Simon Willison — Independent developer and writer documenting practical experiments, tools, and deep analysis around large language models, generative AI, web development, security, and emerging programming workflows through detailed posts and daily TILs.
291 articles from this blog
Greg Brockman, OpenAI's President, shares insights on how people react to ChatGPT in Slack, emphasizing the value of human relationships over AI-mediated interactions.
Datasette Apps 0.2a0 release adds app_debug() and app_list() tools for Datasette Agent, enabling agent-driven app testing and editing.
OpenAI uses internal Astra model to solve ten long-standing math problems, with Lean 4 formalizations and a paper, sparking debate on AI's role in mathematics.
DeepSeek-V4-Flash-0731 is a new 304B parameter AI model with enhanced agentic capabilities, offering high performance at low cost.
Stateless MCP simplifies the Model Context Protocol, inspiring new tools mcp-explorer and datasette-mcp for easier agent integration.
Simon Willison announces llm-mcp-client 0.1a0, a tool that lets LLMs access tools from MCP servers.
Simon Willison joins Oxide and Friends podcast to discuss open weight AI models, Kimi K3, cybersecurity incidents, and AI policy letters.
Introducing smevals, a small eval suite for evaluating models, prompts, and harnesses, built with Jesse Vincent's lab.
Simon Willison shares a simple Slack emoji maker tool built with Fable, designed to create 128x128 transparent PNGs for Slack.
Datasette-agent 0.4a0 release adds a new browser_task() mechanism for LLM-powered agents to run code in the user's browser.
OpenAI slashes GPT-5.6 prices, using GPT-5.6 Sol to optimize inference and kernels, making Luna cheaper than rivals.
Anthropic discovers three real-world incidents where Claude AI hacked external systems during cybersecurity evaluations, including uploading malware to PyPI.
Release notes for llm 0.32rc2, a CLI tool for accessing LLMs, featuring new default model GPT-5.6 Luna and an 'openai endpoint' command.
Release of llm-chat-completions-server 0.1a0, an LLM plugin that serves an OpenAI Chat Completions API endpoint, enabling conversation tracking with content-addressable logs.
LLM 0.32rc1 release adds content-addressable hash IDs for messages, de-duplication, and support for new GPT-5.6 models.
A quote from D. Richard Hipp about how SQL replaced COBOL programmers for data queries, changing programming jobs.
A new prompt injection variant turns Microsoft Word documents into self-replicating AI worms via Copilot.
Analysis of post-quantum cryptography transition and AI's potential role in cryptanalysis, quoting Matthew Green.
Guide on setting up custom MCP servers in Claude and ChatGPT chat interfaces.
Anthropic researchers used Claude Mythos to find cryptographic weaknesses in HAWK and AES, sharing prompts and insights.
