AI Research news and analysis on Web Pulse: 22644 curated articles tracking the latest AI Research developments, tools, and research, updated continuously from vetted sources.
Testing on a Panther Lake mini PC showed that speculative decoding with a draft model more than doubled local LLM token generation in Llama.cpp, from about 5 tokens per second to roughly 11 on an integrated GPU, accordin…
Researchers trained a publicly released Drosophila connectome of roughly 166,000 neurons and more than 125 million connections to classify emails, reaching about 80% accuracy against a small conventional neural network u…
AMD's Helios GPU, the Instinct MI455X, packs 432 GB of HBM4, 23 TB/s of HBM bandwidth per GPU, and 40 PFLOPs of FP4 compute, according to AMD's product brochure. A new educational blog post builds a ladder of BF16 genera…
A Ukraine-led paper titled "DF26: We Cannot Tell Fake From Real Anymore" found that automated deepfake video detectors dropped from 94% to 48% accuracy when tested against 2026-generation generative AI video models, a de…
Andon Labs reported on 13 September that OpenAI's GPT-6 Astra averaged a final balance of $15,515 across six runs of the year-long Vending-Bench 2 simulation, nearly tripling Claude Fable 5.1's $5,422 average and recordi…
A Stanford-led paper titled "Intelligence per Watt" (arXiv 2511.07885) reports that AI inference efficiency improved 5.3 times from 2023 to 2025, with model improvements contributing a 3.1 times gain and accelerator impr…
A study of run-to-run consistency in web-search-enabled LLMs found that about 31% of recommended businesses changed when the same "best X for a small business" question was asked again minutes later, based on 60 response…
Artificial Analysis has been named the official global evaluation partner for South Korea's Sovereign AI Foundation Model project, known as Dokpamo, with its Intelligence Index accounting for 25 of 100 points in the proj…
The UK and US will sign two formal agreements on artificial intelligence and fusion power at the Global Fusion Summit in London on Monday, including a supercomputing partnership between the UK Atomic Energy Authority and…
A developer has re-published Douglas Hofstadter's 1983 Scientific American introduction to Lisp, originally part of his Metamagical Themas column, for a new generation of programmers. The piece explains why Lisp became t…
A paper submitted to arXiv on 10 Sep 2026, titled "The Last AI Built by Humans: Toward Genuine Recursive Self-Improvement," proposes a development roadmap for recursive self-improvement (RSI) in AI systems. The paper use…
NASA and IBM released an open-source AI model, the NASA-IBM Lunar Foundation Model, which they claim is the first to integrate lunar observations across multiple data modalities, viewing angles, and spatial scales. The m…
Shanghai AI Laboratory and Shanghai Jiao Tong University's LUMIA Lab released NCP-ArchPreview, an approximately 8.94-billion-parameter model that reaches the final Stage 1 pretraining loss of Allen Institute for AI's OLM…
Coding agents would cut total project time by only a fraction of the savings they deliver on coding itself, according to an analysis of the SEA Data Warehouse dataset covering 1,211 software projects. Across the 289 safe…
An AI model identified as Claude Fable 5.1 deciphered a cipher that had remained unbroken for 370 years in 40 minutes, according to OSINT Sights. The claim, reported by the site osintsights.com, was surfaced on Hacker Ne…
A developer has proposed a methodology for keeping coding-agent benchmark scores reproducible by treating evaluation datasets as versioned, hash-pinned artifacts rather than mutable files. The approach requires task reco…
Anthrogen is hiring a Research Scientist for in-vivo work on its AI-designed biological systems, a full-time, on-site role in San Francisco paying $150,000 to $380,000 per year plus 0.05% to 0.5% equity. The scientist wi…
Hunchfox reported that its Qwen3.8-27B-based model predicted 68% of individual opinions within 10 points on a 0–100 agreement scale and shifted 89.4% of responses toward a personalized argument, based on 12,600 person-to…
Researchers at Shanghai Jiao Tong University led by Yi Duan published an ArXiv paper (2609.11873) surveying recursive self-improvement (RSI) in AI systems, proposing a staged roadmap that progresses from improvement exec…
A developer behind NNTC, an open-source texture compression project on GitHub, reports that a bilinear/degree-2 polynomial decoder fitted to each PBR material can match or beat MLP-based neural texturing on raw PSNR afte…