<ul><li>Artificial intelligence systems like ChatGPT can provide answers, but they may not reveal gaps in knowledge or uncertainties.</li><li>MIT spinout Themis AI is helping quantify model uncertainty to correct unreliable outputs quickly.</li><li>Capsa platform works with any machine-learning model to detect ambiguity, incompleteness, or bias.</li><li>Themis AI's co-founder Rus emphasizes improving models and offering guarantees of correct functioning.</li><li>The company has worked with telecom, oil, and gas companies, aiming to make AI systems transparent and reliable.</li><li>Rus' lab researches model uncertainty, with past work detecting bias in facial recognition systems and aiding drug discovery predictions.</li><li>Capsa helps large language models self-report confidence and uncertainty for reliable outputs.</li><li>Themis AI is exploring Capsa's potential for improving accuracy in chain-of-thought reasoning in AI.</li><li>The company aims to make AI more impactful and trustworthy, addressing concerns surrounding AI developments.</li><li>Rus sees Themis AI as an opportunity to apply research for societal good and build trust between people and AI technologies.</li></ul>

Teaching AI models what they don’t know

Discover more