Large Language Model News

Researchers at an undisclosed institution introduced clinical reasoning graphs to evaluate LLM diagnostic reasoning across 750 traces from five models on 50 NEJM cases. They found that while LLMs achieved 60-70% diagnost…

Researchers introduced Cortex, a framework that organizes web-scale corpora into an Ontological Corpus Graph (OCG) for structured knowledge organization, addressing the lack of systematic organization in existing corpus …

Researchers at an undisclosed institution developed MemLeak, a benchmark revealing that multimodal AI agents fail to fully delete user facts because retained images allow visual information recovery. The Information Prov…

Researchers introduced MuseBench, a benchmark with 4,016 questions evaluating multimodal large language models on intent-level understanding of audiovisual arts across cinema, visual arts, stage performance, and game des…

The developer of dailyprog, a daily coding puzzle site, built an LLM-based puzzle generator to automate content creation after manual puzzles proved unscalable. The generator uses a structured vocabulary and code validat…

A developer shared a configuration file for an AI tool that connects to OpenRouter to access models like OpenAI's GPT-5.2. The config includes logging, API timeout, and proxy settings, with a placeholder API key.

The New York Times filed a new court motion alleging Microsoft built a supercomputer specifically to help OpenAI train AI models on copyrighted articles without permission, violating copyright law. The Times seeks to ame…

An engineer provides a formula to estimate KV cache memory consumption for large language models, showing that the KV cache often becomes the bottleneck before model weights. For Llama 3.1 70B at 128K context, the KV cac…

A developer tracked AI prompts for 30 days and identified seven that save at least 10 hours per week. The prompts cover code explanation, debugging, test generation, refactoring, code review, documentation, and commit me…

Zhipu AI's open-weight GLM 5.2 model achieved a 39% F1 score in Semgrep's IDOR detection benchmark, outperforming Claude Code's 32% and Claude Opus 4.8. The MIT-licensed model runs locally, enabling secure code analysis …

AI researcher Erik Johannes Husom finds AI agents essential for his work studying them but useless in his personal life due to digital minimalism and a lack of bureaucratic challenges. He notes the technology's rapid dev…

Hugging Face and Pollen Robotics demonstrated a fully local conversational AI pipeline on the Reachy Mini desktop robot, using Silero VAD v5, Parakeet-TDT 0.6B v3, Gemma 4 or Qwen3-4B LLM, and Qwen3-TTS, with no cloud de…

A developer shares five patterns for designing tools that large language models can reliably call, based on lessons from the Claude Certified Architect Foundations exam. The patterns include writing tool descriptions tha…

Engineers adopting AI workflows in production face rising token costs, prompting a shift to evaluate token spend against value creation. A trend is emerging where teams use AI only for steps requiring intelligence, apply…

A developer built a chatbot library in Common Lisp that supports multiple back ends, enabling local LLM prompting. The library features recursive chatbots that can call each other without sharing contexts, reducing conte…

AgentGuard, an open-source static analysis tool for AI agent vulnerabilities, uses regex-based rules to detect prompt injection and other flaws in source code. Its creator, Dockfix Labs, is developing AST-based semantic …

Dockfix Labs released AgentGuard, an open-source static analysis tool that scans AI agent codebases for all 10 OWASP Agentic Security Initiative risks, including prompt injection, tool abuse, and data exfiltration. The t…

Researchers introduced Proxy-KD, a method for distilling knowledge from black-box large language models (LLMs) like GPT-4 into smaller models without accessing internal states. Proxy-KD uses a proxy model to facilitate k…

An engineer warns that large-scale use of AI agents quickly exhausts token and budget limits, leaving users locked out. The developer advocates treating agents as mentees to improve code quality and efficiency, rather th…

MeghRoop Tech Blog reports that after building over 50 AI systems, the team has identified prompt injection as a critical vulnerability in LLMs. Prompt injection exploits the inability of LLMs to distinguish between inst…