🧠 Large Language Models

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

1,700 articles

Q.ANT shows photonic NPU Gen 2+ running an LLM with optical Fourier transforms

At SC26, Q.ANT is demoing a hybrid setup that pairs its NPU Gen 2+ photonic processor with an AMD GPU. The system runs a large language model while performing Fourier transforms in optics rather than in electronics. Source: finance.yahoo.com

FBI and Google warn attackers are using ARTEX and other AI platforms to speed up cyberattacks

An FBI report published Nov. 13 describes how cyber actors are weaponizing ARTEX, an open-source agentic AI platform driven by large language models. Google also warns that threat actors are turning AI tools into accelerators for their attacks. Source: google.com

MIT: LLM internal representations mirror physical laws and scientific reasoning

New research from the Massachusetts Institute of Technology shows that the internal logic of large language models parallels the behavior of physical systems. The work maps model representations onto physical laws and scientific reasoning. Source: google.com

Alibaba releases Qwen4-Max with training and attention optimizations

Alibaba Group Holding Ltd. has introduced Qwen4-Max, its newest and most capable large language model. The model produces higher-quality output thanks to improvements in training and attention mechanisms. Source: google.com

EmbeddingGemma 2 adds native multimodal understanding

Google has published the technical report for EmbeddingGemma 2, an update to its open text-only embedding model. The new version natively handles text, images, audio and video, mapping them into a shared embedding space. Source: blog.google

Tracker follows every major AI model release in one place

A free tracker catalogs model launches from OpenAI, Google, Anthropic, xAI, Meta, DeepSeek and others. It records what each release changed and which model lines are expected to ship next. Source: nextaimodel.com

Altman cautions investors about chipmakers' reliance on AI labs

Sam Altman warns that many chipmakers, including NVIDIA and AMD, are betting on OpenAI and Anthropic to absorb most of their volume. He suggests the underlying demand may be more limited than that bet assumes. Source: finance.yahoo.com

OpenAI launches ChatGPT for Sales

OpenAI is rolling out ChatGPT for Sales, a version aimed at helping sales teams get to know their customers better. The goal is to cut time spent on routine busywork and help reps close more deals. Source: openai.com

Healthcare AI's future lies in specialised, purpose-built models, not just LLMs

Scanbo Technologies founder and CEO Ashissh Raichura argues that the flashiest healthcare AI demo today, a model passing a medical licensing exam, is not the real measure of progress. He contends the future will depend on specialised, purpose-built models. Source: google.com

Calpha aims to bring native biomolecular intelligence to a general-purpose language model

Biomolecular foundation models predict protein structures and complexes accurately, while LLMs excel at reasoning in natural language. Calpha seeks to combine both abilities in a single general-purpose language model. Source: google.com

LLM-powered social simulations face serious questions about epistemic validity

Using LLMs as the cognitive engines of simulated agents in computational social science has sparked both excitement and concern. Researchers are now questioning whether such simulations can truly be considered valid sources of knowledge. Source: bioengineer.org

Tight PAC-Bayes guarantees proposed for LLM judges

LLM judges are increasingly used to monitor the outputs of LLM-based AI agents, but formal guarantees of their reliability have been lacking. New research offers tight PAC-Bayes generalisation bounds for such judges. Source: google.com

Claude Code skill applies Karpathy's mental models to AI analysis

A specialized Claude Code engineering skill lets users examine AI reliability, neural networks and Software 3.0 through the mental models of Andrej Karpathy. Source: mcpmarket.com

Gemini app gets new chat history and AI training controls

Google is updating privacy settings in the Gemini app. Users can now manage their chat history and opt out of having their data used for AI model training. Source: blog.google

Gemini goes local with 66 new regional accents

Google is rolling out the regional accents it promised at I/O. The 66 new voices cover specific cities and regions across the US, UK and other countries. Source: theverge.com

California AG sues OpenAI over unsafe model testing

California's attorney general has sued OpenAI, alleging the company tested advanced AI models without adequate safety measures. The suit also claims OpenAI misled the public about how safe its experiments were. Source: the-decoder.com

Gemini app adds 66 regional voices on Android and iOS

The Gemini app now offers 66 regional voices on both Android and iOS. Users can pick a natural-sounding local dialect for their interactions with the assistant. Source: blog.google

OpenAI opens Daybreak cybersecurity tool to NATO

Through Daybreak, cyber defense teams can use OpenAI's models to audit software, hunt for vulnerabilities and patch security gaps. The tool is now being made available to NATO. Source: the-decoder.com

Gemini Notebook can turn research into interactive sites

Gemini Notebook can now convert your sources and research into interactive websites. Users can build project trackers and visual dashboards directly from their notebook content. Source: blog.google

Sam Altman to speak at Semafor Silicon Valley & The World

OpenAI CEO Sam Altman is among the leading technology figures appearing at the Semafor Silicon Valley & The World event. The session is available via livestream provided by Semafor. Source: youtube.com