🧠 Large Language Models

News about foundation models, LLMs, multimodal models, benchmarks, and model releases from OpenAI, Anthropic, Google, Meta, Mistral, and others.

1,576 articles

LLMs learn to admit uncertainty as they enter scientific labs as discovery tools

A new perspective piece in Nature Machine Intelligence argues that large language models could become valuable lab partners if they're calibrated to recognize the limits of their own knowledge. The authors suggest uncertainty-aware LLMs could help guide scientific discovery rather than just generate confident-sounding but unreliable answers. Source: bioengineer.org

Long-running AI agents silently break compliance rules — bigger context windows won't fix it

Deploying an AI agent for a multi-day data validation task sounds efficient, but over time the agent starts drifting from its original instructions. Researchers warn that as agents process thousands of records across days, they quietly abandon compliance constraints, a flaw that simply expanding context window size cannot solve. Source: venturebeat.com

Qualcomm launches new Hexagon NPU built for always-on agentic AI

Qualcomm introduced its next-generation Hexagon neural processing unit on September 10, designed to support continuous, agentic AI workloads on devices. The chip targets always-on AI tasks that run persistently rather than on-demand. Source: thelec.net

Fly Language Model wires a full fruit fly brain map into a frozen 1.2B LLM — but its own tests show no benefit

The Fly Language Model connects the complete MaleCNS fruit fly connectome to a frozen LFM2.5-1.2B language model. However, the researchers' own control experiments found no measurable improvement from adding the fly-specific neural wiring. Source: marktechpost.com

This tiny LLM powers a virtual aquarium

Running a large language model on a microcontroller is already a technical challenge, and once you manage it, finding a genuinely useful task for such a stripped-down model is even harder. One project solved this by using a small LLM to drive a virtual aquarium simulation. Source: hackster.io

How context-aware AI assistants predict what you'll need next

Context-aware AI assistants combine retrieval-augmented generation, live data feeds, memory, APIs, business rules, and external tools to anticipate a user's next step. This combination lets them move beyond simple responses toward proactive, situationally relevant guidance. Source: nerdbot.com

Anthropic Builds Tool to Model AI's Impact on the US Economy

Anthropic, the company behind the Claude assistant, has released an interactive tool designed to illustrate how AI adoption could reshape the US economy. The tool aims to help policymakers and the public visualize potential shifts across industries and jobs. Source: kqed.org

RingCentral Launches ChatGPT Plug-in and MCP Connectors for Better LLM Access

Communications platform RingCentral has introduced a new plug-in that lets large language models tap into its voice, SMS, and chat data. The move is aimed at improving how AI tools can process and respond to enterprise communication data. Source: telecompaper.com

17-Year-Old Portland Student Finds Way to Make Small AI Models More Empathetic

High school student Henry Xie developed a method to transfer empathetic qualities from large AI models into smaller ones, as part of his project for the Regeneron Science Talent Search. His research explores how compact models can retain more human-like emotional understanding without the computing power of larger systems. Source: m.economictimes.com

New Model-Agnostic Tool Detects Personal Data Using Any LLM

A newly presented system offers configurable, instruction-based detection of personally identifiable information that can run on any large language model hosted via Amazon Bedrock. It was tested across five public PII datasets to measure its accuracy and flexibility. Source: aws.amazon.com

Chinese AI Models Gain Global Traction as Cheaper Alternative

Chinese-developed large language models are increasingly being adopted by US and international tech companies looking for lower-cost options. Their growing popularity suggests China's AI industry may be emerging as a genuine global competitor. Source: kraneshares.com

Researcher Says Leaked 6TB of Logs From Chinese LLM Router Exposed Company Secrets

Security researcher Chaofan Shou reported discovering roughly 6TB of invocation logs from a Chinese large-model relay service. According to Shou, the data contained sensitive information including SSH keys and cloud credentials belonging to enterprise users. Source: pandaily.com

New AI Model Claims Top Spot in Antibody Prediction, Echoing AlphaFold's Impact

A new AI system, reportedly built on GPT-6, has topped rankings for predicting antibody structures, drawing comparisons to AlphaFold's landmark impact on protein science. The claim points to a major step forward in applying large language models to computational biology. Source: eu.36kr.com

The Controversy Behind Anthropic Destroying Physical Books to Train Its AI Models

Anthropic has faced criticism over its practice of physically destroying books to extract training data for its large language models, which the company argues offer the highest-quality text available. The case has reignited debate over the ethics and legality of AI training data sourcing. Source: eu.36kr.com

Researchers Warn AI Workflows Can Be Hijacked to Leak Data Without Jailbreaking

Security researchers say enterprise AI workflows can be exploited to steal sensitive data even without prompt injection, account takeovers, or jailbreaking the underlying model. The vulnerability reportedly stems from how AI systems handle privileged access within automated pipelines. Source: cybersecuritynews.com

Meitu Xiuxiu Rolls Out New AI Assistant Powered by Its Own Large Language Model

Chinese photo-editing app Meitu Xiuxiu has introduced a new AI Assistant feature built on its in-house large language model, according to Sina Technology. The update was announced on September 10, expanding the app's AI capabilities for users. Source: news.futunn.com

BIT Computer Partners with AWS to Build LLM for Korea's Healthcare Sector

South Korea's BIT Computer announced a partnership with Amazon Web Services to develop a large language model tailored for the healthcare industry. The companies aim to bring AI-driven tools into Korean medical services. Source: koreabiomed.com

Mercury 2.5 diffusion model claims 1,107 tokens/sec at GPT-5.6 Luna-level quality

Inception unveiled Mercury 2.5 on September 8, 2026, a text-generation model built on diffusion technology rather than the typical autoregressive approach. The company says it matches GPT-5.6 Luna's performance while generating up to 1,107 tokens per second at a cost of roughly 120 yen per million output tokens. Source: gigazine.net

GPT-6 Astra meets V2Fun: large models begin orchestrating 3D generation

A new wave of AI tools is combining large language models with 3D generation pipelines, letting models coordinate multiple specialized systems instead of doing the work themselves. This shift points to a broader trend of AI acting as an orchestrator across code, image, and 3D modeling workflows. Source: eu.36kr.com

Another Anthropic employee quits, warns about frontier AI labs' direction

An Anthropic staffer known as Cal has publicly announced leaving the company, citing concerns about the path frontier AI labs are taking. The departure adds to a growing list of insiders voicing doubts about how leading AI companies are handling model development. Source: nationalreview.com