nullptr
software engineer, building things for the web.
lately
- thought
- thought
the blind trust that we’re putting on LLM generated code is likely to cause issues in the future. i think formal guarantees of some kind need to be…
- post
can a small model smell a reward hack?
training detectors on qwen3-8b hidden states to catch reward-hacking agents, the three ways the data lied to me, and why early warning mostly didn't work.
- thought
chasing perfection is almost always a sub-optimal strategy. the more you try to seek it - whether it’s by having the most elegant ideas, the ’right’…
- thought
had been researching how benchmarks and rl envs and my agents came up with a clever term while i was trying to balance harness capabilities with the…
- thought
okayish read on RL training frameworks