they're still bounded by training data and how it relates to the real world.
Yes and no. They can extrapolate and build upon the training data, as was the case with the last dozens of math proofs
Yes and no. They can extrapolate and build upon the training data, as was the case with the last dozens of math proofs
To put it differently, if you have some idealized model in front of you that can do anything a team of humans can do, what do you say to it? It's still just a model that takes an input and provides an output.
Someone will still need to ask for something (or maybe it will just run autonomously making things it thinks we'll like) but it won't require any expertise in the field.
For things like sickness, old age and death, generally the lacks are not knowledge but the consistent will of the group. Like global warming, there is no real profound knowledge gap except how to get everyone to agree to mitigating the carbon burn. Perhaps it is likely LLMs would be able to persuade the group to stop burning the carbon before it’s [more] uncomfortable, but the small group of people committed to accelerating the carbon burn can also pay for LLM persuasiveness.
LLMs giving you something novel would be like if you let a model play chess against itself and become the best player in the world this way. Totally impossible.
/sacrasm