🧠 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

StateBench tests how LLMs track evolving state in long texts

StateBench is a new benchmark for long-context state tracking. It measures how well language models keep tabs on entities whose properties change throughout a long document. Source: huggingface.co

Altman calls PC memory sitting idle at home 'quite terrible'

Sam Altman recently said it is "quite terrible" that memory sits in personal computers instead of being used in AI data centers. Source: tech.yahoo.com

CWE-bench v2 expands coverage and publishes 100 open tasks

CWE-bench is built to probe the limits of frontier cyber capabilities in AI models. Version 2 releases a public set of 100 tasks that anyone can run and use to evaluate their own systems. Source: huggingface.co

Google: LLMjacking attacks generate up to $1 million in cloud AI charges

Hijacking of large language models, known as LLMjacking, is producing huge fraudulent cloud charges. Some organizations have lost more than a million dollars. Source: bleepingcomputer.com

Altman suggests consumer PC memory should go to AI instead

In a video interview published last week, Sam Altman said it's pretty terrible that ordinary consumers get to hold large amounts of memory in their own machines. Source: pcworld.com

Economic Times news hub on Stability AI and Emad Mostaque

The Economic Times page collects the latest news, pictures and videos about Stability AI and Emad Mostaque. The snippet itself contains no specific details. Source: economictimes.indiatimes.com

Mandiant: LLMJacking turns stolen cloud and model access into a criminal business

Mandiant details how LLMJacking attacks leave victims with fraudulent cloud bills running into hundreds of thousands of dollars. Underground markets are actively trading stolen LLM credentials. Source: cloud.google.com

LLM jacking hits companies as attackers resell stolen AI access

An analysis warns that LLM jacking, a cyberattack that steals a company's access rights to large language models for unauthorized use or resale, is spreading. Victims are left with mounting usage bills. Source: biz.chosun.com

Emad Mostaque steps down as Stability AI CEO

Stability AI announced that Emad Mostaque resigned from his role as CEO of the company earlier today. He also left his position on the board. Source: stability.ai

How Emad Mostaque built, then lost, his AI empire

Emad Mostaque grew Stability AI into a $1B unicorn before losing it. The story examines exaggerated claims and financial troubles behind the company's downfall. Source: aidiscoveries.io

Transformers formalizes model integration contracts

The Transformers library now supports more than 750 model types. Its test suite is massive, with over 88,000 CPU tests on main and 200,000+ GPU tests in a daily run. Source: huggingface.co

OpenAI readies lobbying push for post-midterm Washington

OpenAI is preparing a political lobbying and PR campaign for after the US midterm elections. The company expects a tougher fight in Washington. Source: the-decoder.com

Developer builds OpenTube as answer to Karpathy's HackerTube

A developer describes building OpenTube after Andrej Karpathy built HackerTube, drawing on 20 years of experience in scalable systems and AI-powered innovation. The post also references a new suggestion from Karpathy about asking AI for output. Source: medium.com

Stability AI: from Stable Diffusion fame to financial troubles

Stability AI was founded in 2019 by Emad Mostaque and Cyrus Hodes and rose to prominence in August 2022 with its open text-to-image model Stable Diffusion. Financial problems in 2023 led employees to leave, and some went on to found Black Forest Labs in 2024. Source: en.wikipedia.org

Cisco pairs network intelligence with OpenAI for faster defense

Cisco is combining its network intelligence with OpenAI models to strengthen enterprise security. The result is a more adaptive defense that reacts at machine speed. Source: openai.com

Add LLM summarization to scikit-learn pipelines with Scikit-LLM

A tutorial shows how to build a single architecture that combines Scikit-LLM estimators for text summarization with traditional scikit-learn methods. Source: kdnuggets.com

Apple goes all in on AI, both its own and rivals'

Apple posted its best fiscal year ever for revenue and earnings per share. iPhone and Mac sales kept growing strongly in the fourth quarter. Source: theverge.com

EU court adviser: Google's Gemini chatbot may breach copyright rules

An adviser to Europe's highest court said Gemini's use of publishers' articles to answer user queries could infringe copyright. The opinion is a setback for Google. Source: reuters.com

Anthropic blocks Claude from cloning existing video games

Anthropic has updated Claude Code's system prompt so the tool can't be used to re-create existing video games, according to Tom's Hardware. Source: theverge.com

Jülich to host EVE language model for Earth observation and Earth science

The Jülich Supercomputing Centre (JSC) has launched a new collaboration with Pi School, an international centre of excellence for AI. The partnership will host the EVE language model for Earth observation and Earth science. Source: hpcwire.com