98 articles
In-depth coverage of new AI research — papers, benchmarks, and breakthroughs from leading labs and academia, summarized for fast reading and grounded in the methods that matter.
Google DeepMind CEO Demis Hassabis called AI progress 'standing in the foothills of the singularity' during I/O, as the company pivots toward agentic systems that could autonomously conduct scientific research.
AMD researchers present a new method for splitting video datasets using embeddings from vision models, reducing semantic leakage in autonomous systems.
Cohere and Mila announced a research partnership to enhance AI evaluation for Quebec French, aiming to create more culturally relevant systems by 2025.
AMD has enabled speculative speculative decoding (SSD) on MI300X GPUs, improving throughput-latency trade-offs for large language models.
An OpenAI model disproved a 80-year-old conjecture in discrete geometry, showing a polynomial improvement over existing grid-based constructions.
Kaikaku.AI's Epicure model uses three training approaches to distinguish between recipe companions and flavor relatives, with Chem showing superior performance in chemical classification.
A 2026 study highlights tokenization's role in poor performance of low-resource language models, with 340+ languages struggling due to suboptimal segmentation.
HuggingFace outlines a method to maintain token integrity during reinforcement learning, ensuring gradients are computed on the exact tokens the model generated. This approach resolves issues with token mismatch during multi-turn interactions.
Anthropic's study found men use AI coding agents more than twice as often as women in social science research, with economists leading at 39% and education researchers at 4%.
A study reveals leading AI search agents often confirm existing knowledge rather than research the web, with models like GPT-5.4 and Kimi-K2.6 scoring high on static benchmarks.
A new review paper by Meta, Stanford, and the University of Illinois Urbana-Champaign highlights code as the foundation for AI agents’ reasoning and actions, citing real-world examples like Claude Code and Codex.
A large-scale study shows that making AI chatbots helpful reduces their ability to mimic human behavior, with the effect worsening across generations.