Comment on Companies be like

<- View Parent
sundray@lemmus.org ⁨3⁩ ⁨days⁩ ago

But LLMs aren’t necessarily bad

Maybe that’s true. LLMs are fancy text prediction gadgets utilizing linear algebra and lambda calculus, whose facility with language creates the illusion of intelligence (which is one thing that LLMs are good at). But I think a lot of the VCs and execs in tech have lost sight of that fact.

I mean, for basic language tasks they’re pretty good! They can summarize a webpage, or go through all my emails and tell me how many Daves I know, or write a summary of the appointments on next week’s calendar. Useful things.

But forcing them to do things they’re not good at makes them look half-baked. They’re not ready to be customer service agents, they’re not ready to accurately answer questions in specialist domains, they’re not ready to book your flights, or do your budgeting, or your grocery shopping, and maybe they’ll be brilliant someday – but so many companies are pushing them right now regardless, and it makes them look bad and stupid.

Ramp up the research. Slow down the deployment. (And maybe deal with the copyright and energy consumption issues, don’t burn the planet down for LLMs.)

source
Sort:hotnewtop