If you have a functioning brain, LLM’s are just a really fancy text generator that gets facts correct about 50% of the time and can be useful for writing emails if you spend the time to proofread the email before slapping the send button. It’s also fairly good at generating functional boilerplate for programming projects. Really handy for small projects. However, LLM code, in my rather limited experience, quickly becomes a trap on more complex projects because the LLM has no capacity to grasp the scope of a project and frequently leads to spaghetti code that is utterly unmanageable. This may be operator error on my part, or just because I’m using local models that are limited by my machine.
The rest of society view a computer as a magic box where their favorite TikTok’s live, and thus are terrified that a machine can imitate human language well, and are too brain rotted to bother proofreading or fact checking a goddamn thing, so they assume LLM’s can accurately and plainly articulate the whole of human knowledge to you at a whim.
I am at a point where if I hear anyone saying how smart “AI” is, I just assume they have a room temperature IQ and lack any critical thinking skills whatsoever. It’s not an AI, it lacks any intelligence whatsoever. An LLM simply generates sentences based on what was fed to it. It has no capacity to understand the words it’s saying. It just knows that based on these tokens, the next set of tokens should be X, Y, and Z. It’s a Chinese room, It’s fucking autocorrect on crack. It’s a waste of time and resources for 90% of things it’s used for.