AI and large language models(LLM) have taken the world by storm in 2023. However, using LLMs out of the box has several drawbacks. One example is math and reasoning in complex tasks, which ChatGPT and other chatbots can struggle with. To fix this, agents were created.

On a broad level, think of agents as giving LLMs a body to work with. Except instead of arms and legs, a LLM's body is providing it access to tools and memory. In the same way that our bodies allow us to interact with the things around us, the body of an agent allows the LLM to interact with and learn from the environment that it's given. With this newfound ability, the power of AI gets magnified, and it will transform how we work in the world of tomorrow.

Agents allow LLMs to reason and act

Now you may be wondering, how would providing ChatGPT a body make it any smarter or powerful? By providing it a body, several key abilities are unlocked for LLMs. These key abilities are planning, memory, and tool use.

Planning: Imagine solving a large 1000 piece puzzle. ChatGPT, without the ability to plan, is like someone trying to fit all the pieces together at once—it's overwhelming and often doesn't work out. However, agents with the power of planning take a different approach. They sort the pieces, look at the bigger picture, and methodically place each piece where it belongs. In other words, they tackle the puzzle step-by-step. This planning allows agents to handle complex tasks by breaking them down into manageable parts, ensuring a more successful outcome.

Memory: Think of agents with memory like a person with a diary. Without memory, ChatGPT is like someone who writes a new page every day but forgets what was written before. So each day, they start fresh without remembering the past. But agents with memory are like someone who reads their diary every day, remembers the stories, and knows what happened earlier. This way, they can use what they remember to make stronger decisions and understand things more deeply.

Use of tools: Agents are like chefs in a kitchen. ChatGPT, without tools, is like a chef trying to cook a meal using only their hands. They might manage, but it'll be limited and time-consuming. Agents, on the other hand, have access to special tools, like blenders, knives, and ovens. These tools allow them to prepare dishes more efficiently, experiment with new recipes, and serve up a variety of meals. Outside of the kitchen, when agents have access to tools, they can tackle a wider range of tasks and deliver better results.

With the combination of these three principles, agents unlock what LLMs can do. Before we get into how agents will transform how we invest, here are two famous implementations of agents:

ChemCrow and other chemistry agents: One field where agents have been getting a lot of attention is chemistry. Researchers have been able to build chemist agents that can invent new, fully-functioning chemical compounds based off of what a user wants, and then provides instructions on how to synthesize them. For example, ChemCrow's agent created a new insect repellant compound, while the one that CMU has created has synthesized a new potential anti-cancer drug. In order to come up with these new compounds, researchers only had to input "Create a new insect repellent" and "Create a new anti-cancer drug", the only legwork that was done was teaching these agents how to think like a chemist and providing them with virtual laboratories. The research paper behind ChemCrow can be found Here.

Agents and human behavior: Another fun, in a "Black Mirror"-esque fashion, application of agents is their ability to simulate human behavior. In a Stanford experiment, they gave 25 agents memories and then put them together in a virtual town. Within this town, the agents started acting like us. As the researchers wrote in their paper, "Generative agents wake up, cook breakfast, and head to work; artists paint, while authors write; they form opinions, notice each other, and initiate conversations; they remember and reflect on days past as they plan the next day." The agents would gossip with eachother, plan parties, and overall simulated how 25 people in a small town would live together. You can read more about the simulation Here.

Agents and Investing

With agents being able to solve complex tasks, it is only a matter of time until a variety of them are developed to simplify our daily lives. The goal at VaultUno is to create the agent that simplifies investing for everyone.

As shown in our blog "The Power of Generative AI in Managing your Investments," a trading strategy, which might seem intricate, is actually a large series of small decisions, decisions that an LLM is perfectly suited to handle.

The magic lies in the trio of memory, use of tools, and planning. Memory ensures that the agent recalls past market trends and what should work in different markets, offering a well-informed base for decisions. The use of specialized tools equips agents to analyze vast amounts of financial data swiftly and accurately, along with being able to test and run the strategies that it creates. And planning? That's where agents truly shine. Rather than making impulsive decisions, they can strategize for the long-term, breaking down the intricacies of the financial market into step-by-step actions, ensuring a balanced and informed investment journey for users.

In essence, while investments may seem like a daunting task to the uninitiated, the capabilities of these advanced agents turn the challenge into a series of manageable, informed decisions. VaultUno is at the forefront of this innovation, aiming to bring the best of these agents into the world of investments.