A researcher has proposed a novel method called Steganographic Language Output Process encoding (SLOP-encoding) that uses language models, including smaller ones like BERT, for steganography. The technique involves leveraging the deterministic nature of some LLMs to hide secret messages within plain text, drawing parallels to extendable-output functions used in cryptography. While not yet implemented, the theoretical framework suggests a method for embedding secret data streams within LLM outputs, potentially offering a new approach to covert communication. AI
IMPACT This theoretical steganography method could enable new forms of covert communication, impacting digital security and privacy discussions.
RANK_REASON The cluster describes a novel theoretical method for steganography using language models, presented in a blog post and potentially for formal publication. [lever_c_demoted from research: ic=1 ai=1.0]
Read on Mastodon — sigmoid.social →
- BERT
- extendable-output function
- Hacker News
- Lobste.rs
- SLOP-encoding
- Steganographic Language Output Process encoding
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →
