Meta's Lucas Beyer finds frontier models can design scientific experiments, but Dimitris Papailiopoulos argues AI lacks intrinsic curiosity
Beyer tested models using experimental setups without leading prompts.
The AI will never be able to drink a glass of water to relieve me from my own thirst!
Dimitris Papailiopoulos@DimitrisPapailEven if all of technical work required to do research is automated, research does not become automated. And that's cool.
Research is not only ideas. It’s questions and hearing feedback and integrating. Seeing how things fit what connects with what; and at the end of the day it’s a process by which you satisfy your thirst for truth. You still gotta drink the water to relieve you from thirst no matter how easy it is to obtain a glass of it. Bottom line is we will never stop asking questions.
Lucas Beyer (bl16)@giffmanaI'm continuously probing this with frontier models btw. Eg when i have an experiment idea, i sometimes tell them the setup without being too leading and ask them for experiments. They are not great yeat, but every now and then they pretty much come up with what i have in mind. So long way of saying, I'm not sure for how long what you say will be right.
@giffmana The AI can’t drink the glass of water for me!!!
Dimitris Papailiopoulos@DimitrisPapailResearch is not only ideas. It’s questions and hearing feedback and integrating. Seeing how things fit what connects with what; and at the end of the day it’s a process by which you satisfy your thirst for truth. You still gotta drink the water to relieve you from thirst no matter how easy it is to obtain a glass of it. Bottom line is we will never stop asking questions.