The Power of Less: How Chain of Draft Makes AI Reasoning Faster and Cheaper

Author(s): MKWriteshere

Originally published on Towards AI.

In today’s AI landscape, large language models (LLMs) like GPT-4 and Claude can solve complex problems with impressive accuracy.

But this capability comes at a cost, both in processing time and computational resources.

What if these AI systems could think just as effectively while writing much less?

That’s the premise behind an innovative approach called “Chain of Draft” (CoD), developed by Zoom Communications researchers.

Let’s explore how this technique helps AI models reason more efficiently by writing less, much like how humans jot down quick notes rather than full paragraphs when solving problems.

When tackling complex problems, modern AI systems often use a technique called Chain of Thought (CoT). This approach encourages the AI to break down problems step-by-step, showing its work in detailed explanations.

While effective, this method leads to extremely wordy responses.

For example, when solving a simple math problem like “Jason had 20 lollipops and gave some to Denny, leaving 12. How many did he give away?”, an AI using Chain of Thought might write:

Initially, Jason had 20 lollipops.After giving some to Denny, Jason now has 12 lollipops.To find out how many lollipops Jason gave to Denny, we need to calculate the difference between the initial number of lollipops and the remaining number.We can… Read the full blog for free on Medium.

Join thousands of data leaders on the AI newsletter. Join over 80,000 subscribers and keep up to date with the latest developments in AI. From research to projects and ideas. If you are building an AI startup, an AI-related product, or a service, we invite you to consider becoming a sponsor.

Published via Towards AI