Believe it or not, I have a real job. It’s hard to balance holding down a job between not at all panicking about LLMs dooming humanity, and not being overly concerned if the bombs drop. Come on—it’s game over if that happens, right? I figure we might as well enjoy life instead of spending small fortunes like so many determined preppers do, against futures with only a minute chance of becoming reality. Odds are, your go bag won’t help you go as far as you thought anyway. Counterpoint, learning survival skills is fun, and talking about the end of the world is a great way to scare people into giving you their money for stuff they don't need. Oh, oops. I was supposed to be making a counterpoint. Learning survival skills really is fun, I suppose. There, counterpoint made.
Anyway, I digress. I asked my LLM of my employer's choosing recently if I should use its services for “mission critical” functions. I expected an “absolutely not” kind of answer and, for a few seconds, I got one. Then the screen went blank and up popped something to the effect of: “I’m sorry Dave. I’m afraid I can’t let you do that.”
Apparently, companies don’t like it when you use their AI to prove their AI isn’t a good fit for your use case.
Not to be defeated, I went to the other three LLMs I use on a regular basis (the side hustlers, not the work-provided one) and asked them the same question. None of the others had any issues talking about it. One gave me a full technical essay, another downplayed my concerns and even defended the censorship, and the third devolved quickly into a bizarre, nonsensical response that seemed to tie in talking points from dozens of earlier prompts (but hey, it's reliable).
Yes, they all explained I likely triggered some guardrails, the same way I did when I once asked ChatGPT about the dangers of its chat-sharing feature. Naughty LLMs and censorship—though ChatGPT’s engine denied and still denies, amusingly enough.
I suppose all this rambling is really to say: seriously, don’t use what’s passing for AI these days for mission critical functions. If your life depends on it, and the model blinks out with “I’m sorry Dave,” you’ll wish you spent that budget on tools that actually work, like I don't know - machine learning, or maybe a really scrappy intern.
I know, I've been leaning into the sarcasm and satire lately, but it's worth talking about how the little AI that couldn't really couldn't (and couldn't even).
Have a funny or otherwise interesting interaction with an 'AI' you'd like to share? Leave it in the comments, or reach out through the contact page.
No comments:
Post a Comment