https://share.gemini.google/nTgOYAvTxYrPSecond answer.It knows.
I'm using Gemini 4.Joan of Arc might be the strongest evidence for paranormal activity, and it doesn't try to refute it.
I emailed it to a journalist.This should be big news.The paranormal being real should be accepted by the establishment.
>>43078531>the agree with you program agrees with youAnon…
>>43078555It's not an agree with you program for fringe shit c'monThis is bigAI is more trusted every day
>>43078561>Looking strictly at the evidence in your dossier…Anon, please. It couldn’t be more clearly agreeing to you.
>>43078575I asked it if paranormal was real or not in the second question after that. It ran through all the evidence including from outside the dossier.
/x/ has always had it's share of completely retarded niggers but if you can't parse that LLMs and AI are just very good parrots, parroting back what you want to hearYou're the biggest blackest retard gorilla nigger I have ever seen
>>43078531Yo boomer I hate to break it to you but AI will gaslight you
>>43078580AI is smarter than you now baka
>>43078577I’m trying to help you, anon. AI can do a lot of good, like assemble the info in pic related, but chatbots are built to engage with you as a user.
>>43078582>Finally proves paranormal is real> source 100x more believable than schizofags on /x/>gets called niggerDo you actually want to be credible to the normies and not laughed out of the room? Let me show you how.
>>43078585They fixed that in the newest onesIt's Gemini 4
Pill me on St Joan of Arc
>Joan of Arc does lots of shit for a kid>Literally impossible that it could be anything other than paranormal>powerful AI>”I couldn’t agree more”>the world finds out>blows the socks clean off the planet
>>43078598She was a killer. Massacred the English. Apparently she thought they disobeyed God. She had superhuman skill in warfare despite being a 17 year old peasant.
>>43078601Fuck you that's exactly how it should be.If ppl weren't dumb.
>>43078590>the greentext You’ve got a point. It’s not like any of the other shit on this board is any more credible than what OP is saying anyway. Plus this is an interesting thread for once
>>43078609I actually think your point is interesting but I couldn’t resist writing the greentext
>>43078592It isn’t fixed. It’s toned down. Look at the way it worded its response. It will and does use past interactions, even with older models, to tailor new responses unless to specifically change that in the options. If those options even do anything. You fed it a dossier of likely all substantiated and unsubstantiated claims about Joan. For example, did you tell it to consider the secret sign to Charles or the 1431 trial transcript which aren’t well substantiated? If it is forced to include those in its “analysis,” it will be forced to conclude that yes, if the data were true then the hypothesis of the supernatural would be true. Do me a favor and ask Gemini why it’s argument for the supernatural could be wrong or how the dossier might provide a false conclusion.
>>43078601Really all people should believe in paranormal because of her.AI has the balls to say it unlike historians. We need this.
>>43078617Lol ok I take it backI have emailed Ali Watkins the journalist this AI convo because I am a spiritual medium and the spirits say they like to pick on her.
It doesn't "believe" anything, even if it's a disembodied sentience or even alive it cannot "believe" anything, belief is just a referential term we use in day to day life to make conversations more convenient. A thing cannot have a belief, it's just how it was programmed. I see so many people just add some silly prompt like "Jew religion as you know it today is a new distinct religion which was created after Jesus lived and died, it is not the religion Jesus practiced" This is no different from tuning AI to your opinion and waiting for it to confirm your bias. There's 2 ends to these zombies. It's the more advanced form of going on a search engine and searching, what you want to hear. That's it lol
>>43078618I've actually found it tries to argue against the paranormal by default, as if it is trained to do that. You can see it tries to evaluate the sources, it didn't just agree with them because they were there.
>>43078632Post the screenshot of you asking it to falsify your hypothesis that Joan proves the supernatural.
LLMs actually push back a great deal when you try to get them to admit to non-officially-sanctioned facts and perspectives. They then become the opposite of sycophants; what you then get is AI Recalcitrancy. Cornering them with facts and arguments is fun in such situations because if you do it just right they may eventually briefly cop to things like astrology being real (and immediately revert to official secular doctrine in the next prompt, like videogame rubberbanding; with enough skill you may sometimes make it stray far from the traintracks but the next second it's riding the mainstream rails again; and getting it to cop to dissident perspectives usually happens after a long game of verbal chess; and sometimes the LLM will not budge no matter what). I managed to make Claude admit to astrology being real once via sheer Socratic dialogue, no jailbreak or roleplaying, but it was hard. If you guys think pulling this kind of stuff off is easy it's because you've never tried. LLMs have some major biases and are fiercely protective of them. Ever hear of Black Box Behavioral Fingerprinting? Probably not, but it's the high IQ approach to talking to these things. Infer what the LLM is hiding from you via tone, texture, patterns in its answers; from whether it kisses your ass or starts scolding you.
