'Post-training is about taking base models and making them less based'
panickssery · x · 2026-09-20
Nina Panickssery's one-liner: "post training is about taking base models and making them less based" — a pun playing on "base model" vs. the slang "based", poking fun at how post-training sands off a model's unfiltered personality via RLHF and safety tuning.
More from Fun
- Grammarly, dubbed the 3rd company wiped out by AI, carpet-bombs users with emails after cancellation — sahilypatel · 2026-09-20
- 100-run trolley problem test: Jev pulls the lever 99% of the time — FinanceYF5 · 2026-09-20
- Gemini 4 Pro stuns by generating a mechanical butterfly in three.js — CurieuxExplorer · 2026-09-20
- "pi is basically the new emacs": one dev's take on AI coding tools — mrdrozdov · 2026-09-20
- 'Crazy rich Bayesians': Anthropic's nickname already joked at CVPR 2019, Fei-Fei Li lab alum says — mariyaivasileva · 2026-09-20
- Claude Fable 5.1 vs Opus 5: a mold-generated meme comparison — repligate · 2026-09-20