AI Briefing#llm
The latest AI and developer news about #llm, with the original source and a short summary.
The latest AI and developer news about #llm, with the original source and a short summary.
ChatGPT is useful, but important decisions require human review and source verification.
To maximize the performance of AI agents, an improvement loop that integrates domain experts' tacit knowledge into workflow and tool design is essential.
AI is rapidly changing how we work, collaborate on teams, and even the labor market itself.
It archives South Korean statutes and precedents as Markdown and Git history, providing a dataset optimized for LLM training and RAG.
GBrain, an agent-based open-source personal knowledge management tool supporting Claude Haiku and MCP, has been released.
The Claw-Eval benchmark, which verifies LLM agents using 139 real-world tasks, has been released.
Meta's internal token leaderboard revealed 60 trillion tokens used in 30 days and the flaws in the metric.
LaCy improves factuality by distinguishing what SLMs should learn from what they should delegate.
Claude Mythos Preview has demonstrated overwhelming cybersecurity capabilities in discovering zero-day vulnerabilities and generating exploits.
GLM-5.1 is a next-generation coding model capable of continuous optimization without performance degradation even in long-horizon agentic tasks.
China's Z.ai has released GLM-5.1, an open-weight model with 754B parameters.
OpenAI pointed out contamination and test flaws in SWE-bench Verified and recommended discontinuing it.
Google has released Google AI Edge Gallery, an optimized app for running the Gemma 4 model on-device.
Introducing nia-docs, which helps AI agents navigate web documentation directly like a file system instead of relying on RAG.
This analyzes the structure of the 'Agent Harness,' the core infrastructure that transforms an LLM into a capable agent.
As AI takes on large-scale SWE tasks that are easy to verify, timelines for AI advancement and R&D automation are expected to shorten.
AI Gateway now supports team-level Zero Data Retention (ZDR) and prompt training control features to strengthen data security.
Google has launched 'Google AI Edge Gallery,' an official app that lets users run Gemma 4 models locally on iPhone.
A Rust-based proxy has been released that compresses CLI output from AI coding tools to reduce LLM token consumption by up to 90%.
Google has released Google AI Edge Gallery, an open-source app for running on-device generative AI.
It's a plugin that uses 'caveman-speak' in Claude Code and Codex to reduce output tokens by up to 75%.
Marc Andreessen predicts that AI is not a mere fad but the culmination of decades of accumulated technology, and that agents will change the paradigm of software.
Meta-Harness is a system that automatically searches for and optimizes model harness code for LLMs through an agentic proposer.
A self-distillation (SSD) technique that uses only the model's own outputs, without a separate teacher model or reinforcement learning, significantly improved LLMs' code generation performance.