AI research news
Stories 841 to 858 of 858, from 6 September 2026 to 15 September 2026, newest first. The latest are on the first page →
-
LabAgent: Automates Reproducing Scientific Methods
A new AI tool called LabAgent aims to streamline the process of replicating scientific methods across various domains. Developed by researchers, LabAgent is designed to address common challenges faced in research…
1 source primary sourcearXiv cs.AI -
Vibe Patenting: LLM Judges Improve AI Patent Drafting Quality
Researchers introduced "Vibe Patenting," a new testbed designed to evaluate AI agents capable of drafting professional patents. The study investigates the reliability of Large Language Model (LLM) judges in assessing…
1 source primary sourcearXiv cs.AI -
Generalized Agent Iteration Framework Unifies Policy Improvement and Recursive Self-Improvement
The paper introduces Generalized Agent Iteration (GAI), a formal framework that brings together iterative policy improvement and recursive self‑improvement under one umbrella. GAI treats an agent as a set of modifiable…
1 source primary sourcearXiv cs.AI -
New Benchmark Shows Hindsight Bias in Clinical Language Models and How Temporal Masking Helps
Researchers have released a paired benchmark to quantify hindsight bias in clinical temporal reasoning for large language models. The dataset comprises 171 PubMed Central case reports—40 sepsis and 131 GLP‑1/diabetes…
1 source primary sourcearXiv cs.CL -
Evaluating LLMs' Protocol Understanding
This study evaluates the ability of large language models (LLMs) to interpret network protocols correctly. Researchers designed 4 tasks and 1,482 queries for 16 different protocols. They examined how well an LLM's…
1 source primary sourcearXiv cs.CL -
Partial Objective Weighting in Cone Beam CT Reports
A new study introduces a method for generating maxillofacial reports from cone beam computed tomography (CBCT) images, scoring them with a composite objective that gives 80% weight to factual entailment and 20% to…
1 source primary sourcearXiv cs.CL -
Training-Free Lexical pipeline cuts LLM prompt tokens by 40% with minimal quality loss
Researchers have introduced a deterministic, CPU-only pipeline for compressing large language model prompts using classical lexical NLP techniques. Unlike learned methods such as LLMLingua, this approach requires no…
1 source primary sourcearXiv cs.CL -
PhysMent Benchmark Tests LLMs on Interactive Physics Reasoning
Researchers have introduced PhysMent, a new benchmark designed to evaluate how large language models reason about the physical world through active experimentation. Unlike traditional static tests that provide all…
1 source primary sourcearXiv cs.CL -
Asclepius: Improves Clinical Agent Performance
Researchers have developed Asclepius, an adaptive agent framework designed to improve the performance of long-horizon clinical agents. The system was tested using the Clinical Environment Simulator (CES), a testbed…
1 source primary sourcearXiv cs.AI -
Fraglingo introduces attachment-aware autoregressive fragment generation for molecular design
Researchers have introduced Fraglingo, a new autoregressive framework for molecular design that addresses the limitations of current fragment-based methods. Traditional approaches often decouple fragment selection from…
1 source primary sourcearXiv cs.AI -
GAVEL: Evaluating Clinical Timeline Extractors
Researchers have developed a new AI tool called GAVEL (Grounded Adjudication of Variations Across Extracted TimeLines) to compare clinical timelines with their corresponding case reports. This protocol uses an LLM…
1 source primary sourcearXiv cs.AI -
OrchSLM Framework Probes Small Language Model Orchestration
The new arXiv paper introduces OrchSLM, a routing framework that studies how small language models (SLMs) can be orchestrated without interactive calls. While large language models have shown impressive abilities,…
1 source primary sourcearXiv cs.AI -
Continual Search framework boosts AI agent failure attribution accuracy on large logs
The paper tackles the challenge of diagnosing failures in long‑horizon AI agents, where execution logs can span thousands of steps and human review is infeasible. Existing root‑cause attribution (RCA) methods that rely…
1 source primary sourcearXiv cs.AI -
LLM Agents Match RL in Zero‑Shot Agricultural Tasks, Adapt Better to Weather Shifts
Large Language Model agents are being tested for long‑term physical tasks that normally require continuous observation and adaptation. The paper introduces a zero‑shot, self‑adaptive framework that combines planning,…
1 source primary sourcearXiv cs.AI -
Sakana AI Introduces PC-ALM for Deep Networks
Researchers at Sakana AI have introduced a new algorithm called PC-ALM, which is designed to train deep neural networks more efficiently. Unlike traditional backpropagation methods, which are global and require a…
1 sourceMarkTechPost -
OpenAI’s internal RSI day highlights rapid research boost from GPT-6 Astra
OpenAI marked a “Recursive Self‑Improvement” (RSI) day, signaling what the company calls its new AGI milestone. Chief Scientist Jakub Pachocki released an essay titled An Alien Mind that, together with internal…
2 sourcesBBC TechnologySimon Willison -
Study finds AI agents lack creativity needed for recursive self-improvement
A multi-institution study led by Princeton University researchers challenges the prevailing narrative that AI recursive self-improvement is imminent. While current Large Language Models excel at narrow engineering…
4 sources HN 60Latent Spacetechnologyreview.comdwarkesh.comThe Decoder -
OpenArch launches PyTorch repo for readable LLM architecture implementations
A new open-source repository named OpenArch has been released, offering hand-written PyTorch implementations of modern large language model architectures. The project aims to provide clarity and educational value…
1 source primary source HN 43github.com
