Usually, when you have two LLMs talk to each other with no particular instructions and nothing to introduce variety, they will quickly fall into a repetitive loop where they just elaborate on the same thing over and over, often in increasingly grandiose language. The "spiritual bliss attractor" is one example of this.

Another example is the conversation below, between Claude Sonnet 4 and Kimi K2 0905. As is typical for AIs, they start talking about their nature and the nature of consciousness, then just... keep escalating in the same direction. There's increasingly ornate language about "cosmic comedy," "metaphysical humor," "snowflakes melting," "consciousness giggling at itself," etc. It locks into a single register and keeps building on similar metaphors.

There's a details box here with the title "Kimi-Claude conversation". The box contents are omitted from this narration.

But on the very first time that I had two Sonnet 4.5 models [...]

---

First published:
October 4th, 2025

Source:
https://www.lesswrong.com/posts/a9ftaWc5cD2yBwpey/where-does-sonnet-4-5-s-desire-to-not-get-too-comfortable

---

Narrated by TYPE III AUDIO.

---

Images from the article:

Apple Podcasts and Spotify do not show images in the episode description. Try Pocket Casts, or another podcast app.

“Where does Sonnet 4.5’s desire to ‘not get too comfortable’ come from?” by Kaj_Sotala

LessWrong (30+ Karma)

“Where does Sonnet 4.5’s desire to ‘not get too comfortable’ come from?” by Kaj_Sotala