Paper proves transformers are inherently succinct and that basic verification problems like emptiness/equivalence are EXPSPACE-complete, making formal LLM verification provably intractable.
HuggingFace ships a multi-agent economic simulation on a 3B parameter model.
NVIDIA demonstrates cold-starting a 120B parameter model in under 5 seconds on Kubernetes.
Research on regret minimization algorithms that adapt to opponent behavior in repeated games.
BRepCLIP applies contrastive multimodal pretraining to CAD boundary representation primitives for geometric understanding.
Tutorial on using MicroPython compiled to WASM as a sandboxed Python execution environment.
TinyTPU is a browser-runnable SystemVerilog systolic array implementation verified against numpy, demonstrating hardware ML accelerator concepts interactively.
Xiaohongshu releases dots.tts, a new text-to-speech model.
KITScenes multimodal dataset release targeting autonomous driving research.
Technical exploration of obscure C pointer arithmetic by security researcher Michal Zalewski.
Discussion prompt about AI systems failing discovery-oriented work by prematurely converging on answers.
Pre-registered experiment finds AI cited a fabricated author correctly within 6 days despite crawler blocks, raising questions about knowledge acquisition.
Exploration of using AI agents for test-driven development with a specify-encode-fulfill methodology.
Simon Willison notes the 0.1a2 release of micropython-wasm.
Developer discovers AI text detectors are unreliable after personal testing.
OpenLumara is a hand-coded, token-efficient AI agent framework designed for local LLMs with a modular architecture.
A personal blog-style walkthrough of how LLMs work, criticized for poor structure and unclear audience.
Documents progress toward AI systems that autonomously improve their own code, architecture, or training procedures without human intervention.
Anthropic releases an open-source framework enabling AI agents to automatically detect and analyze security vulnerabilities in software.
An AI-generated or AI-assisted game, narrative, or visual project set in a samurai-themed urban environment.
Systematically evaluates whether transformer attention mechanisms require all three separate query, key, and value projections, testing reduced variants.
A command-line interface tool that uses AI to automatically review code, flagging issues and suggesting improvements in developer workflows.
Huawei contributes a native vLLM backend that quantizes the KV-cache to reduce memory usage and improve LLM inference throughput.
South Korean legislation or policy mandates AI-based image scanning tools on online forums to automatically detect and censor prohibited content.
LangChain provides a Python/JavaScript framework for building LLM-powered agents and chains by composing tools, memory, and model calls.
Open WebUI delivers a self-hosted browser interface for interacting with local and remote LLMs via Ollama and OpenAI-compatible APIs.
Dify offers a production-ready platform with visual workflow builders and APIs for developing and deploying agentic LLM applications.
Hugging Face Transformers provides standardized model definitions, weights, and APIs for loading and fine-tuning state-of-the-art ML models.
A community-curated collection where users share, discover, and reuse prompts for ChatGPT and other LLM-based chat systems.
Ollama provides a local runtime for downloading and running large language models including Kimi-K2.6, GLM-5.1, MiniMax, DeepSeek, Qwen, and others on personal hardware.