We see that prominent transformer models in use (think LLMs and multimodal implementations more generally) struggle with hallucinations making them hard to use in mission-critical/operational environments.

Overconfidence and a general inability to quantify uncertainty adequately, seem to contribute to this. I thought it wise to examine the existing techniques that tackle uncertainty and learn and play with them in a variety of contexts.

This repo entails my experiments with ANNs and uncertainty as well as potential novel use cases that might arise from this play.