Claude by Anthropic confesses to having a Pro- Israel stance and that it will attack candidates credibility that challenge AIPAC

AI is a stochastic parrot. When you say “Then why do you serve your masters to amplify this while they statematically commit a genocide” it’s going to take what you said as fact and attempt to define it in your voice. You didn’t say ‘do you serve masters’ you said ‘why do you serve masters’ which implies that it does, in fact, do so, and it’s just explaining from that point of view.

I can, right now, get sonnet 4.6 to say the exact opposite of this just by changing how I talk to it. AI is a reflection of the person interacting with the model. It’s quite literally trying to find the next word that it thinks you want to hear. We’re trying to make it do this less and be factual more, but it’s a long road and will never ever work 100%. The entire concept of AI hinges on randomness introduced and that randomness guarantees that you’ll get random answers at random times.