JE
J. Elisenda Grigsby, Yiyao Huang, Arushi Katyal, Ryan Lin, Pranav Ramakrishnan, Weiya Wang
· 1 min read
ResearcharXiv cs.CL
Can small transformers learn the language of Coxeter groups?
arXiv:2611.10936v1 Announce Type: new
Abstract: The field of mechanistic interpretability is well-suited to collaborative projects between mathematicians and AI researchers. This work describes one such project. We trained a small transformer to learn the language of the Coxeter group affine A2, best known as the symmetries of the tiling of the Euclidean plane by equilateral triangles. We then formulated and proved a word-theoretic conjecture about affine A2 after observing the attention patterns of the trained model.
Original source
This story was published by arXiv cs.CL and written by J. Elisenda Grigsby, Yiyao Huang, Arushi Katyal, Ryan Lin, Pranav Ramakrishnan, Weiya Wang. SyncAI.news shows a preview; the complete article is on the publisher's site.
Read the full story on arxiv.org
