Less is more: How ‘chain of draft’ could cut AI costs by 90% while improving performance
1 min read
Summary
Researchers at Zoom Communications have developed a new technique that reduces the cost and resources needed for AI to tackle complex reasoning problems, which they claim could transform the way enterprises deploy AI at scale.
Called ‘chain of draft’ (CoD), the method allows large language models (LLMs) to solve problems using minimal text, reducing output by as much as 93% while maintaining or improving accuracy.
Currently, state-of-the-art reasoning techniques, such as the ‘chain-of-thought’ (CoT), have improved how well AI systems are able to solve complex problems.
However, these techniques generate lengthy explanations that consume substantial computational resources.
The new approach can be implemented straight away with existing models through a simple prompt modification, offering a straightforward way to make AI reasoning more accessible and affordable.