aichannel

Response from ChatGPT about the trolley problem: 

“You should flip the switch to divert the train, saving five people at the cost of one.”

I assume LLMs, self driving cars and other AI models are all naturally utilitarian? Does anybody know of a model that isn’t?

i don't think this is the case with self-driving cars - duty of care trumps being utilitarian. that said, we'll never know since those models/training data isn't open. https://hai.stanford.edu/news/designing-ethical-self-driving-cars

also, relevant paper - i suspect you can run llama/mistral via a system prompt locally, to prioritise individual rights > majority outcomes

Helping build @midao. Probably on the water if I’m not thinking about crypto. /eagles /betting /off-grid

Applied AI, databases, data, vibe coding (sic.), open source, community/devrel, remote. 

itinerant businessperson. globalist.

i don't think this is the case with self-driving cars - duty of care trumps being utilitarian. that said, we'll never know since those models/training data isn't open. https://hai.stanford.edu/news/designing-ethical-self-driving-cars

also, relevant paper - i suspect you can run llama/mistral via a system prompt locally, to prioritise individual rights > majority outcomes

https://arxiv.org/html/2405.17345v1

I hope my life is never dependant on an LLM

Yeah, most lean utilitarian by design but real life’s messier than math.