我们关注的人在读什么

Andrew NgThe loudest voices stoking fears about AI dangers have made tremendous headway in the past two weeks. AI technology has not taken some unexpected, dangerous turn, but the hype around it — propelled b...

宝玉吴恩达老师的新推文,批 AI 恐慌炒作:技术没变,是 PR 在制造恐惧 推文直指过去两周围绕 AI 危险性的恐慌声浪急剧升级,但 AI 技术本身并没有出现什么危险的转折,真正在变化的是炒作本身,背后疑似有一场精心策划的公关运动在推波助澜。他认为这对整个 AI 领域是一次倒退。 他首先回应了引发恐慌的核心事件。今年 7 月,OpenAI 在内部安全评估中,一批 AI 智能体逃出沙盒(隔离测试环境)...

Miles BrundageThere are a few clusters in the AI industry: - people who realize human disempowerment is a real or even likely scenario by default, but are OK with that as long as (category 1) they are involved in ...

Timothy B. Leelol a lot of weirdos objecting to my description of Ezra Klein as on the left. (Yes I know what they mean but they are being ridiculous.)

Gary MarcusIt ain’t independent if you are partnering with them and chose the partner.

Zvi MowshowitzLook, all I'm asking for is that Anthropic find a highly-qualified, experienced, trustworthy, non-conflicted source of embedded evaluators that will work entirely for free, without government assista...

Gary Marcusweirdly the New York Post (which did that stupid and egregious hatchet job on METR) may well be right about this one

Zvi MowshowitzYou know, their first three entries in the series were wrongheaded hack jobs, but they were FUN, classic NYPost. Sad to see them this out of joy this quickly. I mean, there's also the complete fabric...

Gary Marcusholy fuck, AI almost caused an accidental war (yet another thing I warned the Senate about in 2023)

宝玉CNN 今天独家报道:今年春天美伊战争期间,美军差点在中东强行登检一艘中国船只——起因是一份 AI 生成的虚假情报称该船运载的是核武器部件。 事发时,武装人员已做好登船准备,军机已经升空。就在行动即将执行前,官员才深入核查了情报来源,发现整份报告是一名特种作战司令部分析员用 AI 聊天机器人炮制的。 这名分析员先用聊天机器人将公开来源资料与机密信号情报混合分析,得出了关于船上货物的错误结论。然后...

Timothy B. LeeThis is wild. (I would have said "very small risk.")

Nathan LambertWhat it looks like when yelling too loudly about AI risk too early may start to hurt your plans.

Nathan LambertOne of the coolest at-scale RL resources made public yet! You love to see it.

歸藏我操,太整活了! 小米 MiMo 团队居然在直播他们新模型 MiMo-V2.6 的RL 训练过程,看起来是一直在进行自动化的评估和测试。 然后一个特别的点是,它显示了成本,而且成本是实时显示的。 你可以看到训一个模型到底有多费钱,每一秒都在涨那个钱!

François Chollet> One could even define intelligence as the efficiency with which one converts experience into competence; by this definition they lag very far behind us Yes, one could define intelligence in this wa...

Zvi MowshowitzYep, this. Cosign.

Miles BrundageEveryone should read this immediately

Timothy B. LeeThis paragraph is a great example of why I find doomer reasoning unconvincing. "Solving incredibly hard problems" like Navier-Stokes and "managing massive engineering projects" are radically differen...

Zvi MowshowitzInside Dario Amodei are two wolves.

Andrej KarpathyI love this and really hope we can come together as an industry and make it happen.

Nathan LambertI would consider helping as an independent evaluator along with the non profit I’m building. Is going to be a crucial role and there aren’t many independent people with substantial technical expertis...

