Pamela Fox

So now I'm wondering: is there a solar powered webcam that can watch the caterpillars all day long, broadcasting the pupation progress to the web? I have one solar-powered Eufy webcam, but it's only motion-triggered.

Collage of caterpillar, chrysalis, skin

Project Adopt-a-Pupae is going well! I attached the caterpillar to our arbor, and 2 hrs later, they were already a chrysalis! I didn't realize they could transform so quickly.

Terminal with partially written git checkout command

VS Code Intellisense auto-completes git branch names in the terminal! Is this new? How have I never seen this before? I love this feature, as I'm always making new branches and forgetting their exact names.

Collage of pupating monarchs

My handyman just brought over some pupating Monarchs from a house that wanted their pupation-plant removed. I've attached them to my arbor, but will this work? I've never transplanted pupating Monarchs!

Collage of wasp nest pics

Yellowjacket wasp nests are huge! We called vector control on this one - I love most insects, but yellowjackets are both ineffective pollinators and dangerous.

Screenshot of automated PR review

I made a GitHub action using @simon's llm CLI and free tier GitHub Models to review PR descriptions for completeness, so that I don't have to be the one begging people to write decent PRs.

It works, but it nitpicks even harder than I do! I gotta prompt […]

[Original post on fosstodon.org]

Live streaming: playing with llm CLI https://www.youtube.com/watch?v=epZJ3MCPESw

PyLadies of San Francisco | Meetup PyLadies San Francisco is a large chapter of the international PyLadies mentorship group (fiscally sponsored by the Python Software Foundation) with a focus on helping people who identify as women in a way significant to them become active participants and leaders in the Python community in the San

I'm excited to see that PyLadies SF is happening again!
https://www.meetup.com/pyladiessf/
Three upcoming events in next few months.
The meetup scene is heating up again in SF. 🔥

GitHub - arilamstein/streamlit_tutorial: Tutorial for learning Streamlit Tutorial for learning Streamlit. Contribute to arilamstein/streamlit_tutorial development by creating an account on GitHub.

Follow Ari's tutorial here to try it yourself!
https://github.com/arilamstein/streamlit_tutorial

Screenshot of graph of state population change

Thanks to Ari Lamstein for leading a Streamlit tutorial tonight at SF Python!
I learned how easy it is to make plotly graphs that respond to interactively filtered pandas dataframes.
Our 50-line app for visualizing where everyone moved during the pandemic:
https://pamelafoxtutorialnow.streamlit.app/

Today, we're announcing our first hosted infrastructure product: pyx, a Python-native package registry.

We think of pyx as an optimized backend for uv: it’s a package registry, but it also solves problems that go beyond the scope of a traditional "package registry".

Chart of deception metrics

The GPT-5 model card is interesting-
it shows how they evaluated hallucination %, safety, deception, jailbreaks, etc.
https://cdn.openai.com/pdf/8124a3ce-ab78-4f06-96eb-49ea29ffb52f/gpt5-system-card-aug7.pdf

How to build secure and scalable remote MCP servers More context can mean more attack surfaces for your projects. Be prepared for what lies ahead with this guide. The post How to build secure and scalable remote MCP servers appeared first on The GitHub Blog.

There's so much interest in MCP, but also so much risk, so I'm trying to understand the MCP auth story.
Two helpful articles:
https://github.blog/ai-and-ml/generative-ai/how-to-build-secure-and-scalable-remote-mcp-servers/
https://modelcontextprotocol.io/specification/2025-06-18/basic/authorization

Updated my GPT-5+RAG blog post with more findings:
https://blog.pamelafox.org/2025/08/gpt-5-will-it-rag.html

GPT-5 reasoning models are more likely to:
* realize when they don't know enough to answer a question
* use top-level lists when answering
* use smart quotes
* generate wordier queries […]

You can try that out yourself by launching a Codespace using this link:

https://codespaces.new/simonw/codespaces-llm?quickstart=1

Notes on how I got this working: https://simonwillison.net/2025/Aug/13/codespaces-llm/ - plus a detailed TIL on using devcontainer.json to configure Codespaces […]

