Reasoning training
AssessTechniques
Post-training method that improves a model’s step-by-step problem-solving and decision-making abilities.
Why it's here
Placed in Assess: 1 article(s) of evidence from 1 source(s), led by research-stage coverage, with 1 in the last 30 days. Confidence 24%. Low accumulated evidence, so it defaults conservatively pending more signal.
Evidence (1)
- 6The New Stack·10/7/2026researchOpenAI safety lead says releases now add capability and risk weekly
Former OpenAI safety lead David Robinson says the company’s release process has sped up, with new capabilities now coming from reasoning training, tool integrations, and coding agents rather than only from fresh pretraining runs. He argues that this faster cadence is creating new risks more often and makes traditional long safety reports less effective than a live, continuously updated safety dashboard.