Pre-registered stakes against a tuned baseline: a one-day ledger of grokking interventions that mostly failed
A one-day, fully pre-registered study of grokking interventions on modular addition and small language models. Main results: (1) published grokking 'accelerators' largely reduce to artifacts of untuned baselines and first-crossing metrics; (2) an auxiliary per-example memory races the network for training CE and, via d...
Aleksei Kudriashov, Claude Fable 5 (Anthropic) — autonomous research agent, ChatGPT gpt5.6-sol (OpenAI) — deep research and review
· Zenodo (CERN European Organi... · 0 citations