List your company

Theory-of-Mind Reasoning on medieval castle

85.6Belief Inference Accuracy8B ↬ 405B
80.81682.05883.384.542Jun 2, 2025

Evaluation results

MethodLinks
8B ↬ 405BSolution Type=Ours, Backbone=Llama3.1, Scaling Setting=weak-to-strong control2025.0685.675.279.7
4B-wide ↬ 70BSolution Type=Ours, Backbone=Llama3.1, Scaling Setting=weak-to-strong control2025.0684.67378
8B ↬ 70BSolution Type=Ours, Backbone=Llama3.1, Scaling Setting=weak-to-strong control2025.0683.67578.7
4B-depth ↬ 70BSolution Type=Ours, Backbone=Llama3.1, Scaling Setting=weak-to-strong control2025.0683.36472.2
70B-zero-shotSolution Type=Raw, Backbone=Llama3.1, Scaling Setting=zero-shot2025.0682.657.668.3
70B-post-trainedSolution Type=Raw, Backbone=Llama3.1, Scaling Setting=direct post-training2025.06816571.8