Reasoning research’s biggest shift, from symbolic traces to natural language
denny_zhou · x · 2026-07-27
The biggest shift in reasoning research, according to the author, was not a new objective but a new language: moving from symbolic/program-like traces to natural language.
- He says his 2017 work on math and puzzle solving used explicit reasoning encoded in programming languages.
- Early improvements came from training on program traces and reasoning traces generated via rejection sampling, with papers published in NeurIPS and ICLR before 2022.
- The core ideas, he argues, were already present; what changed was how reasoning is represented.
Related event: LLM Reasoning Shifts to Natural Language(2 posts)→
More from Research
- Genome language models uncover new class of reverse-transcriptase mechanisms — BrianHie · 2026-09-23
- Mathematician shares a cheap 4-step heuristic for hyperparameter tuning — dejanseo · 2026-09-23
- Burkov skew AI hype: 'deterministic LLMs' and 'first agents' are old tricks rebranded — burkov · 2026-09-23
- Continuous diffusion beats discrete on random k-SAT, proposed as standard benchmark — ArashVahdat · 2026-09-23
- Grady Booch: Contemporary AI Still Lacks Abductive Reasoning, Just 'Next-Token Prediction' — Grady_Booch · 2026-09-23
- AI solves Navier-Stokes-related problem as machines upend mathematics, New Scientist reports — burny_tech · 2026-09-23