pull down to refresh

279 sats \ 1 reply \ @denlillaapan OP 17 Sep \ parent \ on: Schumpeter Vs Smith on Where Jobs Come From (FT, Soumaya Keynes) Politics_And_Law econ

Fuck, good questions. Will think.

Quick on-the-fly response:

  • my impression, too, is that LLMs are top-50-90 percentile stuff. That is, they improve on most people's stuff — but crucially not the high-end quality. Chat writes well (and shockingly, much better than a year ago) but not better than me. (But Chat is cheaper and faster, so for low-value/high-volume stuff, it makes sense). Same for editing.
  • I kind of think this is Smithian (BC it's incremental and shifting around tasks) even though for the mids it's competing directly with their work. (So seems Schumpeterian/revolutionary)

It's definitely also speeding up my work. Instead of pondering with a draft on an article for days or week, I'll just shove everything I have, unstructured, into an LLM and tell it to write a suitable article in my voice — specifically including x, y, z, and this direct quote from NN. Usually produces decent drafts I can work with

write
compose
232 sats \ 0 replies \ @optimism 17 Sep

I like how you differentiate the subjective here. And maybe this all is subjective in the end. Just because the top 10 in one's field got a raise and are doing better than ever, doesn't mean that the bottom 10% won't lose their jobs (or, as Jack showed us, the "bottom 60%".) You don't need AI for the bottom n-percent to lose their jobs though, that's perfectly capable of happening without AI too, and without automation, and without information technology[1]

I was talking to a young developer the other day, they started their career in 2021. Now they have Claude Code. And they complained: I can do anything now, but when you shoot on my code, I have to ask Claude what you mean. And that's where it becomes harder for me to accept. Because you had this dev that was good within their limits, now pushed the limits to "produce" things that were out of reach before. But because those things not only were, but technically still are out of reach, they can only manage it with a dependency on a subscription to a lab's model. Take that away, and that whole ranking into percentiles will flip. That's also my fear. Why I struggle at times to go through all the shit I need to review that I could ask an LLM to review for me, but the moment I truly depend on it other than for things I know I master without it, I start losing instead of winning?

And perhaps, that's the ultimate challenge, to keep the competitive advantage versus the LLM where we can keep on saying

but not better than me.

  1. I once was hired by a 150+y/o company that every 20 years went through a crisis and would fire half the personnel. Can't say it didn't work for them if they still had money to hire me after 150 years. But it illustrates that you don't need AI to need reorgs. ↩

reply