Planning for AI: What careers will resist headcount reductions?

What Tech careers are most difficult to improve productivity/reduce head with more versatile ML models? What will emerge? How are you planning? I’m thinking security and hard compiled programming languages will be hard to reduce human labor on, but who knows. Of course, some senior experts will always be needed, but that’s not most people.

In my view, the dismissive naysayers about AI seem to be making not very sound arguments and focusing solely on GPT. Some are often nitpicking about understanding or accuracy. They Seem to be enticed by the intelligent sounding contrarian argument. Missing the bigger picture because oh woopsie GPT sometimes returns logically incoherent statements or mistakes facts, it’s not like it’s capable of generating very complex content and changing it to conform to very fine-grained prompts. Not as if it contains access to more words and ideas than any human. Whether it is accurate or coherent, it can generate things at a mind-blowing amount of complexity, and it’s going to get better at coherence, accuracy and precision. In my mind, its versatility would be hampered if it had very strict accuracy controls.

   While I’m no expert, my understanding is that it’s a bit misleading to say it is just “predicting the next word” and “understand nothing”. While this might be technically true, this prediction making has become increasingly complex with transformers, embeddings and attentional heads.  Also, I wouldn’t discount the potential of a really, really good prediction. And how much of this “understand” is tied up in a human notion of an attentional self behind the calculations, a watcher.    

I watch it correct previous mistakes without even pointing specifically to the mistake, simply saying that didn’t work or giving it a broad compiler error. It can translate between languages, style writing in response to paragraph long specifications. Ya, it’s not accurate sometimes and takes many shots to find what you want, but What are the naysayers expecting? Will they keep nitpicking until we have some AGI that runs on a GPU the size of Montana, that reads your thoughts to get prompt, talks to god to check accuracy?

The more I use ChatGPT the more I think, very soon we will be confronted with the serious question of whether humans are really all that much more than predictive generative models with some fancy quirks on top like motivational systems and perceptions, feeling of selfhood. What’s stopping these from being modeled one day?