| I think the ultimate question is how much more productive a software engineer would get compared to baseline. I.e. using GPT for assistive programming instead of using Google, Stackoverflow and some good ol' thinking about problems. Productivity means not just coding, but the ability and rate to successfully complete software projects and maintain software. Factor in that soon, most search engines will also use GPT as a backend (as Bing already does) in some form, to get an explicit answer or better rank search results. Thus, search engines will also improve and lead to more productivity. We might not need less software engineers at all due to the Jevons paradox. [1] Basically, demand for software engineers is constantly increasing currently, in a roughly linear fashion, despite automation being at levels never seen before. Think low-code/no-code API integration platforms, simple to use, high-level and mature programming languages, frameworks & libraries, cloud technology etc. We never measured how many developers these technologies have replaced, because the increase in application possibilities and technology accessibility have only created jobs. Assistive programming with GPT (if/when possible in a professional way, like Copilot is) could only disrupt this if the average productivity increase for every developer in the world over-weighs the average increase of jobs. So, if ceteris paribus there would be 3-4% more developer jobs each year [2], then, if 10% of developers globally started using GPT assistive programming, they would need a 30-40% overall productivity increase in order to thwart that trend. That's a lot, because it's not just being able to create some "x code snippet 20% faster", but generally being able to complete entire software projects 30-40% faster (and maintain them 30-40% more efficiently!) than their non-GPT-user counterparts. If the productivity increase is more like in the 10% range, then on the individual level, time might be better spent leveling up and learning technology instead of learning how to tweak prompts and find bugs in GPT-generated code - as learning any technology as a developer yields to way more than a mere 10% of increase in productivity. [1] https://en.wikipedia.org/wiki/Jevons_paradox
[2] https://www.statista.com/statistics/627312/worldwide-develop... |