Considering the cost, at least in a vacuum, is a few hours of my time setting up a local model to run on a gpu I already had (pretty pixels make me go “Ooooooo Aaaaaahhh”) vs 8+ hours of reading a manual for an appliance I’m going to configure once; I’d say it’s objectively a productive tool in it’s current capacity. If you mean to say productive for society I’ve no clue, and I think there’s far too many variables for any person to reasonably know.
I can think of lots of individually useful and productive cases, yours included, but i mean more the widespread adoption and integration of the current generation of llms may cost more in resources than it puts out economically. Actual intelligence will come when we advance, not by throwing ever greater amounts of compute at what are just slightly better versions of what was available 3 years ago, but by radical and as yet unseen changes to the way the models function.
I meant productive in the sense that they add more value than their cost. I don’t think they are useless.
Considering the cost, at least in a vacuum, is a few hours of my time setting up a local model to run on a gpu I already had (pretty pixels make me go “Ooooooo Aaaaaahhh”) vs 8+ hours of reading a manual for an appliance I’m going to configure once; I’d say it’s objectively a productive tool in it’s current capacity. If you mean to say productive for society I’ve no clue, and I think there’s far too many variables for any person to reasonably know.
I can think of lots of individually useful and productive cases, yours included, but i mean more the widespread adoption and integration of the current generation of llms may cost more in resources than it puts out economically. Actual intelligence will come when we advance, not by throwing ever greater amounts of compute at what are just slightly better versions of what was available 3 years ago, but by radical and as yet unseen changes to the way the models function.