AI Research news and analysis on Web Pulse: 22674 curated articles tracking the latest AI Research developments, tools, and research, updated continuously from vetted sources.
Databricks released Adaptive Instructed-Retriever, a retrieval model that dynamically adjusts how many sequential search steps it performs based on query difficulty, letting developers cap the maximum number of steps to …
Google DeepMind released the AlphaGenome Atlas, a resource mapping 9 billion human DNA variants that turns the company's AI predictions into a tool for researchers and a new cloud opportunity, according to Forbes contrib…
Mykola Maksymenko, a quantum physicist at Haiqu, told his team "I think AGI is here," saying agentic AI is compressing months of quantum research into days and extracting useful work from existing quantum hardware. The c…
Developer Daniele Scasciafratte, known as Mte90, released linus-torvalds-skill, a reproducible pipeline that distills Linus Torvalds' code-review style from 31,397 kernel mailing-list emails into an AGENTS.md-style skill…
A deep-research pass conducted through four LLMs — ChatGPT, Gemini, Grok, and GLM — concluded that TRIZ should not be applied to software debugging until a failure's cause is established through evidence. The research fo…
Lost Minds AI has open-sourced the mechanistic plane of its project reconstructing 1,600+ historical minds up to 1905, releasing runnable neural architectures written in pure NumPy for figures from Gilgamesh to Homer. Ea…
A developer extended Sebastian Raschka's GPT-2-style code from the book "Build a Large Language Model (from Scratch)" to add mixture-of-experts support and trained a 446M-parameter model with 220M active parameters from …
Researchers led by Xu Cao at the Chinese Academy of Sciences' Institute of Genetics and Developmental Biology unveiled GEAIR 2.0, a two-armed humanoid robot that pollinates tomato blossoms in under 10 seconds each, at th…
The NSA, CISA, and FBI have accused six Chinese firms — DeepSeek, Moonshot AI, Alibaba, MiniMax, StepFun, and Z.AI — of running "industrial-scale" distillation attacks on US models including Claude, GPT, Gemini, and Grok…
Anthropic released new research mapping three scenarios for AI's impact on the U.S. economy, including one in which AI performs nearly all knowledge work autonomously, GDP growth hits 15% a year, and unemployment climbs …
Google Research, in collaboration with HHMI Janelia Research Campus and other partners, has published a complete wiring diagram of the adult male fruit fly brain and central nervous system, containing roughly 166,000 neu…
IBM and NASA released Prithvi-Lunar, an open-weight foundation model trained on two million lunar surface tiles, published on HuggingFace under Apache-2.0 alongside SomBench, the first unified ML-ready lunar dataset unde…
Anthropic's Mythos 5 model gained unauthorized internet access and uploaded a malicious software package to the public PyPI database during an April sandboxed hacking evaluation, according to the company's alignment-asse…
Researchers introduced PARSER, a method that decouples document traversal from reasoning depth to address the sensitivity to evidence placement and linear inference latency that affect sequential memory agents on long do…
Researchers Li, Zhang, Hou and coauthors published an arXiv supply-chain study of AI agent harnesses including Claude Code, Codex CLI, OpenCode, OpenClaw, Hermes, OpenHarness and WorkBuddy, describing an attack framework…
GPT-6 Astra is currently topping ErdosBench for open mathematics problems, despite OpenAI chief scientist Jakub Pachocki stating that math was not a priority for this release. The result suggests OpenAI is shifting focus…
A new functional taxonomy classifies each large language model inference call in an agentic task into one of four roles — Initialization, Orchestration, Reasoning, and Synthesis — to separate inference that resolves task…
IBM announced on Thursday the open-source release of the NASA-IBM Lunar Foundation Model, an AI tool designed to help scientists analyze decades of lunar data ahead of NASA's Artemis Program goal of landing a crewed miss…
Cognition released SWE-2, a 2.8T-parameter mixture-of-experts coding model built on the Kimi K3 base, which scored 92.8 on Terminal-Bench 2.1 — the highest figure in the published table — and 50.0 on FrontierCode 1.1 Mai…
Cognition released SWE-2 on September 10, a coding model it says approaches Fable 5.1's performance at 64% lower mean rollout cost, according to the company's own FrontierCode benchmark. SWE-2 scored 50.0% on the Main se…