Leon Lang
@Lang__Leon
Followers
1K
Following
4K
Media
151
Statuses
4K
PhD student at the intersection of information theory and deep learning. Two master's degrees in maths and AI. Interested in AI existential safety
University of Amsterdam
Joined September 2013
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! 🧵
6
75
645
RT @frances__lorenz: Noooo I love a vibes based critique of utilitarianism as much as the next person, but why does it so often lead to mak….
0
2
0
What does the last sentence mean?.
@dystopiangf Wake up, kid, that was happening even before the Internet. US birth rate has been below replacement since ~1971, except for a split second echo of the baby boom. AI will increase birth rates in order to maximize the future light cone of neurotransmitter tonnage.
0
0
1
RT @NeelNanda5: The part I find most bizarre here is the ban on providing any external links, with the highly implausible justification tha….
0
2
0
RT @kayembruno: @NeurIPSConf, why take the option to provide figures in the rebuttals away from the authors during the rebuttal period? Gro….
0
21
0
RT @NathanpmYoung: China might not want to race. I hope we can create some red lines that both powers won't cross.
0
3
0
RT @shengjia_zhao: I am very excited to take up the role of chief scientist for meta super-intelligence labs. Looking forward to building a….
0
333
0
RT @xuandongzhao: #NeurIPS2025 reviews are out, and the authenticity of reviews surprises me again 😟. Two years ago, maybe 1/10 felt AI-ass….
0
10
0
RT @lyang36: Code release! 🚀. Following up on our IMO 2025 results with the public LLM Gemini 2.5 Pro — here’s the full pipeline & general….
github.com
Contribute to lyang36/IMO25 development by creating an account on GitHub.
0
44
0
RT @lyang36: Compared the first version in our paper, this code removes problem specific hints completely. It just works!.
0
3
0
RT @StefanFSchubert: Financial Times analysis: US jobs at a high risk from generative AI have not been more likely to shed young workers si….
0
124
0
RT @AlexKontorovich: Another AI system, ByteDance's SeedProver solved 4 out of 6 IMO problems *with* Lean, and solved a fifth with extended….
0
52
0
RT @plain_simon: I have moved to the University of Duisburg-Essen in Germany, where I will take the amusingly named position of "Akademisch….
0
2
0
RT @OwainEvans_UK: New paper & surprising result. LLMs transmit traits to other models via hidden signals in data. Datasets consisting only….
0
1K
0
This should probably update most people either upward on what current models can do, or downward on what the next generation of unreleased reasoning models can do.
0
0
0
This is pretty interesting. It seems like the prompts and scaffolding did not contain *many* hints about how to solve the problems, and they didn't go beyond "Let us try to solve the problem by induction" and the like.
1
0
0
RT @demishassabis: Official results are in - Gemini achieved gold-medal level in the International Mathematical Olympiad! 🏆 An advanced ver….
deepmind.google
Our advanced model officially achieved a gold-medal level performance on problems from the International Mathematical Olympiad (IMO), the world’s most prestigious competition for young...
0
766
0
RT @littmath: One piece of info that seems important to me in terms of forecasting usefulness of new AI models for mathematics: did the gol….
0
38
0