Tag
The paper challenges the assumption that cosine alignment between supervised latents and visual targets improves accuracy in vision-language models, finding a strong negative correlation. It introduces PRISM diagnostics revealing that answers are decoded downstream from latents, not within them, and that the auxiliary loss reshapes the language model via shared parameters.
Introduces Future-L1, an interleaved latent visual reasoning framework that improves video event prediction by maintaining visual semantics in latent space. Achieves state-of-the-art results on FutureBench and TwiFF-Bench benchmarks.