Lab post-training researchers flood-likes post mocking on-policy distillation reality gap
teortaxesTex · x · 2026-11-14
norpadon contrasted on-policy distillation's paper narrative (beats direct RL, dense signal, solves credit assignment, sample-efficient) with how it actually goes in practice. The post drew likes from many post-training researchers at AI labs, which the reposter reads as a telling silent endorsement of that reality gap.
More from Fun
- Face detection fail: glasses detected as a 250-year-old smiling face — miniapeur · 2026-11-14
- X algorithm zeroes out copy-link score from 20, adds 'copy link bait' spam check — ThePeterMick · 2026-11-14
- Redditor shares "Returning Home", a short AI film made with Seedance — purplenurpleSEB · 2026-11-14
- "You don't use 100% of your brain, so you're an MoE" quip lands in AI circles — gnukeith · 2026-11-14
- Reading glasses vs four 65-inch 4K TVs: the aging geek's choice — RexDouglass · 2026-11-14
- AI image meme pitches 'Jensen the White' — Jensen Huang in a white leather jacket — StefanoGogioso · 2026-11-14