A new framework gives robotic agents human-style long and short-term memory, letting them refine prompts from past successes and failures without retraining.
A new mathematical analysis reveals why the simplest passive walker stays stable, tying eigenvalues of the return map directly to mechanical energy conservation.
GatherMOS uses GPT-5 as a meta-evaluator, combining acoustic descriptors with DNSMOS and VQScore pseudo-labels to predict speech quality without heavy training.
A voice-based chatbot plus clinician dashboard uses chain-of-thought prompting to cover 94% of postoperative check-in questions with zero hallucinations across 200 samples.
Google DeepMind built a unified benchmark testing whether frozen vision backbones can predict what happens next, revealing video beats images and synthesis rivals masking.
A two-stage attack combines context rewriting with hidden-state-guided token edits, boosting jailbreak success by up to 37.74% and transferring across models.
A new research paper proposes replacing static trajectory judges with an active reasoner-actor system that probes the GUI to verify whether agent tasks actually succeeded.
A new hybrid attention design unifies linear RNN memory and sliding window attention under one softmax, matching Transformers with fewer full-attention layers.
New empirical work argues sharp minima are not inherently bad. They often reflect a model learning a more complex function, and can beat flat minima on generalization and reliability.
A new selective freezing method preserves an instruct model's English abilities while teaching it low-resource languages, cutting forgetting from 20% to 3%.