Let's Think Dot by Dot: Hidden Computation in Transformer Language Models

arxiv.org

saved by: 0 people
related read on: GitHub - JuliusBrussee/caveman: 🪨 why use many token when... from the discussion