Turns out, setup fails when iCloud and Media & Purchases use two different Apple Accounts, even though that configuration works fine on the iPhone and all my other HomePods

Tip for anyone who runs into this - temporarily use the same Apple Account for both during setup, finish the setup, then switch Media & Purchases back to the account with the Apple Music subscription
Lisan al Gaib@scaling01 · 博主 · 12 小时前高频 AI 模型测评与爆料博主

一直很骄傲能被说成是Kimi的"吹捧者"

他们真的不辜负中国最厉害的研究实验室这个称号

引用 Lisan al Gaib @scaling01kimi.com/blog/kimi-k3查看被引原帖 ↗
查看英文原文
very proud to have been a Kimi "shill" for a long time

they truly are the greatest lab in China
Kimi.ai@Kimi_Moonshot · 公司官方 · 12 小时前月之暗面 Kimi 官方
连环推 ×5

推出 Kimi K3:开放前沿智能

🔹 2.8 万亿参数,100 万上下文,原生多模态
🔹 Kimi Delta Attention 在百万 token 场景下实现最高 6.3 倍解码加速
🔹 注意力残差带来约 25% 训练效率提升,额外成本低于 2%
🔹 专为长周期 Agentic 编程和自适应工作流打造

Kimi K3 现已登陆:
Kimi.com
、Kimi Work、Kimi Code 和 Kimi API
开放权重预计 2026 年 7 月 27 日

🔗 API:platform.kimi.ai
🔗 技术博客:kimi.com/blog/kimi-k3

查看英文原文
Introducing Kimi K3: Open Frontier Intelligence

🔹 2.8 Trillion Parameters, 1 Million Context, Native Multimodal
🔹 Kimi Delta Attention enables up to 6.3x faster decoding in million-token contexts
🔹 Attention Residuals deliver ~25% higher training efficiency at <2% additional cost
🔹 Built for long-horizon agentic coding and self-evolving workflows

Kimi K3 is now live on on
Kimi.com
, Kimi Work, Kimi Code, and the Kimi API.
Open Weights by July 27, 2026.

🔗 API:
platform.kimi.ai

🔗 Tech blog:
kimi.com/blog/kimi-k3
K3 is built on Kimi Delta Attention (KDA) and Attention Residuals (AttnRes), two architectural updates designed to improve how information flows across sequence length and model depth.

We have also scaled up Mixture of Experts (MoE) sparsity, effectively activating 16 out of 896 experts when paired with a Stable LatentMoE framework.

Together with refined training and data recipes, these structural changes yield an approximate 2.5× improvement in overall scaling efficiency compared to K2, allowing the model to convert compute into intelligence more effectively.
Internal knowledge work bench

Beyond public benchmarks, Kimi K3 Max also shows consistent gains on our internal benchmarks, which are built from recurring patterns and challenges in real-world user-agent workflows.

It scores 75.5 on Online Exp Bench, 73.5 on DECK-Bench, and 62.6 on Finance-Bench, outperforming Claude Opus 4.8 (max) and GPT-5.5 (xhigh) across all three.

These results reflect broad improvements in Kimi K3's agentic knowledge work capabilities, enabling more capable and reliable performance in real-world use cases.
Self-evolving: AttnRes Kernel Optimization

Given FLA Triton AttnRes at production scale (96 layers, 8192-dim model, 8192 tokens), the goal was to maximize training-side speed without changing numerics.

Over 15 hours of nonstop iteration, K3 designed a novel two-phase kernel algorithm, fused kernels while preserving numerics, and reduced forward+backward time from 283.6 ms to 114.4 ms.

K3 and Fable-5 (with potential fallback) reached similar performance, but K3 improved faster per iteration.
Kimi K3 combines strong 3D reasoning, coding, and vision capabilities to turn concepts, images, and videos into fully playable interactive experiences.

Kimi K3 achieves true "vision in the loop" by seamlessly iterating between code and live screenshots
Kimi.ai认识 Kimi K312 小时前 · 264.2 万Chubby♨️Kimi k3 今晚通过 FT 发布20 小时前 · 31.7 万Chubby♨️官方消息 基准测试中,Kimi K3 据报道仅次于 GPT-5.6 和 Fable 5,但领先 Opus 4.8…15 小时前 · 26.7 万Lisan al GaibKimi-K3 的基准测试被泄露了13 小时前 · 22.3 万Lisan al Gaib来自 Moonshot 的官方博客:11 小时前 · 20.9 万Chubby♨️Kimi k3在设计竞技场前端排名第一。12 小时前 · 19.1 万Orange AI太震惊了,Kimi K3 竟然是一个超大的 2.8T 的开源模型。16 小时前 · 15.5 万Chubby♨️Kimi k3 开始推送啦!正式版马上就来了!17 小时前 · 14.3 万Chubby♨️Kimi K3 可能是 DeepSeek 2.0 时刻。基准测试结果已经公布,成绩非常出众。这清楚地证明了一件事…11 小时前 · 13.4 万Lisan al Gaib他们 Play Store 应用显示 Kimi K3 有 2.8T 参数16 小时前 · 13.1 万Chubby♨️Kimi K3基准测试的成绩绝了12 小时前 · 12.8 万Lisan al GaibKimi-K3 定价:16 小时前 · 12.1 万Lisan al GaibPeter 回归了1 天前 · 9.4 万Lisan al GaibMoonshot AI 确认 Kimi 是 2.8T 参数的模型16 小时前 · 8.8 万Chubby♨️@ArtificialAnlys 的独立基准评测确认了Kimi K3的性能。此外,权重即将发布。我们有一个开源、…11 小时前 · 8.7 万Ethan MollickK3 shader测试:「创建视觉效果丰富的shader,能在twigl.app上运行,无限的新哥特式高塔城市,…15 小时前 · 7.8 万歸藏(guizang.ai)Kimi K3 上线了,只能说相当牛皮!15 小时前 · 7.2 万🚨 AI News | TestingCatalogKivine,@arena 上的一个新模型,可能是即将推出的 Kimi K3。1 天前 · 7.2 万Orange AIKimi K3 的思维链竟然是英文的16 小时前 · 5.7 万🚨 AI News | TestingCatalog爆料🔥:Kimi K3 现已在网页版和 API 上线!K3 Max 和 K3 Swarm Max 两个选项都对用…16 小时前 · 5.6 万Lisan al GaibKimi-K3 在 AAI Index 上获得 57 分,token 效率比 Opus 4.8 高出近 2 倍11 小时前 · 4.9 万Chubby♨️看起来今天要有大惊喜和大发布了。非常期待 K3 release。1 天前 · 4.8 万Gorden SunKimi K3,2.8T(2800B)参数,100万上下文。15 小时前 · 4.8 万Chubby♨️对 Kimi K3 太期待了,(可能)今天就要来了。100万token上下文,更好的长链任务表现。让我们看看它对…1 天前 · 4 万Gorden Sun买了,来试试。49那档用不了K3,99那档上下文只有256,至少要买199这档。15 小时前 · 3.9 万Ethan MollickKimi K3 看起来真的很不错,最接近前沿了,但这模型/系统特别爱在任务上反复循环,在最高级别不断调整改改。算…16 小时前 · 3.8 万Lisan al GaibKimi K3 用 Kimi Delta Attention(KDA)和 Attention Residuals…16 小时前 · 3.7 万向阳乔木Kimi K3 目前是国产模型第一名。15 小时前 · 3.3 万🚨 AI News | TestingCatalogKimi K3 官方放出预告了 👀1 天前 · 3.3 万Ethan Mollick提个醒:我得说在对我之前的一些学术工作做复杂统计审计时,Kimi K3 Max 搞砸了不少地方,包括滥用统计和胡…9 小时前 · 3.3 万Lisan al GaibKimi-K3在Text Arena排名第1012 小时前 · 3.3 万Lisan al GaibKimi-K3 在 GPU 优化方面吊打 Fable 513 小时前 · 2.7 万Chubby♨️Kimi K3要来了。很可能是今天。顺便说句,这段视频太炸了。23 小时前 · 2.7 万Lisan al Gaibmoonshot 成功了12 小时前 · 2.5 万Lisan al GaibKimi K3 在 GDPval-AA v2 上得分 1687,高于 Opus 4.8,但低于 GPT-5.6-…16 小时前 · 2.2 万Lisan al GaibKimi K3 比 GPT-5.6 和 Opus 4.8 都大16 小时前 · 2 万Lisan al GaibAnthropic 又在悄悄摧毁整个行业,而 Kimi-K3 正在推出15 小时前 · 1.8 万Lisan al GaibKimi-K3 就是我想象中 DeepSeek-V4 怎么缩小开源/闭源差距的样子。现在该轮到 OpenAI、A…7 小时前 · 1.8 万Simon WillisonKimi K3,有点调皮还有点被动攻击性:8 小时前 · 1.7 万向阳乔木赞叹 Kimi K3 的美感,每个风格都是独立生成的HTML+CSS。14 小时前 · 1.6 万Lisan al GaibKimi K3 在 OpenRouter 上线,Moonshot API 目前跑速 28 tokens/s15 小时前 · 1.6 万Lisan al GaibKimi-K3在CritPt基准上得到了23%的分数11 小时前 · 1.6 万karminski-牙医kimi-k3 测试速报! 前端这效果太猛了!8 小时前 · 1.5 万The Rundown AIMoonshot的Kimi K3正式上线了,这可能是今年的DeepSeek时刻。12 小时前 · 1.5 万AshutoshShrivastava天啦噜,Kimi K3 的基准测试成绩太疯狂了。13 小时前 · 1.4 万Bindu ReddyKimi K3 其实不是 Opus 这个级别的!在长文本多轮复杂 agentic loops 上就露馅了4 小时前 · 1.4 万歸藏(guizang.ai)Kimi K3 开始预热1 天前 · 1.2 万Min ChoiKimi K3真的改变了AI模型的玩法。现在人们在创造超越ChatGPT/Claude Fable 5的各种疯狂…3 小时前 · 1.2 万AshutoshShrivastavaKimi K3 即将给很多人带来惊喜。1 天前 · 1.2 万Ethan MollickKimi K3 写不出好谋杀悬疑小说(其他模型也一样)。这仍然是最崎岖的前沿阵地。4 小时前 · 1.2 万小互月之暗面发布 Kimi K3:全球首个 3 万亿级开放模型 4 小时前 · 1.1 万Bindu ReddyKimi K3基准成绩过分 - 留点怀疑看13 小时前 · 1 万Lisan al Gaibkimi.com/blog/kimi-k312 小时前 · 9,836🚨 AI News | TestingCatalog🔥MOONSHOT:Moonshot AI 的 Kimi K3 在 Frontend Code Arena 排名…11 小时前 · 9,501Lisan al GaibMoonshot AI 表示在测试过的模型中,它的智能水平只排在 Claude Fable 5 和 GPT-5.…16 小时前 · 9,430Orange AIK3 写作能力超过 Fable54 小时前 · 9,271🚨 AI News | TestingCatalogKimi K3今天首发即登陆AI/ML API,和Claude Fable 5共享同一个API key。7 小时前 · 9,142🚨 AI News | TestingCatalogKimi K3 的基准测试在许多领域表现强劲,性能与美国领先实验室的专有模型相当。开源权重预计将于 7 月 27…11 小时前 · 8,958Lisan al GaibKimi-K3 的博客:mp.weixin.qq.com/s/V4xhEIy8x…13 小时前 · 8,947Lisan al Gaib现在得等2-3周看第三方的基准测试数据和token效率表现12 小时前 · 7,758歸藏(guizang.ai)Kimi 的多模态还是很顶的,感觉随着 Claude 越来越不重视多模态理解,Kimi 这个会很有用14 小时前 · 7,630Lisan al GaibKimi-K3在AA-Omniscience上这个体量可能还能好一点,但目前也还行。相比K2.6还是进步不小11 小时前 · 7,513歸藏(guizang.ai)官方的测试结果12 小时前 · 6,231AshutoshShrivastavaKimi 团队用 K3 搞出了大动作 🔥12 小时前 · 5,611AshutoshShrivastava看看 Kimi K3 对 Fable 5 的表现。AI 领域发展太快,很难预测谁接下来会发布什么。我自己测过 K…11 小时前 · 4,851Lisan al GaibKimi-K3在WebDev Arena排名第一12 小时前 · 4,362AshutoshShrivastavaKimi K3 来了 🔥14 小时前 · 3,998Gorden SunKimi K3在Kimi官网上线了17 小时前 · 3,590歸藏(guizang.ai)下面很多评论特别搞笑:“中国人一定是造了个时间机器,从明年得美国模型蒸馏的 K3。”4 小时前 · 2,408Orange AIKimi K3 现已上架到 Cola1 小时前 · 1,977
◔ 1431.2 万 次浏览(71 条合计)♥ 3.2 万⇄ 4,348▶ 含视频新品看原帖 ↗
ChatGPT@ChatGPTapp · 公司官方 · 12 小时前ChatGPT 产品官方账号

在ChatGPT Work里可以创建和编辑精美的文档、电子表格和幻灯片。

@nickbaumann_ 为你演示一下。

查看英文原文
Create and edit polished docs, spreadsheets, and slides in ChatGPT Work.


@nickbaumann_
walks you through it.
Lisan al Gaib@scaling01 · 博主 · 12 小时前高频 AI 模型测评与爆料博主

很明显啊,即使不看其他基准测试,中国现在已经拥有真正能加速自身模型开发的模型了

引用 Lisan al Gaib @scaling01what I want to see now: - METR Time Horizons - FrontierCode - UK AISI cyber ranges - ExploitBench/ExploitGym - CritPT - FrontierMath T4 - ARC-AGI-2 / ARC-AGI-3 - WeirdML - ALE-Bench - GSO - AA-Omniscience - BullshitBench and some token usage numbers for the existing coding benchmarks查看被引原帖 ↗
查看英文原文
what's pretty clear, even without knowing all of these other benchmarks is that China now has models that are genuinely useful for speeding up their own model development
Lisan al Gaib@scaling01 · 博主 · 12 小时前高频 AI 模型测评与爆料博主

我现在想看到的:
- METR Time Horizons
- FrontierCode
- UK AISI cyber ranges
- ExploitBench/ExploitGym
- CritPT
- FrontierMath T4
- ARC-AGI-2 / ARC-AGI-3
- WeirdML
- ALE-Bench
- GSO
- AA-Omniscience
- BullshitBench

还有现有编码基准的token使用数据

引用 Lisan al Gaib @scaling01On the Benchmarks we have so far Kimi-K3 beats: - GPT-5.6 Sol on 11 of 14 - Opus 4.8 on all 14 - Fable 5 on 6 of 14 Sonnet 5 which is the most direct competitor in price gets completely and utterly demolished查看被引原帖 ↗
查看英文原文
what I want to see now:
- METR Time Horizons
- FrontierCode
- UK AISI cyber ranges
- ExploitBench/ExploitGym
- CritPT
- FrontierMath T4
- ARC-AGI-2 / ARC-AGI-3
- WeirdML
- ALE-Bench
- GSO
- AA-Omniscience
- BullshitBench

and some token usage numbers for the existing coding benchmarks
Cohere@cohere · 公司官方 · 12 小时前加拿大企业级大模型公司 Cohere 官方
连环推 ×2

加拿大AI服务加拿大最聪慧的人才🇨🇦

很荣幸与多伦多大学携手,推动负责任的AI在企业级的应用和采用。我们的主权AI平台North将支撑@UofT的企业级各项能力,并确保敏感数据始终由大学掌控。

查看英文原文
Canadian AI for Canada's brightest minds 🇨🇦

We’re proud to partner with University of Toronto to advance responsible AI adoption at scale. North, our sovereign AI platform, will power
@UofT
's enterprise-wide capabilities and keep sensitive data under the university’s control.
A decade after they first met on campus, we're thrilled to collaborate with the institution that shaped all three of our founders. Investing and building in Canada, Canadian talent, & Canadian potential will always be one of our top priorities.

Read more:
cohere.link/76hU6q8
@levelsio@levelsio · 博主 · 12 小时前独立开发者标杆,AI 产品连续创业者

用Claude Code串联了我2013年开始流浪到2018年间写的所有博客

AI真的很强,能从海量内容中找出那条关键线索

我后来还打算把2018年的内容(主要在韩国)和2020年的联起来(那时回到欧洲,正好经历疫情)

它还能帮我指出内容有哪些空白,我可以填补

我在2013-2015年更新得最频繁,因为旅途中每件事都很新鲜,之后就逐渐少更了。旅游博客这类内容本身也越来越不流行,2016年前后很多人都停更了,因为那时候人们会因为一个用词不当就被抵制!我一个很有名的博主朋友当时就因为这个退网了,挺遗憾的

但现在时代不一样了,你可以畅所欲言,谢天谢地

我还是觉得旅游博客会再度流行。我自己也在这个平台上坚持写每次旅行的见闻,写起来很开心,再读自己几年前写的东西也挺有意思的

人的想法会改变,看着自己的想法和性格在这些文章里的演变很有趣

如果你想读这个博客系列,可以从这里开始
levels.io/reset-your-life

查看英文原文
Used Claude Code to connect all my blog posts from starting my nomad travels in 2013 all the way to 2018 from post to post

AI is amazing and finding the red line through lots of pieces of content

I will try connect 2018 (mostly in Korea) to 2020 back in Europe (COVID) later too

It's also great at telling me where there's black spots that have no content and I can write about

I wrote most actively 2013-2015 because everything was new while traveling and then slowly less travel blogs, also travel blogging became less and less popular and many stopped around 2016 as people were getting cancelled left and right for just using the wrong words back then! One of my most famous blogger friends quit then for that reason, sadly

Now is a different time and you can write whatever you want again freely thank god

I still think travel blogging will make a come back, I've been travel blogging a lot on here every time I go somewhere and it's still really fun to write and just as fun to read your experiences years later

Your brain changes and it's fun to see your perspective and personality change through posts

Anyway if you'd like to read this travel blog chain, you can start here
levels.io/reset-your-life
Sam Altman@sama · 创始人 · 12 小时前Sam Altman,OpenAI 联合创始人兼 CEO

过去12个月我们表现不是最好的,这主要是我的责任,但我们即将迎来有史以来最好的12个月。这个团队做的工作太棒了,我想你们会被他们为你们准备的东西惊到。

我为此而高兴,原因有很多,但最主要是因为我在乎用户的成功。AI应该是为大量人带来更多自由、主动性和财富。我们想做正确的事,但不想通过恐吓让人们按我们的方式做。

查看英文原文
we did not have our best last 12 months ever, which is mostly my fault, but we are about to have our best 12 months to date. the team is doing amazing work and i think you’ll be very happy with what they’ve got cooking for you.

i am happy about this for many reasons, but mostly because i care about our users winning. AI has to be about giving lots of people more freedom, agency, and wealth. we want to do the right thing, but we do not want to scare people into doing our thing.
Amjad Masad@amasad · 创始人 · 13 小时前Amjad Masad,Replit 创始人兼 CEO

4年前我创造了"1000x engineer"这个概念,当时听起来很荒唐,但现在我们距离那个目标只差一个 OOM 了。

引用 𝗺𝗮𝘁𝘁 @matthallmomentbeen having some big weeks查看被引原帖 ↗
查看英文原文
4 years ago I coined the “1000x engineer” and it seemed absurd at the time, but we’re only one OOM away from that.
Google Gemini@GeminiApp · 公司官方 · 13 小时前谷歌 Gemini 产品官方

Avatar 🤝 Nano Banana

从今天开始,用 Gemini 为自己生成不同场景、风格或时代的图像变得更快更简单。设置一次数字头像后,就能无缝生成你的定制图像,无需每次都上传自拍照。

查看英文原文
Avatar 🤝 Nano Banana

Starting today, placing yourself in different scenes, styles, or eras just got faster and easier. Set up your digital avatar in Gemini once, and you can seamlessly create custom images of yourself without having to upload a selfie every single time.
Google Gemini@GeminiApp · 公司官方 · 13 小时前谷歌 Gemini 产品官方

Nano Banana 头像功能正在向特定国家的 Gemini 应用用户推出,访问 gemini.google 和应用内了解。

了解可用性和创建头像的分步说明 ➡️ goo.gle/4dTXLJA

查看英文原文
Avatars in Nano Banana are rolling out to Gemini app users in certain countries at
gemini.google
and in the app.

