News about foundation models, LLMs, multimodal models, benchmarks, and model releases from OpenAI, Anthropic, Google, Meta, Mistral, and others.
973 articles
A practical guide moves past basic graph-based retrieval to outline six production-ready architectures. These combine semantic search, knowledge graphs, and LLM reasoning to build more capable retrieval-augmented systems. Source: towardsdatascience.com
China presented what it calls the country's first multilingual, fully multimodal AI-powered input method for the Tibetan language. The system was unveiled on Sunday in Xining. Source: globaltimes.cn
A made-up disease with absurd symptoms and a fictional lab supposedly aboard the USS Enterprise was enough to trick AI chatbots into treating it as real. The fabricated condition reportedly even slipped into other sources presented as legitimate information. Source: acsh.org
TypeSafe AI introduced Jev, a 'System One' model that produces structured, typed decisions with calibrated probability scores rather than free-text responses. The service is priced at $0.042 per 1 million input tokens. Source: marktechpost.com
NVIDIA released AIPerf, a new benchmarking tool built to reliably measure large language model inference speed under large-scale conditions. The tool aims to give developers consistent performance data as LLM deployments grow. Source: quantumzeitgeist.com
Diogo Almeida, a former OpenAI researcher who helped build ChatGPT and later co-developed reinforcement learning from human feedback, grew disillusioned with the chatbot he helped create. He has now returned with a new AI model approach that is generating excitement among developers. Source: techcrunch.com
A Chinese AI startup founded in February by a Tsinghua University professor has quietly reached a valuation of over $1.4 billion. The company raised $400 million in funding while operating largely under the radar. Source: theinformation.com
Researchers are increasingly using LLMs to computationally simulate human behavior in studies. While these models don't explain why people act as they do, they offer a new way to approximate human responses at scale. Source: nature.com
CrowdStrike says the PhantomRaven attack was likely built with the help of an LLM. The malware spread through malicious npm packages designed to steal developer credentials and CI/CD secrets. Source: thehackernews.com
SafeSeal is a new watermarking technology that embeds verifiable, hidden markers into text generated by LLMs. The system is designed to preserve the quality and meaning of the original content while enabling reliable identification of AI-generated material. Source: eurekalert.org
AI lab PrismML may not be widely known yet, but that could soon change. The company is betting that its small-scale LLM can transform how people interact with AI on a daily basis. Source: techcrunch.com
OpenAI has introduced a specialized configuration of its GPT-6 Astra model tailored for Am Law 200 firms and legal tech vendors. The company also outlined co-development partnerships, including one with law firm Sullivan. Source: law.com
As LLM-based agents become widespread, the main engineering challenge has shifted from raw model capability to the economics of running inference. Companies are now focusing on optimizing token usage to keep AI operations affordable at scale. Source: infosys.com
Intel has showcased improved AI inference capabilities through its latest MLPerf Inference v6.1 benchmark results. The company says the results demonstrate meaningful performance gains that could support its future growth prospects. Source: theglobeandmail.com
Cognita, part of Rad Partners' Mosaic Clinical Technologies division, received a $1 million FDA grant to test a novel evaluation approach called 'LLMs as a jury.' The method compares assessments from multiple AI models to judge the quality of AI-generated radiology reports. Source: radiologybusiness.com
In this edition of the Daily 5 roundup for Thursday, September 17, a King's College London study examined large language models and video-language models. The research found signs of bias in how these systems assess pedestrians based on skin color and disability. Source: autonews.com
SUPERWISE, known for its Agentic Management Platform, has launched Sentinel, an enterprise-grade AI gateway. The tool gives organizations a real-time guardrail layer to monitor and govern LLM traffic across their networks. Source: aithority.com
Anthropic CEO Dario Amodei has proposed placing independent evaluators inside leading AI companies, comparing the idea to bank supervision. Critics question whether such oversight would truly remain neutral given industry influence. Source: cnbc.com
A team led by UVA's Jeff Saucerman tested whether LLMs like GPT, Gemini and Claude can accurately explain biological processes. The results showed both useful capabilities and notable weaknesses when applied to complex biomedical reasoning. Source: uvahealth.com
Splunk plans to release another AI model focused on telemetry data analysis via Hugging Face under an open-source license. The move builds on its earlier open-source LLM efforts for log data processing. Source: devops.com