AI research news
Stories 301 to 360 of 858, from 8 October 2026 to 15 October 2026, newest first. The latest are on the first page →
-
Google DeepMind’s Antigravity finds novel antimicrobial peptide candidate
Google DeepMind published a technical walkthrough on October 15, 2026, detailing how its agentic platform, Antigravity, helped researchers identify a new antimicrobial peptide candidate in Streptococcus…
1 sourceUnite.AI -
Fortytwo Engine reduces token use by 24.7% on Qwen 3.8 27B in DeepSWE tests
Ivan Nikitin, CEO and co-founder of Fortytwo, discussed the company's swarm inference technology in an interview with unite.ai. The Fortytwo Engine uses multiple open-weight models to steer reasoning in real time,…
2 sourcesUnite.AI -
Opinion: Rust’s type system and guardrails may help agentic coding
A Rust engineer argues that the language’s strengths—performance by default, fine-grained control over memory, and strong type systems—make it a good fit for agentic development. Agents can now tune performance knobs…
1 source HN 70smallcultfollowing.com -
Researcher maps how many guesses crack AI-generated random numbers
A researcher analyzed the output of open-source AI models to estimate how many guesses an attacker would need to crack their ‘random’ number generation. By reading the logits—the raw scores before softmax—of both base…
1 sourceTowards Data Science -
Open-weight AI differs from open-source AI in licensing and control
Open-weight AI releases trained model parameters under a license, while open-source AI grants broader access to software, data, documentation, and development processes. The article explains that open-weight AI focuses…
1 sourceUnite.AI -
Researchers introduce Seal AI model for gene regulation in developing human brain
A team led by Y. Hao and O. G. Troyanskaya published Seal, a transfer-learning framework that predicts gene expression and variant effects across 26 brain regions, 30 cell types, and 7 developmental stages. Seal links…
1 source primary sourceNature Machine Learning -
Microsoft Research introduces Probabilistic Bias Correction for longer-term weather forecasts
A team led by Microsoft Research published a study in Nature on October 14, 2026, introducing probabilistic bias correction (PBC), a machine learning framework that improves weather forecasts for two- to six-week lead…
3 sourcestbreak.comUnite.AItheverge.com -
TIER IV funds Autoware research center at Technical University of Munich
On October 15, 2026, TIER IV and the Technical University of Munich announced the establishment of the TUM Autoware Research Center through a multi‑year strategic research collaboration. The center unites three TUM…
1 sourceUnite.AI -
MIT professor Eric So warns of AI cognitive atrophy
Eric So, a Sloan Distinguished Professor at MIT Sloan School of Management, released his book The Collision: What AI Does to Us this month. In a Q&A with MIT News, he argues that while AI can speed up work, heavy…
1 source primary sourceMIT News on AI -
Study finds Chatbot Arena rankings may overstate differences between Similar models
A new paper from arXiv examines the statistical foundations of Chatbot Arena, the popular leaderboard for large language models. The authors argue that the Bradley-Terry model, which underpins the Arena's rankings,…
1 source primary sourcearXiv cs.CL -
Researchers show LLMs can link records and infer ethnicity from names without training
A new paper on arXiv demonstrates that pre-trained language models can perform record linkage and demographic inference from names without needing task-specific training data. The author introduces EnsembleLink, an…
1 source primary sourcearXiv cs.CL -
Researchers propose Joint Subspace Sparse Attention for faster long-context decoding
A new paper on arXiv introduces Joint Subspace Sparse Attention (JSSA), a method to speed up long-context decoding in large language models. The technique reduces KV-cache traffic—a bottleneck for long prompts—by…
1 source primary sourcearXiv cs.CL -
Researchers benchmark eight LLMs on English and Bengali summarization
A new study on arXiv compares eight large language models—both closed-source and open-weight—on multilingual abstractive summarization. The authors tested them on size-matched English and Bengali datasets from the…
1 source primary sourcearXiv cs.CL -
Researchers train voice agents to follow unseen debt-collection rules
A team of researchers tested whether small, self-hosted voice agents could adapt to new legal procedures in debt collection without retraining. They used a 9-billion-parameter model fine-tuned on 2,506 conversations…
1 source primary sourcearXiv cs.CL -
Datalab releases OmniParseBench, an open OCR benchmark with 16,288 tests
