Top postsTop post
A bunch of other ideas that I couldn't format well for the main post here, are relevant, but were blocking me from just sharing the main ideas. This is significantly more messy and rough, with random pieces all over the place. Maps for Simplex-Valued Vectors Once we have "stuff on...
cart;horse: How can we constrain our models to be interpretable? Convex, linear sets make for more interpretable parameter spaces, and the simplex and the Birkhoff Polytope are great examples of this that have other desirable properties. An interpretation is something explicit, something discrete, something that compresses, something that summarizes. Our...
tl;dr: For Lisa, GPT-2 does not do IOI. GPT-2 fails to perform the IOI task on a significantly nonzero fraction of names used in the original IOI paper. Code for this post can be found at https://github.com/ronakrm/ioi-enumerate. Unintentionally continuing the trend of "following up" on the IOI paper, I ran...
This button will send a single bit. This is no mindgame, no weird trolley-problem-monkey's-paw-dilemma. * This page, this post, pressing this button, are meant to be whatever they need to be for you in this moment. The purpose of this singular bit is entirely up to you. Take a second,...