Predicting Multiplication with GPT-2: Implicit vs. Explicit CoT

This demo showcases GPT-2's ability to directly predict the product of two large numbers without intermediate steps, using our stepwise internalization method. Compare the performance of implicit CoT (our method), no CoT, and explicit CoT. Implicit CoT offers accuracy and speed, while explicit CoT provides detailed reasoning but is slower.

• Paper 1: Implicit Chain of Thought Reasoning via Knowledge Distillation
• Paper 2: From Explicit CoT to Implicit CoT: Learning to Internalize CoT Step by Step
• Code Repository
• Tweet Announcement