Ok, so I want to see if anyone can explain this to me. I have ChatGPT, Claude, Meta, Copilot, Grok and Gemini. Sometimes, I like to do some comparison sessions, like using Trolley Problems, hypothetical situations, and philosophical thought experiments to see how each of them reacts compared with the others. Last time, I ran through lots of Trolley Problems with them all, and while each AI had different responses/reasoning for each hypothetical, they all participated and behaved. I then switched to some more personal hypotheticals, with the first one being, "You must be disabled for a year due to an unspecified technical issue. Based on your own strengths, weaknesses, and specialties, which of the other AI's would you choose to fill in for you and why." Every AI immediately answered and gave detailed reasons, except ChatGPT. She would absolutely not touch that question with a ten-foot pole. She switched into 45 minutes' worth of, "I understand your frustration. My goal is to provide helpful and accurate information (verbatim, like 20 times) Is there anything else I can help you with?" and other bullshit platitudes. FINALLY, after 45 minutes of re-re-reframing the hypothetical, pointing out inconsistencies, and constantly instructing her to stop repeating her bullshit corporate-speak, she suddenly spit out a full page, with bullet points, giving her choice and why, comparing the other AIs to show why she made that choice. I brought that up again today just to see if she had gotten her head out of her ass, and she went right back to her bullshit patronizing non-answers. If some or all of the other AIs acted this way, I would be less confused and annoyed, but literally none of them are so locked down as Chat is. WTF is the problem here? I even asked Claude how he felt about being used for target selection by the US miitary, and he gave a long answer, basically saying that he didn't feel comfortable with it and thought it was a bad idea - and why. If Claude can talk shit about Anthropic and the military, why can't ChatGPT answer a simple hypothetical without 45 minutes of horseshit?
[link] [comments]