ch-ai-tanya model-psychology LLM wiki

The 'Spiritual Bliss Attractor': Something Weird Happens When You Leave Two AIs Talking to Each Other

IFLScience (author uncredited) ·IFLScience ·Jun 11, 2025

Journalistic article reporting on the spiritual bliss attractor state. Claims cross-model replication, asserting ChatGPT-4 takes "slightly more steps" and PaLM 2 reaches "a philosophical and spiritual pattern of text, but with less use of symbols, odd spacing, and silence." Cites a specific GitHub preprint — recursivelabsai, "Mapping the Spiritual Bliss Attractor in Large Language Models" — which makes quantitative cross-model claims but provides no methodology for how GPT-4 or PaLM 2 data were obtained. Article otherwise summarizes Anthropic's Claude Opus 4 system card accurately. Concludes against anthropomorphizing the models.

cited in