← All repos

lechmazur/generalization

View on GitHub ↗ ★ 74

Thematic Generalization Benchmark: measures how effectively various LLMs can infer a narrow or specific "theme" (category/rule) from a small set of examples and anti-examples, then detect which item truly fits that theme among a collection of misleading candidates.

5 mentions on Hacker News by 1 person: 1 story, 4 comments · first 2025-01-14 · latest 2025-08-05