arxiv.org/abs/2512.19428 Attention Is Not What You Need (2025)
A preprint proposes an attention-free sequence model, the Causal Grassmann layer, which encodes local token pairs as subspaces on a Grassmann manifold. Models with 13 to 18 million parameters land within about 10 to 15 percent of Transformer perplexity on Wikitext-2 and slightly beat a Transformer head on SNLI.
— via Hacker News (top 30), westurner