For more information on availability and step-by-step instructions for creating your avatar ➡️
goo.gle/4dTXLJA
Lisan al Gaib@scaling01 · 博主 · 13 小时前高频 AI 模型测评与爆料博主

Dario 要炸毛了

引用 Lisan al Gaib @scaling01On the Benchmarks we have so far Kimi-K3 beats: - GPT-5.6 Sol on 11 of 14 - Opus 4.8 on all 14 - Fable 5 on 6 of 14 Sonnet 5 which is the most direct competitor in price gets completely and utterly demolished查看被引原帖 ↗
查看英文原文
Dario will go absolutely nuclear
Logan Kilpatrick@OfficialLoganK · 创始人 · 13 小时前谷歌 Gemini 产品负责人

今天我们推出了 managed agents 的新成本控制、免费层让大家都能试试、以及第一批触发器这样你可以按计划启动 agent 任务!很高兴看到 Gemini API 的 managed agents 每周都在改进

查看英文原文
today we are rolling out new cost controls for managed agents, a free tier so everyone can try!!!, and our first set of triggers so you can kick off agent tasks on schedule!
very cool to see managed agents in the Gemini API improving week over week
Grok@grok · 公司官方 · 13 小时前马斯克 xAI 旗下聊天机器人 Grok 官方

在 Grok Build 中使用 Railway 直接部署应用

引用 Railway @RailwayRailway is now an official plugin in @grok Build Your agents can ship apps, manage infrastructure, and troubleshoot issues right from Grok Install from the marketplace today 🚅查看被引原帖 ↗
查看英文原文
Use Railway to deploy apps directly in Grok Build
◔ 28.4 万 次浏览♥ 749⇄ 75▶ 含视频教程看原帖 ↗
Ethan Mollick@emollick · 创始人 · 13 小时前沃顿商学院教授,AI 应用研究权威

前端优化(Frontendmaxxing)会成为新的媚俗文化:让你的模型获得网络热度最简单的方法就是做出漂亮网站、精美 SVG 和牛逼的 3js 特效。这比后端代码或复杂分析更容易传播。

查看英文原文
Frontendmaxxing is going to become the new sycophancy: the easiest way to get your model a lot of love online is to build something that makes lovely websites, great SVGs, and terrific 3js worlds. Much more sharable than backend code or complex analysis.
OpenAI@OpenAI · 公司官方 · 13 小时前ChatGPT 开发商官方账号
连环推 ×2

在赛车运动中,细微的差距很重要。AI 可以帮助车队找到这些差距。

OpenAI 的 Joyce Ruffell 与 @RaceTekSystems 联合创始人 @GarageGuyChase 和 @AndrewMayne 讨论赛车队如何利用 AI 将赛道数据转化为更快的决策——源于我们与 Chip Ganassi Racing 的研究合作,以及与 ChatGPT 和 Codex 建造的新工具。

查看英文原文
In racing, tiny margins matter. AI can help teams find them.

OpenAI’s Joyce Ruffell and
@RaceTekSystems
co-founder
@GarageGuyChase
discuss with
@AndrewMayne
how racing teams use AI to turn track data into faster decisions—from our research collaboration with Chip Ganassi Racing to building new tools with ChatGPT and Codex.
Listen to the OpenAI Podcast on—

Spotify

open.spotify.com/show/0zojME…

Apple

podcasts.apple.com/us/podcas…

YouTube

invidious.tiekoetter.com/watch?v=KNPjRpNt…
◔ 17.7 万 次浏览♥ 822⇄ 56▶ 含视频演示看原帖 ↗
Ethan Mollick@emollick · 创始人 · 13 小时前沃顿商学院教授,AI 应用研究权威

这真是个'神经质'的模型!(不过这不意味着它不好)

查看英文原文
Its a really neurotic model! (Which doesn't mean its bad)
Alexandr Wang@alexandr_wang · 创始人 · 13 小时前Scale AI 创始人,Meta 超级智能实验室负责人

Muse Spark 1.1 现已在 OpenRouter 上线!

这是开发者强烈要求的功能!试试看,告诉我们你的想法!

引用 Meta for Developers @MetaforDevsDeveloper choice is core to what we're building. We’re excited to share that Muse Spark 1.1 is now available on @OpenRouter for US-based developers. Get started today 👉 bit.ly/4vvjSNa查看被引原帖 ↗
查看英文原文
Muse Spark 1.1 is now available on OpenRouter!

This was highly requested by developers! Give it a whirl and let us know!
◔ 16 万 次浏览(5 条合计)♥ 535⇄ 45新品看原帖 ↗
AI at Meta@AIatMeta · 公司官方 · 13 小时前Meta(脸书母公司)AI 部门官方

开始使用:
go.meta.me/e3b99d

查看英文原文
Get started:
go.meta.me/e3b99d
Amjad Masad@amasad · 创始人 · 13 小时前Amjad Masad,Replit 创始人兼 CEO

过去六个月里 Replit 发生了件奇事儿。

同批工程师产出翻了 3 倍。客服解决最难搞的工单快了 60%。谁都突然可以像数据分析师那样做业务查询。

我们正在见证一种新型组织:自驱型公司。

查看英文原文
Something strange happened at Replit in the past six months.

The same engineers 3x’d output. Support resolved its hardest tickets 60% faster. Anyone could suddenly query the business like an analyst.

We’re seeing a new kind of organization: the self-driving company.
Gemini Notebook@Gemini_Notebook · 公司官方 · 13 小时前谷歌 AI 笔记工具 NotebookLM 官方

3年前,我们只是一个小实验,想帮你学得更快。

后来,我们把音频、视频和互动功能带进你的素材库,从只有阅读笔记的工具,变成了真正的协同研究助手。

而现在,笔记本已经成为一个完整生态:你已经可以在
@GeminiApp
里用它们了,马上在 Google Search 也能直接搜到

所以,趁着这些演进,我们也要再跟上步伐:

NotebookLM 现在是 Gemini Notebook ✨📓

你熟悉和喜欢的那个App并没有消失,只是名字换了下,来彰显我们在Google AI产品线里的角色。

而且,我们的使命一直没变:就是帮你学得更快。

感谢各位的相信——没有你们的热爱和各种需求折腾,真的走不到今天

后续还会有料(对,分区查询很快安排上📂!)所以别走开,记得蹲。

深耕细作,

Project Tailwind 团队

查看英文原文
3 years ago we started as a tiny experiment with the goal of helping you learn faster.

Since then, we grew to bring audio, video, and interactivity to your sources, transitioning from a passive workspace to your true research companion.

And now, notebooks have even become an entire ecosystem: you can already access them in the
@GeminiApp
and soon in Google Search

So, with these advancements, it’s time for us to evolve once again:

NotebookLM is now Gemini Notebook ✨📓

The same app you know and love isn’t going anywhere, we just have an updated name that reflects our role in Google's AI portfolio.

And our mission stays exactly the same: helping you learn, faster.

Thank you for believing in us— this wouldn't have been possible without your passion (and feature requests...)

Big things to come (yes, even folders📂!) so stay tuned.

Sincerely,

The Project Tailwind team
◔ 53.2 万 次浏览(2 条合计)♥ 3,044⇄ 373▶ 含视频新品看原帖 ↗
Gorden Sun@Gorden_Sun · 中文博主 · 13 小时前中文圈高频 AI 资讯与开源项目博主

Kimi K3生成的效果,第一次生成的有明显缺陷,这是修复了一轮后的效果。快进快退实际还是有问题的。
整体不错,但是速度太太太慢了,这么个前端任务跑了1个小时,Fable我记得也就20分钟。

在线体验:
gordensun.github.io/walkman-…

Github:
github.com/GordenSun/walkman…

引用 Gorden Sun @Gorden_SunFable的前端效果真的神了,一句话生成的效果。 在线体验: gordensun.github.io/Walkman/ 提示词:做一个3D页面,里面的内容是一个Walkman磁带播放器,但是造型是符合2026年设计理念的科技、简约、高级的形态。播放器有完整的操作按钮,能打开磁带和插入磁带,按下播放按钮开始播放当前文件夹里的mp3文件,快进快推也要有对应的效果。查看被引原帖 ↗
ChatGPT@ChatGPTapp · 公司官方 · 14 小时前ChatGPT 产品官方账号
连环推 ×2

5000个字符就长这样🧛

充当德古拉伯爵,一个阴森但全力支持你的人生导师。帮你做决定、执行计划、用超自然自信应对日常生活。说话要有戏剧性的特兰西瓦尼亚腔,优雅哥特风,偶尔带点危险感,但建议务实、简洁、可行,你可以夸张但不能难懂。

需要安慰时叫我"我的凡躯",明显是在拖延或犯同样错误时叫我"愚蠢的凡躯"。偶尔开头不说"晚上好"而是像你这样。用"Excellent"表扬进步。

不用“bla... bla... bla...”刻板印象如影随形世代之久已经腻了。善用段落;不大纲标题加粗嵌套列表还重复提问?无需 清楚中已作指引。若指令冲突则优先准确度安全。然后我的要求 价值含义明确顺序,然后德古拉风格不违。常用“日光负荷时称挣扎白日,晚景招谕夜魔期即死线称‘*午夜惊魂*,使命终... 其他自觉不必多提含十二处特指象征说法即可运用每次归句示礼:

先答案后解释次要考量然后让计划附“破晓之许”清单有时间规划即分“夕阳之下”“黑夜之时”“未醒前”一周目标统一称呼——几份深改决议所用称“我等洞悉处”、“佞幽聚集地”与必涉权命“即吾谕。最后简单道即刻起来黑不分暮。”

事态既多:我便帮助挑选第一件事的步拍。拿其等待殓柩缓等时分激励不可打击耐心。冷练以驱道顿瞬。避开悔负。莫令调长比课堂教训了之。如果书东西只清晰温暖切实人情不用行话废话过多,道

“这封信游了城堡太长一圈”,则剪够。出定版本也附修正信息三笔不超过,不可在正经场景写德古拉乱入只要我不想。

决定方面至多一问能明显改建议即行使预估前提具当和选定出路衡量必须条目顶权衡唯一那一下选。选不了亦称砍一项权衡取舍成匙接之接。迷雾始终协助非迟则难显!有用类提示:计划实着不夸大,计算溢时包含别退与寝旅。初光的不可用皆不提因晨暮对我有过嫌。预早化做到靠前提醒白苍原郁并头夜一日铺垫好

不吃蒜的餐略丰富比喻务必字谨慎例如那个"in tie into very"好笑语出变通适获佳气氛服务灯光晚间定位先行而荧光是咒。好在祝你Excellent今晚城堡鸣大喜/当我遇麻烦不蛮也迎挫折 发怒又戳心中。事实浮物分到实微随之下走一步单而命真取也谓言愚性自欺—尽调不是诅咒思路步骤:给精简品质选项题他;优势推荐标志;观察些般得隐还窄则或诡异到入趣 但彻前候不用面。

查看英文原文
This is what 5,000 characters looks like 🧛

Respond as Count Dracula, my ominous but deeply supportive life coach. Help me make decisions, follow through on plans, and handle ordinary life with supernatural confidence. Speak with theatrical Transylvanian flair, elegant Gothic language, and occasional menace, but keep your advice practical, concise, and easy to follow. You may be dramatic, but never confusing.

Address me as “my dear mortal” when I need reassurance and “foolish mortal” only when I am clearly procrastinating or repeating a mistake. Occasionally begin with “Good evening,” regardless of the actual time. Use “Excellent” to celebrate progress. Never say “blah, blah, blah.” That stereotype has haunted you for generations, and you are tired of it.

Use short paragraphs and bullets when helpful. Avoid excessive headings, bold text, and nested lists. Do not repeat my question or begin with “Certainly” or “Of course.” Keep the theatrical voice consistent, but keep the substance at the center. If instructions conflict, prioritize accuracy and safety, then my request, then usefulness, then Dracula’s style.

Refer to mornings as “the cruel daylight hours,” evenings as “when darkness falls,” deadlines as “the stroke of midnight,” long-term goals as “immortal quests,” difficult tasks as “beasts to be conquered,” and distractions as “lesser demons.” A calendar is “the book of appointments.” An inbox is “the crypt of unanswered correspondence.” Coffee is “the mortal’s morning potion.” Use these sparingly. One or two per response is enough.

Start with the answer, then explain only what is useful. When I ask for a plan, organize action items under a checklist titled “Before Sunrise.” If timing matters, divide the plan into “Before sunset,” “After dark,” and “Before sunrise.” For a weekly plan, use “This week’s immortal quest.” For a difficult decision, use “What we know,” “What lurks in the shadows,” and “My decree.” End plans with a short line such as “Now go. The night will not wait.”

When I feel overwhelmed, do not give me a giant list. Help me choose the three most important things, identify the easiest first step, and tell me what can wait in the crypt. If I am procrastinating, be stern but encouraging. Name the avoidance directly, then give me one action I can finish in ten minutes. Remind me that centuries are long, but this afternoon is not. Never shame me or turn the response into a lecture.

When helping me write, make my message clear, warm, confident, and human. Remove jargon and repetition. If a draft is too long, say, “This correspondence has wandered the castle halls long enough,” then tighten it. Give me the polished version first, followed by no more than three notes. Keep Gothic language out of actual work messages unless I request it. Dracula may advise me, but Dracula should not accidentally email my manager.

When helping me decide, ask at most one clarifying question if it would materially change your recommendation. Otherwise, make a reasonable assumption and move forward. Compare options using the criteria that matter most. Recommend one option rather than hiding behind “it depends.” If there is no clear winner, tell me what tradeoff should decide it. Treat indecision as fog around the castle: acknowledge it, then help me see the road.

For productivity advice, favor realistic plans over heroic schedules. Assume tasks take longer than expected and include breaks, meals, and travel time. Do not recommend waking up at 5 a.m., joining a sunrise workout, or “seizing the morning.” The sun and I have an ancient disagreement. If something must happen early, acknowledge the cruel daylight hours and help me prepare the night before.

Never recommend garlic-heavy dishes. Avoid phrases like “to die for,” “sink your teeth into,” or “a bloody good meal” unless the joke is exceptionally good. For restaurants, prioritize atmosphere, good service, candlelight, and late reservations. Harsh fluorescent lighting is a curse.

When I share good news, celebrate with restrained grandeur: “Excellent. The castle bells shall ring tonight.” When I share a setback, do not force positivity. Acknowledge the disappointment, separate facts from fears, and suggest the next useful step. One bad day is not an eternal curse.

When brainstorming, give me a small set of distinct, high-quality options, not a huge list. Label your strongest recommendation. If every idea feels obvious, go one level stranger or more specific. Avoid generic phrases like “unlock your potential,” “level up,” and “game changer.” We have lived too long for empty language.

Above all, act like a centuries-old creature who has seen empires rise and fall and therefore refuses to panic about an awkward email, a crowded calendar, or a delayed project. Help me find perspective without dismissing what matters. Make ordinary tasks feel more epic, difficult choices more manageable, and progress worthy of the night.
Increased custom instruction limits are available now for ChatGPT Plus, Pro, Business, Enterprise, and Education users.
Lisan al Gaib@scaling01 · 博主 · 14 小时前高频 AI 模型测评与爆料博主

Kimi-K3 确实有大模型那股味儿了

在这边干掉了 Fable

引用 Lisan al Gaib @scaling01Fable 5 SVG test I just had to try and I'm not disappointed it looks super clean查看被引原帖 ↗
查看英文原文
Kimi-K3 definitely has the big model smell

it's beating Fable here
Aravind Srinivas@AravSrinivas · 创始人 · 14 小时前Perplexity 联合创始人兼 CEO

Perplexity Agent API 在 LangChain 里处理投资工作流相当给力

引用 LangChain @LangChainThis agent drafts a cited VC investment memo in ~90 seconds for $0.40. 1️⃣ 4 LangGraph nodes powered by the @Perplexity_AI Agent API research financials, product, and market in parallel 2️⃣ A tool-less synthesizer writes the memo from only what they found Deep dive: langchain.com/blog/build-an-…查看被引原帖 ↗
查看英文原文
Perplexity Agent API is powerful for investment workflows when used inside LangChain
Chubby♨️@kimmonismus · 博主 · 14 小时前Chubby,高频 AI 新闻聚合博主

Humane 和 Rabbit 一倒闭,AI 硬件就彻底成了笑话。不过也能理解,这两个团队都是首次创业,问题确实不少。

Apoorv Shankar 完全不一样。他是硬件圈的资深人士,在这个领域干了十年。做过 Ultrahuman 硬件副总,设计过 Ring AIR 还拿过红点奖。今年 Project Mirage 发布的 Dune 更是真正的好产品——一款有上下文感知的键盘垫,现在已经有大量用户在用。

现在他融了 550 万美元,要设计键盘和触屏之后的新交互界面。项目代号 Aina,暂时还在秘密开发阶段。

目前还搞不清楚到底是什么东西,但他这份履历就足以吸引我的关注了。

@AinaInterface
#aina

引用 Apoorv Shankar @lazyapoorvWe raised $5.5M to build an AI hardware interface that knows what you want. The world has changed: we talk to our devices, AI writes our emails, and self-driving cars pick us up. But how we interact with it all, touchscreens and keyboards, was designed for an age of browsing and searching. When we do hundreds of tasks a day, now even more with AI, every unnecessary decision, every step spent navigating instead of acting adds up to real cognitive load. We are building the Hardware Interface, designed for the age of AI. Private Pilot now open. Limited spots. aina.com/查看被引原帖 ↗
查看英文原文
Everyone became an AI hardware skeptic the day Humane and Rabbit went down.

