Do your capabilities homework — LessWrong
It seems to me that a lot of technical ai safety people haven't done their capabilities homework - and that's a shame! I'll try to illuminate here ma…
x Do your capabilities homework — LessWrong AI Frontpage 57 Do your capabilities homework by RobinHa 1st Aug 2026 6 min read 11 57 It seems to me that a lot of technical ai safety people haven't done their capabilities homework - and that's a shame! I'll try to illuminate here mainly with an example as to why I think people who care about safety should totally pay more attention to the trends and actively engage with them - the case for safe AI not through an additional loss term but as a consequence of the learning algorithm! RLVR It's now been 1.5 years since R1 came out - the paper which re
saved by
related reading
- Differential acceleration of alignment-relevant capabilities is a bad bet — LessWronglesswrong.com
- Lil'Loglilianweng.github.io
- AI in 2025: gestalt — LessWronglesswrong.com
- State of RL for reasoning LLMs | A. Weersaweers.de
- The case for ensuring that powerful AIs are controlled — LessWronglesswrong.com
- AI Control: Improving Safety Despite Intentional Subversion — LessWronglesswrong.com
- GRPO++: Tricks for Making RL Actually Workcameronrwolfe.substack.com
- GRPO is terrible — LessWronglesswrong.com
- A vision researcher’s guide to some RL stuff: PPO & GRPO - Yuge (Jimmy) Shiyugeten.github.io
- Recommendations for Technical AI Safety Research Directionsalignment.anthropic.com
- Simulated Users & Sad LLMs1a3orn.com
- Models Don't "Get Reward" — LessWronglesswrong.com