>I released the model, the code and I evaluated the model on a huge set of benchmarks and it turns out this horrible, terrible, model is more truthful-yes more truthful-than any other GPT out there