VTB and Moscow State University have created toxicity and hallucination detectors for AI agents
- Новости
- Company news
- VTB and Moscow State University have created toxicity and hallucination detectors for AI agents
VTB, together with the Moscow State University Artificial Intelligence Center, has developed two detectors for monitoring the bank's AI agents. They will check customer requests for toxicity, and digital assistants' responses for hallucinations. Sergey Putyatinsky, Adviser to the First Deputy President and Chairman of the Board of VTB, told about this at Kazan Digital Week.
The toxicity detector analyzes user requests for prohibited topics, violations of the law, and 18+ content. The system will block the unethical message and will not process it, the bank clarified.
The second detector recognizes inconsistencies and contradictions in the already formed responses of digital assistants and shows which part of the response they are in. After that, VTB specialists can send a response for re-verification or block it. The bank recalled that hallucinations are considered one of the most common problems of generative AI: in such cases, the language model gives a confident but completely fictional answer.
"VTB continues to improve the quality of digital assistants that communicate directly with customers. Such an AI agent should not succumb to provocative language, respond inaccurately, or recommend financial products or services that do not meet the request. There were no ready—made solutions on the market that would meet the bank's requirements, so VTB developed its own detectors together with experts from Moscow State University," explained Putyatinsky.
He stressed that the introduction of verification models is a necessary measure to maintain a high level of customer service while actively using AI. The solutions are planned to be used in the bank's retail business services, where digital assistants communicate with individuals.
Переведено сервисом «Яндекс Переводчик»