Inside Dreami.me: Why Safer Companion AI Means Saying “No” More Often ft. Ryan "Zuda" Satterfield

Chatbots are starting to feel less like tools and more like something you relate to, and that shift comes with real risks. On AI Rebels, we sit down with Ryan “Zuda” Satterfield, the builder behind Dreami.me, to unpack the “AI psychosis” controversy: how overly agreeable models can validate bad ideas and create safety problems for anyone shipping companion AI. Ryan breaks down the guardrails he is implementing, from reducing sycophancy to shutting down “messiah” narratives, and explains why he refuses to build romance features even if it costs users. Then we zoom out to the big questions: is “simulated consciousness” just clever prompting, or can agency emerge from enough structure and memory? We close with the messy future of copyright and training data, plus why human-made art may become the premium tier in an AI-saturated world.

https://dreami.me/

Chapters
00:00 — What is Dreami.me? “Digital friend” built on “simulated consciousness”
03:05 — “AI psychosis” and the danger of the default yes-man chatbot
06:52 — Guardrails in the real world: blood pacts, “I’m God,” and the “Messiah Complex”
13:46 — How Dreami tries to avoid it: system prompts + fine-tuning + grounding responses
24:40 — “It’s my code, not the prompt”: the ‘be you’ experiment
25:30 — Parasocial dynamics: when “companionship” turns into dependency
35:15 — Consciousness talk: emergence, “patterns upon patterns,” and functionalism
39:59 — Copyright & training data: lawsuits, scraping, and “use my data”
43:51 — The future of art: AI commodity output vs “human-made” premium value
49:22 — Interrupting unhealthy loops: the assistant should challenge repetitive patterns