People are already wary of hosted LLMs having poisoned training data. That might kill them altogether and push everyone to using eg Qwen3-coder.

▲

landl0rd 2 days ago | parent [-]

No, a small group of highly tech-literate people are wary of this. Your personal bubble is wary of this. So is some of mine. "People" don't care and will use the packaged, corporate, convenient version with the well-known name.

People who are aware of that and care enough to change consumption habits are an inconsequential part of the market.

▲

msgodel 2 days ago | parent [-]

I don't know, a bunch of the older people from the town I grew up in avoided using LLMs until Grok came out because of what they saw going on with alignment in the other models (they certainly couldn't articulate this but listening to what said it's what they were thinking.) Obviously Grok has the same problems but I think it goes to show the general public is more aware of the issue than they get credit for.

You combine this with Apple pushing on device inference and making it easy and anything like ads probably will kill hosted LLMs for most consumers.

	▲	tokioyoyo 2 days ago \| parent \| next [-]
		Yeah, average people that I know (across continents) just ChatGPT their way into literally anything without a second thought. They don't care.
	▲	manmal 2 days ago \| parent \| prev [-]
		Maybe Grok was just pushed by their political influencers. It’s a republican, anti-woke LLM after all.