AI developers are pushing the boundaries with autonomous PT-powered systems.

In a recent article for Motherboard, author Chloe Xiang examines the emergence of autonomous AI systems fueled by OpenAI’s large language model (LLM) GPT. Various developers are working on these systems, which are intended to carry out various tasks without human intervention. This includes executing tasks in sequence, composing, debugging, and developing code, as well as self-assessing and correcting errors.

Auto-GPT, an experimental open-source application created by Toran Bruce Richards, Founder and Lead Developer at Significant Gravitas Ltd., is one such system. Here is how Richards describes Auto-GPT:

Auto-GPT is an experimental open-source application showcasing the capabilities of the GPT-4 language model. This program, driven by GPT-4, autonomously develops and manages businesses to increase net worth. As one of the first examples of GPT-4 running fully autonomously, Auto-GPT pushes the boundaries of what is possible with AI.

Richards explained to Motherboard that he developed Auto-GPT to apply GPT-4’s reasoning to broader, more intricate issues necessitating long-term planning and multiple steps. In a video demo, Auto-GPT can be seen digesting news articles to gain knowledge about a subject to establish a viable business.

Another initiative highlighted in the Motherboard article is a task-driven autonomous agent developed by Yohei Nakajima, a venture capital partner at Untapped Capital. This agent employs GPT-4, a vector database named Pinecone, and an LLM-powered app development framework called LangChain.

In a blog post, Nakajima stated that the system could complete tasks, create new tasks based on completed results, and prioritize real-time tasks. He told Motherboard that his app successfully conducted web research based on input, wrote a paragraph using the search results, and generated a Google Doc containing that paragraph.

While these systems showcase the remarkable potential of AI-powered language models to perform tasks autonomously within diverse constraints and contexts, the Motherboard article also highlights potential issues.

As these agents acquire more capabilities, such as database access and human communication, Nakajima emphasized the growing importance of continuous human supervision. Richards concurs, asserting that this oversight will help guarantee that these autonomous agents “operate within ethical and legal boundaries while respecting privacy and security concerns.”

The article also noted that the pursuit of autonomy in AI research seeks to enable models to simulate chains of thought, reasoning, and self-critique to accomplish a series of tasks and subtasks. However, it appears that LLMs tend to “hallucinate” as they progress further down a list of subtasks.

Finally, it is worth mentioning that Researchers from Northeastern University and MIT recently published a paper exploring using self-reflective LLMs to assist other LLM-driven agents in completing tasks without losing focus.