The fact that you can imagine something will happen doesn't mean it will happen.
Who says the AI will "know the intention of all the code, all the data models, know how users use the application intimately"? Are you aware that language models do in fact have token input/output limitations that will not go away? Are you aware that there is such a thing as diminishing returns when it comes to improvements due to increased number of parameters/training set size that are already evident? Are you aware that the training set of codex pretty much includes all available public code, so it will be impossible to scale it by a factor > 3 in the next several years at least?
Your assertions are full of wild assumptions backed by nothing.
As for photography, the fact is there has been no job apocalypse because your "friends and family" are "pumping out photos". And the point of your initial post, even if it was implicit, was "you are going to be unemployed in 5 years". This will have an impact on your dev flow and will be used by managers to try to reduce salary premiums for software engineering but your wild assumptions stated with so much confidence may never happen.
P.S: At this point, I find Intellicode actually slows me down, that's why it's permanently turned off. Current copilot will at most save me 2-3% of my working time each week if I am coding in a language it can actually do something in (it's worse than useless for Scala).
Never said "you are going to be unemployed in 5 years". Never said anything about a job apocalypse. I have no idea what role humans will play.
> Your assertions are full of wild assumptions backed by nothing.
I use Copilot, you admitted it currently saves you 2-3% of time. Well, that's just Copilot, you think Microsoft will just sit on that? My assertions are based on what is happening today and extrapolating an exponential increase in that performance for tomorrow.
Digital cameras definitely revolutionized photography and made it much more accessible to regular folks. Not everyone wants to be a pro photographer though, and the number of wedding shoots available has not changed. People still need to be paid to take photos because no one is going to do that for free. However, we can all take pro level photos with much more ease than when all we had was 110 and 35mm film with a really crappy lens.
There are more "photographers" than ever, the same number of pro photographers seems reasonable given the burden of people's time to money ratio. So the net result is billions more family and friends photos which previously were not taken, the same will go for art. I want to create art, but I have little skill, but given the opportunity to make a comic strip just by talking to an AI will allow me to do so. I imagine some people will do this extremely well as a profession until it's no longer useful.
I don't know, I understand why you are being dismissive and playing down my wide eyes, but I think you are also wrong and remaining uninterested because it's too "religious" to speculate wild things in light of wild real world changes is head in sand territory.
Who says the AI will "know the intention of all the code, all the data models, know how users use the application intimately"? Are you aware that language models do in fact have token input/output limitations that will not go away? Are you aware that there is such a thing as diminishing returns when it comes to improvements due to increased number of parameters/training set size that are already evident? Are you aware that the training set of codex pretty much includes all available public code, so it will be impossible to scale it by a factor > 3 in the next several years at least?
Your assertions are full of wild assumptions backed by nothing.
As for photography, the fact is there has been no job apocalypse because your "friends and family" are "pumping out photos". And the point of your initial post, even if it was implicit, was "you are going to be unemployed in 5 years". This will have an impact on your dev flow and will be used by managers to try to reduce salary premiums for software engineering but your wild assumptions stated with so much confidence may never happen.
P.S: At this point, I find Intellicode actually slows me down, that's why it's permanently turned off. Current copilot will at most save me 2-3% of my working time each week if I am coding in a language it can actually do something in (it's worse than useless for Scala).