Source item
Open original source ↗RL is very inference heavy and shifts infrastructure build outs heavily
Scaling well engineered environments is difficult
Reward hacking and non verifiable rewards are key areas of research
Recursive self improvement already playing out
Major shift in o4 and o5 RL training
@aj_kourabi is learning so much so fast, he's such a cute little guy
2 segments
Segment 1 · characters 0–273 · Original segment ↗
Segment 2 · characters 275–343 · Original segment ↗
Quoted context
These are the quoted author’s words.
Scaling Reinforcement Learning Environments, Reward Hacking, Agents, Scaling Data Infrastructure Bottlenecks and Changes Distillation Data is a Moat Recursive Self Improvement o4 and o5 RL Training China Accelerator Production https://t.co/jJgRlggxV2
SemiAnalysis_ ↗