A research paper argues that humans and large language models (LLMs) rely heavily on pattern matching for everyday causal reasoning. This challenges the view that human reasoning uses principled, cont
A new study finds that AI-generated language is harder for people to understand because models increasingly favor dense information over human-friendly communication. This matters for anyone reading A
A NeurIPS 2026 paper finds that deleting fewer than 0.05% of an LLM’s weights can match full fine-tuning when adding new capabilities. This matters because the method could make post-training far chea
The paper presents an independent safety evaluation of DeepSeek V4.1 Flash-1017, an open-weight language model with a new architecture, expanded training, and enhanced agentic capabilities. Submitted