GLiNER creator: zero-shot NER models are underestimated; end-to-end RL is next
bclavie · x · 2026-09-16
GLiNER/GLiClass creator Ihor responds that the model family is definitely underestimated. Multi-modality is still on the roadmap, but foundations for end-to-end RL training are already there — and the Jev incident highlighted the potential of such general zero-shot information-extraction models to the public.
More from Models
- Jev model claims 40-400x cheaper than Gemini; dev backtest finds it faster and more consistent — FrankFelixAI · 2026-09-16
- Dev yacineMTB switches from Astra to DeepSeek Flash 4.1, citing much faster responses — yacineMTB · 2026-09-16
- Is distillation really how Chinese models close the gap with the US? One analyst says no — pstAsiatech · 2026-09-16
- Is there an open-source implementation of Jev, the tiny model that can 'play' Doom? — SomewhereAtWork · 2026-09-16
- Speech recognition heats up as Google, Meta and Microsoft vie for the top — DeepLearningAI · 2026-09-16
- Codex users report random outages and ~60% of weekly quota burned in a single day — TissueReligion · 2026-09-16