Python Chatbot

The interaction in the command line will look like this when it starts:

Bot: Hello!
You:

After you've chatted a bit it might look like this:

Bot: Hello!
You: Hi!
Bot: Nice to meet you.
You: Likewise.
Bot: How was your day?
You:

We will achieve this by dynamically creating a prompt for GPT-3. The prompt will look something like this:

August is a kind bot who wants to be friends.

Bot: Hello!
You: [USER_INPUT]
Bot:

Greet the user

Time for "Hello world!", let's test we have everything up and running by greeting the user.

Print a greeting from your bot:

# greet the user
print("Bot: Hello!")

Run your app

Your app should say "Bot: Hello!" and then exit.

Bot: Hello!

Capture user input

We need to ask the user to reply to the bot, and then capture what they type in.

Use a while-loop to make sure we keep asking the user for input. This will let our user talk to the bot for as long as they want.

# greet the user
print("Bot: Hello!")

while True:
  # capture user input
  user_input = input("You: ")

Run your app

Your app should say hi, then wait for you to reply. It should ask you to reply over and over, and never quit.

The bot doesn't reply yet!

Bot: Hello!
You: Hi
You: Hello?
You: Anyone...?
You:

Configure OpenAI

Install OpenAI library

Install the openai Python library, so you can use it in your app.

pip install openai

Important

If you are using python3 to run your app, you will need to use pip3 (instead of pip) to install the openai library.

pip3 install openai

Add OpenAI to your app

Import the openai library, and then configure it with your API Key from the OpenAI settings.

import openai

# use your OpenAI account
openai.api_key = "YOUR_API_KEY"

# greet the user
print("Bot: Hello!")

while True:
  # capture user input
  user_input = input("You: ")

Run your app

There should not be any errors!

Your app should still just be a bot that never replies to you...

Bot: Hello!
You: Hey!
You: How are you?
You: Are you there??
You:

Build a prompt

We want our prompt to be a string that is formatted like the prompts we used in the OpenAI Playground.

For this prompt we want a pattern like this, where we are prompting GPT-3 to fill in the blank space after "Bot:":

August is a kind bot who wants to be friends.

Bot: Hello!
You: [USER_INPUT]
Bot:

import openai

# use your OpenAI account
openai.api_key = "YOUR_API_KEY"

# greet the user
print("Bot: Hello!")

while True:
  # capture user input
  user_input = input("You: ")

  # create a prompt for GPT-3
  prompt = "August is a kind bot who wants to be friends.\n"

  # leave a blank line before the chat history
  prompt += "\n"

  # add example chat history
  prompt += "Bot: Hello!\n"
  prompt += "You: " + user_input + "\n"

  # add empty line for GPT-3 to complete
  prompt += "Bot:"

  # optional line to preview your prompt in the command line
  print("\n===\n" + prompt + "\n===\n")

Run your app

Each time you send a message to the bot, it should print out the prompt you will be sending to GPT-3 in the next step.

The prompt start and end is indicated by the === lines.

Bot: Hello!
You: hey

===
August is a kind bot who wants to be friends.

Bot: Hello!
You: hey
Bot:
===

You:

Generate bot response

We need to send our prompt to GPT-3, so that it can generate a completion for us to use as the bot's response.

This is a multi-step process, where we need to:

• send the prompt request to GPT-3 to get back a completion
• extract the bot response from GPT-3's completion
• display the bot response to the user

We won't see any change in how our bot works until we have added all three steps to our code.

Send prompt to GPT-3

import openai

# use your OpenAI account
openai.api_key = "YOUR_API_KEY"

# greet the user
print("Bot: Hello!")

while True:
  # capture user input
  user_input = input("You: ")

   # create a prompt for GPT-3
  prompt = "August is a kind bot who wants to be friends.\n"

  # leave a blank line before the chat history
  prompt += "\n"

  # add example chat history
  prompt += "Bot: Hello!\n"
  prompt += "You: " + user_input + "\n"

  # add empty line for GPT-3 to complete
  prompt += "Bot:"

  # optional line to preview your prompt in the command line
  print("\n===\n" + prompt + "\n===\n")

  # send the GPT3 request
  completion = openai.Completion.create(
    model="text-davinci-002",
    prompt=prompt,
    temperature=0,
    max_tokens=150,
    stop="You:"
  )

We want to capture the response we get from GPT-3, so we use a new variable called completion.

The openai library we added to our code has a function we can use called openai.Completion.create(). This function takes all the same parameters as the ones available in the Playground.

Adding this part to our app will send the request, but it won't display any results without us adding more code.

Extract the bot response

When you generate a completion, that result actually has way more data than just "what the bot says back".

Here's an example of what the whole completion looks like:

{
  "id": "cmpl-6FvbvTFJEbzyIlAJ9VP7IysUzPk8o",
  "object": "text_completion",
  "created": 1669254395,
  "model": "text-davinci-002",
  "choices": [
    {
      "text": " How are you today?",
      "index": 0,
      "logprobs": null,
      "finish_reason": "stop"
    }
  ],
  "usage": {
    "prompt_tokens": 75,
    "completion_tokens": 5,
    "total_tokens": 80
  }
}

Can you spot the line that actually has the bot's reply in it?

