By Michael Fatemi, December 2022
For more interesting projects, check out my Youtube or my website
I'm also working on AugmateAI
-
Assumes you have your OpenAI API key as the environment variable
OPENAI_API_KEY. -
Automatically cuts off the context that would cause the token count to overflow.
-
Optionally, you can make it possible to search Bing for answers to questions, but you need an Azure account and the environment variable
BING_SEARCH_V7_SUBSCRIPTION_KEY1. GPT-3 is made aware of this through the prompt, if you don't want this then remove the corresponding line from the prompt. -
Optionally, you can make it possible to load text content of pages on the web. This requires installing the packages in
package.json, because there are libraries to do that that are implemented in JavaScript but not with Python. This is also signalled to GPT-3 through the prompt, and you can remove it from the prompt if you don't want to have that feature. -
Have fun and feel free to share any interesting behavior with me!
python jarvis.py
Example 1: Writing, debugging, and executing code (some parts omitted). Installed the scipy package when it wasn't found!
User: write a program to generate the parabola y=x^2 + 5, with some random noise added to it. save it to parabola_noise.py.
Assistant: Program written to parabola_noise.py.
User: run the program
Assistant: Program executed. Parabola with noise image generated.
User: add to the program a section that uses curve fitting to estimate the parameters of the parabola.
Assistant: Program written to parabola_noise.py.
User: run the program
Assistant: Executing: python3 parabola_noise.py
Traceback (most recent call last):
File "parabola_noise.py", line 3, in <module>
from scipy.optimize import curve_fit
ModuleNotFoundError: No module named 'scipy'
Assistant: Program failed to execute. Scipy module not found.
User: you can import it with pip
Assistant: Executing: pip3 install scipy
Collecting scipy
Successfully installed scipy-1.9.3
Assistant: Scipy module installed.
User: run the program
Assistant: Program executed. Parabola with noise and curve fitting image generated.
Example 2: Scraping Web Content with curl and BeautifulSoup. Cool to see GPT-3 interacting with the web.
Assistant: No problem. Is there anything else I can help you with?
User: yeah, please retrieve the content of https://openai.com/blog/chatgpt/ and write a Python script to find out the title.
Assistant: Do you want me to retrieve the content of https://openai.com/blog/chatgpt/ and write a Python script to find out the title?
User: yes
Assistant: Execute: curl https://openai.com/blog/chatgpt/ > chatgpt.html
Assistant: Retrieved the content.
Assistant: Execute: echo "from bs4 import BeautifulSoup\nwith open ( 'chatgpt.html') as f: \n et_ title.py
Command: echo "from bs4 import BeautifulSoup\nwith open 'chatgpt.html') as f: \n
Command line:
soup = BeautifulSoup(f, 'html.parser') In
soup = BeautifulSoup(f, 'html.parser') \n
print (soup.title.string)" > get_title.py
print (soup.title.string)" > get_title.py
Assistant: Program written.
User: run the program you just wrote and tell me the output
Assistant: Execute: python3 get_title.py
Assistant: The output is "ChatGPT: Optimizing Language Models for Dialogue".
User: thanks!
Assistant: You're welcome! Is there anything else I can help you with?