It even generates images using Nano Banana INSIDE the IDE! I provided an overview of the project we're building, discussed some design ideas, and it generated an actual image of the icon, just as it does with files
I've been testing Antigravity over the past few days, and the performance is truly impressive!
I'm using Gemini 3 Pro and Flash, and it's completing every request in one shot. I provide a detailed description of the task, and it resolves smoothly without any bugs
I'm testing GPT-5.2 Codex on Cursor this week, and it's surprisingly good. It's much better than the previous version.
It's more token-efficient, easily handles medium to complex tasks, and is significantly cheaper than the Claude options, at least in terms of consumption
@jakestanex@levelsio@spensersembrat@carrd And this is preferred way by Unplash, because they track how many times an image was downloaded. If you right-click and save it, it won't be tracked.
What do you mean I can't manually set cache control with Next.js?
That's super annoying, I just want to cache the pages for a while and I can't because of dynamic rendering 🥲
@heyimsunflower The usual, having a healthy routine and taking a break when I start feeling stressed.
This helps a lot, I still feel a bit overwhelmed at times, but nothing like the burnouts I had in the past
@goenning@MemoSparkfield Have you tried using Code Spell Checker on VSCode? It helps a lot with typos, and it also support other languages than English
@aTylerRice It's actually pretty simple, I just need to include the model resource in the final build, and somehow use it on my code.
In my case I'll be working with Tauri, so I can embed additional resources. I just need to figure out how to invoke from JS or Rust now
Has anyone tried building a desktop app with an AI model included?
I want to include an open source model on a desktop app, but I have no idea how to use or include the model in the project
#buildinpublic
@goenning By segments of text, I'm breaking down the code by chunk size (512 characters), but it's totally up to you how you'll do the splitting.
This doc section explain pretty well how it works https://t.co/eVS2WKC6hQ
@goenning In summary, the way it works is that you break down your PDF (or any other document) into small pieces of text and save these chunks on a vector store.
Then, when you do a question, you'll do a similarity search for relevant chunks, and send together to GPT to compose the answer
@goenning I'm working on a chat with code tool, almost the same thing as the PDF.
The only difference is how you'll load the document and index it in the vector store. Check out the setup and chat scripts
https://t.co/40uTNhECV6