Cornell professor's Finetuner's Fallacy: early training data leaves hard-to-undo imprints
pratyushmaini · x · 2026-10-06
- Pratyush Maini (joining Cornell Tech) compresses his PhD into one idea: data a model sees early in training leaves imprints on its representations that are very hard to undo later — the "Finetuner's Fallacy."
- The thread spans Rephrasing the Web, Safety Pretraining, TOFU, and Natively Unlearnable LLMs.
- He will discuss two works at COLM and is recruiting PhD students and postdocs.
Related event: Cornell Professor Unveils 'Finetuner's Fallacy' at COLM(2 posts)→
More from Research
- OCBench offers human-like scripted policies for scalable robot BC/RL research — kevin_zakka · 2026-10-06
- Apple research team opens 2027 PhD internships in video models, 3D/4D reconstruction — HildeKuehne · 2026-10-06
- MemAdapter uses counterfactual reasoning to curb memory-induced sycophancy in LLM agents — Ruqing Ning · 2026-10-06
- Peking University's Code2Games gets coding agents to build playable UE5 game worlds — PekingUniversity · 2026-10-06
- QuantCode: domain pretraining + SFT lifts Qwen trading-code pass from 27.8% to 58.2% — Alexey Chernysh · 2026-10-06
- Subsampling and extrapolation keep the Mandelbrot area estimate unbiased near the boundary — geoffreyirving · 2026-10-06