Generative AI has become the conversation topic of these days. Sceptics see the doom of the world rising, visionaires see old dreams come true quicker than expected, and most agree on "this will change everything".
Sure, it is nice to have images generated from prompts. I use this to revamp my old food blog with the ugly images. And surely it is interesting to have a conversation with ChatGPT about any topic you dare to ask. Even if we know that system is lying like a politician briefly before elections, it is also as charming and so we start dialogs about important and personal questions and begin to align with the answers given.
All this AI craze lacks for me of one thing mostly: knowledge about what it really is and what it really can do.
I played around with ChatGPT and Midjourney (as advanced consumer) and Stable Diffusion (mostly following some youtube kids), learnt some of the buzz words (Model, inference, token, semantic vectors, etc.), tried (and succeeded with) some "naive integration" but the more I see, the more questions come up.
How do I build support for new knowledge?
How does input be modelled to get good answers?
How do I ensure some kind of quality side rails?
Which technologies do I select and how do I combine them in a good way?
How can I operate safely, without too many costs?
To get answers, I started this project. I have a "classic" IT background, so I can move somewhat swiftly in the premises, but orchestrating all these new technologies and managing a design vision to reach desired outcomes in this area is new for me, too.
Walk with me first through some exercises to get going with an efficient learning environment.
Open the learning notebook in Colab. You need to set this up in Google. If you already have a Google account, you are done with a series of confirmations. Otherwise create that account first.
This project is a working documentation, not a complete tutorial. If you want to comment or have other ideas, open an issue or create a pull request :)