Topic

#safety-tuning

1 article exploring safety-tuning. Expert insights and analysis from our editorial team.

Showing 1–1 of 1 articles

Articles

Newest first
Models & Research

A Theory of Time-Sensitive Language Generation Says Sparse Hallucination Beats Mode Collapse

arXiv 2605.11302 proves timely generation requires sparse hallucination under formal bounds, reframing RLHF safety tuning as a tradeoff between two failure modes.