Screenshot of a GitHub Codespaces VS Code interface showing a README.md file for codespaces-llm repository. The file describes a GitHub Codespaces environment with LLM, Python 3.13, uv and the GitHub Copilot VS Code extension. It has a "Launch Codespace" button. Below shows a terminal tab with the command "llm 'Fun facts about pelicans'" which has generated output listing 5 pelican facts: 1. **Huge Beaks:** about their enormous beaks and throat pouches for scooping fish and water, some over a foot long; 2. **Fishing Technique:** about working together to herd fish into shallow water; 3. **Great Fliers:** about being strong fliers that migrate great distances and soar on thermals; 4. **Buoyant Bodies:** about having air sacs beneath skin and bones making them extra buoyant; 5. **Dive Bombing:** about Brown Pelicans diving dramatically from air into water to catch fish.

GitHub Codespaces provides a free web-based development environment backed by a Linux container... and it turns out it also includes a GITHUB_TOKEN environment variable that provides free API access to the GitHub Models collection of LLMs

... here […]

[Original post on fedi.simonwillison.net]

Screenshot of MCP.json diff

If you're passing secure values into MCP servers for in mcp.json for GitHub Copilot, use ${input:}. VS Code will prompt you for the values and store them securely, and you won't accidentally check them into source control.

Diff for mcp.json […]

[Original post on fosstodon.org]

🔴 Time for a lil live stream! GPT-5 - Will it RAG?
https://www.youtube.com/watch?v=qUhf0JFQst4

Tables of metrics across GPT models

I ran evals of gpt-5 model family for RAG, interesting differences!
https://blog.pamelafox.org/2025/08/gpt-5-will-it-rag.html

TLDR: gpt-5 reasoning models love top-level lists and “smart” quotes, and are much more likely to say "I don't know" 🤷🏻

Human Who Codes Newsletter - The return of tech specs .ck-link { text-decoration: underline; } @media only screen { .email * { word-break: break-word; } } @media screen and (max-width: 384px) { .mail-message-content { width: ...

Nice post from Zakas about the increasing importance of PRDs (or "design docs", as I call them), now that LLM-assisted coding is more common.
https://ckarchive.com/b/75u7h8h60nvreu6rgg7rlawlrvr66tnhox2e0

Comparison of outputs from two models

I am testing out the GPT-5 models for our RAG apps, and I'm impressed so far: gpt-5-mini avoided a hallucination that every other model generated. Much better for a RAG app to say it doesn't know than to hallucinate!

@simon Looks like you didnt try out -chat, right? Debating which one to use as default for our RAG chat applications. The -chat doesnt support reasoning.

GPT-5: Key characteristics, pricing and model card I’ve had preview access to the new GPT-5 model family for the past two weeks, and have been using GPT-5 as my daily-driver. It’s my new favorite model. It’s still …

I've had preview access to GPT-5 for a couple of weeks, so I have a lot to say about it. Here's my first post, focusing just on core characteristics, pricing (it's VERY competitively priced) and interesting details from the GPT-5 system card https://simonwillison.net/2025/Aug/7/gpt-5/

update: gpt-5-chat is NOT a reasoning model, the other ones are. trying now.

I'm trying out gpt-5-chat now, and am confused as to whether it's a reasoning model.
It does *not* take in a reasoning effort parameter, and I haven't seen reasoning tokens in any of the outputs so far. But I thought they said it sometimes reasons?

GPT-5 announcement happening now: https://www.youtube.com/watch?v=0Uu_VJeVVfo

A step-by-step guide to falling down the stairs:
* Step 1
* Step 2
* Step 5
* Step 8
* ...

(I take no responsibility, this was from my youngest)

Screenshot of ollama running while asitop displayed in other tab

I tried out the new OpenAI gpt-oss:20b using Ollama on my Mac M1 (16GB RAM).
It was 16 GB download, and required ~10 seconds per token, once it finally started thinking.
Not practical for use on my machine. Anyone else tried it?

BART reply on Bluesky

BART replied about what to do! Weapons are not allowed.

Prepping a talk demo based on my struggles to find a good garden hose.
SERIOUSLY THEY ALL BREAK. MY HEART.
Does anyone have a good hose rec? I will pay money if it actually wont break.