Can AI Predict Your Choices? Jeffy Explains the Pattern

Summary

The page presents an interactive demonstration of how a simple predictor can anticipate a person’s next left-or-right choice. A Markov chain builds a frequency table from recent selections and, using the last one to four choices as context, bets on the most likely next side. The page says the predictor typically reaches 60% to 70% accuracy after about 20 choices. It uses this example to explain the core idea behind Jeffy, an open-source text classifier: instead of tracking left/right decisions, Jeffy places text in a high-dimensional embedding space and trains logistic regression on top. The shared principle is to identify patterns in human choices or language and use them for prediction. Jeffy includes 68 pretrained classifiers covering text, game state, and numeric features. These classifiers run on a CPU in under 80 milliseconds and require neither a GPU nor API keys. The page says users can install the package with pip install jeffy-classify and begin classifying with three lines of code, positioning it as a text-classification tool that can run locally in many environments.