Datalab has released OmniParseBench, an open-source benchmark designed to evaluate Optical Character Recognition (OCR) parsers. The dataset consists of 16,288 pass/fail tests derived from 2,937 pages across 2,343…
1 sourceMarkTechPost -
Colossal introduces GenoMorph AI to target dodo gene-editing sites
Colossal Biosciences announced GenoMorph, an AI platform designed to identify gene-editing targets in species with limited genomic data. The Dallas-based company, co-founded by Ben Lamm and George Church, claims the…
1 sourceUnite.AI -
Leash Bio releases Minos, an AI playground for lab-in-the-loop simulations
The tool enables agents to perform verifiable scientific reasoning using tens of thousands of lab-in-the-loop simulations derived from the company’s proprietary data on DNA Encoded Libraries (DELs)—a high-throughput…
1 sourceLatent Space -
Stanford economists argue AI should focus on human collaboration over automation
Gabriel Unger and Arvind Narayanan, economists at Stanford’s Digital Economy Lab, challenge the dominant AI agenda in an essay. They argue that AI research is overly focused on automating human labor, driven by…
1 sourceAI Snake Oil -
Google engineer says self-improving AI remains far from reality
Google engineer Haydn Belfield argues that self-improving AI is not yet achievable, despite automation advances. At Google DeepMind, AI handles over 75% of production code but still requires human oversight for 88% of…
1 sourceThe Decoder -
Meta publishes ten health studies generated with its Muse AI platform
Meta released ten research manuscripts on October 14, 2026, all produced using its Muse AI platform. The studies analyze electronic health records from 4.95 million patients, including data on 156,051 with breast…
1 sourcecryptobriefing.com -
Researchers release Mr.Chatterbox-2.7B-Chat model trained on 19th-century British texts
It is a fine-tuned version of Mr.Chatterbox-2.7B, a language model trained exclusively on British texts from 1800 to 1899. The model simulates a Victorian gentleman’s speech patterns and reasoning, using 97.2% real…
1 source primary sourcehuggingface.co -
UW researchers design AI-built proteases targeting disease proteins
Two teams at the University of Washington’s Institute for Protein Design published papers in Nature on October 14, 2026, detailing AI-designed proteases that cleave specific disease-linked proteins. The studies report…
2 sourcesUnite.AIArs Technica AI -
White Circle says Google and OpenAI image models may be getting worse at blocking explicit outputs
Researchers at White Circle, a public benefit corporation backed by leaders from Anthropic, DeepMind, Hugging Face, and OpenAI, tested the safety filters of leading text-to-image models. In a preprint published last…
2 sourcesArs Technica AIibtimes.sg -
Gladstone AI report outlines how labs may reach recursive self-improvement in 3–12 months
A new report from Gladstone AI, shared exclusively with TIME, details how top AI labs like OpenAI and Anthropic are automating their research pipelines to achieve recursive self-improvement (RSI). Based on interviews…
2 sourcestime.comimportai.substack.com -
Illumina launches BioInsight Platform to unify AI, software and genomic data
Illumina introduced the BioInsight Platform on October 14, 2026, its first integrated software environment for genomic and multiomic data analysis. The platform combines Illumina’s AI models, curated datasets, and…
1 sourceUnite.AI -
Google DeepMind’s AI forecasted Hurricane Isaias’s path 19 miles off
The European model’s AI prediction was 36 miles off, while the American GFS model missed by 119 miles. DeepMind’s accuracy puzzled meteorologists, who noted its strengths in storm tracking and timing but a weaker…
1 sourceyahoo.com -
Hospitals, banks and edge devices use federated Voting for feature selection
A new approach lets hospitals, banks and edge devices collaboratively identify important features for machine‑learning models without exchanging raw data. Instead of sharing rows, each participant casts votes on…
1 sourceTowards Data Science -
Researchers introduce Echo for faster 4-bit LLM training with low error
A team of researchers released Echo, a new method for 4-bit tensor-parallel communication in large language model training and inference. The technique reduces bandwidth costs by nearly fourfold compared to 16-bit…
2 sources primary sourceKDnuggetsarXiv cs.AI -
Ggml-org adds `--spec-draft-n-ubatch` to llama.cpp for batch control
llama.cpp now lets users set a custom physical batch size for draft context via the --spec-draft-n-ubatch flag. The change, merged in a pull request by ggml-org, allows drafts to process tokens in smaller chunks than…
1 source primary sourcegithub.com -
OpenAI solves irrational triangular billiards problem with AI
OpenAI’s AI solved the long-standing irrational triangular billiards problem, proving light rays distribute evenly over time under specific conditions. The breakthrough, highlighted by Two Minute Papers, marks a…
