Perhaps it's wishful thinking, but I think you're correct about this. Did some looking into you and you do your homework, so I trust your journalism here. I'm a programmer and this subject matter has been gnawing away in my head. I'm a daily GitHub Copilot and ChatGPT user, so I am fully aware of it's limitations.

But I also know how to get it to do what I want and it's (honest to God) allowed me to become an entire team of developers. My productivity has increased at least 3-fold.

So I think your assessment is correct. AI will offset some jobs but I think it's only ever going to be a very good productivity tool. (Until we perfect quantum computing, that is.)

As for the inaccuracy problems, that can be fixed without necessarily needing to train with a larger dataset or use more compute power. Already, there are organizations working on multi-agent AI that will be much more accurate. I've experimented a bit myself with this with AutoGPT (a Python application), and I got pretty impressive results, even with GPT-3.

Great article!