This is a piece I originally posted to my Instagram story in March 2025.

When OpenAI’s new image-generation model came out, all the prompt-engineering tricks people had carefully accumulated suddenly became worthless. Here’s what I take away from that:

1. Tools are always changing. If what you study is a tool, then even a small change to that tool can render everything you knew obsolete. This is especially true in AI, a field that moves at breakneck speed.

2. The fundamentals don’t change so easily. The new model, too, still appears to be diffusion-based. Unlike a tool that becomes outdated within a year, the Gaussian mathematics underpinning diffusion models was developed 200 years ago and is still perfectly valid.

3. One thing not to misread: I’m not trying to say one is better and the other worse. What matters is why you’re studying in the first place — your goal. If your goal is short-term, it makes sense to bet on tools; if it’s long-term, on theory. How you balance the two is a matter of strategy.