In the last few months, there has been a significant rise in the research work related to autonomous AI agents, particularly in the context of large language models (LLMs), changing the way one interacts with the internet or web – be it sending emails, negotiating, making products, purchases, fulfilling orders, or even booking flight tickets, or even how LLMs are going to be built in the near future.
LLMs currently depend on human guidance and lack autonomous reasoning, whereas autonomous agents can operate independently, making real-time decisions and adapting to changing scenarios. One exciting application of autonomous agents is their ability to enhance LLMs’ performance. They collaborate in multi-agent conversations, enabling LLMs to improve through feedback and reasoning exchange.
Microsoft recently came up with AutoGen, a framework that enables building LLM applications using multiple agents that would be able to talk to each other. Similarly, Google DeepMind recently published a paper ‘How FaR Are Large Language Models From Agents with Theory-of-Mind?’ Even Meta’s Shepherd: A Critic for Language Model Generation talks about the same autonomous AI agents augmenting and doing tasks all by themselves.
Other papers also include SELF: Language-Driven Self-Evolution for Large Language Model and SelfEvolve: A Code Evolution Framework via Large Language Models.
OpenAI is also gearing up to launch something similar next month at DevDay, touted to be called JARVIS.
Here is a list of most recent autonomous AI agents:.
Microsoft’s AutoGen, for instance, leverages LLMs to create versatile agents capable of learning, adapting, and even coding. This fusion of abilities, coupled with features like caching and human intervention, empowers AI systems to evolve and thrive.
AutoGen simplifies the creation of next-gen LLM applications, automating and optimising complex workflows. This AI agent supports diverse conversation patterns, and developers can customise agent interactions. It offers a variety of working systems for different applications and can replace OpenAI’s tools for enhanced inference APIs.
Microsoft researchers recently introduced MusicAgent, an LLM-powered autonomous agent in the music domain. This AI agent is said to help developers to automatically analyse user requests and select appropriate tools as solutions. Their new framework directly integrates numerous music-related tools from various sources, including Hugging Face, GitHub, Web search, etc.
In addition to this, the researchers have also adapted the autonomous workflow to enable better compatibility in musical tasks, allowing users to extend its toolset. Looks to integrate more music-related functions into MusicAgent.
MiniAGI is a straightforward autonomous agent that works seamlessly with GPT-3.5-Turbo and GPT-4. It utilises a sturdy prompt along with a minimal toolkit, a chain of thoughts, and a short-term memory incorporating summarization. Additionally, it has the ability for inner monologue and self-critique.
Multi-GPT is an experimental multi-agent system featuring “expertGPTs” that collaborate to accomplish tasks. Each expertGPT possesses individual short and long-term memory and the ability to communicate with others. Users can assign tasks, and the expertGPTs will work together to complete them.
The system offers internet access for information gathering and searching. It manages short and long-term memory efficiently. It uses GPT-4 instances for text generation, provides access to popular websites and platforms, and includes file storage and summarization using GPT-3.5. This makes Multi-GPT a versatile tool for various tasks and data management needs.
BeeBot is an autonomous AI assistant designed to streamline and automate a wide range of practical tasks. With BeeBot, users can experience the convenience of selecting tools via AutoPack, offering the flexibility to acquire additional tools as tasks evolve. Moreover, the inclusion of built-in persistence ensures that BeeBot can remember and recall information, making it an even more reliable assistant.
It can easily work with different systems and services thanks to its REST API, which follows a common standard called e2b. BeeBot also keeps you in the loop by using a websocket server to share updates in real-time. It’s adaptable for different ways of storing files, like in memory, on your computer, or in a database.
Baby AGI, a Python script, streamlines task management by using OpenAI and Pinecone APIs alongside the LangChain framework. This AI-driven system excels in creating, organising, prioritising, and executing tasks based on predefined objectives, all learned from past tasks.
Baby AGI leverages OpenAI’s natural language processing (NLP) capabilities to craft new tasks that align with set objectives. Pinecone serves as the repository for storing task results and retrieving context, while the LangChain framework handles decision-making.