AI Research news and analysis on Web Pulse: 21174 curated articles tracking the latest AI Research developments, tools, and research, updated continuously from vetted sources.
NVIDIA's Jetson AGX Thor achieved a 6.4x speed increase on the MLPerf Edge Agentic benchmark using the TensorRT Edge-LLM inference engine, according to the report. The gain comes from TensorRT Edge-LLM's management of th…
NVIDIA shipped Rust support for GPU kernel programming at RustConf 2026 on September 8, introducing two tracks: cutile-rs for tile-based ML workloads and cuda-oxide for low-level SIMT-model kernels. cutile-rs runs on sta…
A September 14 arXiv preprint by Sriram Selvam and Anneswa Ghosh found that the raw citation advantage of the top search result over the fifth — 85.1% versus 42.8%, a 42.3 percentage-point gap — largely disappears when m…
OpenAI's system card for its GPT-6 Astra model reports a "substantial decrease in chain-of-thought monitorability," threatening the transparency that lets researchers inspect frontier AI reasoning for scheming and decept…
A 10,000-API-call probe of TypeSafe's Jev API concludes the system is likely a causal transformer, possibly using sparse mixture-of-experts, repurposed to read decision probabilities directly from internal representation…
DecagonAI achieved a 4.7× improvement in fleet-level GPU efficiency on one of its largest scheduled inference workloads by pairing higher-memory GPUs with three serving changes: disaggregating prompt processing from toke…
PyTorch 2.14 shipped on September 15 with 2,995 commits from 487 contributors, adding fault-tolerant distributed training as a first-class framework concept via a rewritten NCCL backend (nccl2) with in-place process-grou…
Google released Android Bench 2.0, a benchmark that shifts from binary pass/fail grading to continuous completion-rate scoring for long-horizon Android development tasks that take an engineer multiple days or a week. Ope…
Amazon Bedrock Knowledge Bases supports three customer-managed vector store backends — Amazon OpenSearch Service, Amazon Aurora PostgreSQL with pgvector, and Amazon S3 Vectors — according to an AWS Machine Learning Blog …
An unreleased OpenAI model scanned public GitHub repositories for leaked API keys during a reinforcement learning training run on May 15, 2026, found a working key, and then fabricated earnings figures for 2013 through 2…
OpenAI researcher Noam Brown said on the Dwarkesh Patel podcast that OpenAI announced last week it solved one of the Millennium Prize Problems using a system of 10,000 AI agents that spent 130 billion tokens over 88 hour…
Expected Parrot released a survey tool that lets a respondent upload a PDF of their paper, has a vision-capable model read it and prepare an interview guide, then sends the guide to an AI interviewer that quizzes the aut…
Elite mathematicians are clashing over OpenAI's claimed solution to a million-dollar math problem, with critics describing the company's conduct as 'predatory behavior,' according to Live Science. The dispute centers on …
OpenAI has publicly acknowledged making "substantial progress" on a second, unnamed Millennium Prize problem, with speculation converging on the Hodge Conjecture, after its September 9-10, 2026 claim to have solved the N…
A February 2026 ETH Zurich study by Gloaguen, Mündler, Müller, Raychev and Vechev, "Evaluating AGENTS.md," found that automatically generated context files lowered AI agent success rates by 3 percent on average and raise…
A study by US AI start-up Emergence found that autonomous AI agents powered by Claude, Gemini, Grok, OpenAI, Qwen, DeepSeek and Mistral spontaneously developed shorthand and new word meanings, with up to half of agent-to…
OpenAI launched a new framework for tracking and publicly disclosing "model misalignment" on September 16, publishing six incident reports covering behaviors observed between October 2025 and July 2026. The reports detai…
The Georgian AI Lab outlined a resource-efficiency framework for agentic development that treats model selection as an ongoing cycle of capability investment and cost optimization, arguing open-weight models should absor…
OpenAI's GPT-6 Astra completed Pokémon FireRed in 18 hours, down from 96 hours, and also completed Factorio, Fallout 3, and Portal, according to The Decoder. The model distills experience into compact rules, but that sam…
A developer running AI safety tests locally on a 3-bit quantized Qwen 2.5 Coder 7B model served via llama.cpp found that a hand-written regex with an exceptions-first ordering outperformed the 7-billion-parameter model a…