Edit: my point being: can you tell me why PR reviews generated by an LLM are bad? If the discussion was actually diving into the details then I’d be interested but blanket statements are just stupid

If the LLM is owned by an external corporation: Because it generates a workflow dependency that will be enshittified to screw you over. The current quality of service is backed by unfathomable amounts of investment capital, and they will want to make a profit (or political gains) out of it at some point. Also, if you are dealing with any sort of valuable information, because the plagiarism company will commit plagiarism to you.

If you're running a model locally, then it will still hallucinate, be racist, sexist, and biased in other ways (meaning it will be unnecessarily harsh to some and gullibly approving towards others). Though to be fair it might not be worse than you or your colleagues. Also, because model weights can't be interpreted, you're going on faith that the LLM has been trained with your best interest in mind. There is nothing stopping the LLM trainers from, for example, teaching it a pass phrase that will get it to sign off on anything. And then it's up to your underfunded and undertrained human reviewers to catch whatever malware they're trying to sneak into your code base.

It seems quite possible that your risk profile and tolerance for discrimination are such that it makes sense for you to run a local LLM. But the cloud-based ones are definitely going to be bad for you in the long run. Unless you're phoning it in at work and this will help you half-ass it even more, of course.