Original subtitles

- [Instructor] We have learned

that the reasoning power of large language models

is the superpower of agent AI.

This is an evolving field.

Let us learn three basic approaches to reasoning.

Remember, we can ask LLMs to chat with us

or generate anything by prompting.

Using a prompt with the LLM and automating this

is a prompt-based approach.

One of the prompt-based approaches

is to create a set of predetermined prompts in the backend

and let this talk to a LLM.

We will learn some no-code agent AI tools

that are built using predetermined prompts in the backend.

We will even build an agent AI using one of these tools.

Another prompt-based approach

is to do a chain-of-thought prompting at every step

of the step-by-step action plan by the agent AI.

Another approach is called ReAct.

ReAct approach means that the agent AI reasons

and updates its context,

and then makes a plan,

and then takes action to solve the problem step by step.

Contrary to that is ReWOO approach.

With ReWOO approach,

the agent makes a plan upfront,

and then takes action to solve the problem step by step.

Now, you have learned three approaches

to building reasoning in agent AI.