Open the error, copy it and give it to the model. Give the model context to work with (I use Codex locally, create a local repository and work in it, it makes all the changes, you have to FTP the files to your test server and check if the fixes work). If you work with ChatGPT or other model where you can't give direct access to a repo you can paste also the php code from the file that is at fault.
Usua,lly those fixes are very easy (bringing new php compatibility to the code).