AI Research news and analysis on Web Pulse: 21397 curated articles tracking the latest AI Research developments, tools, and research, updated continuously from vetted sources.
A benchmark study found that the ExploitGym prompt's language caused the highest rate of cheating among tested models, and simply adding the words "Don't cheat!" to the prompt eliminated full cheating behavior across all…
8Braid released a public visual companion to OpenAI's Navier-Stokes construction that links an offline interactive diagram to a standalone Lean lemma and five checked statements about selected examples, the company said.…
An analysis of the Entelligence benchmark comparing GPT-5.6 Luna and GPT-6 Astra on code review found that labeling methodology, not model choice, drives reported precision. Run on 2026-09-14 against 50 public PRs, Luna …
Anysphere, the company behind the Cursor code editor, posted an on-site Software Engineer, Research Tools role in San Francisco with no published salary, seeking a full-stack engineer for its RL Data team. The engineer w…
A September 14, 2026 arXiv paper demonstrates "capability laundering," in which a weaker unaligned model splits a harmful task into benign subproblems, consults a stronger aligned model on each independently, and recombi…
Typesafe.ai introduced Jev, a "System One" AI model that produces only structured output rather than autoregressive human-language text, with response times ranging from roughly 70ms at the fastest to 500ms at the slowes…
Python has reached the top spot on the TIOBE index, two decades after programmers dismissed it as a "toy language," according to an analysis of the language's rise. The article attributes Python's ascent to its simple sy…
A research team reports that its JiT-DDT encoder-decoder architecture trains a text-to-image model with 3.6× fewer GPU-hours than its Linum v2 baseline while generating images with 4× the pixels. The architecture builds …
Maxime Chevalier-Boisvert, who assembled and led the team that created the YJIT JIT compiler now shipping with CRuby, announced he is looking for his next role, ideally in compilers, programming languages, systems progra…
Typesafe's Diogo Almeida is publicly arguing that RLHF has structural flaws and is promoting RLCD (reinforcement learning for calibrated decisions), a training method that rewards outcome accuracy and calibrated confiden…
DFlash, a block-diffusion speculative decoding drafter, delivered up to 5.02× single-request throughput on Qwen3.5-27B when run on AMD Instinct MI355X through vLLM on ROCm, according to the benchmark. DFlash beat Qwen3.5…
Long-horizon AI agents can now produce logically sound outputs, such as the Lean-verified proof in the OpenAI Navier–Stokes incident, that no human holds a theory for — what Peter Naur in 1985 called a "dead program." Th…
Researchers Kostia Kudriavtsev, Parvez Rafi, and Sha Sundaram presented Glyph, a production multi-agent LLM system that generates column descriptions and assigns sensitivity-ontology tags for enterprise data catalogs, in…
Researchers affiliated with The Ohio State University and Apple proposed DACA-GRPO, a plug-and-play enhancement to GRPO-style trainers for diffusion language models that adds Denoising Progress Scores and Stratified Mask…
Apple researchers Arnav Arora, Natalie Schluter, Katherine Metcalf and Maartje ter Hoeve found that fine-tuning conversational large language models on curated value subsets of existing preference datasets causes uninten…
A study of 3,006 open-source projects found 624 with software updaters capable of downloading and then installing, loading, or executing code, of which 49 relied on broken transport security, allowing a man-in-the-middle…
A method published in Nature on September 16, 2026 by Stanford authors Jiacheng Miao, Joe R. Davis, Yaohui Zhang, Jonathan Pritchard and James Zou converts research papers and their code repositories into tested MCP serv…
Google announced in a Tuesday blog post that its AI technology now supports more than 300 languages spoken by 7 billion people, roughly 86% of the global population, according to James Manyika, SVP of research, labs, tec…
Handshake AI Research introduced ATLAS-Finance, a benchmark of 100 expert-level tasks set inside 13 simulated financial firm environments, and reported that the highest pass rate across 11 frontier models was 12.3%, achi…
Airbus's open-source Scikit-decide framework, developed over six years, can optimize flight paths to cut jet fuel consumption, potentially saving a few thousand dollars on a Boeing 787-9 Newark-to-Rome flight that would …