Criticism has emerged that METR's AI time horizon graph has serious flaws, including data bias and measurement errors.
A 12.6M-parameter DCGAN was implemented in pure C on a RISC-V microcontroller with 512KB of SRAM.
Providing LLMs with a structural graph of the codebase leads to deeper model exploration, increasing context usage by about 54%.
A CUDA-based FWHT has been added to llama.cpp, improving inference speed when using KV-cache quantization.
Anthropic's new model 'Mythos 1' appears set to be introduced into Claude Code and Claude Security.
A patch is being prepared in llama.cpp to fix VRAM exhaustion and crash issues in Split Mode Tensor when using multiple GPUs.
Delta Attention Residuals has been unveiled, boosting LLM performance by solving the routing collapse problem of existing residual connections.
It guides you through using Git and GitHub with VS Code to manage projects.
An open-source MCP server has been released that tracks citation URLs from major AI engines like Claude and ChatGPT, and compares them with Google Search Console data.
RTPurbo, a technique that converts a Full Attention LLM into a highly efficient Sparse model with only a few hundred short training steps, has been unveiled.
Anthropic's Felix Rieseberg shares concrete examples and his philosophy for using Claude in both work and daily life.
The study analyzes the limitations of LLM agents in adhering to API and architectural constraints when developing production-grade backends.
The Financial Times reported on 'Heretic,' a tool that can remove Llama 3.3's guardrails in just 10 minutes.
Microsoft is canceling external AI coding tool licenses and switching to its own tools to reduce cost burdens.
NuExtract3, a VLM based on Qwen3.5-4B specialized for document data extraction and Markdown conversion, has been released.
Anthropic is finalizing a classified contract with the NSA to provide secret surveillance tools.
A 2-bit quantization technique has been released that compresses KV cache memory usage by about 7x while minimizing performance degradation.
Lavern, an open-source agent-based legal system featuring 67 specialist agents, has been released.
NuExtract3, a 4B-scale vision-language model specialized in document understanding and structured information extraction, has been released.
Apple will unveil iOS 27 at WWDC 2026, significantly enhancing the AI capabilities of Genmoji and Image Playground.
Elon Musk announced that a Grok model with 0.5T parameters will be released next year.
The SDK 'Cider,' which supports W8A8 activation quantization on Apple Silicon by addressing MLX's limitations, has been released.
A checkpoint feature has been proposed to solve the problem of reprocessing the entire prompt whenever the context changes in llama.cpp.
UC Berkeley School of Law will ban AI use for most assignments starting summer 2026 to cultivate critical thinking.