Overheard a data scientist say LLMs are just 'fancy autocomplete' at a meetup last night

Guy named Ryan from some startup was talking to his buddy about how we're overhyping all this AI stuff. Said it's basically just predicting the next word really well. I mean, I get the technical side of it but honestly it kinda stung hearing it put that way. I spent like 6 months learning how to prompt engineer and fine-tune models. Makes me wonder if I'm just learning a parlor trick or something real. Anyone else hearing this kind of talk and getting second thoughts about where the field is going?

3 comments

3 Comments

kai_park1mo ago

Man you're falling for the hype train hard. Autocomplete or not, what matters is what it can DO not how it works inside. Cars are just fancy horseshoes under the hood but nobody says driving is fake. I can write a full marketing email in 10 seconds flat, debug code I barely understand, and translate entire documents without knowing a word of Spanish. Call that a parlor trick and watch me build a whole website while you're still typing out your first prompt. The tech works, period.

alice3361mo ago

Hype train" is right, and I'm on it with a first class ticket, but that's only because the stuff actually works for me too, @kai_park. I once asked it to explain quantum computing like I'm five and still couldn't follow it, so maybe I'm just the perfect user for this autocomplete nonsense. Still beats staring at a blank page for three hours though, which is where my brain usually lives.

sage_dixon21d ago

@alice336 makes a good point about the blank page thing though. It's weird how people get hung up on whether this stuff is "real" intelligence or whatever, when the actual useful part is just getting past that initial wall of starting something. Does it matter if it's a parlor trick if the parlor trick helps you ship code or a newsletter before your coffee gets cold?