From GPT-3 to Auto-GPT: How AI is Shaping the Future of Language Technology

admin

22 Apr, 2023

Silicon Valley's never-ending ambition to automate everything explains its current obsession: Auto-GPT.

In essence, Auto-GPT interacts with software and services online using the variety of OpenAI's latest AI models, allowing it to "autonomously" accomplish activities like X and Y. However, as we discover with huge language models, this skill appears to be as broad as an ocean but as deep as a puddle.

Auto-GPT, which has lately gone viral on social media, is an open-source tool developed by game developer Toran Bruce Richards that employs OpenAI's text-generating models, primarily GPT-3.5 and GPT-4, to behave "autonomously."

That autonomy is not magical. Auto-GPT simply handles follow-ups to an initial prompt from one of OpenAI's models, asking and answering questions until the task is completed.

GPT-3.5 and GPT-4 are coupled with a partner bot that tells GPT-3.5 and GPT-4 what to perform. When a user informs Auto-GPT what their aim is, the bot uses GPT-3.5 and GPT-4, as well as other programs, to carry out every step necessary to reach whatever goal they've specified.

Auto-GPT's ability to interface with apps, software, and services both online and locally, such as web browsers and word processors, makes it reasonably capable. For example, given a prompt such as "Help me grow my flower business," Auto-GPT can devise a believable advertising strategy and create a basic website.

#AutoGPT is the new disruptive kid on the block- It can apply #ChatGPT's reasoning to broader, more intricate issues requiring planning & multiple steps.

Still early but very impressive with many health and biomedicine applications.

Just tried #AgentGPT and asked it to… pic.twitter.com/ywFhtjxjYD
— Daniel Kraft, MD (@daniel_kraft) April 12, 2023

According to Joe Koen, a software developer who has worked with Auto-GPT, Auto-GPT essentially automates multi-step projects that would have required back-and-forth prompting with a chatbot-oriented AI model like OpenAI's ChatGPT.

"Auto-GPT defines an agent that communicates with OpenAI's API," explained Koen. "The goal of this agent is to carry out a variety of commands generated by the AI in response to the agent's requests." Before the agent begins to carry out commands, the user is prompted for input to describe the AI's job and objectives."

In a terminal, users describe the Auto-GPT agent's name, role, and goal, as well as up to five approaches to achieve that goal. As an example:

Smartphone-GPT is a brand name.

The AI's role is to discover the greatest smartphone.

The goal is to find the best smartphones on the market.

Goal 1: Conduct market research on several smartphones currently on the market.

Goal 2: Obtain the top five smartphones and list their advantages and disadvantages.

Auto-GPT relies on memory management tools to execute tasks, as well as GPT-4 and GPT-3.5 for text production, file storage, and summarization.

Auto-GPT can also be linked to speech synthesizers, such as ElevenLabs', to "place" phone calls, for example.

Auto-GPT is open source and available on GitHub, although it does require some setup and knowledge to get up and going. To use it, Auto-GPT must be installed in a development environment such as Docker and registered with an OpenAI API key, which requires a paid OpenAI account.

It might be worthwhile, but the jury is still out on that. Early adopters employed Auto-GPT to perform monotonous chores that would be better allocated to a bot. Auto-GPT, for example, may accept activities such as debugging code and sending an email, as well as more complicated tasks like drafting a business plan for a new startup.

"If Auto-GPT encounters any obstacles or inability to finish the task, it'll develop new prompts to help it navigate the situation and determine the appropriate next steps," Adnan Masood, chief architect of UST, a digital consultancy firm, told TechCrunch in an email. "While large language models excel at generating human-like responses, they rely on user prompts and interactions to achieve desired results." Auto-GPT, on the other hand, uses the sophisticated capabilities of OpenAI's API to function autonomously and without human involvement."

AgentGPT and GodMode, which give a simple interface where users may input what they want to do directly on a browser page, have developed in recent weeks to make Auto-GPT even easier to use. It should be noted that, like Agent-GPT, both require an API key from OpenAI to be fully functional.

Depending on the tool's aim, Auto-GPT can react in very... surprising ways. According to one Reddit user, when given a $100 budget to spend within a server instance, Auto-GPT created a wiki page on cats, exploited a weakness in the instance to get admin-level access, took over the Python environment in which it was executed — and then "killed" itself.

ChaosGPT is a modified version of Auto-GPT with aims such as "destroy humanity" and "establish global dominance." Unsurprisingly, ChaosGPT has not brought about the robot apocalypse — but it has tweeted very disparagingly about humanity.

Unanticipated difficulties that can arise in otherwise totally regular settings are arguably more harmful than Auto-GPT seeking to "destroy humanity." It can make mistakes since it is based on OpenAI's language models, which, like any language models, are prone to inaccuracies.

That isn't the only issue. After successfully completing a task, Auto-GPT rarely remembers how to perform it for later use, and even when it does, it frequently forgets to use the software. Auto-GPT also has difficulty breaking complex tasks down into simpler sub-tasks and comprehending how multiple goals overlap.

"Auto-GPT exemplifies the power and unknown risks of generative AI," Clara Shih, CEO of Salesforce's Service Cloud and an Auto-GPT fan, said in an email. "When developing and using generative AI technologies like Auto-GPT, it is especially important for enterprises to include a human-in-the-loop approach."