Log4Shell required that the attacker be able to generate a log message containing specific parameters. It was a severe exploit, but not one that worked in every Java application with the affected Log4j versions. Attackers had to find calls that logged user submitted text or query paths first. So the list of prerequisites was larger: had to be using Log4j and had to have logging of user generated text.

Still a very large attack surface, but not nearly as ubiquitous as this one.

It's trivially exploitable because it requires no special set up, no special credentials, no privileged access of any kind. If you can call a PHP server and you have the exploit script, you can execute arbitrary code as their `www-data` user.

Usually, attacks are at least conditional and require something that weakens security guarantees.

Yes but that means it is trivially exploitable today since anyone can get access to a LLM and where to point it towards. 2-3 years ago, things were different, that is also correct but that doesn't mean things stay frozen in time.