view reply I did though: state of the art positional encoding = no positional encoding https://blog.hashtron.cloud/post/2025-08-25-the-transformer-that-learned-without-a-gpu/