> Is brain stimulation just not a good domain for GD?

The largest LLMs right now are at best 1% the number of parameters of a human brain.

"At best" if synapses are one parameter each, real ones are probably more than that, but nobody's entirely sure yet.