@pamelafox.bsky.social
1.6K followers 170 following 700 posts
Posts Media Videos Starter Packs
pamelafox.bsky.social
I'm having a "test-off" where I ask different models to write tests for an API, then compare their failures and coverage.
So far:
* GPT-5 has 100% passing tests, 49% coverage.
* Sonnet 4.5 has 95% passing tests, 56% coverage.

What other models should I try?
pamelafox.bsky.social
I love the iNaturalist data!
I made a FastAPI app today that outputs "phenology" charts so you can see the active months for bees in the bay area.
(Most bees peak around May, like the Pyrobombus bumble bees)

Chart of bee activity
pamelafox.bsky.social
For the CS exams at Berkeley in 2021, students hand-wrote their answers, and GradeScope OCR'd them as best it could. We still had to hand-grade many of the digitized answers.. I once graded 500 regular expression answers in one sitting. Never doing that again!
pamelafox.bsky.social
I'm trying Postgres with PostGIS in Python today. Should I use GeoAlchemy, or can I just use SQLAlchemy2 on its own?

geoalchemy-2.readthedocs.io
pamelafox.bsky.social
I only use Streamyard anymore, its surprisingly the most reliable software I've used.
pamelafox.bsky.social
LLMs are slow! But you can make the LLM answer appear faster by *streaming* the response, one token at a time.

Steps:
1) Call LLM API with stream=True
2) Return tokens over HTTP streaming response
3) Re-render in frontend

I wrote more in this post:
techcommunity.microsoft.com/blog/azurede...
The importance of streaming for LLM-powered chat applications | Microsoft Community Hub
Thanks to the popularity of chat-based interfaces like ChatGPT and GitHub Copilot, users have grown accustomed to getting answers conversationally. As a...
techcommunity.microsoft.com
Reposted
globalai.community
Tomorrow 6 PM at GitHub HQ! 🎙️

Valence AI: Emotion detection
Nox: Anticipatory AI
Pol Peiffer (Sierra AI): Voice agents

SF AI community gathers. Arrive early - limited space.

Last call ⬇️
aka.ms/aist-reg/gai...

#AIShowAndTell #Tomorrow
pamelafox.bsky.social
Y si habla español, tendremos una serie para hispanohablantes, con @madebygps.com

🔗 Regístrese ahora:
aka.ms/PythonIA/Serie
pamelafox.bsky.social
My sister's an undergrad ML researcher and trying to choose a focus/lab. She's hearing conflicting advice about the best topics to pursue.
Any tips from those of you in the ML academia space?
pamelafox.bsky.social
Want to learn how to use Python with generative AI models?
Our free livestream series starts tomorrow!

For 3 weeks, every Tues/Wed/Thurs, tune in for:
🔍 LLMs, Vector Embeddings & RAG
🖼️ Multimodal Models
🤖 Agents & MCP
🛡️ AI Safety, Quality & Evaluation

Register @ aka.ms/PythonAI/ser...
Listing of Python+AI series dates and titles
pamelafox.bsky.social
awww and its a pseudo marquee!
pamelafox.bsky.social
yeah it took me a while too. you can PerksPlus it!
pamelafox.bsky.social
Currently reading "Empire of AI" by Karen Hao, a history of OpenAI based off interviews and documents.
bookshop.org/p/books/empi...

Really interesting to get more insight into how prioritization decisions were made. I now understand the OpenAI/Anthropic split better.
Empire of AI: Dreams and Nightmares in Sam Altman's OpenAI
Dreams and Nightmares in Sam Altman's OpenAI
bookshop.org
pamelafox.bsky.social
Yep I saw that! I'm asking on behalf of a dev who wrote their CLIs in Rust, so alas, he can't use that. Love that you have that for Textual though, as I write my CLIs in Textual usually.
pamelafox.bsky.social
Is there an equivalent of playwright/selenium for CLI tool E2E testing?
pamelafox.bsky.social
Great tips from @simonwillison.net on agentic coding loops:
simonwillison.net/2025/Sep/30/...
1) Use safe environments (like Codespaces) to enable YOLO mode
2) Use credentials with fine-grained access and constrained budgets for API calls
3) Setup automated tests so agent knows when it's succeeded
Designing agentic loops
Coding agents like Anthropic’s Claude Code and OpenAI’s Codex CLI represent a genuine step change in how useful LLMs can be for producing working code. These agents can now directly …
simonwillison.net
pamelafox.bsky.social
Added a smoke test to a demo app today, and it immediately caught a bug that I'd introduced an hour ago in a deployment-only path of the code.

Phew! 😅

Workflow file: (using Playwright for the test)
github.com/Azure-Sample...
Screenshot of smoke tests workflow