AWS Automatic Reasoning and Tool-use (ART)

Automatic Reasoning and Tool-use (ART)

Like the self-consistency and ToT prompt techniques, ART is a prompting technique that builds on the chain-of-thought process.

The ART technique, discussed in detail in Bhargavi Paranjape's "ART: Automatic multi-step reasoning and tool-use for large language models," is used specifically for multi-step reasoning tasks.

This technique deconstructs complex tasks by helping you select demonstrations of multiple, or few-shot, examples from the task library.

While using this few-shot breakdown, you use predefined external tools such as search and code generation to carry out the task.

Research shows that ART performs better than few-shot prompting and automatic CoT for unseen tasks.

It matches the performance of handcrafted CoT prompts on many tasks.

With ART, you can more efficiently update information in the task libraries, which helps correct errors and improve performance.