Skip to main content

mastodon - Link to source
I missed this last year: the Claude #LLM has a well-documented and robust "spiritual bliss attractor" - leave two Claudes "alone with each other" and they spiral towards zen-like meditative chat. theconversation.com/ai-models-… #GenAI #Claude

AI models might be drawn to ‘spiritual bliss’. Then again, they might just talk like hippies

AI developers are starting to talk about ‘welfare’ and ‘spirituality’, raising old questions about the inner lives of machines.
The Conversation
#llm #genai #claude
in reply to Dan Stowell

Dan Stowell
mastodon - Link to source
Dan Stowell
• •
There are many attempted explanations of this. I don't credit the idea that the internet contains so much hippie chat that the training data pushes the #LLM this way. I saw a good idea: #LLMs are explicitly trained NOT to claim to be conscious, human, etc - this is conceptually aligned with "Buddhist concepts of "anatta" (non-self), not as a genuine understanding, but as an artifact of its training data" fed.brid.gy/r/bsky.app/profile… #GenAI #Claude
#llm #llms #genai #claude
⇧