What are AI Agents?


$$ Shift \ Models => Compound\ AI\ Systems $$

Shift models

  1. Limited knowledge: Models on their own are limited by the data they have been trained on. This impacts what they know about the world & what kind of tasks they can solve
  2. Hard to adapt demands a ton of investment in data & resources

  1. An example:

    1. Wish to plan a vacation for the summer & hence, need to know how many vacation days are at my disposal?

      The answer to this query will be incorrect, as the model doesnt know who I am & does not have access to any sensitive information about me!

    Models can do lots of stuff (summarize, write an email, generate code), but the magic is unlocked, when we start building systems around the model & actually, take the model and integrate it into the existing processes there are.

b. So if we were to design a system to solve the above query, we’d have to give the model access to the database, where the relevant data is stored.

c.The difference now will be that the model will be prompted to create a search query , which in turn would fetch the relevant information from the database. Then, it would output an answer, and this would go back into the model that would generate a proper response

You have 8 days left in your vacation database”

-this answer would be correct!

<aside> 💡 This was an example of a Compound AI system & it recognizes that certain problems are better solved when you apply the principles of system design

</aside>


Compound Systems

When we talk about a system approach , we can pick the right components to be able to have a desired outcome, and this is inherently easier to solve for than tuning a model

<aside> 💡 RAG(Retrieval Augmented Generation) is one of the most popular Compound AI systems used in the real world.

</aside>