Leon Lang

Reward Learning is just supervised learning, and so should be equally safe, right? Wrong! Our paper “The Perils of Optimizing Learned Reward Functions: Low Training Error Does Not Guarantee Low Regret” shows that policy optimization causes issues. It was accepted to ICML! 🧵

I have moved to the University of Duisburg-Essen in Germany, where I will take the amusingly named position of "Akademischer Rat". Please get in touch if you are around and interested in: - motives - real singularity theory - ML theory - singular learning theory - AI alignment

#NeurIPS2025 reviews are out, and the authenticity of reviews surprises me again 😟 Two years ago, maybe 1/10 felt AI-assisted. Now? It seems 9/10 are AI-modified, beyond grammar fixes to fully generated reviews. As a researcher in AI-generated content detection, I know these…

Night and Day mindsets towards competition: Google Deepmind vs OpenAI

Code release! 🚀 Following up on our IMO 2025 results with the public LLM Gemini 2.5 Pro — here’s the full pipeline & general (non-problem-specific) prompts. 👉 [github.com/lyang36/IMO25] Have fun exploring! #AI #Math #LLMs #IMO2025

Compared the first version in our paper, this code removes problem specific hints completely. It just works!

Financial Times analysis: US jobs at a high risk from generative AI have not been more likely to shed young workers since ChatGPT launched

Another AI system, ByteDance's SeedProver solved 4 out of 6 IMO problems *with* Lean, and solved a fifth with extended compute. This is becoming routine, like when we went to the moon for the fourth time. There is *nothing* "routine" about this!!...

New paper & surprising result. LLMs transmit traits to other models via hidden signals in data. Datasets consisting only of 3-digit numbers can transmit a love for owls, or evil tendencies. 🧵

Official results are in - Gemini achieved gold-medal level in the International Mathematical Olympiad! 🏆 An advanced version was able to solve 5 out of 6 problems. Incredible progress - huge congrats to @lmthang and the team! deepmind.google/discover/blog/…

Pretty fucked up that you have to exercise in order to get more energy

Announcing Workshop Labs, a public benefit company.

One piece of info that seems important to me in terms of forecasting usefulness of new AI models for mathematics: did the gold-medal-winning models, which did not solve IMO problem 6, submit incorrect answers for it? 🧵

These results show we're closer to building systems that can solve more complex mathematics. A version of this model with Deep Think will soon be available to trusted testers, before rolling out to @Google AI Ultra subscribers. Find out more ↓ goo.gle/imo-2025-resul…

An advanced version of Gemini with Deep Think has officially achieved gold medal-level performance at the International Mathematical Olympiad. 🥇 It solved 5️⃣ out of 6️⃣ exceptionally difficult problems, involving algebra, combinatorics, geometry and number theory. Here’s how 🧵

Here's something heartening: Individuals who experienced more economic growth as kids tend to see the world as less zero-sum. That is, they're less likely to see someone's gains as another's loss.

I think OpenAI did a totally fine thing here. The IMO committee doesn’t have a right to prevent other people from talking about their exams, or scoring themselves or their AI systems on it.

🚨 According to a friend, the IMO asked AI companies not to steal the spotlight from kids and to wait a week after the closing ceremony to announce results. OpenAI announced the results BEFORE the closing ceremony. According to a Coordinator on Problem 6, the one problem OpenAI…

I think people generally overrate the IMO as a sign of general progress - IMO Qs are pretty artificial and there are *so* many ways to hack them. But to the degree this was indicative of general reasoning training without specialisation, that's significant