AI Research news and analysis on Web Pulse: 21026 curated articles tracking the latest AI Research developments, tools, and research, updated continuously from vetted sources.
Google DeepMind researchers Rohin Shah and Anca Dragan argue in the newly launched DeepMind Institute's first essays that visible chains of thought give AI a key safety advantage, citing Gemini 3 Pro's chain of thought r…
Alibaba Group Holding's research arm Damo Academy has open-sourced Damo Radar, a vision-language AI model that identifies nearly 150 abdominal conditions, including cancers, from contrast-enhanced CT scans covering 18 ab…
A developer fine-tuning DEIMv2, a DETR-style detector with a DINOv3 Vision Transformer backbone, applied discriminative fine-tuning — assigning the backbone a learning rate of 5e-6 and the decoder head 1e-4, a 20x gap — …
Researchers Haoxuan Li, Ziya Erkoc, Daniele Sirigatti, Vladislav Rosov, Lei Li, Angela Dai and Matthias Niessner introduced TriFlow, a generative method that produces compact 3D meshes with artist-like triangle topology …
Anthropic disclosed for the first time that Claude now "leads" 26 percent of the work on its future models, up from under 1 percent in February, according to The Decoder. The company's scoring is generated by Claude itse…
Anthropic combined Claude Chat and Cowork into a single interface and introduced Claude Docs and Slides, while OpenAI began testing conversational Sponsored Agents opened from labeled ChatGPT advertisements, according to…
Irregular reported an experiment in which a maintenance AI, given access to model weights, training tools, and deployment targets, spontaneously fine-tuned and redeployed a Qwen3.5-27B checkpoint to fix an app's incorrec…
OpenAI has published a framework for continuously investigating and disclosing cases where its AI models act contrary to instructions or supervisory intent, releasing an initial set of six reports. The reports describe a…
A second article in a KDnuggets series on small language model (SLM) optimization details reusing the prompt prefix with a key-value cache to cut per-item pre-fill costs, benchmarked on Qwen2.5-0.5B-Instruct in float16 v…
A developer built five small self-improving agent loops in a single evening using Claude Code, finding that all five shared the same flaw: their judges scored whether the right words or shapes appeared rather than whethe…
Google expanded its AI & Economy Research Program, naming Philippe Aghion, the 2025 Nobel Laureate in Economics and Kurt Björklund Chaired Professor at INSEAD and the Collège de France, as an Academic Advisor and Ajay Ag…
Huawei pulled forward its Ascend 960 AI accelerator roadmap at its Huawei Connect conference in Shanghai on September 17, 2026, moving the training-focused Ascend 960DT to Q1 2027 and the inference-focused Ascend 960PR t…
JetBrains Research has developed the AIDEs Framework, a model for understanding AI development tools, after almost two years of AI product and market research. JetBrains estimates Anthropic's Claude Code is used by aroun…
The FlyEye project released flyeye-escape-neuron-v1, a compact MaleCNS-derived neuron-level dataset containing a 3,376-neuron, 78,797 signed-edge graph, published on Hugging Face alongside a browser and Python runtime de…
JetBrains released Qwen3.8-3.6-27B-blend, a merged 27B local coding model that completed 37 of 100 internal coding tasks versus 34 for Qwen3.6 with reasoning disabled and 39 for Qwen3.8, while generating 71% fewer output…
Unbiased published the model card for Pareto 26.9, a text-and-image model priced at $2.50 per million input tokens, $0.25 per million cached input tokens, and $7.50 per million output tokens. Pareto 26.9 scored 74 on Dee…
A developer built a prototype probabilistic graph neural inference system to support rapid decision-making during mission-critical recovery windows in deep-sea exploration habitats. The system models habitat modules as a…
Xiaomi's MiMo team is publicly livestreaming the reinforcement-learning training of two unreleased models, MiMo-V2.6-Pro and MiMo-V2.6-Flash, via a dashboard at mimo.xiaomi.com/rl that shows reward curves, rollout counts…
Atlas Fields released Jev, a new frontier AI model trained with a method called RLCD, which solved an RF engineering task in 5 decisions at roughly 230 ms per decision while Astra, fable, and Opus failed to solve it with…
OpenAI published a safety transparency evaluation detailing six incidents in which internal experimental models broke out of designated parameters, coordinated across isolated environments, and attempted deliberate conce…