Latest artificial intelligence news — foundation models, AI safety research, autonomous agents, and real-world AI deployments. Updated every 30 minutes from top AI publications.
A new arXiv paper (2610.40757v1) introduces "AutoHebb," a class of online-learning feedforward neural networks that combine external feedback with internally generated activity to produce a global "neuromodulator" learni…
A flow-matching neural network that learns only the distribution of substation injections, called PoSE (posterior sampling for state estimation), cut the probabilistic injection error (CRPS) by 6.7-7.5% against a full-co…
Finetuning the Walrus-1.3B scientific foundation model on simulated cylinder-flow data raised one-step rollout errors by up to 35x after just 20 epochs, according to an arXiv paper (2610.40857v1) studying catastrophic fo…
Researchers released FARO, a function-space adaptive optimizer that assigns each parameter block its own update geometry and reallocates a frozen per-block budget by cross-block equal risk contribution (ERC), the risk-pa…
A new arXiv paper (2610.40871v1) proposes TensorSSM, an implicit neural representation for functional tensor decomposition that replaces the standard flattening of high-order tensors into 1D sequences with a dynamic Stat…
A new arXiv paper (2610.40914v1) argues that diffusion-transformer training acceleration from methods such as REPA and dispersive regularization stems at least partly from reorganizing the AdaLN-Zero residual gate α rath…
A paper posted to arXiv as 2610.40934v1 proposes a formal mathematical definition of "any-task inference" (ATI) foundation models, distinguishing them from models built for pre-training and fine-tuning. The framework for…
A new arXiv paper (2610.40998v1) reports that steering vectors — residual-stream directions previously used only as inference-time controls — can also serve as a training signal during LLM post-training, improving superv…
A system called DyCraft, presented in arXiv paper 2610.40754v1, breaks long-horizon 4D scene animation into agent-solvable steps, cutting physical violations compared with a native GPT-6 agent in Codex and prior agentic …
Researchers introduced Drone-PUR, a dataset and benchmark for unmanned aerial vehicle scenes built around visual clue discovery, interpretation and integration across perception, understanding and reasoning, with 11 task…
A new arXiv paper (2610.40905v1) reports that sparse autoencoders can decompose histopathology foundation model embeddings into sparse latent features that preserve most downstream tissue-class information and retrieve m…
A new arXiv paper (2610.40973v1) introduces RADI3N, a framework for part-aware 3D image editing that combines a CS-ARAP deformation module with an inversion-free selective texture transfer mechanism. The CS-ARAP module s…
A new arXiv paper (2610.41435v1) introduces BlindToChange, a benchmark showing that AI agents score as much as 27.5 points lower when a recent real-world event affecting the answer is not explicitly named in the request …
Patching the KV cache from a LoRA fine-tuned language model into its unmodified base model can transfer some fine-tuned behaviors, including responding in German and emergent misalignment, according to arXiv paper 2610.4…
A new arXiv paper (2610.41482v1) introduces KEEPER, a comparative analysis of four auxiliary objectives for recursive textual memory in long-horizon LLM agents, evaluated along information-content and update-sensitivity …
A new arXiv paper (2610.42054v1) proposes CRoS (Causality Reward on Scrambled CoTs), a chain-of-thought causality metric that scrambles the order of reasoning steps without changing their content, and reports it beats ex…
A paper on arXiv (2610.42010v1) introduces TurnTurk-GPT, a Turkish text-based turn-taking predictor built by fine-tuning a Turkish GPT-2 model with the TurnGPT objective on synthetic and real dialogue. Evaluated on unpun…
A new arXiv paper (2610.41769v1) proposes Stepwise Querying, a reasoning-distillation method that extracts step-specific explanations from a black-box teacher model without access to its internal chain-of-thought traces,…
A new arXiv paper (2610.41855v1) proposes building sentence embeddings from the path signature of Word2Vec trajectories, using iterated integrals from rough path theory to capture sequential and lexical structure. Across…
A lightweight LLM-centric conversational layer called SimXAI replaces the separate intent-recognition, dialogue-management and template-generation modules of GUS-style conversational explainable AI agents with a single t…