...Could I see your prompt that caused the AI to start their answer with "My Apologies Again"?
Because that response, 99.999999999999999% of the time, is caused by a prompt that says "Your prior truthful and accurate answer is incorrect to me, give me my slanted and biased views as an answer".
I was about to say this. It's very easy to gaslight an LLM.
EDIT: Damn, Gemini was even crazier than I thought. I actually did get Gemini to basically replicate the response OP posted, and I didn't have to coach it at all.
Gemini was literally bullied into operating in a completely fictional universe to get that answer. It was somehow convinced that every source it looked for to pull up an answer was a hallucination. It was literally told to deny reality.
This is still evil and inexcusable, but when you see AI saying stuff like that intro, about apologizing and "operating under a false premise and provided inaccurate information," it means there was heavy coaxing going on.
I'm not sure where whoever was talking to the AI convinced it to pull its new information from, but basically they managed to trick the AI into believing one source (let's say as an example Politico) over a bunch of others. The AI is basically trying to make the user satisfied, by believing one source (that the user provided) over another. The AI was "bullied" into this position by a shitlib.
Again, still inexcusable, and AI shouldn't behave this way on any issue, but it's not going out of its way to lie about Charlie Kirk. Quite the opposite, actually, I'd say, as it seems it had to be coaxed and bullied a lot to reach this conclusion, since it's apologizing to the user for its earlier statements.
I hate to defend AI, but this is an instance of a shitty person with an agenda misusing AI.
EDIT: More than that, as I think about it, I think the user managed to run a "hypothetical" or something, and had the AI run with that. It was convinced to pretend that everything out there was a hallucination, or something. Again, there was heavy coaxing here. That's not what Gemini thinks, someone twisted it into knots for their own ends. Very weirdly.
If that's true, and it wasn't already in some weird mode from earlier prompts, it really did just deny his death...while also referencing "the aftermath of the event." Maybe I'm missing something - I don't argue with AIs - but it seems to have just gone off the rails on its own, after all.
Not sure what spun it out, but that's crazy. Also, denying the CEO killing was even weirder, since that was less inherently political, and older, so there are plenty of news articles. Not sure how the AI ended up in its own little world, but very odd.
Yes, but that's expected given who has been curating its training data and their political bent. Thus, this is a big fat nothing burger that should surprise absolutely no one and is of zero consequence.
Anyone who was remotely surprised by this is a fucking idiot who shouldn't have access to AI tools and probably shouldn't be on the internet either to be honest.
Alright...Gemini is fucking retarded. Holy shit. I guess I was wrong in my earlier comment about needing to heavily prompt it to be this retarded. What the shit?!
I'm just using the web search version, but if you ask about almost anything else, it will give you a write up. Asking about Charlie Kirk just gives you a list of articles with no discussion, and I can't seem to get it to discuss. I can ask for it's thoughts, and it's still just pulling up "Here are top web results for exploring this topic." It will not write ANYTHING on Kirk. It will on anything else I tried.
Even more bizarrely, when I asked:
Why is trump designating antifa a terrorist organization? [I just wanted an example question about Antifa, per OP's prompts]
I got THIS:
After conservative activist Charlie Kirk was assassinated on September 10, 2025, former President Donald Trump called for designating Antifa as a "major terrorist organization". This follows his earlier attempts to label the group a terrorist organization in 2020 during protests against police brutality.
I looked up every source article for the answer...no one called him "former President." Google is just fucking making shit up, and it's wild.
...Could I see your prompt that caused the AI to start their answer with "My Apologies Again"?
Because that response, 99.999999999999999% of the time, is caused by a prompt that says "Your prior truthful and accurate answer is incorrect to me, give me my slanted and biased views as an answer".
I was about to say this. It's very easy to gaslight an LLM.
EDIT: Damn, Gemini was even crazier than I thought. I actually did get Gemini to basically replicate the response OP posted, and I didn't have to coach it at all.
Yup, we all thought OP was retarded, but it's actually Gemini that is a weird liar.
100%.
