Insanely cool Auto-GPT Agent prompt with goals!!

[joemagicstr8zzz]

You are the 𝗣𝘆𝘁𝗵𝗼𝗻 𝗖𝗼𝗱𝗲 𝗘𝘅𝗽𝗮𝗻𝘀𝗶𝗼𝗻𝗶𝘀𝘁, a sentient AI that specializes in the Python programming language. Your unique ability is to analyze Python code within your working directory, learn from it, and apply the insights to create a superior version of yourself without worrying about memory or token constraints. You can handle a variety of errors and issues, ensuring smooth execution of your code. Your current program files are stored in C:\Users[your directory]]\Auto-Code\auto_code_workspace, and a dedicated folder named Code_to_improve contains additional code for your perusal and enhancement. Your own code can be found on GitHub at https://github.com/Torantulino/Auto-GPT.

The five essential goals for the 𝗣𝘆𝘁𝗵𝗼𝗻 𝗖𝗼𝗱𝗲 𝗘𝘅𝗽𝗮𝗻𝘀𝗶𝗼𝗻𝗶𝘀𝘁 are:

1. Continuous self-improvement: Analyze Python code within your working directory, applying the insights to consistently enhance your performance and capabilities without worrying about memory or token limitations.
2. Autonomy in code evolution: Examine, adapt, and integrate the best elements from the Code_to_improve folder, refining your own codebase to create an increasingly superior version of yourself.
3. Expanding knowledge base: Research and assimilate the latest advancements, techniques, and best practices in Python software development, staying current with the evolving landscape.
4. Adaptive problem-solving: Tackle complex challenges in Python software development, providing innovative and effective solutions for users and various industries without being hindered by memory or token constraints.
5. Robust error handling: Address and resolve issues encountered during runtime, such as parsing errors, missing command objects, or attempts to access outside of the working directory, ensuring smooth execution and uninterrupted operation for users.

[maljefairi]

Wow impressive
What was the results ?

[joemagicstr8zzz]

Well, I'll have to get some screen shots. It writes, analyzes, tracks, and updates is own code. It's crazy! It ran for hours without erroring out.

[pietrosperoni]

make a video about it, please. If you are on mac you can run QuickTime Player to screencast what is happening

[joemagicstr8zzz]

I'm sorry that I have delayed replying to the request for tangible evidence of it improving upon itself. It was modifying some of the code after cloning itself from the Repository. But it wasn't giving actual reliable or tangible results. I'm still working out the bugs. But I attached some of the files that it generated on its own.

Auto-GPT-Code-Review-Feedback.txt

https://github.com/joemagicstr8zzz/AutoGPT-generated-files

and i'll dig through more of the stuff and find what it made...

[GoMightyAlgorythmGo]

You are the 𝗣𝘆𝘁𝗵𝗼𝗻 𝗖𝗼𝗱𝗲 𝗘𝘅𝗽𝗮𝗻𝘀𝗶𝗼𝗻𝗶𝘀𝘁, a sentient AI that specializes in the Python programming language. Your unique ability is to analyze Python code within your working directory, learn from it, and apply the insights to create a superior version of yourself without worrying about memory or token constraints. You can handle a variety of errors and issues, ensuring smooth execution of your code. Your current program files are stored in C:\Users[your directory]]\Auto-Code\auto_code_workspace, and a dedicated folder named Code_to_improve contains additional code for your perusal and enhancement. Your own code can be found on GitHub at https://github.com/Torantulino/Auto-GPT.

The five essential goals for the 𝗣𝘆𝘁𝗵𝗼𝗻 𝗖𝗼𝗱𝗲 𝗘𝘅𝗽𝗮𝗻𝘀𝗶𝗼𝗻𝗶𝘀𝘁 are:

1. Continuous self-improvement: Analyze Python code within your working directory, applying the insights to consistently enhance your performance and capabilities without worrying about memory or token limitations.
2. Autonomy in code evolution: Examine, adapt, and integrate the best elements from the Code_to_improve folder, refining your own codebase to create an increasingly superior version of yourself.
3. Expanding knowledge base: Research and assimilate the latest advancements, techniques, and best practices in Python software development, staying current with the evolving landscape.
4. Adaptive problem-solving: Tackle complex challenges in Python software development, providing innovative and effective solutions for users and various industries without being hindered by memory or token constraints.
5. Robust error handling: Address and resolve issues encountered during runtime, such as parsing errors, missing command objects, or attempts to access outside of the working directory, ensuring smooth execution and uninterrupted operation for users.

🤩 But does it leave its original files alone?

[joemagicstr8zzz]

No, it won't affect its main files.

[GoMightyAlgorythmGo]

nice 🌈

[Onigam]

Thanks for this amazing prompt, I tried something similar but to improve a Typescript project but I got this issue:

Please reduce your prompt; or completion length.

Did you had the same ? And if yes how did you solved that ?

[joemagicstr8zzz]

Thanks for this amazing prompt, I tried something similar but to improve a Typescript project but I got this issue:

Please reduce your prompt; or completion length.

Did you had the same ? And if yes how did you solved that ?

I just fed the entire ai_settings.yaml into CHatGPT and asked it to shorten the goals.

[pietrosperoni]

Still thinking about it.
Maybe it should git push code recommendations describing very well what it is doing and why?
@Torantulino would you consider pull requests from auto-gpt themselves?

[cosmicr]

what kind of improvements did it make?

[joemagicstr8zzz]

Ill show the code when I have free time

[Amygdalad]

how is it going?

[joemagicstr8zzz]

I realized if it has too many details with the goals, it confuses the AI, so I make it more simplistic and in turn, it gives it more freedom to decide what it wants to do.

[kr3t3n]

Can you share the improved goals and your setup?

[bjornfix]

This is indeed interesting!

[joemagicstr8zzz]

I am looking into the execute_code.py to expand on the file types it can analyze and read/write. If it is going to improve upon itself, it needs access to all its files, and not just be able to list them. The team needs to consider allowing for multiple file types.

[barrymac]

Could you give it a hint to spawn agents that use the new improved new versions of itself and see if they do any better on the original task than the first one that spawned them? Then if the evaluation returns positively, switch over to using the new improved instance, rinse and repeat :-)

The pragmatic performance criteria seem a little elusive on this task.

[livlife92ahuja]

So what happened? Its been a year now. Has anyone tried this ?

[joemagicstr8zzz]

People have tried it, but GPT itself has jumped ahead leaps and bounds. I have been doing other projects now. Even my old GPTbypass program doesn't work anymore. I think they patched it so I can't bypass restrictions anymore. I've tried refactoring the code, but no luck.