Re:Transformers Can Do Arithmetic with the Right Embeddings
by neves on Tuesday May 28, 2024 @10:24AM
(#40501029)
but it makes sense to have a different encoding. Mathematics is a completely different language. Maybe we should have more than one class of encodings.
Re:Transformers Can Do Arithmetic with the Right Embeddings
by PartiallyTyped on Tuesday May 28, 2024 @03:04PM
(#40504295)
There were some recent posts (either here or reddit) supporting the claim that different regions activate when reading programs vs when reading text. If we take that to be true; and squint just enough, one could claim that arithmetic and mathematics should be treated differently to language.
Re:Transformers Can Do Arithmetic with the Right Embeddings
by ckemere on Tuesday May 28, 2024 @04:37PM
(#40505357)
Numeracy is definitely associated with different brain regions than just reading. See, e.g.
https://www.sciencedirect.com/science/article/pii/S105381191...
(Dehaene also has a book, “The Numbet Sense”)
Re:Transformers Can Do Arithmetic with the Right Embeddings
by wrsh07 on Tuesday May 28, 2024 @07:34PM
(#40506709)
I would only find that satisfying (from a snobbish and impractical perspective) if we were able to have the model decide: 1) what encoding should this section use? 2) how should I train this encoding?
A mixture of experts but for encodings is interesting, though!
Maybe there's a clean way to implement
It makes me think that the authors have correctly identified an issue (positional embeddings) but don't propose a general solution.
I'm not sure if such a thing is possible, but if it is, it would feel more complete. (Fwiw, positional embeddings have had issues for a long time! So a general solution to this would benefit more than just arithmetic. Helpfully, we now have a really good specific example to serve as a baseline for any generalization we seek)