Loading activity…
@huggingpapers
Tweeting interesting papers submitted at http://huggingface.co/papers. Submit your own at http://hf.co/papers/submit, and link models/datasets/demos to it!
Science
Anthropic says 17 of 211 designs met binding criteria; it did not test any for neutralization in cells.
AI
Anthropic reports 17 of 211 lab-tested designs met binding criteria, but none were tested for toxin neutralization.
AI
HuggingPapers says Act3D uses outcome-only reinforcement learning to decide when and how to move its camera.
AI
HuggingPapers says humans solve about 100% of its tasks, while GPT-5.6-Sol solves only about 17%.
AI
HuggingPapers says the ICT-CAS benchmark has 328 test cases for long-horizon harness engineering of coding agents.
AI
HuggingPapers claims state-of-the-art zero-shot forecasting across three major benchmarks, with 21 quantiles.
AI
HuggingPapers describes the two-stage system as downscaling precipitation, winds and sea level pressure to 3 km.
AI
HuggingPapers says the deployment model is for SSD picking and includes ONNX graphs and TensorRT engines.
AI
HuggingPapers says the quantization framework matches FP32 accuracy with 6-bit states, cutting total serving memory by up to 68.7%.
AI
A post says success on RoboCasa GR-1 rose from 63.3% to 78.7% as context grew from 0.0 to 19.2 seconds.
AI
HuggingPapers says the method aligns KV caching and conditioning with training to address drift beyond autoregressive video models’ training horizon.
AI
HuggingPapers says FoldingCorpus post-training raised macro-average accuracy from 45.09% to 48.33% across 10 reasoning benchmarks.
AI
HuggingPapers describes the 27B agent as trained on verifiable machine-learning and algorithmic tasks.
AI
HuggingPapers says the method extrapolates RL-induced representation residuals, with code and checkpoints planned for release.
AI
HuggingPapers describes one multimodal model acting as both teacher and student, learning without external supervision.
AI
HuggingPapers describes the reported gain as a post-training effect on unrelated decisions, calling it a “behavioral shadow.”
AI
The InternW0-Δ announcement claims the model outperforms prior methods and says its code, weights and data are open-source.
AI
HuggingPapers describes research in which averaging two texts’ numerical representations lets a language model predict both continuations at once.
AI
HuggingPapers describes a five-platform sandbox where hybrid computer-use agents explore real apps, implement them and visually check their own builds.
Technology
Caixin reports that CXMT claims its new technology increases chip output and improves storage density, power consumption and cost control.
