Fraser Scott@_yonadaApr 24yonadaa.github.ioInjecting country features in Gemma2-2bCircuit tracing lets us identify interpretable features in language models. In this post, I identify country-specific features and test whether they are interchangeable.178
Fraser Scott@_yonadaApr 20yonadaa.github.ioHow does Qwen3-4B recall currencies?After spending the last year designing tasks that are hard for AI, I have become interested in the question of how LLMs think, specifically how they are able to solve novel problems.14213
Fraser Scott@_yonadaMar 25So proud to have helped build ARC-3. Can't wait to see what people throw at it!ARC Prize@arcprizeMar 25Announcing ARC-AGI-3 The only unsaturated agentic intelligence benchmark in the world Humans score 100%, AI <1% This human-AI gap demonstrates we do not yet have AGI Most benchmarks test what models already know, ARC-AGI-3 tests how they learnGIF6198
ARC Prize@arcprizeMar 25Announcing ARC-AGI-3 The only unsaturated agentic intelligence benchmark in the world Humans score 100%, AI <1% This human-AI gap demonstrates we do not yet have AGI Most benchmarks test what models already know, ARC-AGI-3 tests how they learnGIF
