|
|
|
|
|
|
by dvt
9 days ago
|
|
|
Both the Leiden Declaration[1] as well as Commelin et al.'s announcement (Shaping the Future of Mathematics in the Age of AI)[2] are completely common-sense pieces, neither overstating the power of AI nor understating the capability of these tools to be misused (most acutely when it comes to attribution). I do think that ethically, OpenAI and Anthropic, by training their models on the entire corpus of human knowledge, have certainly broken some rules, but these rules were already broken decades ago by Google—unless we really want to start splitting hairs about if indexing + processing is different than training, which is imo a distinction without a difference—, so it's hard to see who to exactly blame. In any case, that cat is out of the bag. (And for the record, I'm technically a stakeholder: I'm part of like two class actions because of a few books I wrote.) But that's neither here nor there. [1] https://leidendeclaration.ai/#declaration [2] https://arxiv.org/pdf/2603.24914 |
|
|
Which Tao is more equivocal about as 2 issues higher on his mind are
1. Funding, short and long term (so he is evangelising "at the bequest of" SAIR)
2. Indigestion from generative abundance (this has always been an issue but now even researchers with permanent positions are affected, over and aside of administrative responsibilities )
here he tries to redirect attention from the "harder" problem of attribution to the "softer" problem of digestion
https://mathstodon.xyz/@tao/116590271196962848
With a side effect of asking the tool-builders to reconsider what is "value"