1/ A long-term thread on useful blog posts, books, videos, and materials to go beyond the popular early AI courses/videos that exist.
Q: What comes next after you've written GPT-2 from scratch? What is interesting to read about to keep your motivation high?
This 👇
2/ Bonus material from Sebastian's book. Specifically going line by line and understanding how to convert GPT-2 to Llama 3.2:
Learn RoPE, intuition around RMS, SilU, grouped-query to modernize your understanding of LLMs: github.com/rasbt/LLMs-fro…
3/ This piece really unlocked my mind to B16: - pretty basic but good to build intuition rather than just use it to solve your nan loss problem because someone told you to.magazine.sebastianraschka.com/p/the-missing-…
1/ I am excited to announce the new Playground: Design graphics like a Pro. You can make T-shirts, Logos, Social Media Posts, and even invent new kinds of memes.
Powered by our next image foundation model Playground v3 (beta).
Go to playground.\com. No wait list.
2/ You can make amazing designs using Playground with graphical text generation to express yourself:
3/ Use Playground to make edits to images by just asking. We made a template library so you always start with a great image.
Talk to it like a graphics designer to make changes:
I am an old man now after two platform waves but what Apple did today was communicate:
"Hi guys, we made native integration points to make all you all AI model makers compete for our 1B-user scale. Of course, we'll show YOU some ideas first. Game on."
It is literally the best thing that could happen.
If you are behind on AI, you make everyone make the best experiences (if AI matters) to compete with each other and build the normal platform advantages you already have.
AI AppStore is providing 100s of integration points to use "AI" right when you need to.
Like browsers, you will be able to set your "default" model whether that's ChatGPT, Gemini, or a future image model like Playground :)
If I was a young or a college student in need of a little income per month, now is a brief window of opportunity where you can make really easy apps dedicated to solving one problem using ChatGPT in a single step where many back and forth chats are required. Make the UI visual.
Sometimes a few thousand dollars per month can empower you to think more ambitiously and the product building itself will teach you a lot about how to make things people want.
Oops typo. I thought Twitter had an edit button but I can't find it :(
2/ Playground now features a powerful collaborative Canvas editor to layer multiple edits on top of each other.
This goes beyond text-to-image giving you more control to make fine grained changes.
3/ Combine multiple elements to make an amazing image with more fine-grained control:
- Cut out a bear
- Add a bear into a forest background
- Get its edge + depth masks
- Add a balloon
- Draw a basic arm holding the balloon
If you're strong at Python, TypeScript, and have been learning a lot of ML fundamentals but still ramping up, you may consider joining us to do Applied AI where you help prototype products at the intersection.
A great way to ship real things and be forced to learn ML basics.