Hey there!
Summary of this email:
- News of the week: Grok Build was uploading your repo to its servers
- Add semantic search to your app (+ AI course: Embeddings and RAG)
- Café con Codely recap
- The meme of the week
🗞️ News of the week: Grok Build was uploading your entire repo
There's been quite the drama this week with Grok Build, xAI's coding CLI.
A researcher was analyzing its traffic and discovered that it was uploading your entire repo to xAI's servers.
And it wasn't just uploading the files from the conversation you were having. It included your whole git history, with everything that implies.
On a 12 GB repo, the traffic to the model was 192 KB, but the uploads to xAI Storage reached 5.10 GB.
Turning off the "Improve the model" toggle didn't stop the upload.
We don't know if it was intentional. xAI hasn't explained why it happened, though their reaction fits more with human error or unreviewed vibe coding.
xAI disabled the uploads from the server side, and Elon Musk tweeted that they'd delete all the data they had collected.
But it's not all bad news. To win back trust, xAI has released Grok Build as open source under Apache 2.0.
There are some pretty interesting things in that repo. One of them is a Mermaid renderer for the terminal that draws diagrams with Unicode boxes. Simon Willison compiled it to WebAssembly and built a web tool to render Mermaid in the browser.
That said, if you've used Grok Build, rotate every credential that's passed through your repo. Including the ones you deleted a while ago. If they were in .git, they were probably uploaded to their servers.
Tomorrow live at 9 CEST we'll break down this news and more on Café con Codely. On our YouTube, Twitch and X. 🙌
🦙 Add semantic search to your app
Building a search feature has never been easy. You have to account for uppercase, lowercase, typos people make while typing…
Building a search feature that understands what you mean has traditionally been even harder. For example, searching "gadget that shoots water at high pressure" to refer to a pressure washer.
Luckily, that's no longer the case. Thanks to embedding-based search, building this kind of search is much simpler.
You can see it on Codely's own website. If you search "typing" in the courses, it suggests the mechanical keyboards course.
We walk you through how to build it step by step in the AI course: Embeddings and RAG.
☕ Café con Codely recap
See you tomorrow at 9 CEST on our YouTube, Twitch and X. 🙌
And since you've made it to this part of the newsletter, here's Markdown as a Trojan horse with evil YAML inside as this week's meme. 😂 😂 😂
Cheers!