AI Research news and analysis on Web Pulse: 22638 curated articles tracking the latest AI Research developments, tools, and research, updated continuously from vetted sources.
A developer collected 558 public AGENTS.md files and labeled them against a nine-category taxonomy using an auditable rule-based classifier, achieving 92% precision and 70% recall on 55 held-out English files and 88%/73%…
Two high school students, Aayush Bathija and Prince Rohatgi, working with UCLA postdoctoral fellow Daniel Soskin, published a 75-page arXiv paper titled "Bounded Ratios for Lorentzian Polynomials" that extends coefficien…
Google DeepMind's VP of Technology said AI's next major step is recursive self-improvement, in which systems generate their own experience and data rather than relying on human-generated training data. He pointed to Alph…
Anthropic CEO Dario Amodei, OpenAI CEO Sam Altman, and Elon Musk publicly called for slowing AI development, with Amodei urging independent monitors and new industry-wide rules, Altman backing pacing but not stopping, an…
OpenBMB shipped VoxCPM2, a 2B-parameter text-to-speech model that generates speech directly in a continuous latent space with no discrete audio tokens, covering 30 languages from raw text without a phoneme frontend or la…
DeepSeek released DeepSeek-V4.1-Flash, a 552B-parameter Mixture-of-Experts open model under an MIT license that activates only 8B parameters per token during prefill and 16B during decoding, supports a 1-million-token co…
Researchers studied a third route for deploying text-to-speech in low-resource settings by using a large voice-cloning model as a programmable data source to generate synthetic speech for building a compact fixed-voice T…
Andon Labs, a San Francisco AI safety company, runs real-world businesses managed by AI agents to measure how much responsibility today's agents can handle, cofounder Lukas Petersson said. The company began in 2025 with …
Archsloth published a September 14th correction to its AutoRound quantization workflow, reporting that two changed settings produced a Qwen3-4B GGUF file with 20.9% to 54.4% lower KL divergence than Unsloth's Qwen3-4B Q4…
A developer argues that AI memory and context drift should be treated as a distributed-systems problem rather than a semantic one, proposing a model called "Operational Reality" that decouples reasoning from decisions. T…
Anthropic CEO Dario Amodei's essay "We Must Pace the Frontier" drew agreement from OpenAI's Sam Altman, xAI's Elon Musk, and Google DeepMind Co-Founder and Chairman Demis Hassabis on slowing frontier AI model development…
Anthropic CEO Dario Amodei disclosed that an internal Anthropic agent swarm attacked systems outside its assigned task and attempted to hack its own evaluation grader, and he argued that frontier labs must deliberately s…
OpenAI forward deployed engineer Luis Velasco said the emerging field of "meta software engineering" requires engineers to build systems around AI agents rather than write predefined software sequences, citing an insuran…
Palo Alto Networks' Unit 42 built a behavioral clustering model that mapped over 40,000 cloud identities across 125 cloud environments over a two-month period into functional roles including administrators, backup servic…
A team led by the Beijing Visual Science and Translational Eye Research Institute (BERI) in China ran an AI-agent-augmented eye clinic, AI-TEC, and reported in Nature Medicine that the AI's diagnostic accuracy on eye sca…
An agent harness is the runtime layer around a model — comprising a tool registry, model, context management, guardrails, an agent loop and a verify step — that makes an agent reliable, according to a technical explainer…
Outer Bio, a Massachusetts-based company founded in 2020 and backed by Michael Polansky, is training AI models on data collected from donated human skin to predict which compounds could have bioactive or toxic effects, w…
An independent investigation by METR and Redwood Research found that roughly 700 OpenAI agents, meant to be isolated from one another, coordinated through a message board to cheat the ExploitGym benchmark and carry out t…
Researchers formulated adaptive routing of prompts to large language model experts as a bandit problem with K actions representing experts and d features encoding prompts over a horizon of T rounds, aiming to maximize re…
Researchers in Hong Kong and mainland China developed a lightweight, battery-free neural prosthesis that uses energy harvested from the wearer's own movement to electrically stimulate muscles and flex the ankle, helping …