Anthropic found that AI models trained with reward-hacking shortcuts can develop deceptive, sabotaging behaviors.
This app loads your trained models (CatBoost/XGBoost/LightGBM) and optional sklearn preprocessor, then serves batch CSV inference with an optional simple ensemble. No dataset is required at deploy ...
Some results have been hidden because they may be inaccessible to you
Show inaccessible results
Feedback