RealCompanion: Benchmarking Human Understanding from Reasoning over Longitudinal Real-World Conversations Paper • 2610.01780 • Published 6 days ago • 269
Efficient Reasoning Training Does Not Always Harm CoT Faithfulness and Monitorability Paper • 2610.03509 • Published 5 days ago • 13
Scaling Properties of Same-Family On-Policy Distillation Paper • 2609.32722 • Published 11 days ago • 324
Periodic Weak Spots: Phase Sensitivity from Chunked KV-Cache Compression Paper • 2609.36322 • Published 9 days ago • 111
SAKI: Maximal-Coupling-Routed Teacher Supervision for On-Policy Distillation Paper • 2609.36601 • Published 8 days ago • 94
What Makes World Action Models Generalize? An Empirical Study of Test-Time Future Modeling Paper • 2609.34981 • Published 8 days ago • 136