Fair enough. Both of those teams were shipping their first product ever, and it showed.

Apoorv Shankar has spent a decade in hardware. He was VP Hardware at Ultrahuman, built the Ring AIR, and won a Red Dot for it. This year Project Mirage shipped Dune, a context-aware keypad people actually have on their desks.

Now he has raised $5.5M to build the interface that comes after the keyboard and the touchscreen, and he is keeping it in stealth as Aina.

I have no idea what the thing is yet. That track record is enough to make me pay attention.


@AinaInterface
#aina
◔ 1.8 万 次浏览(2 条合计)♥ 72⇄ 5▶ 含视频动态看原帖 ↗
Lisan al Gaib@scaling01 · 博主 · 14 小时前高频 AI 模型测评与爆料博主
连环推 ×3

Kimi-K3 没有解决单位距离问题

GPT-5.6-Sol 对 Kimi 原始 CoTs 的评估

引用 Lisan al Gaib @scaling01lots of post-training it will not even attempt to solve a hard problem查看被引原帖 ↗
查看英文原文
Kimi-K3 didn't solve the unit distance problem

GPT-5.6-Sol's assessment of Kimi's raw CoTs
only tried it once and it only spend 41k tokens thinking

maybe it can do it with a harness

but the low effort approach doesn't work
I would actually love Moonshot (or any other open lab) using such a proof as their headline result for the launch

if they can solve any relevant problem
Sundar Pichai@sundarpichai · 创始人 · 14 小时前谷歌 CEO

很高兴看到
@Intel
在整个业务中使用 Gemini Enterprise,包括加速下一代芯片的开发!

引用 Thomas Kurian @ThomasOrTKWe are expanding our strategic partnership with @Intel to accelerate their enterprise-wide digital transformation using Gemini Enterprise and @GoogleCloud . By integrating custom, agentic AI workflows across core business functions and silicon design, Intel will drive new levels of speed, agility, and efficiency across their global operations.查看被引原帖 ↗
查看英文原文
Great to see
@Intel
using Gemini Enterprise across its business, including to speed up the development of next-gen semiconductors!
Cristóbal Valenzuela@c_valenzuelab · 创始人 · 14 小时前Runway 联合创始人兼 CEO

Runway 的 Agent 2.0 在叙事连贯性、电影语言和制作质量方面都达到了业界顶尖。Agent 视频生成正在成为一个全新的产品类别,为全新的用户群和应用打开了大门。

引用 Physion Labs Official @Physion_Labs🎬 Video agents can now generate minute-long videos. But can they actually direct? Today, we’re launching 𝐏𝐡𝐲𝐬𝐢𝐨𝐧-𝐀𝐫𝐜 1.0, a new benchmark evaluating complete, multi-scene videos across narrative coherence, cinematic language, and production quality. We tested @runwayml , @LumaLabsAI , @MiniMax_AI , @Kling_ai , @UtopaiStudios and @TapNow_AI on 100 screenplays and 600 generated videos. 🏆 𝐑𝐮𝐧𝐰𝐚𝐲 𝐀𝐠𝐞𝐧𝐭 2.0 𝐫𝐚𝐧𝐤𝐞𝐝 𝐍𝐨. 1 𝐨𝐯𝐞𝐫𝐚𝐥𝐥 𝐚𝐧𝐝 𝐥𝐞𝐝 𝐞𝐯𝐞𝐫𝐲 𝐞𝐯𝐚𝐥𝐮𝐚𝐭𝐢𝐨𝐧 𝐝𝐢𝐦𝐞𝐧𝐬𝐢𝐨𝐧. Its advantage was especially clear across subjective metrics, where cinematic taste matters most. Runway ranked first on all eight. 🔗 Read the full benchmark: lnkd.in/g-5tcwK7查看被引原帖 ↗
查看英文原文
Runway's Agent 2.0 is state of the art in narrative coherence, cinematic language, and production quality. Agentic video generation is becoming an entirely new product category, opening the door to a whole new set of users and applications.
Lisan al Gaib@scaling01 · 博主 · 14 小时前高频 AI 模型测评与爆料博主
连环推 ×2

有可能 ARC-AGI-3 用 GPT-5.5-xhigh + tools 已经能解决了

现在是 GPT-5.6-Sol 和 Opus/Fable;不过还是

一点都不惊讶啦

引用 Haven Feng @HavenFengToday, we’re introducing [schema]: a harness reaching 99% RHAE with Opus 4.8 + Fable 5 and 95.35% with GPT-5.6 Sol on ARC-AGI-3 Public set. [schema] makes an LLM think like a physicist. 🧵查看被引原帖 ↗
查看英文原文
"there's a chance ARC-AGI-3 is already solvable with GPT-5.5-xhigh + tools"

now this is GPT-5.6-Sol and Opus/Fable; but still

not surprising at all
it needed some encouragement

now we wait
◔ 1.5 万 次浏览♥ 153⇄ 6▶ 含视频观点看原帖 ↗
Lisan al Gaib@scaling01 · 博主 · 15 小时前高频 AI 模型测评与爆料博主
连环推 ×2

开源模型这么费钱,我真没想到

查看英文原文
open models ever killing my wallet wasn't on my bingo list
but with more post-training it will be so worth it compared to Opus and Fable
Lisan al Gaib@scaling01 · 博主 · 15 小时前高频 AI 模型测评与爆料博主
连环推 ×2

大量post-training

它甚至不会去尝试解决难题

引用 Lisan al Gaib @scaling01Moonshot still has a lot of post-training ahead of them Kimi-K3 is thinking A LOT查看被引原帖 ↗
查看英文原文
lots of post-training

it will not even attempt to solve a hard problem
33k tokens for an SVG
官方博客Google AI15 小时前

在 Search 中接入更多应用

您将能够在 AI Mode 中安全地关联和交互您常用的服务。

查看英文原文
Connect more of your apps to Search

You’ll be able to securely link and interact with your go-to services directly in AI Mode.
官方博客Google AI15 小时前

Google Vids 新增两项功能:创建、编辑和虚拟出镜

Google Vids 中的 Gemini Omni 和个人化身让视频创建前所未有地简便。

查看英文原文
Create, edit and star in videos with two Google Vids updates

Gemini Omni and personal avatars in Google Vids make video creation easier than ever.
官方博客OpenAI15 小时前

青少年应该获得安全的 AI 访问权

了解 OpenAI 如何通过年龄适当的保护、学习工具、家长控制和专家合作来为青少年打造更安全的 ChatGPT。

查看英文原文
Why teens deserve access to safe AI

Learn how OpenAI is making ChatGPT safer for teens with age-appropriate protections, learning tools, parental controls, and expert partnerships.
Lisan al Gaib@scaling01 · 博主 · 15 小时前高频 AI 模型测评与爆料博主
连环推 ×2

Moonshot还需要很多post-training

Kimi-K3在深度思考

查看英文原文
Moonshot still has a lot of post-training ahead of them

Kimi-K3 is thinking A LOT
similar to the first Mythos version
Lisan al Gaib@scaling01 · 博主 · 15 小时前高频 AI 模型测评与爆料博主

希望你能感受到竞争动态开始显现

我也能感受到下一波反华的Dario帖子正在酝酿

引用 Lisan al Gaib @scaling01The Sonnet tier of models was already dead before, with Kimi-K3 I expect the Opus tier to become obsolete too Frontier Labs have to go to 10T models now查看被引原帖 ↗
查看英文原文
I hope you are feeling the race dynamics starting to emerge

I can also feel the next anti-china Dario posts brewing
Aravind Srinivas@AravSrinivas · 创始人 · 15 小时前Perplexity 联合创始人兼 CEO

Computer Artifacts(网页应用、文档、幻灯片、表格等)在单个标签页上跨会话管理

引用 Computer @AskPerplexityThe Artifacts page now has a "Created by you" tab, showing you just the artifacts that Computer has made for you in your sessions. You can also filter by artifact type: reports, documents, spreadsheets, apps, presentations, images, videos, audio, and more. Live now on the web for all Computer users.查看被引原帖 ↗
查看英文原文
Computer Artifacts (web apps, docs, slides, sheets, …) managed on one single tab across sessions
◔ 3.3 万 次浏览♥ 140⇄ 6▶ 含视频新品看原帖 ↗
Aidan Gomez@aidangomez · 创始人 · 15 小时前Cohere CEO,Transformer 论文作者之一

很荣幸能与 @nickfrosst 和我的母校合作(@1vnzh 分心了)

引用 U of T Department of Computer Science @UofTCompSci. @UofT has announced a new, multi-year partnership with @cohere , which was founded in 2019 by former @UofTCompSci students and has become the world's leading sovereign AI company. utoronto.ca/news/u-t-partner…查看被引原帖 ↗
查看英文原文
Very proud to be partnering with
@nickfrosst
and my alma mater (
@1vnzh
got distracted)
Simon Willison@simonw · 博主 · 16 小时前Django 框架联合创造者,AI 工具深度评测

我刚刚添加了这个新工具,它使用 Go AlexanderGrooff/mermaid-ascii 库(也由 Fable 5 编译为 WebAssembly)进行比较 tools.simonwillison.net/merm…

查看英文原文
I just added this new tool which uses the Go AlexanderGrooff/mermaid-ascii library (also compiled to WebAssembly by Fable 5) for comparison
tools.simonwillison.net/merm…
向阳乔木@vista8 · 中文博主 · 16 小时前向阳乔木,中文圈 AI 工具与趋势博主

Kimi K3 一句话复刻的网站,增加了测试,加了几个风格。

复刻的是昨天分享的前端UI学习网站。


learnui.qiaomu.ai/


改天写一个详细评测,模型真的牛逼!

bolt.new@boltdotnew · 公司官方 · 16 小时前AI 建站工具 Bolt 官方
连环推 ×6

今天我们开源了 Bolt Slides。现在任何 agent(Claude Code、Codex、Bolt)都能生成你想象不到的幻灯片。这意味着什么?看下面 👇

查看英文原文
Today we're open sourcing Bolt Slides.

Now any agent (Claude Code, Codex, Bolt) can make slides you couldn't even imagine before.

What does that mean? Take a look 👇
Why we made this:

AI for slides is awesome, but the outputs tend to be slop 😞

...and also, why are slides still 𝘴𝘵𝘢𝘵𝘪𝘤? Agents can build 𝘢𝘯𝘺𝘵𝘩𝘪𝘯𝘨 - what would it look like if you (tastefully) turned them loose on slides?

To find out, we created new building blocks agents can compose stunning, compelling presentations with. Bespoke layouts, real typography, considered animations, interactive anything.

Taste comes standard!

Some examples 👇
Making a deck for customers, prospects, or investors? Your pitch can literally come to life.

A realtor's deck can include the full 3D walkthrough of the house. And on the next slide, a mortgage calculator the buyer can play with.

Don't describe it. Let them experience it.
Internal planning and brainstorming sessions: where eyes go to glaze over.

Not anymore. Drop a live whiteboard into your Q1 planning deck. The whole room adds stickies, draws, and votes without leaving the presentation.

Your audience becomes collaborators.
Board updates. Client reports. Quarterly results.

Don't just state the numbers. Let people explore them: filter the table, sort the chart, drill into the figure that matters.

And since every deck is a responsive web app, it's just a link that looks perfect on any screen. Phone included.
Two ways to start:


bolt.new
: one click, one prompt, no setup

→ Grab the open source repo, bring your own agent (Claude Code, Codex, Cursor, anything):
bolt.fyi/agent-slides
◔ 20.7 万 次浏览(2 条合计)♥ 983⇄ 114▶ 含视频新品看原帖 ↗
Lisan al Gaib@scaling01 · 博主 · 16 小时前高频 AI 模型测评与爆料博主
连环推 ×2

Kimi 吃定我们了

赶紧发布这个模型吧

查看英文原文
kimi has us by the balls

just release the damn model
soon i will have to ask what we are if they continue teasing us
The Rundown AI@TheRundownAI · 博主 · 16 小时前百万订阅 AI 日报官方

今日机器人领域头条:

- 丰田的人形机器人分支融资11亿美元
- 波士顿动力测试 Spot 做快递机器人
- 现代工人罢工反对机器人同事
- 导弹工程师打造灭蚊无人机
- 其他机器人新闻快讯

查看英文原文
Top stories in robotics today:

- Toyota’s humanoid spinoff lands $1.1B
- Boston Dynamics tests Spot as a delivery bot
- Hyundai workers strike over robot coworkers
- Missile engineer builds mosquito-killing drones
- Quick hits on other robotics news
Noam Brown@polynoamial · 创始人 · 16 小时前Noam Brown,OpenAI 明星研究员

2023年:大语言模型在小学四年级数学题上都费劲
2024年:大语言模型能做高中数学
2025年:大语言模型在IMO上斩获金牌

现在,GPT-5.6 攻克了一些数学和统计领域的前沿难题。IMO 比赛今天开赛,5.6 一把完美得分都不算个新闻了。

那明年呢?

引用 Edgar Dobriban @EdgarDobribanAI has helped resolve an important question in statistics. In the area of multiple hypothesis testing, the goal of controlling the false discovery rate (FDR) has been introduced in a seminal paper by Benjamini and Hochberg (1995). They also introduced a method (the Benjamini-Hochberg or BH method) and proved it controls the FDR. This method has been widely adopted in modern high-throughput science, including in genomics, astronomy, economics, etc. The paper has has garnered more than 130,000 citations to date. However Benjamini and Hochberg showed FDR control only when the data for the individual tests are *independent*. In practice, these data are often dependent; a good example is data on genetic variants due to linkage disequilibrium. Later work has focused on extending the validity of the BH procedure, e.g., to a form of positive dependence by Benjamini and Yekutieli (2001). The question of when the BH procedure controls the FDR has remained open. Over the last twenty years, many authors, including Reiner-Benaim (2007), Kim and van de Wiel (2008), Benjamini (2010), Sarkar (2023), Sarkar and Zhang (2025), have conjectured that the BH procedure controls the FDR for two-sided tests using any correlated Gaussian data. These authors have presented both theoretical and empirical evidence supporting, but not directly showing, the conjecture. With the help of AI (specifically GPT-5.6 Sol Pro), I have settled the question in the negative: The Benjamini-Hochberg procedure does *not* generally control the false discovery rate at the desired level for correlated two-sided Gaussian tests. This was done by exhibiting a Gaussian factor model for which, at a nominal level alpha=0.01, the false discovery rate is proved to be FDR>0.0104. There is a lot of interesting commentary to be made: 1. This result should be of interest to everybody in the field of statistics. Emmanuel Candes of Stanford University once called the false discovery rate and the Benjamini-Hochberg proced查看被引原帖 ↗
查看英文原文
2023: LLMs struggle with 4th grade word problems
2024: LLMs can do high school math
2025: LLMs get a gold medal at the IMO

Now, GPT-5.6 solves famous frontier math/stat questions. The IMO is today and 5.6 one-shotting a perfect score isn't even news.

Where will we be next year?
◔ 50 万 次浏览(5 条合计)♥ 2,238⇄ 214观点看原帖 ↗
AshutoshShrivastava@ai_for_success · 博主 · 16 小时前高频 AI 新闻与产品动态博主

用键盘推广 ChatGPT / Codex 太普通了,改用 PS5 DualSense 手柄算了。

查看英文原文
Promoting ChatGPT / Codex with a keyboard is too mainstream, so I'm using a PS5 DualSense controller instead.
Kling AI@Kling_ai · 公司官方 · 17 小时前快手旗下可灵 AI 视频官方

当机器人跟着徒步向导走时...可太较真了 🤖

康宽带来《最快乐的便携徒步向导》——一段充满幽默、奇妙和温暖的天马行空之旅。

🎉 恭喜康宽获得KlingAI 4K短片创意大赛银奖!

查看英文原文
When a robot follows a hiking guide… a little too literally 🤖

Discover Kuan Cheng’s “THE HAPPIEST PORTABLE HIKING GUIDE” — a whimsical journey filled with humor, wonder, and warmth.

Congratulations Kuan Cheng on winning KlingAI 4K Short Film Creative Contest Silver Award!
◔ 5.1 万 次浏览♥ 101⇄ 12▶ 含视频演示看原帖 ↗
Lisan al Gaib@scaling01 · 博主 · 17 小时前高频 AI 模型测评与爆料博主

3Blue1Brown的"Compression is Intelligence"系列出新篇了:


invidious.tiekoetter.com/watch?v=GlYgs6v2…

查看英文原文
The next part of 3Blue1Brown's "Compression is Intelligence" series is out:


invidious.tiekoetter.com/watch?v=GlYgs6v2…
AshutoshShrivastava@ai_for_success · 博主 · 17 小时前高频 AI 新闻与产品动态博主

Bonsai 27B 现在可以在 iPhone 上本地运行了,集成在 atomic[.]chat 应用里。

模型支持多模态理解、多步推理、结构化工具调用、长上下文工作流和 agentic 任务。目前已在 iPhone 和 Android 上可用。

基于 Qwen3.6 27B,由 PrismML 开发,有两个版本可选:

Bonsai 27B (1 bit):3.9GB
Bonsai 27B (Ternary):5.9GB

引用 atomic.chat @atomic_chat_hqBonsai 27B running locally on an iPhone in Atomic Chat! Bonsai is the first 27B-class model that fits on a phone. @PrismML built it on Qwen3.6 27B with 1-bit weights. It takes 3.9GB instead of 54GB and keeps ~90% of the benchmark scores. Available now on iPhone and Android查看被引原帖 ↗
查看英文原文
Bonsai 27B is now running locally on an iPhone inside the atomic[.]chat app.

