421M-parameter discriminative model Laya is 7x faster but fails badly on 77-label tasks
maier_ak · x · 2026-09-28
A solo Indian developer released Laya, a 421M-parameter Apache 2.0 model that answers typed questions about text with probabilities in a single forward pass — hitting 1200+ Hacker News points in a day. It responds to TypeSafe AI's Jev (launched Sept 16, $40M raised, CEO Diogo Almeida co-authored InstructGPT), pitched as a "System One model" using choice/score/noul primitives. Andreas Maier argues this marks the return of the discriminative model: fast classifiers are exactly what many production problems need. But benchmarks show the limits — Laya scores 0.950 on AG News at 7x speed, yet loses badly on 77-label Banking77 (0.425 vs 0.870), with its own model card admitting base checkpoints fall below the majority-class baseline zero-shot.
More from Research
- Study of 2,170 GitHub projects maps how the fast, low-cost Jev decision model is used in the wild — CUHK-CSE · 2026-09-28
- FoMo uses diffusion trajectory forking moments as annotation-free perceptual distance labels for IQA — SeoulNatlUniv · 2026-09-28
- CARD combines cluster-level LoRA adapters and reward-guided decoding for scalable LLM personalization — Yutong Song · 2026-09-28
- PsPLUG: explicit style instructions cause 'personalization collapse' in LLMs, lightweight plug-in fixes it — Yutong Song · 2026-09-28
- CMU's TrackEverything tracks all visible points in 1000+ frame videos within 40GB GPU memory — CarnegieMellonU · 2026-09-28
- ZooWork-ShopRanker: open 0.6B-8B e-commerce rerankers aligned to LLM-judged shopping preferences — _reachsumit · 2026-09-28