They also could have been much more interesting in the approach. LLMs can use their token distributions to generate stegotext that read like plausible prose but decode to payloads.¹
They also could have been much more interesting in the approach. LLMs can use their token distributions to generate stegotext that read like plausible prose but decode to payloads.¹
Sure, but the point here is to add a fingerprint from the client.