ai professional use

Posted: June 20th, 2025 23:45 PM

tldr; speed and efficiency in terms of AI is only utilized, at least for the foreseeable future, by a highly competent and contextually informed human prompter.

Dario made a statement recently that is high signal of a moving goal post. He said that in 1-5 years half of white collar jobs could be eliminated raising unemployment by 20%. The estimates from researchers at Epoch AI, to former Open AI employees, to (now) CEOs of these companies ranges between 2045 to early 2027. However, colloquially, the opinion of those mid-to-senior level engineers working with the AI agents, LLM tools, and even those working on things like interpretability and performance engineering of the models themselves widely agree there are still struggles with AI performing the actual task they need.

AI is and will be, until there are breakthroughs in RL, an augment and catalyst for deep technical knowledge. The major bottle necks to having AI actually replace vast amounts of workers is and will continue to be: money, energy, and algorithms. We have limited ability to develop novel algorithms that actually drive AI forward, we have terribly limited energy capacity in the US (some argument that China has the ability to develop enough facilities to generate adequate energy), and the cost of a token is so high that in order to get a good cost-per-token average you would need a human with deep technical knowledge and context nuance to use AI in such a way that it is actually effective.

These are just some observations and you can get great information by listening to the experts in AI on a podcast like the Dwarkesh Podcast.

I think there are many jobs which can and should be easily replaced: customer support, medical billing coders, stenographers, radiologists, accountants, and technical documentation writers. These are all tasks and functions that are not only incredibly linear regarding input and output, but also lack a creative component that would be required for 'novel' or 'new' work. Arguably, there is likelihood that in the not so distant future, AI could be replacing some jobs that create frequent 'novel' work of some kind. This doesn't seem entirely out of the realm of possibility in the near future as RL will likely progress.

My personal take on these kinds of white collar jobs that will be replaced is that livelihood will not be erased, but humans, as we have for centuries, will have to adapt to new evolutionary selective pressures. That being said, it seems that work that is either a bridge of creativity and technicality (video editors, 3d animators, maybe CAD designers) or purely AI related technicality (ML engineer, Research Engineer, ML Ops) will be selected for.

My only counter argument to everything I've said so far is that I don't think it is in the best interest of the powers that be in the world to increase the speed at which AI develops and its overall applications in the world. Investment Banks could be absolved over night because companies can pay $X finite amount of money to get X% exponential increase in innovation that people inherently want. Similarly, governmental powers could be completely reformed because AI is able to perform a majority of functions that a majority of people agree with and prefer.

What I am predicting is that the way we perceive money will fundamentally change over the next 15 years, on the contingency that we achieve AGI/ASI by then. With a secondary prediction that the middle class is completely erased over the next 20 years, on the contingency that AGI/ASI does not significantly decrease the cost-per-token barrier.

just some thoughts.

lesson

living in a somewhat constant state of FUD is a great way to permanently hamstring your own capacity and ability to perform good work. there is something to be said of confidence. a humble trust in oneself.

increase overall utility by learning mathematics and how to reason. human problem solving combined with nuanced taste is how we adapt to future conditions.