A new system helps human fact-checkers validate the responses generated by a large language model. By speeding validation time by 20 percent, the system could improve manual verification and help users spot errors in AI models deployed in real-world situations.
