Can you help me with a few clarifications on the use of AI Agents?
๐๐จ๐จ๐ ๐ฅ๐ ๐๐ก๐ข๐ญ๐๐ฉ๐๐ฉ๐r on ๐๐ ๐๐ง๐ญ๐ฌ is a nicely packaged document for beginners to moderate level AI enthusiasts who want to know more about the current buzzword โAgentsโ.
It talks about the Objectives of AI Agents and different frameworks used for reasoning and planning like
โณ ReACT
โณ Chain-of-Thought
โณ Tree-of-Thought
It goes on to discuss the different components and how they interact in the General Agentic Architecture. These components include
โณ Agent
โณ Model
โณ Tool
It further goes on to explain how agents interact with โoutsideโ Tools like Google Flight API using
โณ Extension
โณ Function
โณ Data Store
The whitepaper uses the example of flight booking to explain how LLMs, Agents and external database (both structured and unstructured) can be used to solve flight booking problems.
And here in lies my confusion.
โณ Why would we want to use a LLM for perfectly logical application like flight booking?
โณ Why would we want to train agents on what kind of inputs to expect and outputs to display by training on diverse set of examples (read data) instead of taking the input directly from the user and thereby save on compute cost and improve accuracy?
โณ Why would we want to train agents on large set of data just to let them chose the right API eg. choose Google Flight API instead of Google Maps API?
I am perfectly fine with the use of LLMs for unstructured data like text, images, audio etc. but I am still trying to understand why use LLMs for solving problems which can be solved with simple query-based input and outputs on structured data like flight booking?
Am I missing something? Comment below
p.s. Happy New Year.
