Compound AI Systems

In the blog post, "The Shift from Models to Compound AI Systems", a group of Berkeley AI researchers notes the difference between using Large Language Models (LLMs) as monolithic standalone systems versus using them as components in larger, compound systems that include both LLMs and other technology. They list the following reasons to use Compound AI Systems instead of relying solely on an LLM:

• Some tasks are easier to improve via system design
• Systems can be dynamic
• Improving control and trust is easier with systems
• Performance goals vary widely

Compound AI Systems & FOLIO

Building on the concept of Compound AI systems, artificial intelligence (AI), particularly LLMs, is providing an approach for the research that will be demonstrated in this presentation.