They infringed nothing. Two judges have already ruled that training on copyrighted data is fair use https://www.whitecase.com/insight-alert/two-california-distr...

But displaying regurgitations of very similar content may not be fair use. Fair use is a very delicate affair. One factor is whether the modified work poses as a market replacement for the original work.

The issue is, in part, a concern that ChatGPT responses are often just simple derivations of the original content in ways that wouldn’t be considered fair use.