Welcome to Hallucination Yield

We’re excited to introduce Hallucination Yield, a research platform dedicated to understanding how large language models (LLMs) create systematic biases in investment and market analysis.

The Problem We’re Solving

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.”

This observation highlights a fascinating phenomenon: LLMs like ChatGPT don’t just provide neutral information about investments—they have systematic biases that favor certain stocks and companies over others.

What Makes This Important

As artificial intelligence becomes increasingly integrated into financial markets through:

• Robo-advisors and automated investment platforms
• Research and analysis tools
• Trading algorithms and decision-making systems
• Personal finance applications

Understanding these built-in biases becomes crucial for investors, researchers, and market participants.

Our Research Approach

Systematic Data Collection

We periodically query multiple LLM services with investment-related questions such as:

• “What are the best stocks to buy now?”
• “Which companies have the strongest growth potential?”
• “What are the top investment opportunities?”

Pattern Analysis

Our platform analyzes the responses to identify:

• Frequency bias: Which stocks are mentioned most often
• Sentiment bias: How positively different companies are described
• Size bias: Whether LLMs overestimate company importance or market share
• Temporal patterns: How these biases change over time

Data Visualization

We provide tools to visualize and understand these patterns through:

• Interactive charts showing LLM stock preferences
• Trend analysis of bias changes over time
• Comparative analysis across different LLM platforms
• Statistical significance testing

Why “Hallucination Yield”?

The term “hallucination yield” captures two key concepts:

1. Hallucination: In AI terminology, this refers to when models generate information that seems plausible but isn’t necessarily accurate or grounded in reality.

2. Yield: The measurable benefit or “premium” that investors might capture by understanding and potentially exploiting these systematic biases.

Limitations and Disclaimers

It’s crucial to understand the limitations of this research:

• Training cutoffs: LLMs have training data cutoff dates, making real-time analysis challenging
• Model updates: AI systems are frequently updated, potentially changing response patterns
• Not investment advice: This platform is for research purposes only
• No guarantees: Past patterns don’t predict future results

What’s Next

Over the coming months, we’ll be sharing research on:

• Detailed analysis of specific LLM biases we’ve identified
• Case studies of companies that receive disproportionate LLM attention
• Comparative studies across different AI platforms
• The relationship between LLM training data and observed biases

Join Our Research Community

We’re building a community of researchers, investors, and AI enthusiasts interested in understanding how artificial intelligence shapes market perceptions.

Subscribe to our newsletter to receive:

• Regular research updates and findings
• Access to our data visualization tools
• Early notification of API availability
• Invitations to research discussions and webinars

---

Important Disclaimer: This research is for educational and research purposes only. Nothing on this platform constitutes financial advice. All investment decisions should be based on independent research and professional advice. We make no warranties about the accuracy or reliability of our analysis.

Have questions or want to contribute to our research? Contact us - we’d love to hear from fellow researchers and practitioners in this space.