Post-training works but is gated; updating parameters risks existing knowledge
coallaoh · x · 2026-08-14
The author responds that post-training is the first answer and works, but it requires data, compute, and expertise that most dependents lack. Updating parameters also risks what the model already knew.
Related event: Model adaptation happens through external engineering, not post-training(2 posts)→
More from AGI Musings
- The Complete Tech Stack for AI Colleagues: From Remembering to Doing — sujingshen · 2026-08-14
- AI Alignment Paradox: Aligning to Decisions We Don't Want to Make — StewartalsopIII · 2026-08-14
- Historian Compares Hugging Face Incident to 1988 Morris Worm — Severe-Internet9948 · 2026-08-14
- Philosopher Chalmers analyzes Anthropic's J-space: not a global workspace — burny_tech · 2026-08-14
- a16z: AI coding could be biggest market, Cursor + SpaceXAI teams iterate fastest — a16z · 2026-08-14
- GPT-5.6 hits 750 tokens/sec in Ultrafast mode, writes 90k-word novel in under 3 minutes — PeterDiamandis · 2026-08-14