This is true today and has been my experience as well -- both to write software as well as doing computational physics.
The interesting question then is to ask how long will this stay true? As the models get better will they eventually not need the human expertise to start adding value?
Just started this project last week and don't have anything written up or properly documented yet.
I am trying to see how far I can push AIs to do a research project in physics. I did a PhD in plasma physics a decade ago - I am using the literature survey I did while looking for a problem as my starting point.
The plan is:
- Using a few research papers to ground the state of the world, see if I can nudge it towards coming up with a few research problems to tackle. The problem needs to be (1) unsolved and (2) something that is probably solvable in a timeboxed fashion.
- Let the AI read more papers to refine the hypothesis, and come up with a plan.
- Let another agent write code to run numerical simulations.
- And finally do synthesis.
I am curious to see how far it's possible to push the models in this direction. At the very least it hopefully helps new grad students in a copilot capacity.
The interesting question then is to ask how long will this stay true? As the models get better will they eventually not need the human expertise to start adding value?