From ChatGPT to AutoGPT -Experiment

Auto-GPT is an open-source application showcasing the capabilities of the GPT-4 language model.?

Unlike ChatGPT that requires numerous detailed prompts, Auto-GPT generates its own prompts and complete the given goals. To complete the goals it will access websites and search engines and generate the result by reading and writing to different files.

What makes it interesting is the ability of self-evaluating and verifying the accuracy of the collected data and discard what’s incorrect to spawn a new subtask & gather more data. Because of such ability to self-generate prompts to complete tasks, it is also referred to as an autonomous AI agent.

Experiment

Initially we gave the AI agent 'a goal' and watch as it thinks, comes up with an execution plan, it scrapes the web for the best information out there, and then it autonomously does the task for us and continues to constantly improve itself.?

We set the following goals - based on the Low-Code/No-Code project we're working on:

• Search the different types of 'No-code' or 'low-code' solutions in Google
• Extract 'top 5 platforms' for 'non-technical professionals' with features such as 'drag and drop'
• Based on the extracted content write a summary of the findings
• Save the written summary as a text file

The AI Agent, NoCodeGPT in our case, starts explaining it's actions, next steps and reasoning.

Next it start working on the goals. In each steps we can control the prompt to be continuous or not. To avoid the infinite loop:

It went through the goals, search the web, gather data, analyse the data and come up with the final result that was written to an external file that can be available in the workspace. The Final results were:

1. Bubble

2. Airtable

3. Zenkit

4. Easysend

5. Kintone

Although it looks impressive, there are some concerns regarding AutoGPT:

1. Price - to complete the tasks the agent requires a call to the expensive GPT-4 model that maxes out tokens to give a better reasoning.
2. Auto-GPT can gets stuck - this includes infinite loop or carrying out actions that would not usually be authorized. This is because AutoGPT learns and makes decisions based on the data it is trained on, which may not always be perfect or unbiased.
3. Data breaches - handling sensitive data, always has a risk of data breaches. Therefore, it is essential to take proper security measures

Final Thoughts

As AutoGPT was just launched on March 30, 2023, we still haven’t seen the full capability of this AI application in different use-cases. Further, it comes with many personal privacy and data security issues that need immediate attention.

The idea of an AI learning from it's mistakes, and delivers good results, while allowing us to optimize our prompts better is exciting and the.