Patronus AI, a San Francisco startup, has launched a self-serve API that detects and prevents AI failures, such as hallucinations and unsafe responses, in real-time. According to CEO Anand Kannappan in an interview with VentureBeat, the platform introduces several innovations, including “judge evaluators” that allow companies to create custom rules in plain English and Lynx, a hallucination detection model that outperforms GPT-4 in detecting medical inaccuracies.
The company has also developed specialized tools like CopyrightCatcher and FinanceBench to provide comprehensive coverage against AI failures. Patronus AI’s pay-as-you-go pricing model aims to increase access to AI safety tools for startups and smaller businesses. Early adopters include HP, AngelList, and Pearson, along with partnerships with tech giants like Nvidia, MongoDB, and IBM.