British company Riverlane, in collaboration with Nvidia, has tested an AI system designed to optimize decoders responsible for error detection and correction in quantum computing. Utilizing the H100 accelerator, the training cycle time was dramatically reduced from 180 seconds to just 2.8 seconds compared to a single CPU core. Furthermore, the error syndrome processing speed reached an impressive 4.2 million per second. In relation to a 32-core processor, the H100 provided a threefold increase in efficiency. This advancement could enable more frequent reconfiguration of decoders in response to changes in qubit operations.