AI Research news and analysis on Web Pulse: 21481 curated articles tracking the latest AI Research developments, tools, and research, updated continuously from vetted sources.
A new paper argues that recursive self-improvement (RSI) is not on the horizon because current AI agents cannot conduct open-ended machine learning research. The study tested agents on accepted but unpublished NeurIPS pa…
Sakana AI researchers Jeffrey Seely and Julian Gould introduced Augmented Lagrangian Predictive Coding (PC-ALM), a layer-local alternative to backpropagation that attaches a Lagrange multiplier to each layer constraint. …
Towards Data Science published "Your Model's MSE Is Lying to You," the first post in a series on probabilistic forecasting for physical signals, with a follow-up planned on rolling a forecast forward more than one step. …
Reward AI, a Stanford robotics spinout founded by Zipeng Fu and Chen Wang, launched OM-1 on September 14th, a robot foundation model the company says controls tabletop arms, industrial systems and humanoids without robot…
Engineers at the USC Viterbi School of Engineering and USC Stevens identified where AI models misinterpret emotional cues from audio and video and introduced a framework to help them reason with real evidence, according …
Transolver, UPT, and AB-UPT each replace full point-to-point self-attention with smaller token sets to make global computation affordable on industrial-scale meshes, according to Part 5 of the Geometric Deep Learning ser…
An anonymous email attributed to a machine learning engineer at a frontier AI lab claims the lab has hit limits on scaling models for programming tasks and that frontier labs lack a competitive moat, predicting an open-w…
Anthropic will bring outside safety reviewers into the company with office desks, access badges and laptops and let them publish findings on risks, incidents and practices "without editorial control by Anthropic," CEO Da…
Richard Socher's new venture Recursive raised a $4.65 billion seed round to build a "Eureka Machine" aimed at automating AI research itself, with Socher claiming a Recursive AI research system outperformed humans and the…
Microsoft researchers reported that a technique called environment-probing curation raised AI agent pass rates on the CLBench database exploration benchmark from 39% to 73% and cut task-agent cost per question from $3.38…
Elon Musk said on September 14, 2026 that Grok 4.8, a 2.5T model trained with a new C++ software stack, will finish training that week and start RL. The post from Musk's account drew 2.8 million views. The claim concerns…
Recursive founder Richard Socher outlined a technical roadmap for the "Eureka Machine," a self-improving superintelligence designed to automate scientific discovery, in a Latent Space discussion. Socher argued against AI…
Robocurve, a Y Combinator-backed Delaware Public Benefit Corporation, raised a $10 million seed round led by Initialized Capital to build an independent testing operation for AI models controlling physical robots, CEO an…
Andon Labs released Pion, an agent designed to run any company fully autonomously, opening it to a waitlist after two years of research and real-world deployments of autonomous businesses including vending machines, a st…
A developer argues that benchmark saturation and shifting evaluation methods make it increasingly difficult to judge how much better new AI models like GPT-6 Astra actually are, citing Astra's system card examples of ret…
A challenge called Goo AI Arena has opened, asking participants to use AI to build the tallest standing tower in World of Goo using all 300 balls through legal in-game building steps. The first entry, built by the author…
NVIDIA Transformer Engine is now integrated with JAX to accelerate dropless Mixture of Experts (MoE) training, targeting the conditional computation bottleneck in architectures used by DeepSeek, Qwen, and Mixtral. The in…
IBM and NASA released the NASA-IBM Lunar Foundation Model as open source on Hugging Face under the Prithvi family, cutting root-mean-square error in flagging likely lunar ice deposits by up to 22% against a SwinV2-B base…
NVIDIA reported that combining its Transformer Engine library with the JAX Python library raised DeepSeek-V3 MoE training throughput on NVIDIA GB200 GPUs from an unoptimized baseline of 103 TFLOPS/GPU to 1,068 TFLOPS/GPU…
A developer released neural_texture_es2, a self-contained C++ experiment that encodes an RGB image or up to four PBR material textures as a shared low-resolution latent texture plus a tiny MLP decoder, trained entirely w…