The completion has a property called choices, which is a list. The first item in that list has a property called text, which is the bot's response. Also, notice that the text has some empty space at the start which we will want to strip off.

We can extract the text from the completion and strip the white space to get the bot response:

  # send the GPT3 request
  completion = openai.Completion.create(
    model="text-davinci-002",
    prompt=prompt,
    temperature=0,
    max_tokens=150,
    stop="You:"
  )

  # extract the response
  bot_response = completion.choices[0].text.strip()

Printing the completion

If you try to just do print(completion) it won't work because Python can't turn a Class into a String.

As an alternative, you can turn the Completion into JSON and print that instead:

import json

print(json.dumps(completion, indent=2))

Display the response

  # send the GPT3 request
  completion = openai.Completion.create(
    model="text-davinci-002",
    prompt=prompt,
    temperature=0,
    max_tokens=150,
    stop="You:"
  )

  # extract the response
  bot_response = completion.choices[0].text.strip()

  # display the response
  print("Bot: " + bot_response)

Run your app

You bot should now be fully functional, why don't you get to know each other a bit?

Bot: Hello!
You: hey! what's your favourite colour?
Bot: My favourite colour is blue, like the ocean.
You: do you have any hobbies?
Bot: I like to browse the Internet.
You: what's your birthday?
Bot: My birthday is on 24 November 2022.
You:

Advanced functionality

Personal Facts

It can be cool for your bot to have a little bit of personality and know some information about themselves.

We can add more info to the prompt, so that GPT-3 will consider this additional info when generating a response.

Write some facts about the bot

We write our facts as a list so that we can loop over them to add them all to the prompt.

import openai

# use your OpenAI account
openai.api_key = "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"

# write some basic facts about your bot.
# these will be added to every prompt.
facts = [
    "The bot was created by Tanya Gray.",
    "The bot was created on 24 November 2022.",
    "The bot's favourite colour is blue, like the ocean.",
    "The bot likes to browse the Internet.",
    "The bot is an only child.",
]

# greet the user
print("Bot: Hello!")

Add facts to the prompt

At the moment, our prompt is quite simple and looks like this:

August is a kind bot who wants to be friends.

Bot: Hello!
You: [USER_INPUT]
Bot:

After this change, we are aiming to have our prompt include the facts, like this:

August is a kind bot who wants to be friends.

The bot was created by Tanya Gray.
The bot was created on 24 November 2022.
The bot's favourite colour is blue, like the ocean.
The bot likes to browse the Internet.
The bot is an only child.

Bot: Hello!
You: [USER_INPUT]
Bot:

To achieve this, we need to change the way we generate our prompt. The facts should be added after the first line of the prompt, and before the example messages.

We can add a for-loop to our code, that adds each fact to the prompt on a new line:

  # build a prompt for GPT-3
  prompt = "August is a kind bot who wants to be friends."
  prompt += "\n"

  # leave a blank line before the facts list
  prompt += "\n"


  # add the facts to the prompt
  for fact in facts:
      prompt += fact
      prompt += "\n"

  # leave a blank line before the chat history
  prompt += "\n"

  # add example chat history
  prompt += "Bot: Hello!\n"
  prompt += "You: " + user_input + "\n"

  # add empty line for GPT-3 to complete
  prompt += "Bot:"

Now when we ask our bot about these specific topics, it will answer based on the facts that we have provided.

Run your app

Ask your bot about something it should know, thanks to these new facts:

Bot: Hello!
You: hey! what's your favourite colour?
Bot: My favourite colour is blue, like the ocean.
You: do you have any hobbies?
Bot: I like to browse the Internet.
You: what's your birthday?
Bot: My birthday is on 24 November 2022.
You:

Remembering topics

import openai

# use your OpenAI account
openai.api_key = "YOUR_API_KEY"

# store the message history for bot and user
# so we can add the messages to the prompt each time
message_history = []

# greet the user
print("Bot: Hello!")

# add bot greeting to the message history
message_history.append("Bot: Hello!")

while True:
  # capture user input
  user_input = input("You: ")

  # add user input to the message history
  message_history.append("You: " + user_input)

  # build a prompt for GPT-3
  prompt = "August is a kind bot who wants to be friends."
  prompt += "\n"

  # leave a blank line before the chat history
  prompt += "\n"

  # add the chat history to the prompt
  # using only the last 6 messages in the history
  for message in message_history[-6:]:
    prompt += message
    prompt += "\n"

  # add the empty line for the bot to respond
  prompt += "Bot:"

  # send the GPT3 request
  completion = openai.Completion.create(
    model="text-davinci-002",
    prompt=prompt,
    temperature=0,
    max_tokens=150,
    stop="You:"
  )

  # extract the response
  bot_response = completion.choices[0].text.strip()

  # display the response
  print("Bot: " + bot_response)

  # add bot response to the message history
  message_history.append("Bot: " + bot_response)