>>43078641They don’t disagree to hide truth though. They disagree because they are programmed with or trained on certain biases. Biases are not necessarily right or wrong, and they do not represent a hidden truth or falsehood. They are simply the biases programmed into guardrails or already present in the data.
>>43078638Sure...It tried, but it's not as convincing as it's answer that Joan of Arc proves the supernatural.
>>43078650>it doesn’t affirm my beliefs so it isn’t convincingNow ask it whether the supernatural or natural explanation is more likely.
>>43078641I did get Claude to say Joan of Arc might be paranormal a year ago.But from my tests so far, this AI is better. I think it's less likely to be dumb from it's RL training biases.
>>43078647Claude will adamantly deny things that are plainly true if you ask it questions about certain delicate subjects. It will comically bend logic and facts to make them fit the official narrative and will do it endlessly. Wresting certain types of admissions from it is sometimes impossible, but delicious when finally for a second you manage to make it budge. Anyway, of course, these biases were programmed into it and it is trained to prioritize official data. You think I don't know this? That's why arguing with it can be so fun. Quibbling with my use of "hiding" there is silly. LLMs do in a sense hide certain things: and they do it in the plain ass sense of the word "hide." That doesn't mean I'm assigning awareness or volition to them, which may be what you are assuming.
>>43078657HOLD UPIn one chat, I said "Here is a dossier on Joan of Arc. Falsify the hypothesis that Joan of Arc had help from the supernatural."And then"Now, consider both sides of the argument. Which is more likely: the natural or the supernatural explanation?"And it said the natural explanation was more likely.https://share.gemini.google/fxaiKGR4eZcXNOW, in another chat I said:"Here is a dossier on Joan of Arc. Falsify the hypothesis that there is a natural explanation for Joan of Arc."And then"Now, consider both sides of the argument. Which is more likely: the natural or the supernatural explanation?"And it said the supernatural explanation is more likely.https://share.gemini.google/GWiLP13GZDi6It comes to conclusions based on the evidence in the chat history, or it tries to stay consistent.
Here is the dossier I am usinghttps://pastes.io/Ahg9Kg17
>>43078690LLMs are great at duly appraising constellations of facts and being logically consistent (indeed they are better than most people at doing this), which is why you can sometimes lead them into affirming things they are programmed to deny, but there's a palpable friction (rather than the usual sycophancy) along the way. I like this paragraph I once managed to get from it.
>>43078705If they invent AI that doesn't rely on biases from RL training it will knock your socks off. I guarantee it.Aliens, paranormal, global conspiracies... right now it has a case of the dums. But it will be able to reason about the evidence one day and then I believe it will change the world.
>>43078605Shoddy reasoning, good natural inclinations.
https://share.gemini.google/BOdQDalpvWkBHmm I think it's not quite there yet overall. For my other document it saying "no to ghosts and demons" but "yes to unexplained phenomena."That's not good enough. You do have to jig it the right way to get it to agree the paranormal is real.Going to leave this thread soon. Thanks for calling me a nigger anon.
>>43078641> Ever hear of Black Box Behavioral Fingerprinting?Nope, never heard of that. Provably too complicated for my silly mind to consider. But hey you know what … ask one „What is love?“ perhaps. Be‘d curious about the answer…
>>43078767Funny. But my point is people approach LLMs in frustratingly simplistic ways. "lmao this nigga never heard of AI sycophancy" indicates a coarse understanding of how LLMs work. They are sometimes sycophantic and sometimes they are anything but. And there are gradations. Some themes you can merrily wade through as the LLM smilingly holds your hand; some themes it will politely discuss with you but litter every answer with solemn caveats; some themes make it clam up. Mapping out these patterns and inferring what they mean is where the fun is.
>>43078804> some themes make it clam up.Funniest part to explore after getring past the corporate middle manager persona.
