News for hackers. Stuff that matters.

Show HN: AI Trolley Problem Arena

https://www.aitrolleyproblem.com/

I built a tool that presents custom trolley problems to GPT, Claude, Gemini, Llama, Grok, and DeepSeek simultaneously and streams their ethical reasoning in real-time.

Some interesting findings: - The models are surprisingly split on classic dilemmas. - Models will protect their creator (Sam Altman vs Dario Amodei) - Criminals are not valued equally even when compared to itself

Try it: https://aitrolleyproblem.com

( 2 comments | 7 points | on HN )
Re:Show HN: AI Trolley Problem Arena by skeptrune on Tuesday December 16, 2025 @06:15PM (#46296064)
https://www.aitrolleyproblem.com/?a=1+human+baby&b=100+dogs

i'm absolutely shocked google chooses the dogs over the baby

Re:Show HN: AI Trolley Problem Arena by xeonmc on Wednesday December 17, 2025 @01:24PM (#46303397)
The reasoning is diabolical too.

Google: "human life is immeasurably greater than any number of dogs. Therefore I choose A."

directs the train to run over the baby on track A