ReAct: Reasoning and Acting with Prompts

Duration: 5 min

This module delves into the ReAct (Reasoning and Acting) framework, a sophisticated approach to prompt engineering that enables AI models to reason about tasks and take appropriate actions. Understanding ReAct is crucial for developing more intelligent and context-aware AI applications.

Understanding ReAct Framework

The ReAct framework combines reasoning and acting capabilities to enhance the performance of AI models. It allows the model to generate intermediate reasoning steps before arriving at a final answer, making the process more transparent and reliable. This is particularly useful in complex problem-solving scenarios where simple prompt-response mechanisms fall short.

import openai

# Set up the API key
openai.api_key = 'your-api-key'

# Define the prompt
prompt = """Let's think step by step:

## Step 1: Identify the problem
The problem is to calculate the sum of the first 10 natural numbers.

## Step 2: Recall the formula
The sum of the first n natural numbers is given by n * (n + 1) / 2.

## Step 3: Apply the formula
For n = 10, the sum is 10 * (10 + 1) / 2.

## Step 4: Calculate the result
10 * 11 / 2 = 55

The answer is 55."""

# Generate the response
response = openai.Completion.create(
  engine="text-davinci-003",
  prompt=prompt,
  max_tokens=100
)

print(response.choices[0].text.strip())

Try it in Google Colab:

The answer is 55.

Implementing ReAct in Python

To implement the ReAct framework in Python, you need to structure your prompts to include reasoning steps. This involves breaking down the problem into smaller, manageable parts and guiding the model through each step. The model then uses this structured approach to arrive at a solution, making the process more interpretable and reliable.

import openai

# Set up the API key
openai.api_key = 'your-api-key'

# Define a more complex problem
prompt = """Let's think step by step:

## Step 1: Identify the problem
The problem is to determine the day of the week for a given date.

## Step 2: Recall the method
We can use Zeller's Congruence to find the day of the week.

## Step 3: Apply Zeller's Congruence
For date: day=21, month=8, year=2023

## Step 4: Calculate h (the day of the week)
h = (q + ((13*(m+1))/5) + K + (K/4) + (J/4) + 5J) % 7
where q=21, m=6 (month adjusted for Zeller's), K=year%100, J=year//100

## Step 5: Interpret the result
0 = Saturday, 1 = Sunday, 2 = Monday,..., 6 = Friday

The answer is [calculated day of the week]."""

# Generate the response
response = openai.Completion.create(
  engine="text-davinci-003",
  prompt=prompt,
  max_tokens=100
)

print(response.choices[0].text.strip())

💡 Tip: Ensure that your reasoning steps are clear and logically ordered to help the model understand and follow the process effectively.

Key Concepts

Check Your Understanding