About Hallucination Yield

The concept of "hallucination yield" emerged from a fascinating observation: ChatGPT and other large language models show consistent biases when discussing investments and stocks. As noted by @goodalexander:

"I'm a recently converted fan of 'hallucination yield' - the idea that ChatGPT arbitrarily likes certain stocks or thinks they're bigger than they are. Hard to backtest because of training cutoffs but... yeah, it works."

Large language models are trained on vast amounts of internet data, which creates systematic biases in how they perceive and recommend investments. These biases aren't random - they reflect patterns in the training data that can be measured and analyzed.

Our platform systematically queries multiple AI services (specifically large language models or LLMs) to:

• Identify stocks that AI models consistently favor or overweight
• Measure the "premium" AI models assign to specific companies
• Track how these biases change over time
• Provide data-driven insights into AI market sentiment

As AI becomes increasingly influential in financial markets - through robo-advisors, research tools, and decision-making systems - understanding these built-in biases becomes crucial. The "hallucination yield" represents a new form of market premium that traditional analysis might miss.

Our research is informed by several key philosophical concepts that help explain how AI can influence markets:

We take a rigorous, data-driven approach to understanding AI market biases:

Hallucination Yield is fundamentally a free and open research project. Our mission is to gather a comprehensive dataset on AI investment opinions and make it accessible to researchers, journalists, and the public to foster a deeper understanding of this emerging phenomenon.

To ensure the project's sustainability, we may introduce a premium tier in the future aimed at professional analysts and traders. Potential premium features include:

• Powerful screeners to filter assets by nuanced AI sentiment metrics.
• Complete historical data access for robust backtesting of trading strategies.
• Real-time API access for systematic trading and analysis.

Our immediate focus remains on building a world-class, publicly available dataset. This will serve as the foundation for future tools and research into AI's reflexive impact on markets.