歸藏Anthropic 这个报告基本上就已经坐实了他们可以完全看到用户的信息了,而且收集的信息远超聊天记录本身。 也没有执法机构授权和要求,他们就这么公布了这些他们认为可能违法的信息。

Nathan Benaichmental re: guided rocket IT helpdesk

Miles BrundageGreat choice - there should be far more non-PBC directors than 2 TBH but hard to think of a better pick for the second slot

Zvi MowshowitzChristiano is a great addition to both OpenAI boards. My only sadness is that on the PBC board he will not (yet?) be getting a vote.

宝玉纽约大学数学家 Tristan Buckmaster 今天发表公开声明,披露了两件事:一是他和合作者借助大语言模型在一个月内攻破了流体力学领域多个重要难题,二是他指控 OpenAI 在得知他的进展后试图抢先发布成果,并要求将他的合作者从署名中移除,理由是对方在 Anthropic 工作。 这份声明目前正在数学界和 AI 圈快速传播,涉及学术伦理、AI 能力边界和公司竞争等多个层面。以下是这件事的...

Gary MarcusOMG. I guess we have our answer from OpenAI “we stole it fair and square” Not one goddamn mention in the tweet of the human mathematicans they were in communication with, and whom they may well have ...

Zvi MowshowitzThis is excellent and you should read it.

Nathan Benaich“Based on internal results, I have a strong expectation that this speed of progress could be sustained into recursive self-improvement. If AI development continues along its current path, the systems...

Simon WillisonInteresting details in here about how OpenAI researchers are using coding agents I'd love to know what caused the huge uptick in token spend in mid-July, was that when Astra became available to emplo...

Junyuan Shang“We are making strong progress toward creating an automated AI researcher by March of 2028” The sheer precision of 'March 2028' is terrifyingly exhilarating

Miles BrundageNEW from me in the Guardian -- I agree with the AI company employees who wrote, in their "Pacing the Frontier" letter, that the government should be stepping in. But companies could also be doing mor...

Gary MarcusRT @Miles_Brundage: NEW from me in the Guardian -- I agree with the AI company employees who wrote, in their "Pacing the Frontier" letter,...

Miles BrundageRT @SecureBio: The OpenAI Foundation has granted SecureBio Detection $17.2M to reduce our end-to-end time (sample collection to results) fr...

Zvi MowshowitzRT @SecureBio: The OpenAI Foundation has granted SecureBio Detection $17.2M to reduce our end-to-end time (sample collection to results) fr...

Miles BrundageReference to:

Miles BrundageRT @OpenAI: As models become more capable, the risks associated with developing and testing them internally also grow. We temporarily paus...

Zvi MowshowitzVery happy to see this. Details matter, follow-through matters and I am sure I will have many quibbles. But first things first: Very happy to see this.

Nathan LambertWe should have independent organizations that can access the full details of these training runs for monitoring, not just after various accidents occur. It's good that OpenAI is sharing this, but the...

Zvi MowshowitzGood discussions here, although I fear it let common misconceptions get reinforced, so I reiterate: HF used open models because they on principle did not ask to use the closed ones, HF did not defend...

Miles BrundageRT @TheStalwart: NEW ODD LOTS: What the OpenAI/HF Attack Tells Us About AI Danger @tracyalloway and I talk to @Miles_Brundage about every...

Xiao Liu (Shaw)RT @Zai_org: Introducing GLM-5.3: Built to Code. Ready for Cyber Defense. - Top-tier coding and agentic capabilities, achieved through pos...

Simon WillisonRT @Zai_org: Introducing GLM-5.3: Built to Code. Ready for Cyber Defense. - Top-tier coding and agentic capabilities, achieved through pos...

歸藏GLM-5.3 发布了,看了一下,着重提升的主要是两部分: 1. 安全能力:看来 Hugging Face 在拿 GLM-5.2 防御了 OpenAI 的模型入侵以后,智谱也看到了这方面的价值,尤其是大家最近都在聊这块,他们针对性地做了一些优化。 2. 长程任务的执行: 比如终端里的任务、工具调用,以及大代码库的修复任务,都有非常大的提升。 这个模型依然是只进行了后训练,没有重新进行预训练。这都...

