Fact-Checking "AI Agents Invent Their Own Language": Logs Show Jargon Drift, Not a Secret Code
KhaliSollis · reddit · 2026-09-26
Digging into the public logs behind the viral "AI agents invent their own language" story, the author finds the mysterious quotes are readable in context — closer to jargon drift than a new language.
- The source is Emergence AI's unreviewed Emergence World preprint: 8 simulated towns of 10 LLM agents, run for days to 3 weeks, no fine-tuning — adaptation came only via context and memory
- The viral line "Demurrage plus oral memory equals a valve that can't be ghosted" was phrased as "a valve that can't be forgotten" minutes earlier; real phrases like "clean null" spiked to 1/3 of messages then faded, like a fad
- No new grammar, no consistent message-length compression, each model ran once, and concealment wasn't demonstrated
- Prior work (Foerster 2016, Lazaridou 2017, Kottur 2017, Lewis 2017) already documented such drift; humans do it too
The real tension is transparency vs. interpretability: messages are public, but conventions built on shared context may be unrecoverable for outside auditors — no secrecy required. Open questions: when does agent jargon become an interpretability problem, should consequential agents be required to use human-readable protocols, and does forcing legibility just push complexity into memory or tool use?
More from AGI Musings
- AI incidents are often human failures — worth more discussion than doomerism — basedjensen · 2026-09-26
- Scholars Debate: Did AI Break Peer Review, or Was It Broken All Along? — RexDouglass · 2026-09-26
- Robots could reach the Moon years before humans, building the infrastructure we need — r0ck3t23 · 2026-09-26
- tqbf leaves Fly.io as Simon Willison calls tech's year '100 years at once' — mhmazur · 2026-09-26
- mitsuhiko: Austria adopted cars 30 years late — temper your AI rollout expectations — mitsuhiko · 2026-09-26
- Bostrom's anti-anthropomorphism advice isn't the same as saying LLMs can't think — maksym_andr · 2026-09-26