1 sourcegeeky-gadgets.com -
Google.org funds 17 AI-driven projects
On October 14, 2026, Google.org announced $36 million in funding for 17 organizations applying AI to scientific research. The announcement was authored by Maggie Johnson, Global Head and VP of Google.org. Recipients…
1 sourceUnite.AI -
Anthropic’s Claude finds CRISPR-like DNA repeats in bacteriophage, but function remains unknown
Anthropic’s AI model Claude identified a repeating DNA pattern in a bacteriophage genome that resembles CRISPR, a gene-editing tool. The discovery sits beside a reverse transcriptase gene, hinting at potential utility…
1 sourcescientificamerican.com -
Researchers hide encrypted capability in AI model, fine‑tuning fails
On October 9, 2026, four researchers published a paper titled “AI Models Can Provably Hide Arbitrary Capabilities” on arXiv and announced its acceptance at NeurIPS 2026. The study shows that a small neural network can…
1 sourcemaccus.fr -
Asset learns to compress speech encoders
It uses reinforcement learning (RL) to adaptively compress audio input sequences for better performance in large language models (LLMs). The framework divides the encoded segments into variable-duration parts and then…
1 source primary sourcearXiv cs.CL -
Researchers find Qwen3’s reasoning gains may stem from ‘thinking leakage’
A new paper on arXiv examines how reinforcement learning (RL) training affects hybrid reasoning models like Qwen3. The study suggests that much of the reported improvement in reasoning performance—when models switch…
1 source primary source HN 8arXiv cs.CL -
Researchers test AI agents improving LLM training strategies with up to 13.34% gains
Their Agentic Data Engineering (ADE) framework lets agents experiment with data selection, reward design, and curricula, using feedback to adjust strategies. The study found agents achieved up to a 13.34% improvement…
1 source primary sourcearXiv cs.CL -
Researchers model how AI and humans time new words in text
A new paper on arXiv explores how information appears over time in natural language. The authors propose using neural temporal point processes to predict when new words or high-surprisal tokens appear in text…
1 source primary sourcearXiv cs.CL -
Researchers propose Next-Concept Modeling for faster text generation
Researchers introduced Next-Concept Modeling (NCM), a two-stage framework for faster text generation. The method combines a multiscale residual VQ-VAE encoder with a transformer that predicts discrete codes in…
1 source primary sourcearXiv cs.CL -
Researchers test Typed-Decision models on Indonesian regional languages
A new study evaluates typed-decision models—fast, low-cost systems that return labels and probabilities in one step—on Indonesian regional languages. The paper, posted on arXiv, compares the commercial Jev model and…
1 source primary sourcearXiv cs.CL -
Researchers propose Confidence-Aware Temporal Reasoning for safer future predictions
A team of researchers introduced ChronoRegime, a new task designed to test how well large language models handle uncertain future events. The framework, called Confidence-Aware Temporal Reasoning (CTR), evaluates…
1 source primary sourcearXiv cs.CL -
Study finds capable LLM agents leak more privacy in dialogue
A new arXiv paper introduces CIMEMORIES DIALOGUE, a benchmark of 4,410 two-agent dialogue instances designed to test privacy leakage in LLM agents. The study is grounded in Contextual Integrity theory, which assesses…
1 source primary sourcearXiv cs.CL -
Researchers release PsyBer-Agent, a psychology-driven recommender system simulator
Researchers introduced PsyBer-Agent, a new simulator designed to improve recommender systems by modeling user dynamics like fatigue and shifting preferences. The system uses a Psy-Engine with three evolving latent…
1 source primary sourcearXiv cs.CL -
Researchers introduce ConicFlow for constrained flow matching model sampling
It reframes constrained sampling from flow matching models as an optimal control problem at inference time. The approach balances deviations from the pretrained trajectory against a reward function while enforcing hard…
1 source primary sourcearXiv cs.AI -
Researchers introduce pullback directions for varied AI image generation
A new paper on arXiv proposes a training-free method to improve diversity in AI-generated images without sacrificing quality or prompt alignment. The technique, called controllable diverse sampling through pullback…
1 source primary sourcearXiv cs.AI -
Mosaic learns sparse image-text latents for VLMs
Researchers introduced MOSAIC, a new method to understand vision-language models (VLMs). It creates a shared sparse latent space between images and text concepts. This enables localized influence attribution across the…
