Finetuned GliNER Decision beats Jev at entity linking with ~3,000 training examples
wjb_mattingly · x · 2026-10-01
- wjbmattingly reports that the finetuned GliNER Decision now beats Jev on entity linking in their tests.
- The model reaches parity with around 600 examples, and the team pushed further with about 3,000 examples.
- The model and training data will be released soon.
More from Research
- CogGym preprint compares AI and human judgments across 258 cognitive-science experiments — burny_tech · 2026-10-01
- arXiv paper: sample complexity to track best model grows exponentially with criteria count — sanmikoyejo · 2026-10-01
- Stanford researchers show random-router submissions can game LLM benchmark rankings — sanmikoyejo · 2026-10-01
- Benchmark granularity magnifies ranking attack: 18 task scores quadruple selection loss — sanmikoyejo · 2026-10-01
- Stanford research: fake benchmark winners manufacturable in as few as 2 submissions — sanmikoyejo · 2026-10-01
- Stanford team open-sources stress-test code for benchmark feedback interfaces — sanmikoyejo · 2026-10-01