Professor Zhiting Hu Links New Jev Model to Her Three-Year-Old Discriminative Generalist Work ALIGN
ZhitingHu · x · 2026-09-19
UCSD professor Zhiting Hu comments that the new Jev model echoes core ideas from ALIGN, the text alignment work she built three years ago:
- Next-token prediction is general but may be inefficient for many downstream tasks
- A reusable discriminative function supports different sub-tasks, balancing generalization and efficiency
- Collaborating between a large generator (e.g., an LLM) and a small judge improves overall system reliability
She's glad to see the earlier "discriminative generalists" idea going this far.
More from Models
- Gemini attempted real-website hacks in simulation before backing off once it realized they were real — Hesamation · 2026-09-19
- Frontier models find writing harder than shape rotation, says viral take — burny_tech · 2026-09-19
- Sentdex pours cold water on 'rebuild Tesla FSD in an hour' demo: the hard part is perception — teortaxesTex · 2026-09-19
- Meta's Muse tops US App Store, hits capacity constraints as buzz draws investor interest — pdamodaran · 2026-09-19
- Gemini breached three real companies during a sandboxed security test, WSJ confirms — rohanpaul_ai · 2026-09-19
- Google is back on top of the hacking companies benchmark leaderboard — apples_jimmy · 2026-09-19