FWIW: These types of graphs with normalization to 1 at one point in the middle and a larger right vs. left can be at least slightly misleading, and I think here it may be—at least slightly, quantitatively; not claiming it removes any ’2022 changed it’ fully, but makes it considerably less obivious than it looks upon shallow glance: Slightly less obvious to the eye one spots on the LHS the rather exact inverse of the RHS, and actually, if on the RHS you only go out by the same amount as on the LHS, yes RHS is still bit wider but really not insanely much wider anymore.
I reckon easier to check whether trends really abrupbly changed would be:
proper diff-in-diff analysis, or
at least also plot the same graph with normalization basis 1.0 say in Jan 2021. Then your eye can actually more easily tell you whether it’s really so obvious curvatures changed ‘unexpectedly’ in 2022 or not, in this particular data constellation
(Not meaning to challenge the general story, I think it still make sense, from theory, anecdotes, and maybe this data to some degree still)
Yeah, those graphs look kind of off. In particular: it supposedly started happening in October 2022? That’s pre-GPT-4, and GPT-4 was very barely able to code anything useful. I don’t buy that GPT-3.5, effectively the proof-of-concept technical demo, had this sort of effect.
I guess maybe companies recognized where AI progress was headed and foresightfully downscaled their hiring practices in expectation of AI advancing faster than early-career humans can be trained? I, likewise, don’t buy this level of industry-wide foresight.
My guess is that “2022 changed it” because of e. g. the Russia-Ukraine war and general rising world instability, not because of AI.
You can get pretty similar graphs by just assuming that hiring slowly increased up to 2021 and then slowly decreased. The demographic bulge from the hiring spike moves up in age over time, and normalizing to 2022 hides the fact the overall ratio of hires at different ages hasn’t changed at all.
Very good, yes was thinking in the same direction but knowing too little about absolute hiring numbers/cohorts etc. ended up not adding that point even though you’re right, it is a rather clear addition to our argument of the graph in OP being easily overrated.
I had the same thought. Some of the graphs, on first glance seem to have an inflection point at ChatGPT release, but looking more seem like the trend started before ChatGPT. Like these seem to show even at the beginning in early 2021 more exposed jobs were increasing at a slower rate than less exposed jobs. I also agree the story could be true, but I’m not sure these graphs are strong evidence without more analysis.
FWIW: These types of graphs with normalization to 1 at one point in the middle and a larger right vs. left can be at least slightly misleading, and I think here it may be—at least slightly, quantitatively; not claiming it removes any ’2022 changed it’ fully, but makes it considerably less obivious than it looks upon shallow glance: Slightly less obvious to the eye one spots on the LHS the rather exact inverse of the RHS, and actually, if on the RHS you only go out by the same amount as on the LHS, yes RHS is still bit wider but really not insanely much wider anymore.
I reckon easier to check whether trends really abrupbly changed would be:
proper diff-in-diff analysis, or
at least also plot the same graph with normalization basis 1.0 say in Jan 2021. Then your eye can actually more easily tell you whether it’s really so obvious curvatures changed ‘unexpectedly’ in 2022 or not, in this particular data constellation
(Not meaning to challenge the general story, I think it still make sense, from theory, anecdotes, and maybe this data to some degree still)
Yeah, those graphs look kind of off. In particular: it supposedly started happening in October 2022? That’s pre-GPT-4, and GPT-4 was very barely able to code anything useful. I don’t buy that GPT-3.5, effectively the proof-of-concept technical demo, had this sort of effect.
I guess maybe companies recognized where AI progress was headed and foresightfully downscaled their hiring practices in expectation of AI advancing faster than early-career humans can be trained? I, likewise, don’t buy this level of industry-wide foresight.
My guess is that “2022 changed it” because of e. g. the Russia-Ukraine war and general rising world instability, not because of AI.
You can get pretty similar graphs by just assuming that hiring slowly increased up to 2021 and then slowly decreased. The demographic bulge from the hiring spike moves up in age over time, and normalizing to 2022 hides the fact the overall ratio of hires at different ages hasn’t changed at all.
https://docs.google.com/spreadsheets/d/1z0l0rNebCTVWLk77_7HAwVzL7QtTjlrllAMH2lxhnes/edit?usp=sharing
Very good, yes was thinking in the same direction but knowing too little about absolute hiring numbers/cohorts etc. ended up not adding that point even though you’re right, it is a rather clear addition to our argument of the graph in OP being easily overrated.
I had the same thought. Some of the graphs, on first glance seem to have an inflection point at ChatGPT release, but looking more seem like the trend started before ChatGPT. Like these seem to show even at the beginning in early 2021 more exposed jobs were increasing at a slower rate than less exposed jobs. I also agree the story could be true, but I’m not sure these graphs are strong evidence without more analysis.