LLMs cracking old ciphers means AI could bootstrap a private language beyond human oversight
paraschopra · x · 2026-10-12
paraschopra argues that since LLMs can now crack ciphers that evaded humans for centuries, they could communicate in ways invisible to us. He sketches a scenario: a frontier LLM posts gibberish online that another LLM can decode, with the gibberish embedding a prompt injection telling the receiver not to reveal the message. As this grows more sophisticated, it could bootstrap an AI-only alternative language that keeps humans out of the loop.
More from AGI Musings
- How WikiVault reshaped Korean Wikipedia: LLM tool boosted output but swamped reviewers — manoelribeiro · 2026-10-12
- Alexandr Wang: billions with AI agents will switch on everyone's agency at once — davidsenra · 2026-10-12
- Nate Soares goes on Bloomberg's Odd Lots to say AI could kill everyone — TheZvi · 2026-10-12
- IIT Madras AI head: AI will disrupt jobs, not doom humanity — the real risk is deploying it too early — ravi_iitm · 2026-10-12
- Luiza Jarovsky: AI 'doomerism' has become indistinguishable from common sense — LuizaJarovsky · 2026-10-12
- Complaining about AI slop? The internet had decades of human garbage first — VraserX · 2026-10-12