The elephant in the room is that this young man is indeed doing the right thing, he is just too poor to buy the right tools.
How do you think the "real scientists(TM)" work : they use AI tools too.
Do you really think you can design a tokamak Stellarator with pen and paper ?
What good engineers do is click the "topological optimization" button on their physical simulator, and then they build the machine according to the plan the computer make.
Do you really think deep-seek can't use your COMSOL or Ansys multiphysics tool ?
Finite element Method was invented in the 1950s, our more modern AIs use some variants of Physics-Informed Neural Networks to solve the differential equations of physics.
LLM without reinforcement learning won't invent your flying saucer from reading stuff on the internet, but let your local AI play for a day with a MHD simulator https://www.jp-petit.org/science/mhd/m_mhd_e/m_mhd_e.htm and the sky is not the limit.
> Do you really think you can design a tokamak Stellarator with pen and paper ?
No, but you can draw a very nice strawman on pen an paper.
> The "topological optimization" button on their physical simulator,
A engineering tool for topological optimization is similar to "AI tools" only in the sense that a big amount of numbers are crunched together. Saying that they "are the same" is like saying that a shark is the same as a kangaroo.
Besides, for tokamak and others, I doubt that off-the-shelf tools were enough for them. I would bet that they had to build their own tools anyway.
> Do you really think deep-seek can't use your COMSOL or Ansys multiphysics tool
A monkey can "use" those tools, as long as they have bright buttons and he's conditioned to press them for food. That doesn't mean that I would trust the results it comes with.
> but let your local AI play for a day with a MHD simulator ... and the sky is not the limit.
The limit is still death, which can come much much faster than the sky.
How do you think the "real scientists(TM)" work : they use AI tools too.
Do you really think you can design a tokamak Stellarator with pen and paper ?
What good engineers do is click the "topological optimization" button on their physical simulator, and then they build the machine according to the plan the computer make.
Do you really think deep-seek can't use your COMSOL or Ansys multiphysics tool ?
Finite element Method was invented in the 1950s, our more modern AIs use some variants of Physics-Informed Neural Networks to solve the differential equations of physics.
LLM without reinforcement learning won't invent your flying saucer from reading stuff on the internet, but let your local AI play for a day with a MHD simulator https://www.jp-petit.org/science/mhd/m_mhd_e/m_mhd_e.htm and the sky is not the limit.