John Carlos Baez on Nostr: wrote: "We must advocate for models and their training data to be open source, and ...

wrote:

"We must advocate for models and their training data to be open source, and demand that “compute” not be concentrated in private hands by socially dangerous methods of financing, in addition to being environmentally sustainable."

I agree. But these are dramatic changes from the current trajectory - very difficult to achieve. What do we plan to do in the meantime? Do we use LLMs in our research from Monday to Friday and spend Saturday and Sunday advocating these nice things? Do we forswear using LLMs? (A promise to do that would be an interesting declaration that goes way beyond the Leiden declaration.)

"And, finally, we must encourage the current generation of students to see through the transparent misconception that a new tool, however powerful, should represent a threat to the mathematical pursuit. We should not underestimate students’ abilities to absorb and leverage technological advancements to obtain ever greater insights that will push mathematics forward. Universities would do well to fill their chairs with such resilient, post-AI mathematicians."

I have too much to say here. Let me try.

At some point AI may be better at math than humans in all ways. At that point - or, preferably, before it - we'll need to decide between keeping math human and letting math go its own way. But for AI to be better at math than humans are, it would need to be fundamentally different than current LLMs. So we're not there yet.

For now, good math requires humans. I hope students can wisely decide whether and how they want to use the latest technologies. Alas, our educational system is not ready to help them much. We should be working on this!

(3/n)