John Siracusa

It's just, it's not, you know, the energy usage of the AI feature in Photoshop is not significant in the grand scheme of human energy usage.

And I think they're, if they're telling the truth, which I believe they are, about only training their models on data that they owned or legally licensed,

There you go.

I think it's perfectly fine.

And I think that is possible for coding agents.

Right now, it's not happening because, you know, even if you have like an MIT license open source thing, does the MIT license say anything about AI training?

I bet it doesn't because it didn't exist when the MIT license was written.

So even the most permissive open source license probably doesn't explicitly allow AI training.

So it's an open question.

But you can imagine a future

in which I know Stack Overflow is doing this.

Stack Overflow is essentially licensing their data to AI people.

And I think people are grumpy about that, but legally, you know, probably good ethically, whatever.

I mean, Stack Overflow is dead with AI.

I know, but like the only asset they have are all their questions that people train the models on.

So they're selling those to, they're selling access to those.

I know, but the AI coding agents could not exist at their current level without training on things like Stack Overflow.

So, you know, get well beginning is good.

But also for open source projects, you can imagine a future in which licenses are made that say explicitly things about whether or not AI training is allowed and under what conditions.

And if that's the case, I can imagine a future Adobe style coding agent that is entirely trained on licensed code.