AI Research news and analysis on Web Pulse: 21294 curated articles tracking the latest AI Research developments, tools, and research, updated continuously from vetted sources.
Salesforce AI Research published DarwinX, a framework that evolves AI agent harnesses without modifying model weights, raising pass@1 on the 1,260-task WebArena-Infinity benchmark from 43.5% to 93.0% and cutting invalid …
Google DeepMind opened its first dedicated research hub outside its existing offices, an AI Campus in Seoul built with South Korea's Ministry of Science and ICT, as part of a year that also included a five-year genomics …
Google DeepMind announced the launch of the DeepMind Institute, a new interdisciplinary effort the company says is needed to understand the implications of artificial general intelligence as it nears. The institute's int…
Anthropic CEO Dario Amodei published a 3,800-word essay on September 12 calling for the AI industry to slow down, with OpenAI's Sam Altman, xAI's Elon Musk, and Google DeepMind's Demis Hassabis signing on within days, ac…
Demis Hassabis, co-founder and CEO of Google DeepMind, publicly endorsed slowing down frontier AI development in September 2026 to let safety research catch up with advancing capabilities. His position follows the July 2…
Researchers submitted a paper to arXiv on 15 Sep 2026 proposing Episode-Normalized Conformal Prediction (ENCP), a method that rescales a nonconformity score by a policy's residual confidence and calibrates one maximum sc…
A paper submitted to arXiv on 15 Sep 2026 argues that agentic societies — collections of AI agents coordinating autonomously across trust boundaries on behalf of different principals — need a "social harness" for inter-a…
Researchers released ScienceBuddy, an interactive scientific research workspace built on a paradigm they call recursive-in-recursive self-improvement, submitted to arXiv on 15 Sep 2026. The system couples harness evoluti…
OpenAI's GPT-6 Astra spent 141 hours in a Minecraft test run by evaluation firm Vals AI, progressing further than any AI system Vals AI had previously tested before a Creeper explosion destroyed its chest and bed. Vals A…
IonQ, Oak Ridge National Laboratory, Nvidia, and the University of Tennessee developed DQAOA-GPT, a hybrid framework that uses a pre-trained generative AI model to synthesize efficient quantum circuits in a single pass r…
Google researchers, with collaborators from Google DeepMind, the University of Maryland, and the University of Virginia, published a paper describing Dream-RSI, a system that recursively improves an AI agent's exploratio…
AI Model Radar published measurements on 16 September 2026 showing that the standard rule-of-thumb for estimating context-cache memory in open-weights models can be off by a factor of four, and by roughly forty times for…
A new data page tracks release dates and training cutoffs for 20 current AI models across 8 labs, finding that only 9 of the 20 models have a training cutoff their lab actually publishes. The page lists models including …
Google Research introduced R4T, an RL-trained diffusion retriever that optimizes fan-out query sets for diversity and coverage, achieving 12x to 20x latency speedups over autoregressive language models. The 53.9M-paramet…
Tron founder Justin Sun announced the "Justin Sun Prize" on X on September 16, offering up to $1 million per "Pinnacle" mathematics problem to whoever solves it first, with no distinction between human and AI-generated p…
The Download newsletter highlighted AI's trillion-dollar gamble, citing University of Pennsylvania finance professor Jessica Wachter's assessment of AI's impact on the economy, alongside OpenAI's bid for biology data. Th…
IBM published a guide on building effective evals for AI agents, detailing how to design clear tasks, choose graders for reasoning, tool calls and final results, and build reliable eval harnesses. The guide notes agents …
A study titled "ImpossibleRubrics: Stress-Testing Generated Rubrics as Reward Signals" examines whether language model-generated rubrics used as reward signals for rubric-based reinforcement learning, LLM-as-a-judge eval…
Qwen3.8-Flash, running inside the closed-source host-enforced AIC software engineering runtime, passed all 17/17 canonical checks on the DeepSWE v1.1 updo-policy-alerting task at an operator-reported model/API cost of $0…
Google shipped Gemini 3.8 Live and an Extended Thinking variant that rank first on speech-to-speech and voice-agent benchmarks, and the models can now call tools mid-conversation without breaking the dialogue. The releas…