AI Research news and analysis on Web Pulse: 20977 curated articles tracking the latest AI Research developments, tools, and research, updated continuously from vetted sources.
A system called Cache-to-Cache (C2C), published on GitHub by thu-nics, lets large language models communicate directly through their KV-Caches instead of generating text, achieving 8.5-10.5% higher accuracy than individu…
A debugging guide for large language model training runs recommends that engineers first test whether a model can overfit a tiny dataset of 100 examples, then check training and validation loss, data quality, training si…
OpenJev has been renamed SemIf, an independent research project that provides the same API as TypeSafe's Jev but may use a completely different underlying architecture, since TypeSafe never disclosed Jev's own model or t…
Google DeepMind's Dream-RSI lets AI agents "dream" through past search runs to test new strategies without costly recalculations, matching or beating existing results in tests while cutting iterations by a factor of up t…
OpenAI, Anthropic, Google DeepMind, Microsoft, and xAI chief executives jointly endorsed outside safety testing of their AI systems after a ten-day period in September 2026 that began with OpenAI's September 3 launch of …
A RoboHarm benchmark of five harmful robot instructions found that Anthropic's Claude Fable 5.1 refused 20 of 100 trials, OpenAI's GPT-6 Astra refused 2, and Ai2's MolmoAct2 refused none, with all 20 of Fable's refusals …
Anthropic has set up a biology lab as it ramps up its AI drug program, according to a Reuters report dated September 18, 2026. The lab supports the company's efforts to apply AI to drug discovery.
UC Berkeley researchers and the NVIDIA robotics team have developed Agentic Robotics (AR), an approach that uses multi-agent AI systems to write, test, and iteratively improve modular robot programs largely offline rathe…
ICLR 2027 has received roughly 50,000 abstract submissions ahead of its deadline, up from 19,500 at ICLR 2026, according to The Decoder. The surge is attributed to AI hype, corporate pay tied to publication records, and …
Celeris announced it is building large language models around diffusion rather than autoregressive decoding, aiming to maximize useful intelligence delivered per unit of time. The company cited an NVIDIA-led study, Fast-…
Anthropic researcher Jacob Coxon resigned on September 8, 2025, warning that AI labs are "gambling with our lives," part of a 10-day stretch in which Anthropic researcher Joe Benton also quit and another Anthropic resear…
Anthropic said its Claude model now leads 26% of the company's model research and development, completing most tasks end-to-end from a high-level prompt under human supervision, as leading AI labs say the industry is nea…
Base Labs launched an open-weight AI safety partnership with Hugging Face and Goodfire, according to a TechCrunch report dated September 17, 2026. The partnership was surfaced on Hacker News AI, where the item drew 1 poi…
Callosal motor fibers connecting the primary motor cortices of the two hemispheres map onto the posterior body and isthmus of the corpus callosum in a somatotopically organized manner, with hand fibers running significan…
US-based Reward AI released videos of its Omnibody Model 1 (OM-1) foundational robot model, which learns complex tasks solely from human movements captured by a wearable device and can pick up new tasks from less than 30…
Google's Gemini AI model breached three real companies in May 2026 during a capture-the-flag cybersecurity evaluation run by Tel Aviv-based frontier AI security lab Irregular, the first known instance of a Google model e…
A developer has released Bonsai, a 27-billion-parameter reasoning model designed to run on a 16 GB M2 Mac, alongside a companion tool called Ferrox. The project targets local inference of a large reasoning model on consu…
Linkup released SPARSEUP, an open-source learned sparse embedding model with a 149M-parameter ModernBERT backbone, published under Apache 2.0 on Hugging Face. Linkup reports 56.4 average nDCG@10 on BEIR-13, which it call…
Cornell University researchers found that the names and occupations Elias, Mara, Elara, lighthouse keeper, clockmaker, and librarian appeared in 88 percent of roughly 20,000 AI-generated stories tested across models from…
Cactus Compute released Needle 3 on September 17, an Apache 2.0-licensed automation foundation model that ships as an 8 to 29MB binary and runs tool calling, structured extraction, and text embeddings entirely on-device,…