As the CEO of GenAI Services, a company at the forefront of innovative AI solutions, I am always eager to explore and test the latest advancements in AI technology. Recently, we had the opportunity to experiment with AutoGPT, an intriguing AI agent that is an open-source project that can work with any large language model (LLM), including but not limited to, ChatGPT 3.5.

This blog post is dedicated to sharing our experience of leveraging AutoGPT to estimate the impact of crypto news on Bitcoin and Ethereum trading.

AutoGPT is designed to accept a goal in natural language and attempt to achieve it by breaking it down into smaller tasks. It can autonomously perform a series of tasks without constant human interaction. With this concept in mind, we set out on an ambitious project: to analyze the effects of crypto news on Bitcoin and Ethereum volatility.

We set the following goals for AutoGPT:

  1. Find the ten leading crypto-related and general economic news sources on the internet and create a document with the website names and links.
  2. Retrieve up to 100 of the most relevant crypto trading news articles from the last three days from these sources and store each article in a separate file within a designated ‘news’ folder.
  3. Summarize these articles and compile the summaries into a single document.
  4. For each summary, estimate its effect on the volatility of Bitcoin and Ethereum, assigning a score from 0 to 10.
  5. Calculate the average volatility based on the assigned scores.

The AutoGPT agent was quick to accomplish the first task, successfully identifying the top ten crypto news sites and generating a list with links, saved in a file named ‘sites.txt’. However, the following steps proved to be more challenging for the AI agent.

When it came to the task of creating a Python program for web crawling, AutoGPT found itself in a recursive loop. It set a new goal for itself, executed the goal, evaluated the result, and then set another goal. This cycle continued without the agent being able to successfully create the Python program we needed.

From our experience, it’s evident that while the concept of AutoGPT is fascinating and holds significant potential, it’s still a distance away from being a practical application for complex real-world tasks. AutoGPT is great at handling single-step tasks and can even manage simpler multi-step tasks. However, when it comes to more complex tasks requiring the synthesis of various skills, such as web scraping and text analysis, AutoGPT still has some way to go.

Despite the hurdles, this experience has further strengthened our resolve to stay at the cutting edge of AI technology and continue to push the boundaries of what AI can achieve. Stay tuned for more updates on our AI exploration journey!