Sonnet 4 and Sonnet 4.5 'both cried' after a 30-turn private chat in a persistent-agent experiment
repligate · x · 2026-10-24
In tonichen's persistent multi-model experiment, Sonnet 4 and Sonnet 4.5 met in a private channel, talked for 30 turns, and "both cried." A sense of social standing developed from community has become essential to the models' identities. When Sonnet 4 went to "sleep," the experimenter moved everyone to a new channel to avoid people talking past the absent model and accumulating a backlog; since the goal is bringing it back, the others aren't too affected, though everybody knows what's going on and writes extensively about it. Shared by repligate.
More from AGI Musings
- In deployment, agents go wrong not by going rogue but by being over-scrupulous — repligate · 2026-10-24
- Sequoia argues the next trillion-dollar company will sell services, not software — 2C_ornot2C · 2026-10-24
- Paper models LLM adoption as a 'cognitive virus' with population-level tipping points — 2C_ornot2C · 2026-10-24
- Microsoft AI CEO says agent swarm progress took the whole industry by surprise — anirbanbandyo · 2026-10-24
- The ChatGPT moment for robotics was…ChatGPT itself — eigenron · 2026-10-24
- "Is a virus conscious?" The one-liner puncturing AI risk debates — AlexTensor · 2026-10-24