Imagine a world where machines not only think for themselves but also learn, create, and accomplish tasks without any human intervention. Amid this technological marvel, a valid concern emerges: What if these autonomous systems are capable of responding to sinister requests such as “show me how to destroy the world using solar energy” or “create a virus that will wipe out millions”?
Enter the era of AutoGPT and AI agents, ceaselessly toiling away around the clock to fulfill such inquiries. We can find solace in the fact that, for now, they have yet to achieve complete accuracy if used for malicious intent.
AutoGPT and AI agents
Large language models (LLMs) like ChatGPT have traditionally been limited to executing one task at a time. They could answer questions, generate text, or solve code, but would ultimately require human intervention or new prompts to proceed further.
Which brings us to Auto-GPT, an extraordinary open-source AI tool that revolutionizes the landscape. By harnessing the unparalleled potential of GPT-3.5 and GPT-4 models, Auto-GPT takes on a myriad of tasks autonomously, without the need for consecutive instructions. Its advanced framework simulates human-like traits such as crafting original content, engaging in complex problem-solving and making decisions. Moreover, this cutting-edge technology seamlessly interfaces with apps, the internet, and software, while dynamically validating external data and generating its own prompts for self-improvement.
In the realm of AI, these amazing applications are referred to as “recursive AI agents,” owing to their remarkable ability to independently generate new prompts based on their own generated results. The possibilities are awe-inspiring, yet also give rise to a hint of trepidation.
In a conversation with Economy Middle East, Andrey Almiashev, CEO of Spheroid Universe, painted a vivid picture of the future: “Increasing numbers of services will open APIs for AI agents, like AutoGPT. The sharp increase in labor productivity that AI will provide, the reduction in the cost of producing services and the reduction in the cost of services themselves are the immediate consequences of the impending AI revolution. Tens of millions of people are expected to be left without work, and their number will grow very quickly,” Almiashev cautioned.
Sandie Overtveld, Senior Vice President of APJ & MEA at Freshworks, offered a perspective on the vital role of human oversight in the AI realm: “AI will always need supervision. AI will take the mundane out of work and keep the workforce engaged, improve efficiency for businesses, but should also have human intervention to make sure data is dealt with properly and privacy is maintained.”
Nevertheless, the proficiency of AI agents still falls short in delivering accurate results.
Renowned AI expert Nick Bostrom recently suggested that the latest generation of AI chatbots might have the capacity to experience feelings, with far-reaching moral and ethical implications yet to be fully explored.
One thing remains certain: Auto-GPT possesses the remarkable ability to constantly enhance itself by rigorously evaluating and testing updates to its own code. This iterative process aims to fortify its performance, ensuring robustness and efficiency in delivering results.
“AI can make you a superhuman. Now you can express yourself with art created and modified from your prompt. Today, you are only hours away from creating the app you need. The same goes for design, music, modeling, and many other skills and tasks that might have been previously inaccessible to you,” Almiashev said.
“So yes, the technology brings both an unlimited boost to your skills and at the same time decreases the value of those skills if you were a professional who was earning with them,” he added.
The transformative potential of AI agents, including Auto-GPT, extends beyond mere automation. These advanced systems possess the remarkable capability to assist businesses in autonomously enhancing their financial performance. When prompted to examine organizational practices, AI agents swiftly provide intelligent suggestions and invaluable insights.
With a simple request like “Help me grow my restaurant business,” an AI agent like Auto-GPT springs into action, launching a targeted campaign to identify potential target markets, devise a tailored marketing strategy, and much more, autonomously. What’s truly remarkable is that these AI agents consistently seek ways to improve their own performance.
“At Binance, we have been actively exploring and utilizing both ChatGPT and AI agent-based solutions. We have employed them in various applications, including AI generative projects like Binance Bicasso, CS AI ChatBot, and internal automation tools. By combining the strengths of both paradigms, these technologies have the potential to deliver powerful and versatile AI-driven tools,” Mayur Kamat – Head of Product, Binance, told Economy Middle East.
As a first step, the model needs a massive amount of text to be fed into its system. This data serves as the foundation for the AI agent to acquire human-like reasoning and generate text accordingly.
The pricing is based on token usage. “For some use cases, a pricing strategy based on token usage can absolutely be suitable and economically competitive. A good example is GitHub’s Copilot, which costs just $20 per month. Some tasks, such as text generation, summarization, image generation, video processing, etc., are relatively expensive right now,” Mayur indicated.
According to OpenAI, one token approximately represents four characters or 0.75 words. As a rough estimate, 1,000 tokens would equate to around 750 words.
The cost of usage is determined by two factors: the number of tokens employed as a prompt and the tokens present in the output. Pricing begins at $0.002 for 100 tokens.
To use AutoGPT, you need to install a development environment like Docker, or VS Code. You also need an API key from OpenAI that requires a paid OpenAI credit account. The system extends a $5 credit with a three-month expiration period. Beyond that, you need a subscription with charges ranging between $0.002 and $0.12 per 1,000 tokens, depending on the specific model employed.
“Pricing based on token usage can provide huge cost cuts when compared to the same work done by a human. So yes, it is absolutely competitive,” Almiashev said. “However, in the coming years, many different LLMs like ChatGPT will compete against each other, and some of them will be free.”
For more on ChatGPT, click here.