ChatGPT is awesome, but this conversational model is constantly waiting for the user to make requests and questions through prompts. There are developers who are already working on more autonomous models that are capable of performing several consecutive actions by themselves without waiting for human interaction, and one of them is causing real excitement (and a bit of terror) on the internet. Its name is Auto-GPT.
Auto-GPT. The new trend in the field of generative AIs is “autoprompting”. Instead of a user entering the requests in the form of prompts, it is the AI model itself that feeds itself based on the initial input. Auto-GPT (or AutoGPT) is an experimental Open Source application — the code is in the GitHub repository — developed in Python by Toran Bruce Richards. It is designed to “develop and manage business ideas to generate income.” Anyone can install it on their computer, and while it can be used with a GPT-3.5 API, it’s best to use it with a GPT-4 API, which is only available to paid ChatGPT Plus subscribers.
He cooks it, he eats it. Once we set it a goal like “create a business that generates income from a $100 investment,” Auto-GPT comes up with just a multi-step roadmap that it then develops in detail. That these plans work later, of course, is something else, but the surprising thing about the model is that it is quite autonomous.
Advantages over ChatGPT. The system makes use of ChatGPT —or rather, its API— to generate text, but it has the advantage of being able to access the Internet to collect data —as Bing does with ChatGPT— and it has both short- and long-term memory, a clear ( and intentional ) limitation of current models. It is even capable of integrating 11 Labs, an AI system that generates a synthesized voice from text so that you can “talk” with Auto-GPT.
Does ChatGPT get old? Some experts are already talking about how developments like Auto-GPT are the next evolution of this unique field. Andrej Karpathy, former director of the AI division at Tesla and now at OpenAI, made it clear: “In my opinion the next frontier of prompt engineering: AutoGPTs.”
Alternatives. In recent days, other similar projects have been appearing, such as Baby-AGI, Pinecone or JARVIS, whose objective is to automate complex tasks and “let generative AIs do it”. The language model acts as a kind of controller that uses certain additional language models and various tools automatically to achieve the intended goal of the user who made the original input. OpenAI itself raises something similar with the recent (preliminary) support of plugins in GPT-4, for example.
Fear at the beginning of an IAG . These types of systems that are capable of operating autonomously have a significant risk: that they end up making unforeseen and, in some way, toxic or dangerous decisions. Some describe Auto-GPT as an approach to Artificial General Intelligence (IAG or AGI), and that is also generating an important debate.