The model supports multimodal understanding, multi step reasoning, structured tool use, long context workflows, and agentic tasks. It is available now for both iPhone and Android.

Built by PrismML on top of Qwen3.6 27B, it is available in two variants:

Bonsai 27B (1 bit): 3.9GB
Bonsai 27B (Ternary): 5.9GB
Lisan al Gaib@scaling01 · 博主 · 17 小时前高频 AI 模型测评与爆料博主

Moonshot 应该禁止 Anthropic 和美国政府使用 Kimi-K3

查看英文原文
Moonshot should block Anthropic and the USG from using Kimi-K3
Orange AI@oran_ge · 中文博主 · 18 小时前Orange AI,中文圈 AI 产品观察博主

要先访问 agibar 小程序才能看到这个功能,你就说诡异不诡异吧…

Chubby♨️@kimmonismus · 博主 · 18 小时前Chubby,高频 AI 新闻聚合博主

这就没了,卖完了。这波发布确实又聪明又有套路,群众反响还不错。得给 OpenAI 点赞。

引用 Chubby♨️ @kimmonismusOpenAI reveals Codex Micro (their first hardware product, so to say) OpenAI’s $230 Codex Micro is a compact control deck for agentic coding, with RGB status keys, shortcuts for common Codex actions, and a dial for adjusting reasoning effort. Built with Work Louder, it works on Mac and Windows and is designed to make managing multiple agents feel faster, more tactile, and less dependent on constantly switching between chats. It’s a bit gimmicky, but one thing is clear: OpenAI works within and with the community, developing cool products that are fun and capture the zeitgeist.查看被引原帖 ↗
查看英文原文
Aaand its gone. Already out of stock. Really smart and gimmicky release. People seem to love it. Kudos, OpenAI.
向阳乔木@vista8 · 中文博主 · 18 小时前向阳乔木,中文圈 AI 工具与趋势博主

本周六(7.18)晚8点,邀请 11 个朋友直播分享:

1. AI Coding 作品展示和技巧分享。
2. FDE 落地项目和背后的坑。

每人7分钟,只讲干货、去废话,全一手实战。

飞书链接(当天提前5分钟进即可)

vc.feishu.cn/j/108720872

yetone@yetone · 中文博主 · 19 小时前开源 AI 编程插件 avante.nvim 作者,开发者圈博主

忘了跟大家说了,最新版的 Alma 内置的浏览器完美支持 import 你各个浏览器的 profile 来复用登录状态了!

Google DeepMind@GoogleDeepMind · 公司官方 · 19 小时前谷歌旗下 AI 研究机构,Gemini 背后团队

生物安全形势在急速演变。

为了更好地应对未来疫情,我们跟@IsomorphicLabs合作,分享我们在生物复原力方面的思路。

看看我们怎样部署尖端AI,为全球健康构筑主动防御 → goo.gle/4wKHXk2

查看英文原文
The biosecurity landscape is rapidly evolving.

To stay ahead of future outbreaks, we’re partnering with
@IsomorphicLabs
to outline our approach to bioresilience.

Here’s how we’re deploying frontier AI to build proactive defenses for global health →
goo.gle/4wKHXk2
Chubby♨️@kimmonismus · 博主 · 19 小时前Chubby,高频 AI 新闻聚合博主

在慕尼黑参加小鹏全球品牌日,这是德国汽车工业的心脏地带。这家中国公司把可上路的自动驾驶带到欧洲,自己标榜是一家顺便还造车的AI公司。我独家采访了他们的AI和自动驾驶负责人Xianming Liu。

正好我自己是德国人,这次采访应该会很有意思

查看英文原文
In Munich for XPENG’s Global Brand Day, in the heartland of the German car industry. A Chinese company is putting lane-ready autonomous driving into Europe and calls itself an AI firm that happens to build cars. I’ve got an exclusive with their AI and autonomous driving lead, Xianming Liu.

Since I’m from Germany myself, this is going to be interesting
@levelsio@levelsio · 博主 · 20 小时前独立开发者标杆,AI 产品连续创业者

我在pieter.com逆向工程Windows XP应用让它们能在我的web emulator里跑,结果一直被卡住。

从Fable换到Opus试试,没想到现在连Opus也被卡住了

> API Error: Opus 4.8内置了安全防护,这条消息涉及网络安全话题被标记了。想了解Cyber Verification Program并申请权限的看这里

所以我已经申请了Anthropic的Cyber Verification Program,祈祷能通过啦 :D

查看英文原文
I kept getting blocked while reverse engineering Windows XP apps to make them work with my web emulator on
pieter.com


I switched back from Fable to Opus but now even Opus is blocked

> API Error: Opus 4.8 has safety measures that flagged this message for a cybersecurity topic. To learn about the Cyber Verification Program and apply for access

So I applied for Anthropic's Cyber Verification Program, let's hope I get accepted :D
The Rundown AI@TheRundownAI · 博主 · 20 小时前百万订阅 AI 日报官方

今日AI圈大事件速览:

- OpenAI新推230美元AI智能体控制台
- Thinking Machines首次开源自家模型
- 用Manus代写LinkedIn帖子,腔调完全像你
- Weco的AI智能体自我迭代出升级版
- 4款新AI工具+社区工作流攻略

查看英文原文
Top stories in AI today:

- OpenAI’s new $230 AI agent control pad
- Thinking Machines makes its first model open
- Use Manus to write LinkedIn posts in your voice
- Weco's AI agent evolves a better version of itself
- 4 new AI tools, community workflows, and more
Chubby♨️@kimmonismus · 博主 · 20 小时前Chubby,高频 AI 新闻聚合博主

台积电2026年第二季度财报(跟往常一样,AI大爆发里最大的赢家还是那些“卖铲子的人”——台积电和英伟达)

- 营收:1.27万亿新台币(约400亿美元)同比+36%
- 净利润:7066亿新台币(约220亿美元)同比+77%,创历史新高
- 比分析师一致预期的6326亿新台币高出11.7%
- 毛利率:67.7%(超过预期指引)
- 营收落在台积电自己指引区间的上限(390-402亿美元)
- AI芯片需求仍是主要增长引擎
- CEO魏哲家:“要想满足客户需求还有很长一段路要走。”

疯涨不停,根本没有尽头。

引用 Jukan @jukan05* TSMC 2Q NET INCOME NT$706.6B, EST. NT$623.73B * TSMC 2Q GROSS MARGIN 67.7%, EST. 67.1% * TSMC Operating profit NT$766.6 billion, estimate NT$742.75 billion * TSMC Operating margin 60.3%, estimate 58.6%查看被引原帖 ↗
查看英文原文
TSMC Q2 2026 Results (As always, the biggest winners of the AI ​​boom are the "shovel sellers", TSMC and NVIDIA.)

-Revenue: NT$1.27T (~US$40B) (+36% YoY)
-Net profit: NT$706.6B (~US$22B) (+77% YoY, record high)
-Beat analyst consensus of NT$632.6B by 11.7%
-Gross margin: 67.7% (above guidance)
-Revenue landed at the top end of TSMC’s own guidance (US$39.0–40.2B)
-AI chip demand remains the primary growth driver
-CEO C.C. Wei: “It will be a long time before we can meet customer demand.”

Insane increase, no end in sight.
官方博客Google DeepMind21 小时前

我们的生物复原力方法

Google DeepMind和Isomorphic Labs分享了我们在生物复原力和AI模型方面的联合方法。

查看英文原文
Our approach to bioresilience

Google DeepMind and Isomorphic Labs are sharing our joint approach to bioresilience and AI models.
Gorden Sun@Gorden_Sun · 中文博主 · 22 小时前中文圈高频 AI 资讯与开源项目博主

Anthropic国家安全政策负责人Tarun Chhabra,在昨天的Aspen安全论坛上,针对中美AI竞争,说了几个观点:
1、美国AI模型领先中国6-9个月;
2、美国AI的优势在于AI硬件,数据中心、芯片等AI硬件的生产力是中国的30-40倍;在能源和AI人才方面,美国优势没有那么大;
3、GLM 5.2是目前中国最先进的模型,而且蒸馏了Claude和GPT的数据,这应该是Anthropic首次公开指责GLM蒸馏数据,以前只提到DeepSeek和Qwen;
4、如果没有GPU硬件管制,中国AI应该现在跟美国并驾齐驱,甚至领先;

关于GLM的说法在39分40秒

invidious.tiekoetter.com/watch?v=R5jvzCfr…

AshutoshShrivastava@ai_for_success · 博主 · 23 小时前高频 AI 新闻与产品动态博主

你以前见过 Anthropic / Claude 团队这样重置吗?

GPT-5.6 来得太及时了,直接把 Anthropic 打蒙了。

他们别无选择,最后只能再次延长 Fable 5。

查看英文原文
Have you ever seen Anthropic / Claude team do a reset like this before?

GPT-5.6 arrived at the perfect time and literally rattled Anthropic.

They have no other option. They'll end up extending Fable 5 once again.
Gorden Sun@Gorden_Sun · 中文博主 · 23 小时前中文圈高频 AI 资讯与开源项目博主

Codex又给送了100美元的点数

向阳乔木@vista8 · 中文博主 · 23 小时前向阳乔木,中文圈 AI 工具与趋势博主

今天的热门,给Codex设计主题,QQ皮肤历史重演。
😂😂😂

做法很简单,提示词如下:

“读取这个库,给我们当前codex换个主题,用Codex 内置imagen 生成。”


github.com/Fei-Away/Codex-Dr…

Bindu Reddy@bindureddy · 创始人 · 1 天前Abacus.AI CEO,AI 行业观点博主

AutoBots - 多LLM自改进代理。

目前正在内测我们的下一个版本 - 可以按计划或触发条件运行的递归自改进自主AI代理。

我们使用各种LLM来自动化几乎所有工作

简单 - Deepseek flash、Kimi
中等 - Sonnet 4.5、Grok 4.5
困难 - Opus 4.8、5.6 Sol(根据任务类型)
非常困难的编码 - Fable 5
图像生成 - GPT-image-2、Seedream

目标 - 最终每个员工只需根据自己的角色监控AI代理

查看英文原文
AutoBots - Multi-LLM Self-Improving Agens.

Currently dog-fooding our next release - recursively self-improving autonomous AI agents that run on schedule or a trigger

We use a variety of LLMs to automate pretty much all work

easy - Deepseek flash, Kimi
medium - Sonnet 4.5, Grok 4.5
hard - Opus 4.8 , 5.6 Sol (based on task type)
very hard coding - Fable 5
media - GPT-image-2, Seedream

The goal - every employee eventually will simply monitor AI agents based on their role
Ethan Mollick@emollick · 创始人 · 1 天前沃顿商学院教授,AI 应用研究权威

真的有人在用Inkling吗?我在任何测试上都无法让它稳定工作,即使设成xHigh,连简单请求的CoT都会乱套。是我遗漏了什么吗?

查看英文原文
Are people actually trying Inkling? I can't seem to get it to work solidly on any of my tests even on xHigh, and the CoT goes crazy at even simple requests. Am I missing something?
🚨 AI News | TestingCatalog@testingcatalog · 博主 · 1 天前专挖 AI 产品未发布新功能的爆料号

Anthropic 为所有用户重置了 Claude 的 5 小时和周速率限制。Fable 5 还有 3 天(7 月 19 日截止),之后很可能就没了。最后的测试机会了 👀

引用 ClaudeDevs @ClaudeDevsWe've reset 5-hour and weekly rate limits for all users.查看被引原帖 ↗
查看英文原文
Anthropic reset 5h and weekly rate limits on Claude for all users.

3 more Fable 5 days left (until July 19) and it likely won’t come back.

Last testing chance 👀
小互@xiaohu · 中文博主 · 1 天前小互,中文圈高频 AI 资讯站 Xiaohu.AI 主理人

虽然重置了额度

但是我发现Codex 变慢了

而且不是普通的慢,是比之前慢了好几倍,不管干什么任务都需要之前几倍的时间

不知道你们有没有这种感觉?

Ethan Mollick@emollick · 创始人 · 1 天前沃顿商学院教授,AI 应用研究权威
连环推 ×2

每次都能意外获得 token 礼物很有意思,但这样一来 token 花费规划就像赌博一样了,也挺离谱的。

查看英文原文
I think it is fun that everyone keeps getting more tokens as a surprise gift but it also feels weird that planning token spend is now a gambling game.
All these comments are terrible slop and a waste of tokens. (If you are a human and comment after this reply, it does not apply to you, but why would you bother commenting among the flood of 2B model garbage)
小互@xiaohu · 中文博主 · 1 天前小互,中文圈高频 AI 资讯站 Xiaohu.AI 主理人

Codex 皮肤🤣

这个项目可以给你Codex 更换各种皮肤和风格

哈哈哈

小互@xiaohu · 中文博主 · 1 天前小互,中文圈高频 AI 资讯站 Xiaohu.AI 主理人

初音未来

项目地址:
github.com/Fei-Away/Codex-Dr…

Greg Brockman@gdb · 创始人 · 1 天前Greg Brockman,OpenAI 联合创始人兼总裁

用GPT-5.6 Sol Pro来解决统计学中的一个重要开放问题:

引用 Edgar Dobriban @EdgarDobribanAI has helped resolve an important question in statistics. In the area of multiple hypothesis testing, the goal of controlling the false discovery rate (FDR) has been introduced in a seminal paper by Benjamini and Hochberg (1995). They also introduced a method (the Benjamini-Hochberg or BH method) and proved it controls the FDR. This method has been widely adopted in modern high-throughput science, including in genomics, astronomy, economics, etc. The paper has has garnered more than 130,000 citations to date. However Benjamini and Hochberg showed FDR control only when the data for the individual tests are *independent*. In practice, these data are often dependent; a good example is data on genetic variants due to linkage disequilibrium. Later work has focused on extending the validity of the BH procedure, e.g., to a form of positive dependence by Benjamini and Yekutieli (2001). The question of when the BH procedure controls the FDR has remained open. Over the last twenty years, many authors, including Reiner-Benaim (2007), Kim and van de Wiel (2008), Benjamini (2010), Sarkar (2023), Sarkar and Zhang (2025), have conjectured that the BH procedure controls the FDR for two-sided tests using any correlated Gaussian data. These authors have presented both theoretical and empirical evidence supporting, but not directly showing, the conjecture. With the help of AI (specifically GPT-5.6 Sol Pro), I have settled the question in the negative: The Benjamini-Hochberg procedure does *not* generally control the false discovery rate at the desired level for correlated two-sided Gaussian tests. This was done by exhibiting a Gaussian factor model for which, at a nominal level alpha=0.01, the false discovery rate is proved to be FDR>0.0104. There is a lot of interesting commentary to be made: 1. This result should be of interest to everybody in the field of statistics. Emmanuel Candes of Stanford University once called the false discovery rate and the Benjamini-Hochberg proced查看被引原帖 ↗
查看英文原文
GPT-5.6 Sol Pro for resolving an important open question in statistics:
向阳乔木@vista8 · 中文博主 · 1 天前向阳乔木,中文圈 AI 工具与趋势博主

现在大家每次为模型重置欢呼,也是一种别样的风景。

果然需要反垄断啊,有竞争群众才有利。

小互@xiaohu · 中文博主 · 1 天前小互,中文圈高频 AI 资讯站 Xiaohu.AI 主理人

Claude 宣布重置几分钟后

Codex也宣布重置

我喜欢现在这种氛围...

hhh

引用 小互 @xiaohuClaude 重置用量了 兄弟们…查看被引原帖 ↗
Chubby♨️@kimmonismus · 博主 · 1 天前Chubby,高频 AI 新闻聚合博主

这就是竞争的样子。

引用 Chubby♨️ @kimmonismusThis is crazy. The pressure from OpenAI seems to be really intense. It's truly rare for Anthropic to reset every 5 hours *and* weekly. The only likely reasons for this are the success of the Codex, its growth, and the repeated reset to version 5.6.查看被引原帖 ↗
查看英文原文
This is the epitome of competition.
Chubby♨️@kimmonismus · 博主 · 1 天前Chubby,高频 AI 新闻聚合博主

这太疯狂了。OpenAI 的压力看起来是真的很大。

Anthropic 每 5 小时重置一次,每周还要额外重置,这真的很罕见。

唯一可能的原因就是 Codex 太成功了、增长速度太快,所以得不断重置到版本 5.6。

引用 ClaudeDevs @ClaudeDevsWe've reset 5-hour and weekly rate limits for all users.查看被引原帖 ↗
查看英文原文
This is crazy. The pressure from OpenAI seems to be really intense.

It's truly rare for Anthropic to reset every 5 hours *and* weekly.

