@thanius@peertube.chuggybumba.com I use llm's to find stuff or solutions for items where I am able to verify the output immediately (or with minimal effort). Example: last week I had a problem with my network interface. I prompted a (locally running) llm for the 'nmcli' commands which I hadn't used for years because I don't work as a sysadmin. The output was usable and saved me quite some 'manual-reading time'. I was able to verify the output because my internet connection was working again :-)
It is my opinion that you should limit the scope of llm's there. If you start to blindly accept the output of llm's (which happened on a big scale with the php2shell cve this week, for example) you're just using the wrong tool for the job. Accepting the output of an llm without verification is only acceptable if an incorrect output is acceptable.
The big problem with llm's is how you can abuse this technology. From a sociological point of view, it is the 'best tool ever for the (flaws in) meritocracy': it is a perfect tool to cheat your way up to a higher social class. And this is, in my opinion, the real problem. As long as llm's help people reach a better life standard by generating fake art/articles/cv's/programs/solutions to math problems/science it will be abused on the current massive scale. I have not found a solution for this problem, but I believe we should not discuss if 'ai' is good or bad ('ai' is just a subset of 'programs'. Some are useful, some are less useful) but ask 'where is, for example, an llm useful?' and learn each other to strictly limit the scope.
To make it more clear what the scope of an llm is, I would suggest to remove the word 'intelligence' because an llm has nothing to do with intelligence whatsoever (but it is a very good marketing buzzword because a user can feel himself implicitly intelligent by using this technology). The scientists who make the great podcast 'mystery ai-hype 3000' published a nice attempt: https://buttondown.com/maiht3k/archive/how-to-talk-about-ai-without-adding-to-the/. They replaced the word 'artificial intelligence' with 'probabilistic automation'. This sounds technically much more accurate, but obviously, any marketing team would rate this 0/10 😁

I think AI can be safe and useful only when used in a certain way.

For example, prompting AI to write a program that does something is bad. Even if prompting it to do smaller things like write a function that does something small is not ideal.

However, asking AI what command is ideal for performing a task then the programmer reads the docs on that command and implements it, is safe and helpful.

In any case the programmer should not take the AI as a source of truth. The AI provided program or function will most likely be buggy and/or insecure. An AI provided command recommendation can still be complete bullshit.