1 source primary sourcearXiv cs.AI -
World model in agent's value function discovered
Visualizing the World Model in an Agent's Value Function arXiv:2610.14186v1 Announce Type: new Abstract: Could model-free reinforcement learning algorithms actually be learning a model of the environment? In this work,…
1 source primary sourcearXiv cs.AI -
Vetnet launches benchmark datasets for veterinary AI research
VETNET is a new initiative introducing standardized benchmark datasets and evaluation tools for AI in veterinary medicine. The project aims to bridge the gap between human health AI advancements and veterinary…
1 source primary sourcearXiv cs.AI -
LHTB 2.0 benchmarks AI agents over three-hour tasks with 77 new scenarios
Researchers released Long-Horizon-Terminal-Bench 2.0 (LHTB 2.0) on arXiv, expanding a test suite for AI agents that measure sustained progress over long workflows. The update adds 31 extended tasks—bringing the total…
1 source primary sourcearXiv cs.AI -
Researchers propose Correction step to improve diffusion model sampling quality
A new paper on arXiv introduces a correction step for diffusion models to address error accumulation during iterative sampling. The method trains a corrector on the base model’s predictions, refining intermediate…
1 source primary sourcearXiv cs.AI -
Researchers introduce ClawLoop for self-improving AI agents
A new framework called ClawLoop enables AI agents to improve themselves without relying on external teacher models or human input. The system works in three phases: Task Synthesis, which creates varied training tasks…
1 source primary sourcearXiv cs.AI -
arXiv paper shows silent failures in agentic data science
Large language models have lowered the barrier to writing code, which has spurred the creation of data science agents that promise to automate the design and execution of data science tasks. Existing evaluations of…
1 source primary sourcearXiv cs.AI -
7 Best Resources to Learn About Self-Evolving AI Agents
The article defines these systems as those that adapt by learning from failures, improving prompts, or refining tool-use strategies. It recommends starting with Hugging Face’s free Agents Course, which covers…
1 sourceKDnuggets -
Is 'Attention' the Essence of Intelligence?
In 2014, a 23-year-old intern at the University of Montreal, Dzmitry Bahdanau, developed the attention mechanism—a mathematical operation that let neural networks process long sentences by dynamically focusing on…
1 sourcequantamagazine.org -
China leads in AI researchers and publications but trails US in compute and model capabilities
China now has 51,000 active AI researchers compared with 14,000 in the United States, according to Digital Science data cited by Nature. Chinese undergraduates made up 57% of NeurIPS 2025 authors, up from 46% three…
1 source primary sourceNature Machine Learning -
Arena Physica’s Heaviside EM model joins Microsoft Discovery for faster design checks
Arena Physica announced on October 13 that its foundation model for electromagnetism, Heaviside, is now available on Microsoft Discovery, a platform for agentic engineering. The model predicts full 3D electromagnetic…
1 sourceUnite.AI -
AI agents write CUDA kernels faster than PyTorch, author tests on DGX Spark
An author on Towards Data Science reports that AI agents can now generate CUDA kernels that outperform PyTorch's built-in implementations. The testing was conducted on an NVIDIA DGX Spark system. The article emphasizes…
1 sourceTowards Data Science -
Google invests €4.5m in AI-powered weather forecasting for Irish schools
Google announced a €4.5 million investment through its philanthropic arm Google.org to boost AI‑driven weather forecasting and climate research in Ireland. The four‑year initiative will deploy 560 solar‑powered weather…
1 sourcerte.ie -
GitHub Security Lab automates C/C++ fuzzing with AI-driven taskflows
GitHub’s Security Lab released Fuzzing Taskflow, an autonomous pipeline for vulnerability discovery in C/C++ repositories. By specifying a GitHub repo, the system—built on its Taskflow Agent—handles entry-point…
1 sourcenote.com -
OpenAI says its Navier-Stokes proof does not follow Spanish mathematicians' approach
OpenAI posted a 166-page solution to the Navier-Stokes problem in early September, initially claiming it followed an approach by Spanish mathematicians Diego Córdoba and Luis Martínez-Zoroa. Experts now say the proof…
1 sourcescientificamerican.com -
Alibaba’s QwenGyre speeds up long-horizon agent training by up to 1.85x
Alibaba’s Qwen team introduced QwenGyre, an elastic reinforcement learning framework designed for extreme-long-horizon agent training. It targets tasks running for hours and processing nearly 1M tokens per rollout,…
1 sourceMarkTechPost
