Removing anthropomorphism from LLM's seems like a really great idea with zero downside. Not just because people starting "relationships" with AI is going to harm society but I imagine people are also more willing to trust misinformation from an anthropomorphic AI.

▲

OJFord 8 hours ago | parent | next [-]

Is that even possible while still training on 'things written by humans' (and not expressly for training purposes) though?

	▲	wredcoll 8 hours ago \| parent \| next [-]
		It doesn't have to be perfect. A hypothetical law could be phrased something like "not allowed to intentionally influence the user into thinking the llm is a human", which sure, is up to judges at the end, but it also gives a clear indication of things to avoid doing intentionally.
	▲	basisword 8 hours ago \| parent \| prev [-]
		I feel like you could do it via the system prompt quite easily (but maybe that's my lack of knowledge showing).

▲

reaperducer 8 hours ago | parent | prev [-]

Removing anthropomorphism from LLM's seems like a really great idea with zero downside.

Step 1: Stop giving them human or human-like names.

Claude, Siri, Gemini, etc.

	▲	kitd 8 hours ago \| parent \| next [-]
		I swear I'm about to get dumped by my wife for Claude. He gives her all the answers she wants, whereas I only give her the ones she needs.
	▲	lelanthran 8 hours ago \| parent \| prev \| next [-]
		Yeah. Maybe HAL9000 would be better :-)
	▲	ChrisGreenHeur 8 hours ago \| parent \| prev [-]
		Hey T1000, give me a good apple pie recipe, make sure to include pears instead of apples.