Zen master on what on-policy distillation actually transfers
tokenbender · x · 2026-08-18
A zen-style dialogue about on-policy distillation in RL. When asked what is actually transferred, the master states it is "the teacher's bias over the variances the student already walks," noting that "the variance was never the missing thing."
More from Fun
- AI researcher notes the polarization of public opinion on AI — iScienceLuvr · 2026-08-18
- User gaslights Claude by mimicking its verbose filler words in conversation — wynwyn87 · 2026-08-18
- Marc Lou flies to Singapore to visit Network School, finds campus moved to Kazakhstan — marclou · 2026-08-18
- I am a machine that turns 3 Wikipedia tabs into 146 — CatAstro_Piyush · 2026-08-18
- User mocks Google: 'Vibe coding' Chrome releases? — basedjensen · 2026-08-18
- AI Agent hallucinates user's name as 'Darren' — hugobowne · 2026-08-18