Comment by minimaxir
2 years ago
This post was originally made in 2022. In early 2023, I spent a lot of effort training a model with a new approach that made the generation quality extremely good...
Then the ChatGPT API came out and made all my effort obsolete.
In one hour, I was able to create a Notebook (https://colab.research.google.com/github/minimaxir/chatgpt_a... ) that was able to create mechanically valid and relatively balanced cards given a natural language prompt, even extremely absurd ones ("Create ten variations of Magic cards based on Spongebob Squarepants and ancient Roman history"). In all cases it's more stable and accurate than my hand-made solution.
That notebook is now obsolete too, due to ChatGPT's structured data support now allowing for even more control and stability. I need to create an updated MtG card generator at some point.
Yes. Very scary. This is a very good example of the speed of AI advancement. You used AI in 2022 to make a cool thing, then by 2023 GPT could do it better faster.
Everyone is saying the AI craze is hype and already dying away.
Meanwhile I can just talk to GPT4 conversationally and it will do a project for me in minutes that would have taken days.
It is not hype, and change is still rapidly occuring.
To clarify, I'm not the author of the original article (although I was aware of the work).
I've been working with generating AI Magic cards since RoboRosewater was popular.
Impressive anyway. I was looking for a deck generator that can generate standard, historic and explorer decks for MTG arena. I try to ChatGpt using context for the last sets, and prompt but quality wasn't what I was expecting. Sometime fails to add 24 lands or create decks Wich are copies of old decks. Try to fix it in prompt with no luck.
Amazing work. This is a real treasure. Thank you.
What structured data support are you referring to? I'm out of the loop, did they add something new in the API?
ChatGPT's June update added support for "function calling", which in practice is structured data I/O marketed very poorly: https://openai.com/blog/function-calling-and-other-api-updat...
Here's an example of using structured data for better output control, lightly leveraging my Python package to reduce LoC: https://github.com/minimaxir/simpleaichat/blob/main/examples...
How has your experience with function calls been? We tried doing code generation and making ChatGPT generate diffs and it seems to perform worse than the March edition.
1 reply →
Still incredible in 2023. Goes to show UX and maybe first to market will be the differentiator