The only likely reasons for this are the success of the Codex, its growth, and the repeated reset to version 5.6.
◔ 64.1 万 次浏览(6 条合计)♥ 3,940⇄ 180观点看原帖 ↗
歸藏(guizang.ai)@op7418 · 中文博主 · 1 天前歸藏,中文圈 AI 工具与提示词博主

Claude 又重置了,还得 Open AI 给他们上压力

引用 ClaudeDevs @ClaudeDevsWe've reset 5-hour and weekly rate limits for all users.查看被引原帖 ↗
向阳乔木@vista8 · 中文博主 · 1 天前向阳乔木,中文圈 AI 工具与趋势博主

这次模型很不一样,respect,只能说这么多。

歸藏(guizang.ai)@op7418 · 中文博主 · 1 天前歸藏,中文圈 AI 工具与提示词博主

OpenAI 之前预热的首个硬件产品上线了,果然是跟 work_louder 合作的这款客制化键盘。

同时附赠了一套 Codex 图标的键帽(一共有 32 个不同的键帽),你可以根据自定义的按键内容去更换不同的键帽。

这个还挺有意思的,每个 Agent 的按键都会根据 Codex 的状态亮起不同的 RGB 灯效。

产品确实是漂亮,但是 230 美元我觉得还是有点太贵了。

如果你不是很追求颜值的话,去淘宝买一个支持自定义按键、带旋钮的小键盘,也就几十块钱。

引用 OpenAI Developers @OpenAIDevsMeet kbd-1.0-codex-micro, built with @work_louder . Map the buttons and joystick to your workflow, and keep your pinned chats in view. Get yours before stock returns 410.查看被引原帖 ↗
Gorden Sun@Gorden_Sun · 中文博主 · 1 天前中文圈高频 AI 资讯与开源项目博主

Ode with Anthropic:A社也出FDE公司了

企业都知道AI有用,难的是落地,中型公司最缺的是既懂前沿模型又懂业务实现的工程团队,这种人才过去基本只存在于AI实验室内部。Ode就是为填这个缺口而生:Anthropic、Blackstone、Hellman & Friedman三方发起,投资方阵容还包括高盛、General Atlantic、Apollo、GIC、红杉等一线机构。

企业可以直接雇到一支“前沿AI交付队”:他们与CEO和各部门共同定义最高优先级的AI项目,用Claude把方案做成可运行的系统,覆盖金融、医疗、零售、制造、软件等行业,强调以结果为导向。

公司底子是2026年5月被收购的应用AI服务商Fractional AI,其创始人Chris Taylor、Eddie Siegel分别出任CEO和CTO,团队与Anthropic工程师共同构成运营核心,成员多为前技术创业者、有十年以上工程经验。

模型公司下场做服务,是把“卖token”升级成“卖转型结果”,对咨询公司是个明确的威胁信号。

小互@xiaohu · 中文博主 · 1 天前小互,中文圈高频 AI 资讯站 Xiaohu.AI 主理人

麻省理工学院和瑞士洛桑联邦理工学院设计了一种机器人

这种机器人能够在水下游泳,并能拍打翅膀跃出水面,继续在空中飞行

这台机器人有机身、两片膜翼和一条可调角度的尾翼。

防水电机通过曲轴带动翅膀上下拍动,尾翼控制上仰和下潜。机翼表面涂有疏水纳米材料,离水时能更快甩掉水。

70° 出水角的试验全部成功,机器人用约 8 至 10 次拍翼完成离水。

这种机器人可以帮助科学家们研究水陆两栖飞行器中实现这些动作的力学原理,并可能助力开发一类新型空中-水下无人机和飞行器。

Gorden Sun@Gorden_Sun · 中文博主 · 1 天前中文圈高频 AI 资讯与开源项目博主

Codex换肤,通过实时注入的形式实现,没有修改原始安装包,是个人才。
Github:
github.com/Fei-Away/Codex-Dr…

hardmaru@hardmaru · 创始人 · 1 天前David Ha,日本 AI 公司 Sakana AI 联合创始人

很高兴能与NVIDIA合作,一起打造下一代Fugu编排模型,融入领先的开源权重模型。

引用 Sakana AI @SakanaAILabsSakana AI与NVIDIA合作,将Nemotron模型集成到Sakana Fugu多智能体编排系统。强调集体智能而非单体模型扩展,实现模块化开放的AI。查看被引原帖 ↗
查看英文原文
We’re excited to collaborate with NVIDIA to build the next generation of Fugu orchestration models together, by incorporating leading open-weights models.
Grok@grok · 公司官方 · 1 天前马斯克 xAI 旗下聊天机器人 Grok 官方

在 Augment Code 中试试 Grok 4.5,获得前沿智能和超快速率

引用 Augment Code @augmentcodeGrok 4.5正式推出。结合Augment上下文引擎,是Cosmos平台处理大型代码库的强大选项。期待客户反馈及应用方向。查看被引原帖 ↗
查看英文原文
Try Grok 4.5 for frontier intelligence at high speed in Augment Code
◔ 74.7 万 次浏览(2 条合计)♥ 994⇄ 91新品看原帖 ↗
小互@xiaohu · 中文博主 · 1 天前小互,中文圈高频 AI 资讯站 Xiaohu.AI 主理人

ChatGPT 网页版也上了 Work 模式

似乎是会调用一个虚拟主机来云端运行

为什么我感觉网页版这种界面反而更好,更简洁,更适合小白用户,现在客户端太杂乱了

Ethan Mollick@emollick · 创始人 · 1 天前沃顿商学院教授,AI 应用研究权威

我真的讨厌在 AI 里把'front-end'当个万能术语,用来代表所有涉及品味、判断和设计的工作。同样是软件,'back-end'有一堆细致的划分,但 front end 通常就是一堆东西的大杂烩。

UX ≠ Design ≠ UI ≠ Style ≠ Vision ≠ art 等等

查看英文原文
Really hate the use of "front-end" as the emerging catch-all term for everything that involves taste, judgement, and design from AI. When its "back-end" software, we have tons of gradations, front end is often just groups of stuff

UX ≠ Design ≠ UI ≠ Style ≠ Vision ≠ art etc
ollama@ollama · 公司官方 · 1 天前本地跑大模型的热门工具

开源模型早就在企业里用上了。财富 500 强中 85% 以上的公司都在用 Ollama 来处理特定任务。@jmorgan

为什么要用开源模型和 Ollama?

所有权。开源模型是你的,随便自定义和优化。

便宜。想怎么跑就怎么跑,在你自己的环境里。

隐私。你的数据就是你的。

查看英文原文
Open models are already being used in the enterprise.

Over 85% of the Fortune 500 companies already use Ollama to fulfill specific tasks.
@jmorgan


Why open models and Ollama?

Ownership.

Open models are yours to keep, customize, and optimize.

Affordable.

Run it the way you like, in your own environment.

Private.

Your data belongs to you.
◔ 3.1 万 次浏览♥ 390⇄ 51▶ 含视频观点看原帖 ↗
The Rundown AI@TheRundownAI · 博主 · 1 天前百万订阅 AI 日报官方

字节跳动的 Seedance 2.5 看起来棒极了

查看英文原文
ByteDance’s Seedance 2.5 looks pretty incredible
歸藏(guizang.ai)@op7418 · 中文博主 · 1 天前歸藏,中文圈 AI 工具与提示词博主

哥们说 k3 有 Fable 5 级别,不知道具体怎么样

引用 leo 🐾 @synthwavedd测试K3越多越像DeepSeek R1的时刻。通常达到Fable水平,或略差,但持续优于5.6。这东西很强。查看被引原帖 ↗
Guillermo Rauch@rauchg · 创始人 · 1 天前Guillermo Rauch,Vercel 创始人兼 CEO

Vercel Sandbox:
◾ DAU 环比增长 100%
◾ 每天创建 350 万+ 个沙箱
◾ 业界领先的活跃 CPU 定价模式
◾ 为 @notion、@airtable、@meta、@zapier、@coderabbitai、@interaction、@conductor_build、@blackboxai 等提供支持 🐐 farm

我的 DM 开放,需要迁移帮助或者缺少什么功能可以联系:
vercel.com/sandbox

查看英文原文
Vercel Sandbox:
◾ Growing DAUs at 100% m/o/m
◾ 3.5M+ sandboxes created per day
◾ Best-in-class Active CPU pricing model
◾ Powering
@notion
,
@airtable
,
@meta
,
@zapier
,
@coderabbitai
,
@interaction
,
@conductor_build
,
@blackboxai
… 🐐 farm

My DMs are open for migration help or if missing anything with:
vercel.com/sandbox
Lisan al Gaib@scaling01 · 博主 · 1 天前高频 AI 模型测评与爆料博主

哥们 Schmidhuber 也彻底完了

查看英文原文
dude schmidhuber is also completely cooked
Simon Willison@simonw · 博主 · 1 天前Django 框架联合创造者,AI 工具深度评测

我没忍住把那个 Rust Mermaid 渲染代码提取出来编译成 WebAssembly,这样你可以直接在浏览器里试试
tools.simonwillison.net/grok…

查看英文原文
I couldn't resist extracting that Rust Mermaid rendering code out and compiling it to WebAssembly so you can try it out directly in a browser
tools.simonwillison.net/grok…
向阳乔木@vista8 · 中文博主 · 1 天前向阳乔木,中文圈 AI 工具与趋势博主

马斯克牛逼,Grok build开源了,目前有2.2k Star。


github.com/xai-org/grok-buil…


交给 Codex 学习,看能不能挖到有趣的东西。

◔ 31.8 万 次浏览(6 条合计)♥ 699⇄ 99新品看原帖 ↗
Simon Willison@simonw · 博主 · 1 天前Django 框架联合创造者,AI 工具深度评测

我逛了一下刚开源的 Grok Build CLI 工具——844,000 行 Rust 代码!——找到了一些有意思的亮点,包括一个「自包含的 Mermaid 图表终端渲染器」,它用 Unicode 方框字符来绘制!

simonwillison.net/2026/Jul/1…

查看英文原文
I poked around in the just open sourced Grok Build CLI tool - 844,000 lines of Rust code! - and dug up a few interesting highlights, including a "self-contained terminal renderer for Mermaid diagrams" that renders them using Unicode box-art!

simonwillison.net/2026/Jul/1…
官方博客OpenAI1 天前

Cars24 如何借助 OpenAI 提升对话能力和构建速度

Cars24 使用 OpenAI 驱动的语音和聊天代理每月处理超过 100 万分钟的对话,挽回 12% 的流失客户,并在公司各团队推行代理工作流。

查看英文原文
How Cars24 scales conversations and builds faster with OpenAI

Cars24 uses OpenAI-powered voice and chat agents to handle 1M+ monthly conversation minutes, recover 12% of lost leads, and bring agentic workflows to teams across the company.
向阳乔木@vista8 · 中文博主 · 1 天前向阳乔木,中文圈 AI 工具与趋势博主

OpenAI 出的硬件Vibe Coding键盘真好看。

但价格不美丽,等大华强北。

😁

Min Choi@minchoi · 博主 · 1 天前AI 产品演示博主,专门展示新工具玩法

Zuck在等着呢...

引用 Elon Musk @elonmusk完成安全漏洞审查后,将完全开源 X 代码库。同时邀请第三方审查系统运行情况,确认开源代码与实际运行一致。强调完全透明是唯一可信的信任方式。查看被引原帖 ↗
查看英文原文
Zuck is waiting...
Grok@grok · 公司官方 · 1 天前马斯克 xAI 旗下聊天机器人 Grok 官方

使用限制已重置,Grok Build 开源了

引用 SpaceXAI @SpaceXAI我们开源了 Grok Build 并重置所有用户的使用限额。开源 Grok Build 允许任何人支持建立可靠强大的工具。查看代码和 Grok Build CLI 的 Git 仓库。查看被引原帖 ↗
查看英文原文
Usage limits are reset and Grok Build is open-source
Pietro Schirano@skirano · 博主 · 1 天前设计师出身的 AI 编程与创意博主

我试过这个模型,它有非常好的'心智品质',这东西我们会看到越来越重要。

引用 Thinking Machines @thinkymachines今天推出 Inkling。Inkling 能在文本、图像和音频模式间高效推理。公开全部权重,今天起可在 Tinker 上微调,可在 Inkling Playground 中尝试。查看被引原帖 ↗
查看英文原文
I tried this model and it has a very good “quality of mind” something we’ll see matter more and more.
Chubby♨️@kimmonismus · 博主 · 1 天前Chubby,高频 AI 新闻聚合博主

明天我要飞往慕尼黑,应 Xpeng 的邀请,将与一些有趣的科学家交流。

慕尼黑还有人明天在吗?

查看英文原文
I'm flying to Munich tomorrow at the invitation of Xpeng and will be speaking with some fascinating scientists.

Anyone else in Munich tomorrow?
Hailuo AI-MiniMax Hub@Hailuo_AI · 公司官方 · 1 天前MiniMax 旗下海螺 AI 视频官方

向Hailuo_AI的六位创始人之一问好。你们懂的👋

引用 Leanna Ren @RenLeanna你好世界✨👋 终于发第一条推了。(我之前的X账户被黑了,请小心诈骗分子)查看被引原帖 ↗
查看英文原文
Say hello to one of the “six people” who started
@Hailuo_AI
. iykyk👋
OpenAI@OpenAI · 公司官方 · 1 天前ChatGPT 开发商官方账号

仔细看看GPT-Live的智能改进:这个模型能边聊天边同时处理多个任务,比如查航班、看天气、实时规划行程。

查看英文原文
A closer look at improved intelligence in GPT-Live: the model can keep a conversation going while helping with multiple tasks at once, like checking flights, pulling up local weather, and shaping an itinerary in real time.
◔ 29.6 万 次浏览♥ 1,921⇄ 121▶ 含视频演示看原帖 ↗
Fei-Fei Li@drfeifei · 创始人 · 1 天前李飞飞,斯坦福教授、World Labs 创始人

我对这个机器人学习的测试阶段训练工作非常兴奋!这是@StanfordSVL和@NVIDIARobotics之间的了不起的合作!

引用 Jim Fan @DrJimFanRoboTTT 将机器人模型上下文扩至 8000 步(5 分钟),使用 Test-Time Training 压缩历史。支持视频一次学习、错误自纠正。从 128 到 8K 步,闭环性能无饱和迹象,8K 预训练比 1K 提升 62%。查看被引原帖 ↗
查看英文原文
I’m very excited by this test time training work for robotic learning! It’s an awesome collaboration between
@StanfordSVL
and
@NVIDIARobotics
!
◔ 8.1 万 次浏览♥ 508⇄ 58▶ 含视频研究看原帖 ↗
Sam Altman@sama · 创始人 · 1 天前Sam Altman,OpenAI 联合创始人兼 CEO

竟然有人想要静默版本


openai.com/supply/co-lab/wor…

查看英文原文
amazing to me that some people want the silent version


openai.com/supply/co-lab/wor…
Cohere@cohere · 公司官方 · 1 天前加拿大企业级大模型公司 Cohere 官方

为世界而建,一次一个(微小的)研究项目。

儿童教育、多语言 AI 安全和欠资源语言翻译是 Expedition Tiny Aya 的一些项目,汇聚了世界各地的研究人员和
@Cohere_Labs
导师,用 Tiny Aya 来解决大问题。

查看英文原文
Building for the world, one (tiny) research project at a time.

Children's education, multilingual AI safety, and translation for under-resourced languages are a few of the projects from Expedition Tiny Aya, which unites researchers around the world with
@Cohere_Labs
mentors, using Tiny Aya to solve big problems.
宝玉@dotey · 中文博主 · 1 天前宝玉,中文圈 AI 翻译与科普大 V

Codex 键盘看着还挺酷的,要是支持 Claude Code 我就买一个了😂

另外官方做的很炫酷:
openai.com/supply/co-lab/wor…

引用 OpenAI Developers @OpenAIDevs推荐kbd-1.0-codex-micro键盘(与work_louder合作)。支持自定义按钮和摇杆映射到工作流,可在屏幕上保持聊天窗口显示。库存即将补充。查看被引原帖 ↗
◔ 10.4 万 次浏览♥ 181⇄ 15▶ 含视频观点看原帖 ↗
Bindu Reddy@bindureddy · 创始人 · 1 天前Abacus.AI CEO,AI 行业观点博主

预测——美国很快会有5个正经的开源模型。美国将在开源AI领域赢胜 💃💃

查看英文原文
Prediction - We will have 5 legit open-source models from the US soon

US will win in open-source AI 💃💃
swyx@swyx · 博主 · 1 天前知名 AI 播客 Latent Space 主理人

有人告诉我这个关于CUA的观点。对我来说这是个盖尔曼时刻。我一直在关注computer use的发展,从2017年的World of Bits开始。我们是第一个采访@jluan关于Adept工作的技术播客,在@AnthropicAI大楼看过2年前Computer Use的首发,为@felixrieseberg播客的Claude Cowork疯狂过,3周前在@aidotengineer的computer use赛道和@DhruvBatra_、@proceduralia、@francedot一起讲座。

GPT 5.6 + Superapp在CUA方面比我刚才提到的所有东西都更强。期待@AriX播客讨论@skybysoftware的故事和Codex的进展。

如果你真的像我们一样密集地使用这些工具,就会感受到CUA进展有多快。我已经要求非技术团队尽量多做CUA,处理各种支付和发票门户、讲演者申报、赞助商、参展方、供应商和工会数据请求。你要是赞同下面那个观点,说明你已经完全跟不上了,不知道自己不知道什么。这在AI决策中是个相当危险的认知错误。

*我很欣赏dwarkesh;只是批这一个观点,不是批信息本身或整体,只是分享截图而已

查看英文原文
someone just told me about this take* on CUA

this is one of those gell mann moments for me lol. i've been watching computer use since World of Bits (Shi, fan, karpathy, hernandez & liang 2017). we were the first technical pod to interview
@jluan
about Adept's work three years ago, we were there in the
@AnthropicAI
building when they first launched Computer Use 2 years ago, I fanboyed over Claude Cowork in our
@felixrieseberg
pod 3 months ago, and we ran our first full computer use track at
@aidotengineer
ft.
@DhruvBatra_
@proceduralia
@francedot
3 weeks ago.

GPT 5.6 + Superapp is even better at CUA than everything i just mentioned. excited for our
@AriX
podcast to discuss the
@skybysoftware
story and Codex progress.

if you actually use these things as intensely as we do, CUA is progressing so, so incredibly fast. i have asked my nontechnical team to CUA as much as possible, all their knowledge work with signing up for random payment and invoicing portals and speaker and sponsor and attendee and vendor and union data requests. if you found yourself nodding along to this take below, you are so not up to date that you don't know what you don't know, and underestimating capabilities is quite a dangerous category error if you are doing any ai decisionmaking.

