Like many people, you may have been recently excited by the possibility of ChatGPT and other large language models (LLM) like the new Bing or Google’s Bard.
For anyone who hasn’t come across them – which you probably haven’t since ChatGPT is reportedly the fastest growing app of all time – here’s a quick recap:
LLMs are software algorithms trained on large text datasets, enabling them to understand and respond to human language in a very lifelike way.
The best-known example is ChatGPT, a chatbot interface powered by GPT-4 LLM that has dominated the world. ChatGPT is able to communicate like a human and compose everything from blog posts, letters, and emails to fiction, poetry, and even computer code.
As impressive as they are, until now, LLMs have been limited in a significant way. They tend to only complete a task, such as answering a question or constructing a piece of text, before requiring more human interaction (known as “prompts”).
This means they are not always good at more complex tasks that require multi-step instructions or depend on external variables.
come in Auto-GPT – a technology that tries to overcome this obstacle with a simple solution. Some believe this could be the next step towards the “holy grail” of AI – the creation of general, or strong, AI.
Let’s see what this means first:
Strong AI vs. weak AI
Current AI applications are generally designed to perform one task, getting better at it as they are fed more data. Some examples include analyzing images, translating languages, or navigating self-driving cars. Because of this, they are sometimes referred to as “special AI,” “narrow AI,” or “weak AI.”
A general AI is one that is theoretically capable of performing many different types of tasks, even if they were not originally created to perform, much like the way a naturally intelligent being (such as of one person). This is sometimes called “strong AI” or “artificial general intelligence” (AGI).
AGI is probably what we traditionally envisioned when we envisioned what AI would look like in the days before machine learning and deep learning made AI weak/narrow as an everyday reality at the beginning of the last decade . Think of the science fiction AI shown by robots like Data in Star Trek that can do almost anything a human can do.
So what is Auto-GPT?
The simplest way to look at it is that Auto-GPT is able to perform more complex, multi-step procedures than current LLM-powered applications by creating its own signals and feeding them back to itself, which creates a loop.
Here’s one way to think about it: Getting the best results from an application like ChatGPT requires careful thought in the way you phrase the questions you ask it. So why not let the application generate the question itself? And while it’s at it, also ask what the next step should be – and how it should go about that … and so on, creating a loop until the task is completed.
It works by dividing a larger task into smaller sub-tasks and then spinning up independent Auto-GPT instances to execute them. The original instance acts as a kind of “project manager,” which coordinates all the work performed and consolidates it into a finished result.
As well as using GPT-4 to generate sentences and prose based on the text it has studied, Auto-GPT has the ability to browse the internet and incorporate the information it finds there into its calculations and output. In this respect, it is more similar to the new GPT-4-enabled version of Microsoft’s Bing search engine. It also has better memory than ChatGPT, so it can generate and remember longer sets of commands.
Auto-GPT is an open-source application that uses GPT-4 and was created by one person, Toran Bruce Richards. Richards said that he was inspired to develop it because traditional AI models “, while powerful, often struggle to adapt to tasks that require long-term planning, or are unable to autonomously adjust their strategies based on real-time feedback. “
It is one of a class of applications called recursive AI agents because they have the ability to automatically use the results they generate to create new prompts, combining these operations to complete complex tasks.
Another agent is BabyAGIcreated by a partner in a venture capitalist firm to help him with daily tasks that are too complex for something like ChatGPT, such as researching new technologies and companies.
What are some applications of Auto-GPT and AI agents?
Although apps like ChatGPT have become popular due to their ability to generate code, they tend to be limited to relatively short and simple programming and software designs. Auto-GPT, and possibly other AI agents that work in a similar way, can be used to build software applications from start to finish.
Auto-GPT is also able to help businesses automatically increase their net worth by analyzing their processes and making intelligent recommendations and insights about how they can be improved.
Unlike ChatGPT it can also access the internet, meaning you can ask it to carry out market research or perform other similar tasks – for example “find me the best range of golf clubs in worth less than $500.”
A very disturbing task it sets out is the “destroy humanity” – and the first sub-task it assigned itself to do this was to begin researching the most powerful atomic weapons of all time. Since its output is still limited to producing text, its creator assures us that it won’t really be up to this task – hopefully.
Auto-GPT can also likely be used to improve itself – its creator says it can create, check, analyze and test updates to its own code that could potentially make it more capable and efficient.
It can even be used to create better LLMs that can form the basis of future AI agents, by speeding up the model making process.
What could this mean for the future of AI?
Since generative AI applications began to emerge, it is clear that we are only at the beginning of a very long journey, how AI will evolve and impact our lives and society.
Are Auto-GPT and other agents that follow the same principles the next step of that journey? It certainly seems likely. At the very least, we can expect AI tools that allow us to perform more complex tasks than the relatively simple things that ChatGPT can do, to start becoming the norm.
Soon, we’ll start seeing more creative, sophisticated, diverse and useful AI output than the simple text and images we’re used to. These will undoubtedly eventually have a greater impact on the way we work, play and communicate.
Other potential positive impacts include the reduced cost and environmental impact of creating LLMs (and other machine learning-related activities) as autonomous, recursive AI agents find ways to make the process more efficient.
However, we also have to consider that it won’t really solve any of the problems associated with generative AI. This includes the variable (to put it nicely) accuracy of the output it creates, the potential for abuse of intellectual property rights, and the possibility of its use to spread bias or harmful content. In fact, by building and running many more AI processes to achieve larger tasks, it could exacerbate these issues.
The potential problems don’t stop there – noted AI expert and philosopher Nick Bostrom recently said he believes that the latest generation of AI chatbots (such as GPT-4) are even starting to show signs of sentience. Which could create a whole new moral and ethical dilemma if as a society we plan to start creating and operating them on a large scale.
follow me Twitter o LinkedIn. Look my website or some of my other work here.