AI’s Newborn Auto-GPT, Explained!

A spate of developments are underway in the global artificial intelligence (AI) space. The launch of ChatGPT in November 2022 proved to be a seminal moment in the world of computing and technology, as it ignited a tech war between giants Google and Microsoft over what seems to be the future of technology.

But, while ChatGPT seemed to be a breakthrough, allowing users to hold human-like conversations with a chatbot, not many expected a newer upgraded version within months that would spawn a new era of technology. 

The launch of OpenAI’s GPT-4, in the form of both API and a feature of ChatGPT Plus, in March this year changed all that. Created by developer Toran Bruce Richards, Auto-GPT is an open-source platform that deploys OpenAI’s GPT-4 large language model (LLM) to automate multi-step projects that would have otherwise required multiple human efforts if worked directly with GPT-4.

The new open-source platform aims to make GPT-4 entirely self-governing and has seen rapid adoption from YouTubers to company executives in the past month. But, the game-changing part is that the platform has shown sparks of artificial general intelligence (AGI). 

Simply put, AGI could herald the era of artificial intelligence where AI models such as GPT could possess the ability to define and perform tasks autonomously to achieve objectives without any human intervention. 

Launched on March 30 on GitHub, the launch of Auto-GPT sent the entire internet into a tizzy as users found usage in multiple domains, including autonomous coding and debugging, internet searching, and operating as an independent Twitter bot. But this was just the tip of the iceberg. 

What Comprises The DNA?

Auto-GPT’s architecture is broadly based on the GPT-4 and GPT-3.5 language models, which serve as the backbone of the platform, allowing it to reason and process. On top of that, the new LLM autonomously iterates and learns from its past mistakes to improve its results. 

The new platform is integrated with vector databases, enabling Auto-GPT to preserve context and easily fetch previous iterations to make better decisions. 

“This is like equipping the robot with a long-term memory to remember past experiences,” said an AI platform Jina AI while describing the platform’s memory management.

But, the biggest facet of Auto-GPT appears to be its multi-functionality capabilities, which makes it flexible and robust compared to other language models in the market. The new AI chatbot can directly perform functions such as file manipulation, web browsing, and data retrieval, enabling it to perform a broad range of tasks. 

As per the GitHub page, Auto GPT also comes loaded with other features such as internet connectivity for scraping and gathering information. Leveraging GPT-3.5’s summarisation capabilities, Auto GPT also has the ability to manage long-term and short-term memory, and access to GPT-4 instances for text generation and data-driven decision-making.

Internet Loses It, And So Does ChatGPT

Close on the heels of the launch, developers across the globe deployed Auto GPT for a slew of use cases. Some used it to build a personal stock analyser while others automated product  reviews to buy headphones. A developer even created a podcast from scratch using Auto GPT.

The aftermath also saw users taking to social media to automate processes such as copywriting, building tutorials on edtech platforms and even creating affiliate marketing content. 

Far from just the text generation capabilities of ChatGPT, Auto GPT allows users to generate test cases and automate workflows. The new language model can run through large amounts of data to identify patterns and streamline operations. 

Using data, Auto GPT can also analyse upcoming trends and suggest potential business opportunities and ideas. This can help users cut time and effort in going through verbose and enormous amounts of data at the click of a button. 

On the coding front, it can effectively debug codes, fix issues and pinpoint exact issues with codes that sometimes run into thousands of lines. On the business front, it can help executives optimise their operations via effective cost-cutting measures and subsequently compete better with peers.

It could also have uses in cybersecurity, drug discovery, human resources, translation and what not.

The new tool requires some fundamental language of coding and requires Python 3.8 or higher libraries. As per the developer of Auto-GPT, the project is still not fully polished and could still be limited in certain real-world scenarios. 

Amidst all the hullabaloo, Auto GPT continues to be plagued by certain basic issues which have been flagged by developers online. While many have pointed out that the project was still subject to high processing costs, which could render it outside the reach of a common person. 

Then there is the issue of Auto-GPT frequently getting stuck in a loop while solving a given problem. 

“I don’t think Auto-GPT will solve any problem. Just like I give gpt4 a coding problem and the output code has (a) bug. I feed it back the runtime error, and it will just give the wrong code again and again. Repeating such (a) loop over 20 times 30 times doesn’t solve it at all,” a developer tweeted.

When all is said and done, the new platform truly has marked a big development in the world of AI. The new platform has shown sparks of AGI, which, as per Microsoft researchers, was not supposed to happen for at least the next 20 years. 

In just over a week, Auto GPT has led to the discovery of a new world of possibilities and with the ongoing spate of developments in the space, it remains to be seen what barrier the technology breaks next. For now, the world waits with bated breath for what comes next.

Meanwhile, the predecessor of GPT-4, ChatGPT, had an interesting take on Auto-GPT that shares its DNA with GPT-4.

When asked ‘what is Auto-GPT’ and ‘how does it work’, the chatbot denied recognising the existence of its own kind. 

“Auto-GPT is not a widely used or recognized term or technology. It is possible that it could refer to a specific implementation or use case of GPT (Generative Pre-trained Transformer) language models… However, without further context or information, it is difficult to provide a more specific answer,” the chatbot said.

Note: We at Inc42 take our ethics very seriously. More information about it can be found here.