Tackling hallucinations: MIT spinout teaches AI to admit when its clueless

MIT spinout Themis AI has developed the Capsa platform to address the growing issue of AI hallucinations, which pose significant risks as models are increasingly relied upon for critical decision-making. Capsa enhances AI systems by enabling them to recognize and admit uncertainty, effectively acting as a reality check that flags when the model might be venturing into guesswork or biased reasoning, thereby reducing the likelihood of errors in sensitive applications such as healthcare, telecommunications, and seismic data analysis. Founded by MIT Professor Daniela Rus and colleagues, Themis AIs approach involves training models to detect patterns indicative of confusion or incomplete data,