I’m not sure whether I agree with you or disagree with you here. What you say seems to just not address the core reasons why the shoggoth metaphor is allegedly a good one.
The core reasons IMO are: --Shoggoths are weird difficult-to-understand and plausibly-dangerous aliens. LLMs don’t come from outer space but other than that they fit the bill. They certainly aren’t human or close cousins to humans (like, say, neanderthals or chimpanzees would be) --RLHF doesn’t change the above, but it appears to. It trains the LLM to put on the appearance of a helpful assistant with similar personality to a human remote worker. It’s like putting a friendly face mask on the shoggoth.
I’m not sure whether I agree with you or disagree with you here. What you say seems to just not address the core reasons why the shoggoth metaphor is allegedly a good one.
The core reasons IMO are:
--Shoggoths are weird difficult-to-understand and plausibly-dangerous aliens. LLMs don’t come from outer space but other than that they fit the bill. They certainly aren’t human or close cousins to humans (like, say, neanderthals or chimpanzees would be)
--RLHF doesn’t change the above, but it appears to. It trains the LLM to put on the appearance of a helpful assistant with similar personality to a human remote worker. It’s like putting a friendly face mask on the shoggoth.