Machine-Learning

We trained a model on 1.5 million criminal records to predict race using mugshots and names achieving 92.76% accuracy. An accurate linear model trained on biased data learns the true signal, with deviations indicating mislabeling rather than model error. 29% of predicted Hispanics are misclassified as White. Correcting this increases Hispanic criminal record rates by 20-31% and decreases White rates by 4-6%.

We use data from millions of Substack posts and tens of thousands of publications to figure out what actually predicts success – from post frequency and pricing to word counts and category choice

Introducing PedoAI. The first deep learning driven physiognomy model built to distinguish predatory pedophilia from a sample of 100,000 pedophiles and 100,000 non-pedophilic criminals.

The UN’s methodology for forecasting fertility has remained largely unchanged. For many countries these forecasts are consistently inaccurate. Here, we attempt to produce a successor. The results could be described as unexpected.