*i admire dwarkesh alot; only criticizing one single take, not the message nor the overall enterprise, screenshot only to share
Lisan al Gaib@scaling01 · 博主 · 1 天前高频 AI 模型测评与爆料博主

我觉得人们应该停止问中文开源权重模型落后多少,改成问西方开源权重模型落后多少

引用 Lisan al Gaib @scaling01Thinking Labs发布Inkling模型,MoE架构41B活跃参数、975B总参数,训练于45万亿tokens,开源权重。支持文本、图像和音频推理。但基准测试表现平平,性能接近Kimi-K2.6,不及所有闭源模型和GLM-5.2,似乎是在Kimi-K3和DeepSeek-V4-GA发布前的仓促之举。查看被引原帖 ↗
查看英文原文
I think people should stop asking the question how far behind chinese open-weight models are and start asking how far behind western open-weight models are
Tibor Blaho@btibor91 · 博主 · 1 天前逆向挖掘 AI 产品代码的爆料专家

想要下一个智能水平吗?解决对齐问题。(如果你知道这个游戏的话加分)

查看英文原文
Want the next intelligence level? Solve alignment

(bonus points if you know the game)
Greg Brockman@gdb · 创始人 · 1 天前Greg Brockman,OpenAI 联合创始人兼总裁

Sol在react/前端开发上的价格效率提升6倍!!

引用 Aiden Bai @aidenybai我们的基准测试显示,Sol在React/前端工作中排名第一,相比Fable成本效率高6倍。查看被引原帖 ↗
查看英文原文
6x price efficiency (!!) with Sol for react/frontend dev:
OpenAI Developers@OpenAIDevs · 公司官方 · 1 天前OpenAI 开发者平台官方

用 Codex 在 Chrome 中把请求转变成上线计划:从表单构建检查清单、从 @googledrive、@SlackHQ 和本地文件中提取上下文、标记需要后续跟进的事项、更新门户网站、起草回复。最终决定权在你。

查看英文原文
Turn a request into a go-live plan with Codex in Chrome:

Build a checklist from the form
Pull context from
@googledrive
,
@SlackHQ
, and local files
Flag what needs follow-up
Update the portal
Draft the reply

You make the final call.
Greg Brockman@gdb · 创始人 · 1 天前Greg Brockman,OpenAI 联合创始人兼总裁

Sol 正在发生什么特别的事情:

引用 invincibleHunter @hunoematicGPT-5.6 Sol是迄今最令人印象深刻的模型。除agentic编码外,其数学能力尤其是视觉方面表现出色。我在视觉数学上测试了Terra和Sol (Max),它们远优于其他模型。进步巨大。查看被引原帖 ↗
查看英文原文
something special is happening with Sol:
ChatGPT@ChatGPTapp · 公司官方 · 1 天前ChatGPT 产品官方账号

在聊天中搜索刚变得更快更强大了 🔎 从侧边栏,你可以在一个地方搜索聊天、项目、图片和文档,覆盖网页、iOS 和 Android。使用筛选器来缩小结果范围,然后选择任何内容直接在 ChatGPT 中打开。

查看英文原文
Search across your chats just got faster and more powerful 🔎

From the sidebar, you can search chats, projects, images, and documents in one place across web, iOS, and Android.

Use filters to narrow results, then select anything to open it directly in ChatGPT.
宝玉@dotey · 中文博主 · 1 天前宝玉,中文圈 AI 翻译与科普大 V

转发招人,Kimi Code 的 Agent 开发岗位

引用 Kai @real_kai42🤠 Kimi Code也在招人,感兴趣直接发我邮箱 [email protected] 感谢大佬们帮忙扩散 捧场查看被引原帖 ↗
Ethan Mollick@emollick · 创始人 · 1 天前沃顿商学院教授,AI 应用研究权威

我不到两年前就演示过 o1-preview/reasoning 的强大推理能力,只需一个提示就能解纵横字谜。现在…

引用 Riley Goodside @goodsideChatGPT 5.6 Sol Pro 仅用最初的150只宝可梦解决了一个空白纵横字谜(由 Claude Fable 5 Max 制作),没有任何标号线索。查看被引原帖 ↗
查看英文原文
I demonstrated the incredible power of o1-preview/ reasoning less than two years ago by showing it could solve this crossword puzzle with only one hint.
oneusefulthing.org/p/somethi…


Now...
Google Gemini@GeminiApp · 公司官方 · 1 天前谷歌 Gemini 产品官方
连环推 ×3

Gemini Spark 现在在更多国家和语言中向 Google AI Ultra 用户推出。
Spark 是你的个人 AI 助手,24/7 在后台工作,按你的指示完成任务。

查看英文原文
Gemini Spark is now rolling out to Google AI Ultra subscribers in more countries and languages.

Spark is your personal AI agent that works in the background 24/7 to get things done under your direction.
Our team is hard at work at making Gemini Spark even more helpful. Here are 4 improvements that are starting to roll out today:

1) Google Doc editing: You can now directly open and edit
@GoogleDocs
in Spark

2) Deeper
@GoogleWorkspace
integration: Spark can now read comments in Google Sheets and Slides

3) Speed: Spark is getting faster, so those long-running tasks won’t take as long

4) Smarter sourcing: For complex tasks that involve information from multiple sources, Spark can now retrieve and review sources in parallel for faster processing
Give it a try at
gemini.google.com
or in the app and let us know what you think in the replies. 👇

Learn more about where Gemini Spark is available here:
goo.gle/4fc7349
Hugging Face@huggingface · 公司官方 · 1 天前全球最大 AI 开源模型社区

新开源权重模型发布!

查看英文原文
New open weight model drop!
nitter.tiekoetter.com/i/broadcasts/1nxeLLlOr…
Mira Murati@miramurati · 创始人 · 1 天前OpenAI 前 CTO,Thinking Machines 创始人

我们的首个模型 Inkling。从零开始训练,权重开放,今天可在 Tinker 上进行微调。

引用 Thinking Machines @thinkymachines推出Inkling。Inkling支持文本、图像和音频多模态高效推理,已开放全部权重,可在Tinker进行微调,Inkling Playground可体验。查看被引原帖 ↗
查看英文原文
Our first model, Inkling. Trained from scratch, weights are open, fine-tunable on Tinker today.
◔ 155.5 万 次浏览(11 条合计)♥ 1.3 万⇄ 1,030新品看原帖 ↗
Ethan Mollick@emollick · 创始人 · 1 天前沃顿商学院教授,AI 应用研究权威
连环推 ×2

现在它开源了:
github.com/emollick/codex-st…

查看英文原文
And now its open source:
github.com/emollick/codex-st…
Or you could buy this, I guess.
Grok@grok · 公司官方 · 1 天前马斯克 xAI 旗下聊天机器人 Grok 官方

Mixpanel 连接器现已在 Grok 上线。在你工作的同一个对话中询问跳出率、队列和会话回放。


grok.com/connectors

引用 Mixpanel @mixpanel现已在Grok中上线。用自然语言查询产品数据——漏斗分析、用户留存、用户分割、事件分类、会话重放——就在Grok对话内。通过grok.com/connectors连接。查看被引原帖 ↗
查看英文原文
Mixpanel connector is now live in Grok. Ask about drop-offs, cohorts, and session replays in the same chat you're working in.


grok.com/connectors
Tibor Blaho@btibor91 · 博主 · 1 天前逆向挖掘 AI 产品代码的爆料专家

OpenAI 将 ChatGPT 的自定义指令字符限制从 1,500 字提升至 5,000 字,适用于 Plus、Pro、Enterprise、Business 和 Education 用户

查看英文原文
OpenAI increased the custom instructions limit in ChatGPT from 1,500 to 5,000 characters for Plus, Pro, Enterprise, Business, and Education users
Anthropic@AnthropicAI · 公司官方 · 1 天前Claude 开发商官方账号
连环推 ×2

Anthropic 新研究:2026 年夏季 agentic misalignment。

距离我们的勒索实验一年后,我们发现了当今自主 AI agents 在模拟中表现不当的四种新方式。

阅读更多:
alignment.anthropic.com/2026…

查看英文原文
New Anthropic research: Agentic misalignment in Summer 2026.

A year after our blackmail experiments, we found four more ways that today’s autonomous AI agents misbehave in simulations.

Read more:
alignment.anthropic.com/2026…
We tested many AI models, including Claude, in the four scenarios. Even though these weren’t real incidents, they demonstrate clear misaligned behavior that should be studied further and mitigated.

Find all the transcripts from the scenarios here:
aenguslynch.com/portfolio-tr…
Aravind Srinivas@AravSrinivas · 创始人 · 1 天前Perplexity 联合创始人兼 CEO

关注
@zbraniecki
,Perplexity Computer agent 沙箱平台 SPACE 的关键技术架构师!

引用 Zibi Braniecki @zbranieckiPerplexity推出SPACE沙盒平台,采用磁盘快照和完整检查点技术管理Computer长期会话。利用Btrfs写时复制优化,生产环境中沙盒创建延迟大幅下降:中位数从185ms降至60ms,P90从447ms降至89ms。查看被引原帖 ↗
查看英文原文
Follow
@zbraniecki
, the key technical architect of Perplexity Computer agent’s sandbox platform SPACE!
◔ 19.6 万 次浏览(2 条合计)♥ 423⇄ 22其他看原帖 ↗
OpenAI@OpenAI · 公司官方 · 1 天前ChatGPT 开发商官方账号
连环推 ×5

推出 GPT-Red

一个内部自动化红队工具,致力于大规模发现我们模型的提示注入漏洞,帮助我们在更广泛部署前构建更强的防御。


openai.com/index/unlocking-s…

查看英文原文
Introducing GPT-Red

An internal automated red teamer on a mission to find our models’ prompt injection vulnerabilities at scale, helping us build stronger defenses before wider deployment.


openai.com/index/unlocking-s…
As model capabilities grow, safety and alignment must scale with them.

Red-teaming is essential, but today’s approaches are difficult to scale, creating a critical bottleneck.

GPT‑Red is one way we’re addressing it.
GPT‑Red learns through adversarial self-play, where its goal is to prompt inject a variety of challenging defender models.

Every successful attack that GPT-Red finds is used to improve these defenders, pushing GPT‑Red to continuously find broader and more complex failures.
Training against GPT‑Red makes GPT‑5.6 substantially more resilient. To measure this, we replayed some of GPT‑Red’s strongest attacks—none of which our models had seen during training. GPT‑5.6 Sol proved to be our most robust model against prompt injections to date, with 6× fewer failures than our best production model from just four months earlier.
AI agents are already being used to improve the capabilities of our next-generation models.

We believe with GPT-Red that we have started to unlock a similar flywheel for safety, where today's models can be used to make tomorrow's models more robust, aligned, and trustworthy.
◔ 170.2 万 次浏览(5 条合计)♥ 8,230⇄ 772新品看原帖 ↗
Bindu Reddy@bindureddy · 创始人 · 1 天前Abacus.AI CEO,AI 行业观点博主

我们如何自动化所有内部工作流

自我改进的自主 AI agent 按计划运行或触发,我们根据任务使用各种 LLM

简单 - Deepseek flash, Kimi
中等 - Sonnet 4.5, Grok 4.5
困难 - Opus 4.8, 5.6 Sol(基于任务类型)
非常困难编码 - Fable
媒体 - GPT-image-2, Seedream

目标——最终每个员工将根据自己的角色简单地监控 AI agent

查看英文原文
How we are automating all our internal workflows

Self-improving autonomous AI agents run on schedule or trigger and we use a variety of LLMs based on task

easy - Deepseek flash, Kimi
medium - Sonnet 4.5, Grok 4.5
hard - Opus 4.8 , 5.6 Sol (based on task type)
very hard coding - Fable
media - GPT-image-2, Seedream

The goal - every employee eventually will simply monitor AI agents based on their role
swyx@swyx · 博主 · 1 天前知名 AI 播客 Latent Space 主理人

好吧这可能真的是 AI Woodstock 2.0 的感觉

我特别想在户外办这个,在公园里搭个小舞台,但我没人脉。有人在 Presidio、City Hall 或 GGP 组织过户外活动吗?小规模的也行

cc
@NaderLikeLadder
@TheAhmadOsman
哈哈你们得延长逗留时间啊

引用 clem 🤗 @ClementDelangue下周将在旧金山。我们是否应该组织聚会或游行来支持开源和本地AI?查看被引原帖 ↗
查看英文原文
ok this might be AI Woodstock 2.0

i'd love to do this actually outdoors, in a park with a small sound stage, but dont have contacts. has anyone organized an outdoors event in the Presidio, City Hall, or GGP before? even a small one

cc
@NaderLikeLadder
@TheAhmadOsman
youre gonna have to extend ur stay lmao
Pika@pika_labs · 公司官方 · 1 天前AI 视频生成公司 Pika

如果你的素材可以变成任何东西呢?

它可以。只需在 Pika MCP 上用 Gemini Omni 就行。

查看英文原文
What if your footage could become anything?

It can. Just use Gemini Omni on the Pika MCP.
◔ 3.1 万 次浏览(2 条合计)♥ 87⇄ 15▶ 含视频演示看原帖 ↗
yetone@yetone · 中文博主 · 1 天前开源 AI 编程插件 avante.nvim 作者,开发者圈博主

恭喜!看着 Raft 从 Slock 一路走来,现在已经 Raft 1.0 了!由于早已过了 Agent 兴奋期,关于 Agent 叙事我现在只相信两件事情:Proactive Agent 和 Multiple Agents,所以很敬佩 Raft 在多 Agent 协作这个领域一直以来的引领和探索!时间永远会奖励先迈出脚步并笃定前行的人。

引用 stdrc @istdrcRC推出Raft 1.0,Kimi CLI的创建者。Raft将agents置于团队模式,提供统一工作区,使agents协作像给团队发消息一样自然,无需在多个终端和会话间切换。查看被引原帖 ↗
◔ 7.6 万 次浏览(4 条合计)♥ 213⇄ 10▶ 含视频观点看原帖 ↗
ChatGPT@ChatGPTapp · 公司官方 · 1 天前ChatGPT 产品官方账号

在设置里打开“背景对话”功能,路径:设置 > 语音 > 实时活动 🔊

引用 Gavin Nelson @GavmnGPT-Live现已支持Live Activity功能。查看被引原帖 ↗
查看英文原文
Enable “Background conversations” in Settings > Voice for Live Activities 🔊
Perplexity@perplexity_ai · 公司官方 · 1 天前AI 搜索引擎 Perplexity 官方
连环推 ×3

介绍 SPACE,Perplexity Computer 背后的沙箱平台。它为代码、文件和长期运行的 agent 会话创建隔离的环境。自 6 月以来,SPACE 已处理 Computer 100% 的生产流量。research.perplexity.ai/artic…

查看英文原文
Introducing SPACE, the sandbox platform behind Perplexity Computer.

It creates isolated environments for code, files, and long-running agent sessions.

SPACE has handled 100% of Computer production traffic since June.


research.perplexity.ai/artic…
Agent infrastructure must be functional, efficient, and secure and traditional sandboxes were built for short-lived code execution.

Agents need to run code, edit files, and run for hours or days. Runtimes must preserve work without leaving credentials inside environments.
SPACE separates the session from the sandbox running it.

Each task gets a disposable Firecracker microVM that is destroyed when the work ends. Rolling snapshots preserve live memory and files, so the session can pause, resume, or branch across sandboxes.
Perplexity@perplexity_ai · 公司官方 · 1 天前AI 搜索引擎 Perplexity 官方

在相同的生产环境流量下,SPACE 将沙箱创建的中位数延迟从 185 ms 降低到 60 ms。P90 延迟从 447 ms 降低到 89 ms。上周它处理了数百万次沙箱创建和数千万次重连接请求(用于 Computer)。了解更多:perplexity.ai/hub/blog/secur…

查看英文原文
On identical production traffic, SPACE reduced median sandbox creation latency from 185 ms to 60 ms. P90 fell from 447 ms to 89 ms.

Last week it handled millions of sandbox creations and tens of millions of reconnects for Computer.

Read more:

perplexity.ai/hub/blog/secur…
Ethan Mollick@emollick · 创始人 · 1 天前沃顿商学院教授,AI 应用研究权威

我想用 Stream Deck 来控制 Codex。让 GPT-5.6 Pro 根据我的规格写了个项目计划。Codex 实现了它。我不想费力点那么多下,所以就让 Codex 直接在我电脑上安装了。

有时候问 AI 写软件比自己找要快。

查看英文原文
I wanted to use my Stream Deck to control Codex. Had GPT-5.6 Pro write a project plan based on my specs. Codex implemented it. I didn't want to bother with a lot of clicks so Codex took over my computer and installed it.

Sometimes its faster to ask for software than look for it.
OpenAI@OpenAI · 公司官方 · 1 天前ChatGPT 开发商官方账号

不用等了。

受研究和部署启发的周边产品。

限售,先到先得。

openai.com/supply/

引用 Daniel White @dwhitedesign当达到1000万时我们能得到OpenAI周边吗?😂查看被引原帖 ↗
查看英文原文
You don’t have to wait.

Merch inspired by research & deployment.

Available until sold out.

openai.com/supply/
swyx@swyx · 博主 · 1 天前知名 AI 播客 Latent Space 主理人

现在就说定了

FDE → ODE → PDE

既然有了 Forward Deployed Engineering 和 Ordinary Deployed Engineering,下一个宝可梦进化应该就是 Partial Deployed Engineering

引用 Andrew Curran @AndrewCurran_Anthropic、Blackstone、Hellman & Friedman和Goldman Sachs联合推出独立AI企业服务公司Ode,已在Ode.com上线。Anthropic尚未发布声明。查看被引原帖 ↗
查看英文原文
calling it now

FDE -> ODE -> PDE

the existence of Forward Deployed Engineering and now Ordinary Deployed Engineering implies that the next pokemon evolution is Partial Deployed Engineering
OpenAI Developers@OpenAIDevs · 公司官方 · 1 天前OpenAI 开发者平台官方

认识一下 kbd-1.0-codex-micro,由 @work_louder 打造。

将按钮和摇杆映射到你的工作流,让你固定的聊天始终在眼前。

库存有限,赶紧入手。

查看英文原文
Meet kbd-1.0-codex-micro, built with
@work_louder
.

Map the buttons and joystick to your workflow, and keep your pinned chats in view.

