LLMs can learn (some) multi-step reasoning on filler tokens

(lesswrong.com)

2 points | by gmays 9 hours ago ago