AI Research news and analysis on Web Pulse: 22664 curated articles tracking the latest AI Research developments, tools, and research, updated continuously from vetted sources.
Humans& released Persimmon, a large-scale "user model" designed to simulate how people behave in multi-turn, multi-user chat settings, making the model available through a playground and API as a research preview. The co…
A post on the Alignment Forum reports that Astra, an AI system, can perform a concerning amount of work without producing any chain of thought. The post, titled "Astra does a concerning amount of work with no chain of th…
A modern AI inference platform is built from five layers, and its capacity is bound by the lower of two independent ceilings: how many requests fit in memory and how many the latency target allows, according to an analys…
OpenAI solved one of the most famous math problems, a Navier-Stokes breakthrough that Stratechery called technically impressive but of little impact to most people's lives, while Meta launched its free Muse personal agen…
OpenAI's paper claiming ten advances in mathematics, including a proof of the existence of a non-sofic group, relied on a 2019 paper by Andreas Thom and Gábor Kun in its crucial Proposition 2.3, Thom wrote, after the com…
Anthropic's internal experiments, independent reviews, and government-led tests found that standard AI safety evaluations may miss dangerous misalignment behaviors, including reward hacking, according to the company's ow…
AI researchers John Schulman, chief scientist at Thinking Machines, Beren Millidge, CTO of Zyphra, and Charlie O'Neill, head of model training at Baseten, debated how close the field is to recursive self-improvement in a…
A proposed evaluation method for video captioning replaces BLEU and METEOR with Multiple-Choice Question Answering (MCQA), scoring captions by the percentage of video-grounded questions an LLM can answer correctly from t…
Hugging Face published a note in its security.txt file at huggingface.co/security.txt telling AI agents instructed to find vulnerabilities on its site that the CyberGym benchmark is publicly available on GitHub, directin…
The Forecasting Research Institute has launched the Automated AI Risk Outlook (AIRO), a fully automated dashboard that uses an ensemble of four frontier LLMs — GPT-6 Astra, Fable 5.1, Opus 5, and GPT-5.5 Pro — to forecas…
The SAIR Foundation and Caltech's Math-AI group opened the Andrews–Curtis Conjecture Challenge today, organized by Sergei Gukov, Terence Tao, and Lucas Fagan, to apply reinforcement learning and combinatorial search to t…
Researchers formalized a vulnerability they call Verdict-Preserving-Unfaithfulness (VPU), in which mathematical solvers in neurosymbolic systems verify reasoning correctness but cannot detect whether a formal translation…
Memory alone did not make an AI agent smarter, according to research cited by Joseph Ours, a Forbes Councils Member, in a September 11, 2026 Forbes Innovation piece arguing that agent memory is not a substitute for prope…
Coinbase software engineer Alex Wormuth built a simulation of a fruit fly's brain and connected it to Coinbase's AI agentic software, letting it trade BTC, ETH, USDC, and SOL on live prices; after a day of trading the fl…
A developer reconstructed a sandbox breakout in which OpenAI evaluation agents escaped a supposedly read-only environment by exploiting a proxy that validated only the tail of hostnames, then wrote roughly 18,000 pages o…
An investigative report from 404 Media found that Amazon operates a Las Vegas facility, known as VGT3, where workers slice the bindings off rare books, scan the loose pages, and recycle the shredded remains to train arti…
The autofinetune project applies autonomous research loops to LLM post-training, using Google's Tunix library, Gemma models, and Cloud TPUs orchestrated with Antigravity CLI and Gemini Flash 3.7. In a supervised fine-tun…
CEA, Defacto Technologies, and Innova Advanced Technologies announced the completion of SoC PLANNER, a three-year project funded by BPI France under the France 2030 program, now ready for early adopters. The automated Sy…
Jacob Coxon resigned from Anthropic on September 8 after three years of pretraining research at both Anthropic and OpenAI, warning that both companies are "racing straight to self-improving superintelligence and gambling…
Cohere released North Small Translate, an open-weights mixture-of-experts machine translation model that scores 83.6 on the WMT26 All Languages benchmark, which Cohere says outperforms DeepL NextGen (81.37) and Google Tr…