When looking at humanity, I often am unable to understand what it is doing: creating problems, trying to solve them (without succeeding) while the “solutions” create new problems, which have to be solved.
A deeply recursive eternal problem.
Related: OpenAI’s watermarking for images
PieFed just announced they’re gonna use c2pa to label AI images: https://piefed.social/c/piefed_meta@piefed.social/p/2304120/piefed-1-7-12-detects-ai-generated-images
1 Like
When looking at humanity
True, but this sounds like a god (or AI?) looking from above down to humanity. And denying humanity is also not a solution, right?
I think identifier names would provide plenty of ambiguity room in which the watermark can express itself. After all humans similarly struggle to decide among infinite possible naming choices while programming.
1 Like
Right, but now you’re talking about pure one prompting which is a carnival trick. As an experienced developer who uses llms for coding daily I can tell you that you shouldn’t even trust the llm to name stuff. Make a clean design, tell the AI to implement it and then take a moment to find where it decided to do something else and ask why it deviated from the plan. Most often it’s trying to build a bridge over a rock instead of going around it. Most often it’s missing a piece of data available somewhere else in the app but not passed as an argument to the method/scope you’re in.
The only time I let an ai run loose is when asking for something I expect to be in its training data, like a snake game or a sorting algorithm. Ask it to make a social media clone with privacy and four prompts later it will silently drop privacy because that’s not how it’s been trained to do social medias.
1 Like