Researchers introduced HALO, a hybrid adaptive latent-refinement method that improves frozen pretrained language models by selectively applying extra computation to tokens. On MMLU-Pro and GPQA-Diamon…
A new study testing 67 frontier language models from 21 providers found that combining multiple models rarely outperforms the single best model, with gains capped by a 'co-failure ceiling' where all m…
Researchers propose SuperThoughts, a method that compresses pairs of consecutive Chain-of-Thought tokens into single latent representations and decodes two tokens per step, doubling inference throughp…
Researchers at Duke University introduced Reward-Variance Policy Optimization (RVPO), a risk-sensitive alignment method that penalizes inter-reward variance to prevent language models from neglecting …