Lossless Compression of LLMl-Generated Text via Next-Token Prediction