LLMs represent numbers on a helix and manipulate that helix to do addition.
-
Updated
Feb 4, 2025 - Jupyter Notebook
LLMs represent numbers on a helix and manipulate that helix to do addition.
Cross-modal SAE interpretability: are sparse features causally tied to a foundation model's self-difficulty? Tested identically on Pythia (LLM) and Chronos-T5 (time-series), predictive null in both, causal signal in the LM only.
Add a description, image, and links to the mechanistic-inter topic page so that developers can more easily learn about it.
To associate your repository with the mechanistic-inter topic, visit your repo's landing page and select "manage topics."