Skip to content

Baby AGI is cool, but why write so much code when it could just be a single GPT4 call?

License

Notifications You must be signed in to change notification settings

personx000/OneStepBabyAGI

 
 

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

4 Commits
 
 
 
 
 
 

Repository files navigation

OneStepBabyAGI

Baby AGI is one the fastest liked GitHub repos ever.

Very cool, but way too much code.

By the time I even got through 10 loc I was already tired, and I didn't want to bother reading the the rest of the code.

Instead I thought "why not just let GPT4 hallucinate the entire program?"

So, I took the docstrings, didn't bother reading anything in the repo and now I present: OneStepBabyAGI, which performs the same high level function as BabyAGI but in a single GPT4 call!

GPT4 is a linguistic programming machine

An interesting note about framing to understand what's happening - GPT3 was a linguistic function executor. GPT4 is a linguistic program executor - it can understand, save, and create complex intermediate states.

Will probably write a blog post about this at some point, but for now you have this two sentence knowledge drop.

Usage

Send new user messages through the ChatGPT api so simulate each step, e.g.

messages = [{"role": "user", "content": "simulate `run_baby_agi_one_step("create the next viral twitter post on AGI")`}]"

Which yielded the following output

Objective: create the next viral twitter post on AGI
Priors: ""

1. Execution Agent:
   Input: objective="create the next viral twitter post on AGI", task="generate a catchy tweet"
   Output: "Breaking News: Artificial General Intelligence takes over the world! 🌍💡🤖 Are you ready for a future powered by AGI? #AGIRevolution #TechTrends"

2. Task Creation Agent:
   Input: objective="create the next viral twitter post on AGI", result={"tweet": "Breaking News: Artificial General Intelligence takes over the world! 🌍💡🤖 Are you ready for a future powered by AGI? #AGIRevolution #TechTrends"}, task_description="generate a catchy tweet", task_list=[]
   Output: [{"task": "find relevant AGI article"}, {"task": "create a Twitter thread"}, {"task": "engage with users"}]

3. Prioritization Agent:
   Input: objective="create the next viral twitter post on AGI", task_list=[{"task": "find relevant AGI article"}, {"task": "create a Twitter thread"}, {"task": "engage with users"}]
   Output: 
   1. Find relevant AGI article
   2. Create a Twitter thread
   3. Engage with users

Here's a quick screenshot in the Open AI Playground

image

About

Baby AGI is cool, but why write so much code when it could just be a single GPT4 call?

Topics

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published