A question for the #LLM specialists around here. Does this pattern have a name yet?
When a LLM generates a response to a prompt but then erases it partially or entirely after having it displayed for a few seconds. I'm even thinking this is not the LLM censoring itself a posteriori, but another program watching its output.
Anyone knows what terminology is being used to describe this behavior?
According to this article and others I've seen online, they might just call this: "output censorship" or generally "llm censorship". https://arxiv.org/abs/2307.10719
Add comment