I wonder how much undiscovered knowledge already exists within published papers, just because no one has gone and connected it yet. Training LLMs to predict results seems like a good way of moving towards helpful science agents.

I'm curious as to your opinion on using GenAI for coding tasks. It's the one area that I consistently notice it helps me, and it does appear to be getting much better.
Much of the hype is BS, but this is one area I could see it being extremely useful.