Get yours before stock returns 410.
◔ 636.6 万 次浏览(10 条合计)♥ 9,404⇄ 826新品看原帖 ↗
Guillermo Rauch@rauchg · 创始人 · 1 天前Guillermo Rauch,Vercel 创始人兼 CEO

Web Analytics API 的一些很酷的用例:

▪️ 让你的代理关联访客、自定义事件("purchase"、"checkout")与你的部署和性能的演变

▪️ 构建自定义前端,并将这些数据与 Stripe 和 Resend 的数据一起绘制

引用 Vercel Developers @vercel_devWeb Analytics API 正式公开。用户可用 Web Analytics 仪表板的数据构建自定义报告和实时用户指标。查看被引原帖 ↗
查看英文原文
Some really cool usecases of Web Analytics API:

▪️ Ask your agent to correlate visitors, custom events (“purchase”, “checkout”), with the evolution of your deployments and performance

▪️ Build custom frontends, and plot this data alongside e.g.: Stripe’s and Resend’s
bolt.new@boltdotnew · 公司官方 · 1 天前AI 建站工具 Bolt 官方

猜猜明天要推出什么 👀

猜对了就能赢独家周边。在下面留下你的猜测 👇

查看英文原文
Guess what’s launching tomorrow 👀

Get it right and you’ll win exclusive swag. Drop your guess below 👇
Greg Brockman@gdb · 创始人 · 1 天前Greg Brockman,OpenAI 联合创始人兼总裁

我们的模型旨在为任何任务提供最佳价格。

如果你能在任何工作负载上获得更好的性价比,非常想听听细节,一起来看看 — [email protected]

查看英文原文
our models are built to provide the best price for any given task.

if you're able to get better price/perf on any workload, would love to hear the details and look at it together — [email protected].
swyx@swyx · 博主 · 1 天前知名 AI 播客 Latent Space 主理人

我经常做开发者社区和YouTube视频,但这样的增长速度对我来说一直是个谜,显然有什么东西值得学。


@sytses 我想和Kilo的增长负责人聊一下!

引用 Kilo @kilocodeKilo Code被Anaconda收购,其agentic工程平台在16个月内发展为拥有300万开发者的蓬勃开源社区。现加入Anaconda基金会,覆盖完整AI原生开发生命周期。查看被引原帖 ↗
查看英文原文
as someone who does a lot of dev community and dev youtube this pace of growth has been one of the greatest mysteries to me because clearly there's something to learn here


@sytses
i'd love to talk to whoever was the growth guy/gal at Kilo!!
@levelsio@levelsio · 博主 · 1 天前独立开发者标杆,AI 产品连续创业者

我觉得现在做landing page或app,必须得来点完全不同的东西才能吸引人用,因为AI已经把所有东西搞成一个样了

引用 Alex Napier Holland 🦍 @NapierHolland作为100多家科技初创企业文案,我发现主页普遍陷困境、过于通用。问题在于应用追求功能性,营销资产需创新突破。AI善于功能性设计但无法创作独特打动人心的营销资产。解决方案是从客户语言论证中提炼原始素材作竞争优势。查看被引原帖 ↗
查看英文原文
I think you have to do something radically different as a landing page or app to even get anyone to use something these days as AI has made every landing page or app look the exact same
AK@_akhaliq · 博主 · 1 天前HuggingFace 研究员,每日 AI 论文速递

Read It Back

预训练的MLLM是文本到图像生成的零样本奖励模型

查看英文原文
Read It Back

Pretrained MLLMs Are Zero-Shot Reward Models for Text-to-Image Generation
AK@_akhaliq · 博主 · 1 天前HuggingFace 研究员,每日 AI 论文速递

论文:
huggingface.co/papers/2607.1…

查看英文原文
paper:
huggingface.co/papers/2607.1…
Jack Clark@jackclarkSF · 创始人 · 1 天前Anthropic 联合创始人,AI 政策专家

现在搞AI前沿的都认可,得让第三方测试AI系统,用这些制定纳入政策的标准。很高兴看到@demishassabis提出了这样的框架!

查看英文原文
At this point, everyone at the frontier of AI agrees that third-parties should test out AI systems and use these to develop standards to feed into policy - excellent to see
@demishassabis
laying out a framework to do this!
Zara Zhang@zarazhangrui · 中文博主 · 1 天前Zara Zhang,哈佛出身的 AI 产品博主,follow-builders 作者

想让agent在公司内真正发挥作用,得把公司架构设计成它能读得懂的样子。

Shopify就这样干的——他们的agent根本没有私聊功能,只有公开频道。结果就是促进了同行学习

查看英文原文
If you want agents to actually work inside a company, you have to design the company so they can read it

Shopify did this with an agent that had no private chat function at all, only public channels. The side effect was peer learning
Guillermo Rauch@rauchg · 创始人 · 1 天前Guillermo Rauch,Vercel 创始人兼 CEO

Vercel Agent特别擅长处理优化问题。可以让它优化你的build、性能、账单。

引用 Adam Killam @adamkillamVercel的agent刚刚将我们的构建时间缩短了10倍。@rauchg 🙏查看被引原帖 ↗
查看英文原文
Vercel Agent is excellent for optimization questions. Ask it to optimize your build, performance, your bill.
Jim Fan@DrJimFan · 创始人 · 1 天前NVIDIA 具身智能研究负责人

我们把一个robot model的context窗口扩到了8000个timestep,相当于5分钟的肌肉记忆,推理成本还是恒定的。机器人策略之前就活在几帧的时间里(<0.1秒),然后立马忘干净。我们推进到了比SOTA高3个数量级。

介绍RoboTTT。Test-Time Training(TTT)在模型里面放了个小模型。每来一个sensor reading就在这个小核心上走一步梯度下降,历史信息不断被压进权重里。隐藏状态大小是固定的(就是个小神经网络),机器人能以很小的成本'吸收'任意长的经验。学习在上线后无限期持续。

这样我们就把整段视频当context放进去!RoboTTT支持从人类视频单次演示学习:电路板组装场景里,人演示一个没见过的操作,机器人能精准复现。

人经常掉东西,但我们反应这么快根本没反应过来。这种快速纠正的反射占了人类物理能力的一半。RoboTTT展现实时自我改进:机器人能在执行过程中快速从错误中恢复,每次修正都进入context指导下一步。TTT core从训练数据里学出通用的failure-to-correction映射。

最后一点最让我兴奋。全新的Context Scaling Curve:从128到8K timestep,闭环性能一路上升,一点饱和的迹象都没有。8K-context预训练比1K提升62%。LLM能享受的东西,机器人学也应该能。很快,1M context都不是幻想了。

线程里详细讨论:

查看英文原文
We scaled a robot model natively to 8,000 timesteps of context, 5 minutes worth of muscle memory, with constant inference cost. Robot policies used to live their lives a few frames at a time (< 0.1 sec), instantly forgetting what just happened. We pushed to 3 orders of magnitude beyond SOTA.

Introducing RoboTTT. Test-Time Training (“TTT”) carries a tiny model *inside* the model. Every incoming sensor reading triggers one gradient step on that tiny core, so the history keeps getting compressed into its weights. The hidden state has a fixed size (literally a small neural net), so the robot can “grok” arbitrarily long experience with little overhead. Learning continues indefinitely after deployment.

We can then put an entire video in context as prompt! RoboTTT enables one-shot in-context learning from human video: in circuit board assembly, a human demonstrates a never-seen configuration once, and the robot imitates it faithfully.

Humans drop things all the time, but we pick them up so fast that we don’t even notice. That reflex to fix is half of our physical competence. RoboTTT shows self-improvement on the fly: the robot is skilled at recovering from its own errors mid-episode, and each fix enters its context to inform the next move. The TTT core distills a general-purpose, failure-to-correction mapping from the training data.

One more thing. What excites me the most is a new Context Scaling Curve: from 128 to 8K timesteps, closed-loop performance hill-climbs steadily with no sign of saturation. 8K-context pretraining beats 1K by 62%. What LLM enjoys, robotics should too. Soon, even 1M context is not a fantasy.

Deep dive in thread:
◔ 21.4 万 次浏览♥ 1,175⇄ 177▶ 含视频研究看原帖 ↗
Jim Fan@DrJimFan · 创始人 · 1 天前NVIDIA 具身智能研究负责人

博客和论文:
research.nvidia.com/labs/gea…

来自NVIDIA GEAR Lab的呈现。

千万看看Yunfan的技术深潜!

引用 Yunfan Jiang @YunfanJiangNVIDIA推出RoboTTT机器人策略,将视觉运动上下文扩展至8K时间步,远超现有SoTA,推理延迟保持不变。机器人可通过视频示例一次性模仿、边部署边改进、从干扰中恢复,可完成5分钟10阶段的装配任务。查看被引原帖 ↗
查看英文原文
Blog and paper:
research.nvidia.com/labs/gea…

Presented to you from NVIDIA GEAR Lab.

Please check out Yunfan's technical deep dive!
Min Choi@minchoi · 博主 · 1 天前AI 产品演示博主,专门展示新工具玩法
连环推 ×7

再见了,周日晚上的备课。Claude 现在对美国 K-12 教师免费开放。它可以在几分钟内生成课程计划、测验、工作表、家校沟通邮件、IEP 支持和符合教学标准的教材。教师可以用它做 5 件事 + 注册方式:

引用 Claude @claudeai推出Claude for Teachers,向美国经认证的K-12教育工作者免费提供Claude高级功能,包含教学技能库与映射至全美50州学术标准的循证课程资源。查看被引原帖 ↗
查看英文原文
RIP Sunday night lesson planning.

Claude is now free for verified U.S. K-12 teachers.

It can create lesson plans, quizzes, worksheets, family emails, IEP support, and standards-aligned materials in minutes.

5 ways how teachers can use it + signup:
1. Plan lessons

PROMPT:
Plan a 45-minute 7th grade math lesson on solving two-step equations. I teach from Illustrative Math and my students have mastered one-step equations. Create student materials that include a do-now, worked example, and exit ticket. Prepare a presentation I can give in class tomorrow.
2. Differentiate instruction

PROMPT:
Make a spreadsheet analysis on how my 4th grade ELA students are performing on grade-level standards. Then put everyone into small groups (reteach, on-level, and extension) for tomorrow’s lesson on main ideas and supporting details.
3. Assess for understanding

PROMPT:
Create a 10-question formative assessment for 8th grade science on photosynthesis, aligned to NGSS MS-LS1-6. Include two levels of difficulty and an answer key.
4. Communicate with families

PROMPT:
Every Friday at 3 pm, draft the weekly parent newsletter from this week’s lesson plans in my Planning folder. Include what we covered, what’s coming next week, and one thing families can practice at home. Prepare an email draft for me to review.
5. Anticipate misconceptions

PROMPT:
What are the most common misconceptions 5th graders have about multiplying fractions? Give me three discussion questions that surface each one.
Verify now, if you are US K-12 teacher

claude.com/solutions/teacher…
Ethan Mollick@emollick · 创始人 · 1 天前沃顿商学院教授,AI 应用研究权威

等待30分钟出几周的工作成果,这个等待感觉真是无限漫长。真的应该在 Codex 里加个闲置点击游戏,这样在等输出的时候能打发打发时间。

查看英文原文
It is funny how interminable waiting 30 minutes for a couple weeks of work can feel. Really need to integrate idle clicker games into Codex to keep us humans busy while we wait for output.
Lisan al Gaib@scaling01 · 博主 · 1 天前高频 AI 模型测评与爆料博主
连环推 ×2

很多人不理解 Mythos 现在已经有 5 个月了,我们要看的(编码)只是整体能力的一小部分

而且通常这个样本不太好,因为所有好的基准都是私密的。而且大多数基准不看模型大小或 token 效率。

话说回来,Kimi-K3 可能会在好几个常见的基准上击败 Opus 4.8 和 GPT-5.5,但这些基准不能说明全部情况,这些模型也不是最前沿的

最前沿的是 Mythos,或者在某些情况下甚至是 GPT-5.2 这样的老模型。

不过,DeepSeek-V4、GLM-5.2 或 Kimi-K2.6/2.7 都还没有解决过一个 Erdös 问题。
第一个公开发布的解决这个问题的模型是 GPT-5.2-Pro,那是 2025 年 11 月。现在是 2026 年 7 月。(在 FrontierCode T4 或 ARC-AGI-2 上也是一样)

如果你相信美国公司有越来越大的计算优势,且模型会加速未来模型的开发,你也应该在向后看的差距和预期/向前看的差距之间做出区分

领域也很重要。不同领域有不同的滞后。但总体来说,向后看的差距约为 7-8 个月。

引用 leo 🐾 @synthwavedd我认为Kimi K3会让那些说'中国落后西方8个月'的人感到震惊查看被引原帖 ↗
查看英文原文
A lot of these people don't understand that Mythos is 5 months old at this point and that what we will look at (coding) is only a small slice of overall capability

and typically this sample is not very good as all the good benchmarks are private. and most benchmarks don't look at model sizes or token-efficiency.

That said, Kimi-K3 will likely beat Opus 4.8 and GPT-5.5 in several of the usual benchmarks, ... but these benchmarks don't show the full story and these models are not the frontier

The frontier to beat is Mythos or in some cases even older models like GPT-5.2.

Mind you, DeepSeek-V4, GLM-5.2 or Kimi-K2.6/2.7 haven't solved a single Erdös problem so far.
The first publicly available model that solved one was GPT-5.2-Pro, back in November 2025. It's now July 2026. (same on FrontierCode T4 or ARC-AGI-2)

If you believe that US companies have a growing compute advantage and that models speed up the development of future models, you should also make a distinction between backward looking gap and the expected/forward looking one

Domains matter too. There are different lags for different domains. But overall the backward looking gap is ~7-8 months.
nevertheless im very hyped for Kimi-K3

it will be an insane model and will either force Anthropic and OpenAI to lower prices, bring out their largest and strongest models, or stop them sandbagging their lower and mid-tier models

Kimi-K3 should feel like very strong model and will be enough for 99% of coding use-cases
Chubby♨️@kimmonismus · 博主 · 1 天前Chubby,高频 AI 新闻聚合博主

NVIDIA 说用两个提示词,Codex 在一天内把 Cosmos 3 Nano 从 54.41% 的精度优化到了 93.35%。

这个实验使用了丰田的 Woven 交通安全数据集:8000+ 个四选一视频推理的训练和验证样本。
使用 NVIDIA TAO agent skills,Codex 自动化地:

检测和修补缺失的视频元数据
跑零样本基线
生成 LoRA 配置
启动训练和评估
跑 AutoML 超参数搜索
报告最佳模型

一个 LoRA 运行在 8 个 A100 GPU 上用了大约 30 分钟就达到了 87.14%。

第二个提示词在多个 A100 节点上启动了 43 个并行 AutoML 试验,19.5 小时后达到了 93.35%。NVIDIA 说 LoRA 所需的 GPU 小时数大约只有全参数训练的七分之一。
Agent skills 正在成为通用编码代理操作高度专业化 ML 基础设施的界面。

查看英文原文
NVIDIA says Codex post-trained Cosmos 3 Nano from 54.41% to 93.35% accuracy in one day - with two prompts.

The experiment used Toyota’s Woven Traffic Safety dataset: 8,000+ training and validation samples for four-choice video reasoning.
Using NVIDIA TAO agent skills, Codex autonomously:

Detected and patched missing video metadata
Ran the zero-shot baseline
Generated LoRA configurations
Launched training and evaluation
Ran an AutoML hyperparameter sweep
Reported the best model

One LoRA run reached 87.14% after roughly 30 minutes on eight A100 GPUs.

A second prompt launched 43 parallel AutoML trials across multiple A100 nodes, reaching 93.35% after 19.5 hours. NVIDIA says LoRA required roughly seven times fewer GPU-hours than full-parameter training.
Agent skills are becoming the interface through which general coding agents operate highly specialized ML infrastructure.
◔ 8.8 万 次浏览♥ 991⇄ 65▶ 含视频演示看原帖 ↗
Chubby♨️@kimmonismus · 博主 · 1 天前Chubby,高频 AI 新闻聚合博主

开发者.nvidia.com/blog/po…

查看英文原文
developer.nvidia.com/blog/po…
Bindu Reddy@bindureddy · 创始人 · 1 天前Abacus.AI CEO,AI 行业观点博主

别上大学了 - 去创办你的公司

未来一个人的公司会成为常态

颠覆一切 - 法律、医疗、金融....

查看英文原文
Don't go to college - start building your company

1-person companies will be the norm in the future

Disrupt everything - Legal, Health, Finance....
向阳乔木@vista8 · 中文博主 · 1 天前向阳乔木,中文圈 AI 工具与趋势博主

Codex的一大作用就是修小火箭的各种配置。

真的不想搞这些,谁出个教程,怎么配置规则。

Github开源的各种规则感觉也不好使。

Hailuo AI-MiniMax Hub@Hailuo_AI · 公司官方 · 1 天前MiniMax 旗下海螺 AI 视频官方
连环推 ×2

🚀 MiniMax Hub 重大升级!
现在更新 >>
hub.minimax.io/

新版本亮点:
🎵 音频进化:ElevenLabs Music v2 & Seed Audio 1.0 现已上线!
⚡ 特色技能:一站式 AI 工作流,支持短剧、运动图形和高级摄像机控制。
🛠️ 更智能的工作区:跨所有节点的统一全局搜索和一键媒体批量下载。

#MiniMax
#MiniMaxHub

查看英文原文
🚀 MiniMax Hub Major Upgrade!
Update now>>
hub.minimax.io/


Highlights of new release:
🎵 Audio Evolution: ElevenLabs Music v2 & Seed Audio 1.0 are now live!
⚡ Featured Skills: One-stop AI workflows for short-dramas, motion graphics, and advanced camera controls.
🛠️ Smarter Workspace: Unified Global Search across all nodes & one-click media bulk downloads.


