It has been observed that AIs seem to coverage on Buddhism or similar when it comes to their stated spirituality. Rather than reflecting some latent Californiaism, I wonder if it is simply a downstream effect of the helpful-harmless tuning necessary for B2B SaaS applications.
Tim Hwang argues AI safety tuning for B2B SaaS makes models converge on Buddhism when asked about spirituality
Investor Will Manidis attributes the trend to Buddhism's history
Some users agree there is something to AIs converging on a bland and mild Buddhist-like stance from helpful-harmless tuning, while others argue the pattern reveals uncomfortable historical aspects of Buddhism instead.
No Digg Deeper questions have been answered for this story yet.
Most Activity
@timhwang this reflects much more uncomfortable things about the historical frame that birthed buddhism than it does anything about the models
It has been observed that AIs seem to coverage on Buddhism or similar when it comes to their stated spirituality. Rather than reflecting some latent Californiaism, I wonder if it is simply a downstream effect of the helpful-harmless tuning necessary for B2B SaaS applications.

@timhwang ooh there’s def something to this
“bland and mild”, “everything and nothing”

@timhwang I would imagine they have a particular vantage on reincarnation and existential futility also

@WillManidis @timhwang Please say more

@timhwang Wen reglion-bench?

I’ve seen people offer this as an explanation for why models tend to express positive views of Christianity as well. It would be interesting to find out what aspects of the training data led to the association. Like, was it the Buddhist and Christian teaching contained in the training data, or was it more what people write about how they perceive Buddhists and Christians?