David Smith

Cc @sajilicious.bsky.social

It’ll be back soon, if only as a static site so the links don’t break. TypePad, the blog host shut down with only a few weeks warning. Read the story of the rescue operation here: medium.com/@gcordidoa/m...

It should hopefully be coming back soon as a static site. TypePad didn’t give us much warning but we were able to download all the posts and images before it shut down.

Join the Azure AI Foundry Discord Server! A discussion space for developers and interested individuals looking to build with Microsoft AI tooling. | 35477 members

Join the AMA at 4PM SF time, here: discord.gg/YZbkuKb3?eve...

Check out the awesome-copilot repo, here: github.com/github/aweso...

Devs: looking to up your GitHub Copilot game? In 60 mins, I'll be hosting an AMA with Aaron Powell, maintainer of the Awesome Copilot repo, your source of prompts, chat modes and more!

Bio pic of Simon Willison, wearing glasses and standing against pale brick wall, with his name in big letters and the Talking Postgres podcast square thumbnail in bottom left.

🎙️ New @talkingpostgres.com #podcast Ep30 just published!

Guest @simonwillison.net talked about AI for data engineers. We skipped hype & boosterism to chat about Postgres permissions, alt text, 150-line SQL queries, & pelicans on 🚲

🎧 talkingpostgres.com/episodes/ai-...
📺 youtu.be/8SAqeJHsmRM?...

simonw/codespaces-llm I found out today that GitHub Codespaces come with a GITHUB_TOKEN environment variable... and that token works as an API key for accessing LLMs in the GitHub Models collection, which …

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

codespaces.new/simonw/codes...

Notes on how I got this working: simonwillison.net/2025/Aug/13/... - plus a detailed TIL on using devcontainer.json to configure Codespaces: til.simonwillison.net/github/codes...

Looks like something Simon Stålenhag would paint! Where is that?

Chatbots — LLMs — do not know facts and are not designed to be able to accurately answer factual questions. They are designed to find and mimic patterns of words, probabilistically. When they’re “right” it’s because correct things are often written down, so those patterns are frequent. That’s all.

They Asked an A.I. Chatbot Questions. The Answers Sent Them Spiraling.

This @kashhill.bsky.social piece is wild and disturbing:

Consciousness exists in the observer’s mind

HMMER is literally *the* go-to tool for profile HMMs in bioinformatics. I've used it countless times on a wide range of projects spanning viruses, retrotransposons, etc. Absolute insanity! hmmer.org

I wish journalists would stop framing political news as wins or losses for politicians or political parties. It’s an abrogation of our mission. Why do we keep doing this? Is the spending bill a win or loss for the American people? Who does it help? Who does it hurt? Is it sound policy? Fiscally?

Just signed you up for 1000 newsletters.

Outstanding!

Hmm, looks like not yet, which is unusual. Maybe it’s because it’s only available as “latest” in OpenAI’s API, not a specific version number.

Quite a bit. Content safety filters (important for brands), provisioned throughput (guaranteed TPM), AI Foundry integration. OpenAI models are usually available in Azure same day, not sure about gpt4o you reference there.

For one thing, data is contained to the region you specify rather than the entire globe — important for various compliances.

It’s humbling to realize how many innovative/novel things in tech are actually often triggered by existing, proven ideas or practices: and using them in a practical setting.

Latest discovery: MCP in-part inspired by LSP (Microsoft’s Language Server Protocol)

Did you know that if you put hot chocolate in a thermos and forget about it, it turns into a bomb?

My new hobby is asking ChatGPT to name the new cocktail my partner creates each evening. It did a good job with this one.

Life hack: write down stuff you’ve already done on your to-do list and slam a checkmark on it

X post by Timothy Snyder (@TimothyD ...), posted 6 hours ago. The post reads, "The people Attorney General Bondi is calling 'terrorists' have had no trial and no chance to defend them. We do not know they are. Ignoring the Constitution does not make us safe. It puts us all in peril."

I know this first hand. When I was a child, my entire community was branded as spies and sabateurs, justifying our internment inside barbed wire camps for years without due process, trial or charge.

I’m crying, this is so funny

WEAPONIZED MANSPLAINING FOR ESPIONAGE ☠️☠️

Well worth a watch if you're feeling helpless and confused these days.

Coincidentally, 65536 is the maximum number of rows for the Excel 97-2003 file format

GPT-4o? GPT-o3? Llama 3.3? Command-R? Phi-4?

If you're not sure what Generative AI model to use (and when!) you're not alone — there are thousands to choose from. Tune in to Model Mondays at 10:30AM to learn now. First episode streams live in a couple of hours! github.com/microsoft/mo...

Big props to @hnky.nl for all of his work behind the scenes to make this all happen, and to colleagues at @microsoft.com for their sponsorship of Global AI Community!

Global AI Bootcamp 2025 The Global AI Bootcamp is an annual event that occurs worldwide, where developers and AI enthusiasts can learn about AI through workshops, sessions, and discussions. Local chapters of the Global AI Co...

And if you don't see one nearby -- it's not too late to host and speak at your own Bootcamp! Information for organizers is at globalai.community/bootcamp#org...