AutoGPT - The Next Big Thing in AI
The excitement around ChatGPT and GPT-4 hasn’t even cooled down and we already have a new shiny toy. But does AutoGPT live up to the hype?
The excitement around ChatGPT and GPT-4 hasn’t even cooled down and we already have a new shiny toy. Autonomous LLM-based agents, or AutoGPT in short, are the new next big thing in the AI world.
Let’s take a closer look at AutoGPT - what it is, what it can do and what are its limitations.
Imagine you’ve been tasked to create a report comparing different products and listing their pros and cons. To make the task complete quicker, you use ChatGPT and run a series of prompts back and forth, using the previous response as a base for the next prompt, until the report is complete.
Now imagine this entire process was fully automated. From your perspective, you just described what you need and moments later you get the result.
This is in a nutshell what AutoGPT is.
The idea behind AutoGPT is quite simple - you give an AI a complex task (like creating that report) and under the hood, AutoGPT asks itself how to accomplish this task, breaks it into smaller tasks and executes them.
Before executing each step, AutoGPT writes back what it is exactly planning to do alongside with reasoning behind the plan and the criticism. The AI then waits for the human to approve the step and then it proceeds to execute the plan.
AutoGPT’s popularity skyrocketed in April and became a hot topic in the AI community.
The most popular AutoGPT projects - Auto-GPT and BabyAGI - are open-source and available on GitHub. If you know how to use terminal, Docker and GitHub, you can run AutoGPT on your computer and experiment with them.
There are also online services like AgentGPT or GodMode available which have done the setup for you.
For both local and online services, you will need an OpenAI account and provide an API key. You get a $5 credit when you create an OpenAI account and this should be enough to experiment with AutoGPT.
We need to acknowledge that there is a lot of hype in the AI space. AutoGPT is now at the centre of this hype and the entire hustle culture has been created around it.
All the mentioned projects use GPT-4 and GPT-3.5 (or other models, like LlaMa) to run the queries. The same limitations and quirks that apply to large language models also apply to AutoGPT. AutoGPT can hallucinate facts and make up things that do not exist in reality. It may introduce bias into the results and be very confident about it.
AutoGPT functionalities are not currently baked in ChatGPT but I won’t be surprised to see them being introduced sometime in the future (after being proven to be safe for general use).
AutoGPT is another step in the pursuit of automating as many things as possible. It is still an experimental technology but it has the promise to be a powerful tool and to be the next big thing in AI.
Things are going to be getting interesting in the next couple of years.