Original subtitles

- [Narrator] Remember the definition of agent AI?

It is automation of AI

using the reasoning power of large language models.

What does reasoning in LLMs mean? Let's find out.

LLMs are summarizing texts

and writing marketing copy for us.

They chat with us for customer support or legion.

Are these LLMs reasoning?

No, they're not reasoning in these examples.

What is reasoning?

Write your thoughts on what is reasoning on a notepad.

Reasoning is the ability of the LLM

to make decisions like a human

in the absence of the human.

When we chat with an intelligent chatbot,

is the chatbot learning live from our conversations?

When you ask a chatbot about a product or country,

it might give incorrect answers sometimes.

When you chat back saying that is not right,

it will apologize and try to get you another answer,

but the chatbot does not learn from the past mistakes.

Do you know why an intelligent chatbot

cannot learn from its mistakes

even if it apologizes in the chat?

Write your thoughts in your notepad.

The chatbot is not learning from its mistakes

because it is built upon an LLM model that is not changing.

When you chat with the chatbot, the model is fixed

and the chatbot is an application built on top of it.

One of the abilities of making a LLM do reasoning

is by building agent AI

and giving it the ability to reason

and correct its mistakes.

To do this, the agent AI has to access

new data sources and services

and debate or ponder its decisions

to choose the correct course of action.

That is the ability of reasoning.

Next, we'll learn how LLM reasoning can be achieved.