Author(s): Zofia Jose Prakash
Originally published on Towards AI.
How I moved from “reply” to “action” with recovery, tools, and agent loops
Large-language models (LLMs) arise as stable, albeit powerful, generalist engines of the brain. Stable in the sense that although their outputs may be magical, they are limited on the range of knowledge obtained at the time of training. It’s a fact of life that any developer leveraging these models quickly encounters when asked about anything beyond their training: current events, the latest internal policies, or proprietary information that isn’t in their pre-training data. This recognition led me (and many engineers like me) down the path of what I would consider three big evolutionary steps toward elevating LLMs from mere answering machines to active partners: retrieval-augmented generation (RAG), tool calling, and autonomous agents. In this article, I share my insights I’ve learned about why these three things are important, and the changing paradigms arising from each.
The article discusses the development of large-language models (LLMs) capable of performing a variety of tasks, from static tools to dynamic agents. It explores three key developments: retrieval-augmented generation (RAG) that enhances LLM by providing real-time, contextual data, tool calling that enables models to interact and perform external tasks, and autonomous agents that can observe, plan, and perform tasks with varying degrees of complexity. Each level increases user engagement and capabilities while presenting new challenges and considerations, particularly with respect to security and the need for responsible AI development.
Read the entire blog for free on Medium.
Published via Towards AI
