Profile avatar
simon.fedi.simonwillison.net.ap.brid.gy
Open source developer building tools to help journalists, archivists, librarians and others analyze, explore and publish their data. https://datasette.io […] [bridged from https://fedi.simonwillison.net/@simon on the fediverse by https://fed.brid.gy/ ]
985 posts 8,657 followers 2 following
Regular Contributor
Active Commenter
Posts 22 Comments 28

I'm now calling Claude Code "honey badger" on account of its voracious appetite for crunching through code and tokens looking for the right thing to fix https://simonwillison.net/2025/May/23/honey-badger/

submitted 3 hours ago • 0 comments

Once again, if your LLM system combines access to private data, exposure to malicious instructions and the ability to exfiltrate information (through tool use or through rendering links and images) you have a nasty security hole This time, GitLab […]

submitted 7 hours ago • 1 comment

Started a live blog for today's Claude 4 release at Code with Claude https://simonwillison.net/2025/May/22/code-with-claude-live-blog/

submitted 1 day ago • 1 comment

If your library doesn't have any documentation, it can't have any bugs https://simonwillison.net/2025/May/22/no-docs-no-bugs/

submitted 1 day ago • 1 comment

Today's other big new model is Devstral, an Apache 2.0 licensed LLM that specializes in code and seems very good from my initial experiments It's a 14GB download from Ollama, notes here https://simonwillison.net/2025/May/21/devstral/

submitted 1 day ago • 0 comments

I got access to Gemini Diffusion, Google's first diffusion LLM, and the thing is absurdly fast - it ran at 857 tokens/second and built me a prototype chat interface in just a couple of seconds, video here: https://simonwillison.net/2025/May/21/gemini-diffusion/

submitted 2 days ago • 0 comments

ChatGPT's new dossier-from-your-chats feature is a huge change to how it works, and as a power user who tries to control all of the model's input I don't like it at all “30 messages are good interaction quality (25%); 9 messages are bad interaction quality (7%)” […]

submitted 2 days ago • 3 comments

There was a pelican riding a bicycle in today's Google I/O keynote! https://simonwillison.net/2025/May/20/google-io-pelican/

submitted 3 days ago • 4 comments

I think the hardest problem in computer science may be "if I add 'PyMuPDF' (AGPL) to my open source Python library as a dependency in setup.py/pyproject.toml does that mean my entire library needs to be licensed AGPL? And if it's a plugin (loaded via pluggy/entry points) does that impact the […]

submitted 4 days ago • 6 comments

I built a new LLM plugin that can turn a PDF into an image-per-page for feeding into vision models, and in testing it found that GPT-4.1 mini hallucinates WILDLY if you feed it a blank white rectangle followed by a blank black rectangle https://simonwillison.net/2025/May/18/llm-pdf-to-images/

submitted 5 days ago • 3 comments

Qwen 2.5 VL is out on Ollama - I tried the 6GB one and it worked for describing a photo but gave me "It looks like the image you provided is a jumbled and distorted text" when I tried using it for OCR Anyone had any luck with it for that? https://simonwillison.net/2025/May/18/qwen25vl-in-ollama/

submitted 5 days ago • 4 comments

If you're at #PyConUS I would LOVE to talk to you about my various projects - I'll be stood next to my poster in Hall A tomorrow from 10am to 1pm, please drop by and say hi My poster is the one with the "teenage bedroom" aesthetic, I decided to fill […] [Original post on fedi.simonwillison.net]

submitted 6 days ago • 0 comments

Here's the full workshop handout plus annotated slides from "Building software on top of Large Language Models", a three hour tutorial I presented yesterday at PyCon US #PyConUS https://simonwillison.net/2025/May/15/building-on-llms/

submitted 8 days ago • 2 comments

I've been working on this for a while... llm --functions ' def multiply(x: int, y: int) -> int: """Multiply two numbers.""" return x * y ' 'what is 34234 * 213345' -m o4-mini https://simonwillison.net/2025/May/14/llm-adds-support-for-tools/

submitted 9 days ago • 1 comment

I asked o4-mini-high about OpenAI's core engineering principles and it either leaked or hallucinated the URL of OpenAI's internal engineering handbook https://simonwillison.net/2025/May/13/launching-chatgpt-images/

submitted 9 days ago • 0 comments

Made some notes on how Cursor works under the hood based on their security documentation - it turns out an organization's list of subprocessors offers a loose form of "view source" for their infrastructure! https://simonwillison.net/2025/May/11/cursor-security/

submitted 12 days ago • 0 comments

llama.cpp shipped new support for vision models this morning, including macOS binaries (albeit quarantined so you have to take extra steps to run them) that let you run vision models in a terminal or as a localhost web UI My notes on how to get it running on a Mac […]

submitted 13 days ago • 1 comment

Gemini 2.5 now applies the 75% cached token discount automatically - previously you had to manually configure it Potentially big cost savings here for applications that run prompts against the same long context, or continue existing conversations […]

submitted 14 days ago • 0 comments

Some notes on the gemini-2.0-flash-preview-image-generation model that's now available via the Gemini API https://simonwillison.net/2025/May/7/gemini-images-preview/

submitted 15 days ago • 1 comment

Published some notes on Microsoft's phi4-reasoning model, an 11GB download (via Ollama) which may well overthink things... it produced 56 sentences of reasoning output in response to my prompt of "hi" https://simonwillison.net/2025/May/6/phi-4-reasoning/

submitted 17 days ago • 2 comments

New Gemini 2.5 Pro preview model today, released in advance of Google I/O. Google claim it is state of the art in both frontend code generation and video understanding. https://simonwillison.net/2025/May/6/gemini-25-pro-preview/

submitted 17 days ago • 2 comments

New release of LLM, accompanied by a new plugin - you can now use llm-video-frames to turn a video file into a sequence of JPEGs and feed those into a long-context vision model like GPT-4.1-mini https://simonwillison.net/2025/May/5/llm-video-frames/

submitted 18 days ago • 1 comment