On: GPTBot – OpenAI’s Web Crawler

[jwr]

I wonder — when GPTBot crawls my website, which has a number of translations performed using GPT, will it use all that data for training future models? That doesn't seem like a good idea, but I don't know how they could tell.

[Escapade5160 → jwr]

Models trained on the data from another model eventually leads to model collapse.

[kasdi → Escapade5160]

It’s like incest.

[thenonsequitur → kasdi]

That's actually an excellently apt analogy.