Amir Hadifar

[你不要胡说 @Hi558575331]

全网最尴尬的可能就是北大。

不是王虹拒绝回国任教，也不是王虹离开北大才获奖，而是王虹在采访中亲口说：在北大，长期觉得自己“能生存下来就不错了”。

​这句“能生存下来就不错了”，王虹说了不止一次。

在法国《世界报》专访里她补充道：“在我的班级里，有一些人真的非常优秀。我并不确定自己的能力……我没有觉得自己拥有和他们一样的天赋。”

​说这话的人，16岁以653分考入北大，大二从地球与空间科学学院转入数学科学学院。可进了数院，她发现自己周围全是奥数金牌得主和竞赛大神。没受过系统竞赛训练的王虹，一度是班上的“小透明”。

​她的同学、后来成为北大博雅特聘教授的孙鑫回忆得特别直接：“坦率地说，那时我还没有意识到王虹会从事严肃的数学研究。”

连同学都不看好她。

北大���授田刚也承认：“她不是竞赛生，在班上的表现也不是最亮眼的。”

​一所中国最顶尖的大学，一个后来拿下菲尔兹奖的人，当年在这所大学里只觉得“能生存下来就不错了”。这话搁谁听了不扎心？

​王虹后来去了法国。

一个在北大连生存都觉得吃力的人，换个环境反而找到了自己的路。

2025年2月，她与合作者发表127页论文，彻底证明了困扰数学界逾百年的三维挂谷猜想。

2026年7月23日，她成为菲尔兹奖90余年历史上第三位女性获奖者、首位中国籍女性得主。

​问题来了：王虹在北大“能生存下来就不错了”，是她的问题，还是北大的问题？

​有人说这是谦虚，天才都这样。可王虹的原话是“一直在挣扎”。挣扎和谦虚是两码事。一个真正享受数学的人，在北大感受到的不是滋养而是挣扎，这所大学的人才培养机制是不是该反思一下？

​王虹获奖后，北大发文祝贺。可如果她当年没熬过来呢？如果她在法国那一个月真的转了行呢？

有人说“是金子总会发光”，可问题是，有些土壤会埋没金子，让金子发不出光。

作者：静静的顿河

[Jürgen Schmidhuber @SchmidhuberAI]

Meta learning and recursive self-improvement are old ideas. Foundation models breathe new life into them. Our new survey, “Self-Improvements in Modern Agentic Systems,” reviews how the concepts are continuing to evolve.

Paper: https://t.co/59oXCMVUkD
Project: https://t.co/sZwFYdGetH
Github: https://t.co/7OFgJUCN3a

[Andrew Case @attrc]

To summarize: HuggingFace got autonomously compromised by a model from an American company. HF then tried to use American frontier model(s) to defend themselves, but were blocked by guardrails. HF then had to turn to open source Chinese models to defend themselves from another American company.

[Mihail Eric @mihail_eric]

The new syllabus for my Stanford course 𝗧𝗵𝗲 𝗠𝗼𝗱𝗲𝗿𝗻 𝗦𝗼𝗳𝘁𝘄𝗮𝗿𝗲 𝗗𝗲𝘃𝗲𝗹𝗼𝗽𝗲𝗿 is coming along nicely. The AI coding world has changed a lot since last September. The new class will be bigger, more difficult, more fun.

More exciting things coming soon!

[Muratcan Koylan @muratcan]

The prompt engineering here is super impressive!

Such a great example of agent prompting:

[John Schulman @johnschulman2]

People sometimes ask why fine-tune when general-purpose models keep getting better. Bridgewater's work is a good reminder that with the right data -- here, expert judgements -- you can beat prompting-only approaches by a lot.
@ddkang and the Bridgewater AIA Labs team are great -- glad to see them sharing this.

[Greg Kamradt @GregKamradt]

obvious in retrospect but I had no idea there was a black market for tokens

[Tim Rocktäschel @_rockt]

@gneubig In my view @jaseweston & @j_foerst laid out the spectrum well.

https://t.co/KMNs9iVmkq

[Silvia Sapora @silviasapora]

After interviewing for Research Scientist roles at DeepMind, Isomorphic, Meta, Cohere and more, I wrote up everything I learned. Technical prep, logistics, negotiation, and emotional breakdowns. Check out my guide: https://t.co/eLh20ggMHW

[Amir Hadifar @hadif4r]

Yet, it still makes mistakes

[Amir Hadifar @hadif4r]

Fable is the only model that can predict hours correctly.

[Lukasz Kaiser @lukaszkaiser]

I believe that exploring and making mistakes is key to learning and research.

[Amir Hadifar @hadif4r]

Still think we are a couple of years away from autoresearch (5+).

[Intology @intology]

Can coding agents do research?

We release NanoGPT-Bench, an internal eval we’ve used to test agents on an AI R&D problem with months of human progress

Codex, Claude Code, Autoresearch recover only 9.3% of human progress, mostly tuning hyperparams & ignoring algorithmic research

NanoGPT-Bench is built on the NanoGPT Speedrun, a popular LLM pretraining competition to minimize the training time of a GPT-2 style model. Existing human submissions constitute nearly 2 years of work. To control for dependencies and contamination in frontier models, we standardize evaluation to a 5-month window of world records. Evaluation is fully autonomous and end-to-end, with no human intervention or internet access. 🧵

[François Chollet @fchollet]

Most human tasks are not Markovian, the optimal next action cannot be determined solely by looking at the current state. It depends heavily on the past trajectory, the original intent, and context constraints. An agent that cannot compress and track its past trajectory with absolute fidelity is maybe 20% as useful as one that can.

[Prime Intellect @PrimeIntellect]

They struggle with novelty though. We ran a phase where every idea had to pass a novelty check, neither agent managed to improve the baseline.

We release all the generated ideas here:

Codex: https://t.co/Dda3YzuC5Q

Claude: https://t.co/t1r3qdwacY

[Peyman Milanfar @docmilanfar]

Career "advice":

Live modestly, save money
Put your family and health first
Keep learning -don't live on your laurels
No jerks: don't work for 'em, don't hire 'em

[Yuchen Jin @Yuchenj_UW]

I’m so glad AI killed LeetCode interviews.

For 10 years, tech companies made every engineer grind the same puzzles and prove they could invert a binary tree from memory.

Today, the dumbest AI model can walk in and one-shot the entire interview.

Thank you, AI.

[Gabriele Berton @gabriberton]

Really cool

When they trained GPT3 they had loss spikes because they scraped from a subreddit of microwave noises

That training batch was literally text like "mmmmmmmmmmmmmmm"

[Ahmad Hijazi @_AhmadHijazi]

When people talk about god... sometimes they just want to say : "My tribe is better than yours!"

[Both types]

A story of Tribes, Conflation, and Dissonance.

Podcast v

[Amir Hadifar @hadif4r]

Businesses rush into AI assistants (e.g., TripAdvisor, Cardoen), but they are often not well secured! Along with Mythos, this becomes something important in the future..