Based on some feedback, definitely would have been nice to flesh out more both the actual capabilities of GPT-N for each N, and also talk about the state of other models and how good they are/what they’re used to.
I think the question of when does large scale job automation is a big deal and will be a key turning point in public relationship to AI. It’s not that clear to me when it’ll happen. Habryka made the point to me that economies move slowly, and human labor is still pretty cheap for lots of things.
Some interesting things that came out since I drafted the above:
Agentized LLMs (not something I’d thought about enough, they’re implicitly in what I describe, but I would have written something different if I’d been writing after the hype/attention on agent LLMs)
I don’t think job automation will be as dramatically disruptive as we might think or at least not just from current technology, anyway. I predict it’ll be a mostly smooth transition for most industries (unless there’s a new relatively cheap tech that massively outperforms the rest). Already we see in polls and surveys that millenial/gen z workers are more picky about what jobs they’ll accept, are willing to sacrifice pay for more interesting work/better balance, don’t want repetitive jobs, etc. (It partially explains the current worker shortage in many industries). I think the rate at which workers exit/refuse those unwanted jobs will mostly keep up with the rate at which automation spreads to those jobs/industries. To put it simply: in five years it will be common knowledge that most call center jobs are automated and won’t be on job seekers’ radar. This is an arbitrary example, it could be only 3 years but a transitional period of five years seems like a safe bet.
Some quick musings
Based on some feedback, definitely would have been nice to flesh out more both the actual capabilities of GPT-N for each N, and also talk about the state of other models and how good they are/what they’re used to.
I think the question of when does large scale job automation is a big deal and will be a key turning point in public relationship to AI. It’s not that clear to me when it’ll happen. Habryka made the point to me that economies move slowly, and human labor is still pretty cheap for lots of things.
Some interesting things that came out since I drafted the above:
Palantir putting AI to military use
Agentized LLMs (not something I’d thought about enough, they’re implicitly in what I describe, but I would have written something different if I’d been writing after the hype/attention on agent LLMs)
I don’t think job automation will be as dramatically disruptive as we might think or at least not just from current technology, anyway. I predict it’ll be a mostly smooth transition for most industries (unless there’s a new relatively cheap tech that massively outperforms the rest). Already we see in polls and surveys that millenial/gen z workers are more picky about what jobs they’ll accept, are willing to sacrifice pay for more interesting work/better balance, don’t want repetitive jobs, etc. (It partially explains the current worker shortage in many industries). I think the rate at which workers exit/refuse those unwanted jobs will mostly keep up with the rate at which automation spreads to those jobs/industries. To put it simply: in five years it will be common knowledge that most call center jobs are automated and won’t be on job seekers’ radar. This is an arbitrary example, it could be only 3 years but a transitional period of five years seems like a safe bet.