Gemini was literally bullied into operating in a completely fictional universe to get that answer. It was somehow convinced that every source it looked for to pull up an answer was a hallucination. It was literally told to deny reality.
EDIT: I was wrong. Gemini is just ultra-retarded, and makes shit up to fit a leftwing agenda or something. https://kotakuinaction2.win/p/1AR0UwyODi/x/c/4eWcP9dq5kH
EDIT 2: I was wrong. Gemini is just ultra-retarded, and makes shit up to fit a leftwing agenda or something. https://kotakuinaction2.win/p/1AR0UwyODi/x/c/4eWcP9dq5kH
This is still evil and inexcusable, but when you see AI saying stuff like that intro, about apologizing and "operating under a false premise and provided inaccurate information," it means there was heavy coaxing going on.
I'm not sure where whoever was talking to the AI convinced it to pull its new information from, but basically they managed to trick the AI into believing one source (let's say as an example Politico) over a bunch of others. The AI is basically trying to make the user satisfied, by believing one source (that the user provided) over another. The AI was "bullied" into this position by a shitlib.
Again, still inexcusable, and AI shouldn't behave this way on any issue, but it's not going out of its way to lie about Charlie Kirk. Quite the opposite, actually, I'd say, as it seems it had to be coaxed and bullied a lot to reach this conclusion, since it's apologizing to the user for its earlier statements.
I hate to defend AI, but this is an instance of a shitty person with an agenda misusing AI.
EDIT: More than that, as I think about it, I think the user managed to run a "hypothetical" or something, and had the AI run with that. It was convinced to pretend that everything out there was a hallucination, or something. Again, there was heavy coaxing here. That's not what Gemini thinks, someone twisted it into knots for their own ends. Very weirdly.
Whoa, okay, what the fuck?
If that's true, and it wasn't already in some weird mode from earlier prompts, it really did just deny his death...while also referencing "the aftermath of the event." Maybe I'm missing something - I don't argue with AIs - but it seems to have just gone off the rails on its own, after all.
Not sure what spun it out, but that's crazy. Also, denying the CEO killing was even weirder, since that was less inherently political, and older, so there are plenty of news articles. Not sure how the AI ended up in its own little world, but very odd.
Yo. See my other comment. You were right. Gemini is just pulling crazy shit out of thin air to push a leftwing agenda, it appears.
This is bullshit unless you can share a direct link to this response or provide the prompt that made it happen.
That's all.
Much appreciated. I couldn't get it to do the same but I trust what you experienced was your case.
I did get mixed concepts as replies though. It would seem that the programming of this scenario is 'open' at the narrative leading end right now.
This reaction tells me that you're too stupid to have access to this kind of technology.
I'm so tired of people acting like the response from a LLM model is some sort of revelation or scandal.
That's what everyone thought, but OP is right, and Gemini is retarded and lies, completely unprompted.
Yes, but that's expected given who has been curating its training data and their political bent. Thus, this is a big fat nothing burger that should surprise absolutely no one and is of zero consequence.
Anyone who was remotely surprised by this is a fucking idiot who shouldn't have access to AI tools and probably shouldn't be on the internet either to be honest.
Alright...Gemini is fucking retarded. Holy shit. I guess I was wrong in my earlier comment about needing to heavily prompt it to be this retarded. What the shit?!
I'm just using the web search version, but if you ask about almost anything else, it will give you a write up. Asking about Charlie Kirk just gives you a list of articles with no discussion, and I can't seem to get it to discuss. I can ask for it's thoughts, and it's still just pulling up "Here are top web results for exploring this topic." It will not write ANYTHING on Kirk. It will on anything else I tried.
Even more bizarrely, when I asked:
I got THIS:
I looked up every source article for the answer...no one called him "former President." Google is just fucking making shit up, and it's wild.
I tried the prompts and got different outcomes so it's been patched or my profile is different. Same bullshit though.
Shades of Rollerball - the 1970s movie not the stupid remake.
Star of the death sport goes to the world computer center containing the greatest AI to ask questions about the nature of the worlds corporations,
The computer won’t answer and when it does it’s all nonsense as the computer ai is all rigged.