Reproduce “Latent Feedback Is a Compute and Data Amplifier” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Benhao Huang, Chaitanya Dwivedi, Xuezhe Ma, Alexander Moreno, Hector Liu
The paper introduces latent feedback, which feeds each token’s top-layer state into the next token’s input. Against a tuned Transformer trained on the same unique data, it measures compute needed to reach the same validation loss. At 60M, 400 unique tokens per parameter, and two epochs, feedback uses about 48% less compute; at scarce data, it can cost more.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.