Task-driven Autonomous Agent or the first-step towards the infinite paperclips AI machine

I came across this “whitepaper” written by ChatGPT: https://yoheinakajima.com/task-driven-autonomous-agent-utilizing-gpt-4-pinecone-and-langchain-for-diverse-applications/

It was based on the prompts:

I would like to incubate a startup founded by GPT4. Please write instructions and code for a code that runs intermittently to create, complete, and prioritize tasks. This should use Python and GPT4 API, and store tasks in Pinecone. Use Langchain agents with ChatGPT plug-ins to complete tasks. Each loop should consist of the following steps: (1) review task list and complete first task by retrieving relevant context from Pinecone and using Langchain agent with ChatGPT plug-ins, (2) enrich result of task with further context and store in Pinecone, (3) create new tasks based on results, add do task list, and reprioritize task list for next iteration.

THEN:

You are an Al science researcher. Analyze the following code and write a scientific research paper describing what it’s for, how it works, the novel approaches it takes, and the significance of this new approach for Al and the world.

THEN:

Analyze the following code and show a flowchart of how it works by writing it in Graphviz.

THEN:

Write a tweet thread based on the following research paper:

The best part about it, Chatgpt can produce Graphviz… of course it can - nice.