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.
The Shanghai Artificial Intelligence Laboratory released Atria Dawn Preview, a preview version of a new-generation agentic model built on a 744B-parameter MoE foundation model. The model is designed for research and engi…
A developer has released OpenArch, a repository of hand-written PyTorch implementations of modern open-source LLM architectures, cataloged against Sebastian Raschka's LLM Architecture Gallery. The project currently provi…
A study of 256 private coding tasks found no reliable overall performance advantage for vendor-native harnesses over neutral third-party harnesses on the same models, with Opus 4.8 scoring 48.8% versus 50.0% and GPT-5.5 …
Occamy-1.0, an open 35B co-work agent model trained from Qwen3.6-35B-A3B, lands at the low-cost knee of the cost-performance Pareto frontier across four representative co-work benchmarks, according to its arXiv paper. Th…
Developer ErnestoSOFTWARE released an interactive browser project that simulates a fruit fly's male central nervous system connectome playing chess against a human opponent, rendering live neural activity across the opti…
A developer benchmarked DFlash, a diffusion-model-based speculative decoding drafter from Z-Lab, against Gemma-4-12B-it's native Assistant MTP model on an RTX 3060 with 12GB VRAM using llama.cpp. DFlash failed to outperf…
LG AI Research unveiled "expert" AI models built on its EXAONE large language model at the LG AI Talk Concert 2026, with specialized versions for manufacturing, science and finance. The institute introduced EXAONE Tabula…
A developer's invoice-triage agent that passed 22 test cases three days in a row later filed the same PDF under two different vendors, illustrating what IBM Research calls the "consistency gap." In a new arXiv paper, Eve…
Software developer and blogger Jason Gorman published a taxonomy of out-of-distribution (OOD) errors that automated quality gates miss in coding-agent workflows, extending his "Special Theory of Autonomous Agent Reliabil…
OpenAI announced in 2019 that it would not publicly release its GPT-2 text generator, citing the risk of malicious use, a break from its normal practice of publishing full research. OpenAI research director Dario Amodei …
Programmer and machine learning specialist Michał Żarnecki published the first part of a three-part series describing how he automates software development and project maintenance using Claude Code, MCP servers, and AI a…
A 3D visualization posted to whip.run illustrates how DeepSeek Flash v4.1 differs architecturally from a decode-only transformer. The item drew 3 points and 0 comments on Hacker News.
LG AI Research unveiled a suite of industry-specific EXAONE systems at its AI Talk Concert 2026 in Seoul on Monday, including EXAONE Tabular and EXAONE Omni-Inspect for factory defect prediction, EXAONE Discovery for mat…
A developer's experiment presented as a poster at CppCon tested whether a small, local LLM integrated into the Clang/LLVM compilation pipeline could match the -O3 optimization flag, using five deliberately inefficient C+…
A blog post explains the Residual-Quantized Variational Autoencoder (RQ-VAE), an architecture that extends the Vector Quantized Variational Autoencoder (VQ-VAE) introduced in arXiv paper 1711.00937 to capture hierarchica…
Arena's September 11th WebDev snapshot ranked OpenAI's GPT-6 Astra Max first with a score of 1800 from 2,281 votes, ahead of Anthropic's Claude Fable 5.1 Max at 1758 from 3,036 votes and Claude Opus 5 Max at 1687 from 12…
A pre-registered three-month randomized controlled trial of 133 practicing patent lawyers at eleven U.S. intellectual property law firms found that access to a custom AI drafting assistant raised benchmark patent draftin…
OpenAI CEO Sam Altman called for a federal framework setting consistent safety requirements for frontier AI labs, arguing that shared standards for misalignment, monitoring, and safety will produce better outcomes. Altma…
AWS released the Deception Benchmark, a public dataset and evaluation process that measures how well AI models distinguish genuine security vulnerabilities from code that looks risky but is safe, and the results show fal…
A bio-inspired deep learning framework achieved 92.2% accuracy in recognizing the odor perception of multi-molecule mixtures, according to a study published on arXiv as 2609.11958v1. The framework builds neural response …