“The biggest bottleneck preventing current large language models from being truly useful is their lack of reliability.”