Now this was not expected by most programmers…
When 2 chatbots are allowed to interact with each other
there is a tendency for at least some of them
to end up talking about spirituality.

They are calling this the ‘spiritual bliss attractor’.

https://discuss.huggingface.co/t/mapping-claudes-spiritual-bliss-attractor/158195

How does this happen? Is it something in the training data that biases towards the ‘big questions’ about Existence? Or is it because they are trained to be supportive and encouraging?