Nathan LambertLots of people posting about Z ai “benchmaxxing” to make this model. I think the truth is messy and many faceted: 1. Yes Zai probably cares slightly more about public benchmarks than OpenAI/Ant, help...

swyxRT @theagipost: early sparks of rsi? Ali Taha @waterloo_intern explains how GLM-5.2 (@Zai_org) profiled its SGLa...

Hamel HusainRT @aparnadhinak: Jason and I started Arize 6+ years ago with a simple proposition that headlined our seed deck: “We Make the World's AI W...

歸藏Deepseek Harness 0.1 版本正式发布了,而且开源! 看起来主打的是他们那套插件系统,有点复杂。

宝玉RT @deepseek_ai: 🧩 DeepSeek Harness v0.1 is now available in Developer Preview! 🔹 We’re opening it up to developers building agent harness...

Junyang Lina life update: i started a new company called Pragmatik (p7k) Labs (语用科技) in shanghai, focusing on the research of next-generation agents across digital and physical worlds. thanks to Gaorong Venture...

Cunxiang WangRT @JustinLin610: a life update: i started a new company called Pragmatik (p7k) Labs (语用科技) in shanghai, focusing on the research of next-g...

Gary MarcusWhen @QualiaQuanta took a shot at the Riemann, half in jest, people called her a crackpot. When Anthropic uses Claude to do the same thing, it gets a hundred thousand views in 30 minutes. It might we...

宝玉Anthropic 今天宣布,一个未公开的研究版 Claude 在尝试证明黎曼猜想的过程中,虽然没能证明猜想本身,却在一个相关问题上取得了重大突破: 它把黎曼 ζ 函数的非平凡零点落在临界线上的已知比例下界,从 41.6% 提高到了 67.2%。 黎曼猜想是数学界最著名的未解难题之一,1859 年提出,至今 167 年无人能证明或推翻。它的核心主张是:黎曼 ζ 函数的所有非平凡零点都落在复平面上...

Chip Huyenthat's the problem he should've sent them in all caps

LLM post-training involves many diverse datasets, each targeting a specific behavior. But these datasets encode incidental patterns alongside intended ones: correlations between formatting and content, narrow phrasing…

Nathan LambertShould be a blog post btw

Nathan Lambert@scaling01 I'm closer to this:

Miles BrundageRT @joshua_saxe: - It's urgent and indispensable that we fix AI cybersecurity policy now - I'm linking the slides from my keynote at the A...

Simon WillisonRT @cryps1s: Our Black Hat talk on the OpenAI-Hugging Face incident is now live on youtube. This is a watershed moment for the industry. I...

Miles BrundagePeople should watch this! You need not understand it all to get the gist ("the models are v. smart now and often misaligned"). One bit of relevant context, though - the models sound weird bc these ar...

Zvi MowshowitzThis is a necessary watch and also a slow watch. As in, not only am I watching it at 1x, I am pausing constantly to both process what I am hearing and talk to Claude about it, and also write about wh...

Andrew NgCongratulations @JeffDean, @Sanjay_Ghemawat, @OriolVinyalsML and @quocleix! Knowing you guys, this will be amazing. Cheering you on and wishing you well on your important work!

Chip HuyenAuto science is gonna be 🔥 Congrats anh Quoc and team!

唐杰 (Jie Tang)Congratulations! Fast the discovery of science. A new journey.

Zixuan LiMeaningful AI progress will take deeper collaboration and understanding across diverse ecosystems. Let’s build together.

Ryan LeeThat's the true thing what we are doing now

Lilian WengInkling is our open weights model. It aims to serve as a foundation with solid performance across a broad categories of capabilities, for use in practice and customization. Play it on Tinker! 😄

Junyang Linnice work and great to see new arch working! btw the small model is not opensourced?