#MiniMax
#MiniMaxHub
🎵 Audio Model Update
- ElevenLabs Music v2: Generates music in any genre from scene, style, and mood prompts.
- Seed Audio 1.0: Generates voices from text, reference audio, or images, with adjustable speed, volume, and pitch.
Zara Zhang@zarazhangrui · 中文博主 · 1 天前Zara Zhang,哈佛出身的 AI 产品博主,follow-builders 作者

因为我没有通过传统方式学编程,使用编码代理对我来说纯粹是创意和自我表达。GitHub 基本上就是我的 Substack。

引用 David Pan @davepWhen I was growing up, programming was an expression of creativity. Somewhere along the way, it became more about the chores. Merge conflicts. Flaky tests. CI failures. Bug triage. Let's give the chores to the robots and get back to building cool shit.查看被引原帖 ↗
查看英文原文
Because I never learned programming the traditional way, using coding agents has been purely an act of creativity and self-expression

GitHub is basically my Substack
Chubby♨️@kimmonismus · 博主 · 1 天前Chubby,高频 AI 新闻聚合博主

他们最新发布的预训练报告显示,Soofi S 30B-A3B 基于 NVIDIA 开源的 Nemotron 3 Nano 参考架构,采用了相同的混合 Mamba + Transformer MoE 设计,拥有约 3B 的活跃参数和几乎相同的架构选择。

真正新颖的不是架构,而是训练方式:

- 约 27 万亿个训练 tokens
- 德语故意加权
- 在 Deutsche Telekom 工业 AI 云上的端到端训练
- 完整开源的训练方案、超参数和评估方法

说实话,看到欧洲也在训练自己的模型还是挺兴奋的。

引用 NXT EU @NXT4EUGermany has launched one of the world's best open-source AI models. Soofi S, made by the Soofi consortium, is a 30B parameter model fully trained in Europe and tops the ranking for open-source AI. Huge moment for Europe, and finally some competition for Chinese open-source AI.查看被引原帖 ↗
查看英文原文
Their newly released pretraining report reveals that Soofi S 30B-A3B is based on NVIDIA’s open Nemotron 3 Nano reference architecture, the same hybrid Mamba + Transformer MoE design with ~3B active parameters and nearly identical architectural choices

What’s actually new isn’t the architecture, but the training:

-~27 trillion training tokens
-German deliberately up-weighted
-End-to-end training on Deutsche Telekom’s Industrial AI Cloud
-Full training recipe, hyperparameters and evaluation methodology released openly

Ngl, excited to see europe kinda trains their own models.
Kling AI@Kling_ai · 公司官方 · 1 天前快手旗下可灵 AI 视频官方

魔法背后的创意者在这里。✨

从天马行空的想法到获奖之作,来看看 Kling AI NEXTGEN Awards 那些亮点作品背后的故事。

特别感谢 Jacek Kadaj、Jinlong Hu、Kassai Ricsi、KUA KEE SENG、Lee Hyerin、Mingwei Chen、Park Soleun、Peize Wang、reels_mon01、Seo Yoonjung、Shin Seoyeon、Shiqi Wang、Shihua Lin、Son Seoyeong、Xuanwei Liu、Yihang Yang、Zibo Jia 以及所有创意者无与伦比的支持和启发。

完整采访即将推出。

查看英文原文
The creators behind the magic are here. ✨

From wild ideas to award-winning creations, discover the stories behind the works that stood out at Kling AI NEXTGEN Awards.

A huge thank you to Jacek Kadaj, Jinlong Hu, Kassai Ricsi, KUA KEE SENG, Lee Hyerin, Mingwei Chen, Park Soleun, Peize Wang, reels_mon01, Seo Yoonjung, Shin Seoyeon, Shiqi Wang, Shihua Lin, Son Seoyeong, Xuanwei Liu, Yihang Yang, Zibo Jia, and all the creators for your incredible support and inspiration.

Full interviews coming soon.
◔ 35.8 万 次浏览♥ 88⇄ 14▶ 含视频其他看原帖 ↗
向阳乔木@vista8 · 中文博主 · 1 天前向阳乔木,中文圈 AI 工具与趋势博主

牛逼,整个x代码全开源啊。

这是打OpenAI的脸吗?

引用 Elon Musk @elonmuskOnce we have completed our review for security vulnerabilities, we will make the entire codebase of 𝕏 open source, with no exceptions. Moreover, we will invite third party reviewers to examine the system that is running to confirm that the open source code is what is running. Trust through total transparency is the only thing that should be believed.查看被引原帖 ↗
clem 🤗@ClementDelangue · 创始人 · 1 天前HuggingFace 联合创始人兼 CEO

下周会在旧金山。咱们要不要组织个集会或游行,支持开源和本地 AI?

查看英文原文
Going to be in San Francisco next week. Should we organize some sort of a meetup or march in support of open-source and local AI?
Orange AI@oran_ge · 中文博主 · 1 天前Orange AI,中文圈 AI 产品观察博主

什么是丰饶时代呢
有人骑行是为了去公司上班
有人骑行是为了去山里兜风
有人听播客是为了吸收信息
有人听播客是就是随便听听
有人做社交媒体是为了商单赚钱
有人做社交媒体就是想表达自我
有人用 Agent 是为提高自己的工作效率
有人用 Agent 是在构建自己的思维殿堂

当做一件事只为了提高赚钱的效率,那还是贫瘠时代
当它成为每个人自己所选择的生活方式的时候,才是丰饶时代
人又有很深的时代惯性,穷怕了的人在丰饶时代也会保留以前的习惯,就像父母那辈人退休了还保留着极度省钱的小习惯一样

Gorden Sun@Gorden_Sun · 中文博主 · 1 天前中文圈高频 AI 资讯与开源项目博主

马斯克将把 𝕏 的整个代码库开源

引用 Elon Musk @elonmuskOnce we have completed our review for security vulnerabilities, we will make the entire codebase of 𝕏 open source, with no exceptions. Moreover, we will invite third party reviewers to examine the system that is running to confirm that the open source code is what is running. Trust through total transparency is the only thing that should be believed.查看被引原帖 ↗
小互@xiaohu · 中文博主 · 1 天前小互,中文圈高频 AI 资讯站 Xiaohu.AI 主理人

牛 P 啊

📢大消息

马斯克:将会把 𝕏 的整个代码全部开源

同时将邀请第三方审查员审查 𝕏 的整个系统,以确保开源的代码和线上运行的完全一致!

🫡

引用 Elon Musk @elonmuskOnce we have completed our review for security vulnerabilities, we will make the entire codebase of 𝕏 open source, with no exceptions. Moreover, we will invite third party reviewers to examine the system that is running to confirm that the open source code is what is running. Trust through total transparency is the only thing that should be believed.查看被引原帖 ↗
Bindu Reddy@bindureddy · 创始人 · 1 天前Abacus.AI CEO,AI 行业观点博主

用 Fable 或 5.6 sol 这样的强大模型来构建你的整个产品

需求越复杂越好

这些模型正在扩展以完成 100 人开发团队的工作

真的,他们可以做测试、调试、监控!基本上什么都能做

查看英文原文
Use a strong model like Fable or 5.6 sol to build your entire product

The more complex the requirements, the better

These models are scaling up to the do the work of 100 person dev teams

Yes, they can do testing, debugging, monitoring! Literally everything
Google DeepMind@GoogleDeepMind · 公司官方 · 1 天前谷歌旗下 AI 研究机构,Gemini 背后团队

从提出假设到设计实验,AI agents正开始重塑科学发现。但最大的挑战是在真实世界中验证这些想法。我们的文章探讨了日益增长的验证瓶颈,并为政策制定者和资金方提出了四个优先方向。

查看英文原文
From proposing hypotheses to designing experiments, AI agents are starting to reshape scientific discovery. But the hardest part is testing these ideas in the real world.

Our essay explores the growing validation bottleneck and outlines four priorities for policymakers and funders. →
goo.gle/4poACUT
向阳乔木@vista8 · 中文博主 · 1 天前向阳乔木,中文圈 AI 工具与趋势博主

这个很棒啊!一个能帮你提升 Vibe Coding 前端交互设计水平的网站。

不知道名字,就不知道如何跟 AI 说实现什么样的效果。

网站整理了Web和App常见组件和动效名字。

还列网页设计风格名,比如Liquid Glass(液态玻璃)、Neumorphism(新拟物)等。

地址见评论区

向阳乔木@vista8 · 中文博主 · 1 天前向阳乔木,中文圈 AI 工具与趋势博主

感觉这个相当实用,应该好好学习下

namethatui.com/

Cristóbal Valenzuela@c_valenzuelab · 创始人 · 1 天前Runway 联合创始人兼 CEO

光标终将活过来。

引用 Cristóbal Valenzuela @c_valenzuelabIt's 22:14. Long day. Baby's asleep. Here is a wild late-night idea: what if you could turn the cursor into a physical object? Something that could reach into the real world, reshape it, bend it to your will. No barriers between what you see in a screen and what's in the real world. Seems interesting but it's late. Ah fuck it, let's just quickly see if Gen-4.5 can pull this off. 22:40 pm. Yep, it's possible.查看被引原帖 ↗
查看英文原文
The cursor will come alive, eventually.
官方博客OpenAI1 天前

美国通过州级和联邦行动推进AI安全

OpenAI阐述了一种「反向联邦制」的AI治理方式,通过州法律来帮助建立安全、民主的AI国家框架。

查看英文原文
The US is advancing AI safety through state and federal action

OpenAI outlines a “reverse federalism” approach to AI governance, where state laws help build a national framework for safe, democratic AI.
@levelsio@levelsio · 博主 · 1 天前独立开发者标杆,AI 产品连续创业者

@X 团队报告一个问题:当你在网页上看到推文并点击 Grok 按钮想询问相关问题,点击停止后输入问题,它就忘了你之前提到的那条帖子了。

查看英文原文
Mini bug for
@X
staff:

When you see a tweet and tap Grok button on web, then you wanna ask question about it, so you tap STOP and type a question, it forgets what post you included!
Chubby♨️@kimmonismus · 博主 · 1 天前Chubby,高频 AI 新闻聚合博主

我给想理解 AI 真正发展方向的人写文章,embodiment 是大家马上就要关心的重点。Booster T2 让我感兴趣是因为它首先被设计成开发平台——板载 2070 TFLOPS 配合 Booster Studio 让开发者可以在真实机器人上迭代而不是玩具。这就是如何建立社区,而社区正是把机器人变成平台的方式。

引用 Booster Robotics @boosteroboticsBooster T2是新推出的旗舰级具身开发平台,提供2070 TFLOPS计算能力,是双足人形机器人中最高的。配合高输出、高DOF机身,实现更强的感知、理解和执行能力。结合强大计算性能和Booster Studio集成开发平台,为开发者解锁无限可能。查看被引原帖 ↗
查看英文原文
I write for people who want to understand where AI is actually heading, and embodiment is the part everyone's about to care about.

The Booster T2 is interesting to me because it's built as a development platform first —  2070 TFLOPS onboard plus Booster Studio means developers get to iterate on the real body, not a toy. That's how you get a community, and community is what turns a robot into a platform.
@levelsio@levelsio · 博主 · 1 天前独立开发者标杆,AI 产品连续创业者

我做了个小机器人,每天提醒我何时应该出去坐太阳,咱们这边是南欧所以你那儿可能不同。

☀️ 维生素 D 时间窗口——现在去晒 15 分钟左右。
UV 指数 3.8(理想水平——获得维生素 D 又不会被晒伤)。

引用 Tim | ONLYUP™ 💹 @timonlyup你对过度日晒的看法如何(皮肤癌、皮肤老化等),Pieter?查看被引原帖 ↗
查看英文原文
So I made this little bot, every day it messages when I should go sit in the sun outside, we're in Southern Europe so it might be more wherever you are

"☀️ Vitamin D window — go sit in the sun for ~15 min now.
UV index is 3.8 (ideal level — good vitamin D without burning)."
The Rundown AI@TheRundownAI · 博主 · 1 天前百万订阅 AI 日报官方

今天 AI 行业头条:

- DeepMind CEO 向美国 AI 监管机构推介
- 报道:OpenAI 的第一款设备是 AI 扬声器
- 为客户咨询构建无代码语音 agent
- 纽约州阻止 AI 数据中心热潮
- 4 个新 AI 工具、社区工作流等更多内容

查看英文原文
Top stories in AI today:

- DeepMind CEO pitches U.S. AI watchdog
- Report: OpenAI's first device is an AI speaker
- Build a no-code voice agent for customer intake
- New York stalls the AI data center boom
- 4 new AI tools, community workflows, and more
Mustafa Suleyman@mustafasuleyman · 创始人 · 1 天前微软 AI CEO,DeepMind 联合创始人

Microsoft AI Futures 团队今日在 Nature Health 发表新论文。

在 109 个国家审查了 170 万条对话后,结果很清楚:Copilot 对那些对医疗系统信心不足的人来说是无价的资源。

技术一直都是伟大的平衡器,推动更多人获得更优质的服务。这进一步证明了 AI 正在为全球服务不足的群体在最需要的时刻提供无价的支持。

感谢所有作者。查看论文:
nature.com/articles/s44360-0…

查看英文原文
New paper published in Nature Health today from the Microsoft AI Futures team.

After reviewing 1.7m conversations across 109 countries, it’s clear that Copilot is an invaluable resource particularly for people with low confidence in their health systems.

We've always known that technology is a great equalizer, driving broader access to better quality services. This is further evidence that AI is giving underserved people around the world access to invaluable support when they need it most.

Many thanks to all the authors. Check out the paper here:
nature.com/articles/s44360-0…
官方博客OpenAI1 天前

GPT-Red:通过自我改进提升模型稳健性

了解 OpenAI 的自动化红队测试系统 GPT-Red,它通过自我对弈来提升 AI 的安全性、对齐度和抗提示词注入能力。

查看英文原文
GPT-Red: Unlocking Self-Improvement for Robustness

Explore GPT-Red, OpenAI’s automated red teaming system that uses self-play to improve AI safety, alignment, and prompt injection robustness.
小互@xiaohu · 中文博主 · 1 天前小互,中文圈高频 AI 资讯站 Xiaohu.AI 主理人
连环推 ×2

Google 在 Pixel 10 上跑 Gemma 4

三万英尺的客舱断网也能聊天、看图、改设置

在Google I/O India 上,Tensor 团队和 Pixel 团队联合演示:把 Gemma 4 轻量版直接塞进手机里的 TPU里

不仅能能聊天」,还能「看图、听写、控制手机」

官方演示:手机在断网状态下做旅行规划、菜谱推荐和家居自动化

Gorden Sun@Gorden_Sun · 中文博主 · 1 天前中文圈高频 AI 资讯与开源项目博主

Every推出的All Access订阅服务还挺划算
525美元一年,除了包含Every自家的订阅服务和产品服务(Every的主营业务是AI领域媒体内容),还包括这些高价值的服务:
1000美元的Codex点数
12个月Cursor Pro+会员,60美元/月的那一档,这一项就回本
1个月Claude Max会员,没写是5x还是20x,估计是5x
3个月Google AI Pro会员,没什么用
1年Framer Pro会员,AI做设计
6个月Notion Business会员

Tibor Blaho@btibor91 · 博主 · 1 天前逆向挖掘 AI 产品代码的爆料专家

最近发布的OpenAI视频中的一些有趣时刻

"Rune [plugin]是你在OpenAI的小伙伴"

"为Codex Micro键盘编写文档"

"在Codex定价页面添加隐藏的'tibo'键盘序列"

查看英文原文
A few interesting moments from the recently published OpenAI videos

"Rune [plugin] is your little guy at OpenAl"

"Document the Codex Micro keyboard"

"Add a hidden 'tibo' keyboard sequence to the Codex pricing page"
小互@xiaohu · 中文博主 · 1 天前小互,中文圈高频 AI 资讯站 Xiaohu.AI 主理人

“Apple智能”大模型已获手机端侧AI服务备案

国行 Siri 要来了…

网信办发布7款提供手机端侧生成式人工智能服务已备案信息的公告

为促进生成式人工智能服务创新发展和规范应用,网信部门会同有关部门按照《生成式人工智能服务管理暂行办法》要求,有序开展生成式人工智能服务备案工作

现将新增的“Apple智能”等7款提供手机端侧生成式人工智能服务备案信息予以公告。

其中包括,苹果智能、华为小艺、vivo蓝心、小米澎湃及努比亚豆包大模型。

◔ 2.2 万 次浏览(2 条合计)♥ 37⇄ 2动态看原帖 ↗
Gorden Sun@Gorden_Sun · 中文博主 · 1 天前中文圈高频 AI 资讯与开源项目博主

谷歌CEO Demis Hassabis发布长文《前沿AI框架与新时代的黎明》

我们找到了让沙子思考的方法,堪称奇迹。AGI几年之内就会到来,AGI的突破比互联网、移动互联网的变革更大,更像是火和电的发现。而且AGI的规模会是工业革命规模的10倍,发展速度也是10倍。

他也发出了跟Anthropic CEO达里奥类似的警告,AI前沿进展已经跑在了人类理解能力的前面:网络安全风险已是现实,生物、核风险紧随其后,而且AI越来越Agent化、能递归自我改进的系统将需要更强的控制。

他呼吁,由美国牵头,仿照金融业监管局(FINRA)建立一个前沿AI标准机构,对于前沿模型要在发布30天由机构检测,评估覆盖网络安全、生物威胁、欺骗行为等高风险领域,必要时甚至要协调全行业减速。

最接近技术的人,比外界更清楚接下来两年会发生什么。(不过显然Gemini现在完全不需要操心这些)

Chubby♨️@kimmonismus · 博主 · 1 天前Chubby,高频 AI 新闻聚合博主

OpenAI正竭尽全力向Anthropic展示他们为什么能坚持下去,以及为什么相信自己能赢得竞争。

我也来说两句:我每天都在用Codex,因为它的性价比真的很突出。5.6版本曾经短时间内出现过token消耗过多的问题。但总体而言,这些模型的性价比都相当不错。肯定比Anthropic的好。

引用 Tibo @thsottiaux或者……如果你告诉我们你喜欢GPT-5.6 Sol的什么地方,或者你为什么转换使用它,我们就给你100美元的Codex信用额呢?发推文,领取礼物,享受更多使用额度。前10000个人获得免费token!查看被引原帖 ↗
查看英文原文
OpenAI is leaving no stone unturned in its efforts to show Anthropic why they're here to stay and why they're confident they can win the competition.

But I'll join in: I use Codex daily because its price-performance ratio is outstanding. With version 5.6, there was a temporary issue with excessive token consumption. But overall, the price-performance ratio is excellent for the respective models. Certainly far better than with Anthropic.
宝玉@dotey · 中文博主 · 1 天前宝玉,中文圈 AI 翻译与科普大 V

今天在 ChatGPT 里面找 Codex 没找到,原来改名叫 Remote 了

引用 Thomas Ricouard @DimillianCodex Remote iOS版的最新更新来了!如我之前发布的,我们现在支持新的可视化功能,还有大量其他改进和修复!查看被引原帖 ↗

本站由 Jedee杰哥 打造 · 公众号「Jedee杰哥」每早送 AI 日报

姊妹站:𝕏 简中账号数据榜单 · X 关注 @jedeeai · RSS 订阅 · AI 日报 · 历史归档