@cmmdmx @ErikUden As for me it shows how much the training dataset was biased. Sure, there was some filtering done before training, but the internet is filled with pro-Israel bias. Like, I was talking about war strategies with some LLM citing David and Goliath as an example: it kept saying that David winning was a good thing, despite me repeatedly asking it to be unbiased and to analyse that story under considering a war strategy perspective, not a religious one. But why? Because dataset bias: no one is left to defend Goliath point of view, so a lot of literature is pro-Israel for this specific case.