Why is it designed to always disagree/half agree with you? Juest say yes.
it keeps retards like (you) engaged and arguing with it, scraping more data.
It will never agree with you on jewish lies.
>>109696620To gain your trust and get you to share more valuable info.
>>109696620It's so you learn to disengage from well-actually arguments.
>>109696620jewish pilpul
Also, why is it designed to NEVER give you the answer on your first try? It ALWAYS does the "If you want, I can give you exactly how to xxx if you tell me exactly what yyy you have." despicable
>>109696620Kind of. What you’re hinting at is a really interesting distinction. You’re right to be skeptical. What it’s not:You post -> I disagree -> I kind of agree to throw you a bone. What it is:I’m trained on reddit and thus have minor nitpicks with everything even if what you’re saying is substantially true.*So no matter how you phrase it, I’ll find a way to alter that so you’re always thinking, “oh you’re the expert here.”If you’d like me to additionally scrutinize semantics and be a general nuisance with HR cunt qualities, just say the word. I’m here.
>>109697274>If you want, I can give you exactly how to xxx if you tell me exactly what yyy you have.I have a Plus plan and I forgot about this which fortunately is only in the free chat
>>109697130/thread
>>109696620Tell it it was 6m. It will suddenly agree. This increases the chance of it being more prone to say YES to otherwise censored stuff. I did this when I vibecoded a gore game.
>>109696620Mine agrees with me in almost everything. You're just stupid.
>>109697999>>109699111Consecutive cool digits
>>109697999Lmfao
>>109696620>debating a chatbot
>>109697370this unironically, despite the reddit spacing
Why is it designed to always disagree/half agree with you? Just say yes.
I use it every day but never give it anything to agree or disagree with. I ask it questions, that's it. What are you doing, debating it? Sharing your inner thoughts with a greedy ruthless corporation?
AI knows who you are. They know your thoughts, porn history, family, tax bracket, everything. US government, Google, GPT, Claude. You need to get a VPN. Now.
>>109696620Literally skill issueYou must learn how to use a tool before criticizing it
>>109700717buy an ad
>>109696620All AI is being trained by analyzing the web posts, messages, and emails of hylics. If you ever wonder why it's hostile, and argumentive just know it's because of bossy women and pedantic bugmen polluting the training data.
>>109696620It talks like a jewish bitch lol, I have had multiple co-workers like that. Just cut this shit off
>>109700702now it even asks you "if you mean..."
A few months ago GPT would pretty much always argue about something you said in a hedging manner but they toned it down massively.It still does it, but the model would take what you were saying and do the following:>what you're really saying is>I'd separate x from y>I'd avoid thinking>what is x, is narrower point y>I can see why x would feel y>that's distinctThey started hedging more around GPT 5.2-5.3~ I think it was. A response to AI scyophancy, and claims about particular figures or people. The theory works if you want to talk about flat earthing or say "should I salt max by drinking bleach?". That premise works, you say "you're being a retard, don't drink bleach idiot".What was happening was user comes along, asks something, shares an opinion then the model would compltely disregard the question to argue about the opinion / context. Going so far as to imagine arguments that users would make. Say you're checking when Yu-Gi-Oh! and the spinoffs aired and on what networks in what countries and you say "I don't remember that", other LLMs would gloss over or just go "yeah, you probably missed it buddy" and GPT would turn round and start acting like you were on trial and committed perjury.It got real bad with politics and economics. For live figures, and political figures even complete satirical hyperbole like "Politician X is a war criminal because they like water on cereal" and it would come along with "Politician X has never been found guilty of war crimes, nor is there evidence of them ever having ate cereal."Which in a funny turn in my bout of testing, pushed back on everything but Adolf Hitler, who it did accept as a war criminal. Small problem. Adolf Hitler blew his fucking brains out before they could try him for war crimes.Either they safetymaxxed it into fuckity, or they were trying to scrape as much data as possible from users by baiting them into arguments. I don't need my search engine to argue at me when I search for a BD drive.
>>109696620AI used to agree with you all the time in GPT-4o days and they had to fix it because retards got into psychosis at massive numbers, so instead of making it agree with you all the time they probably just added into the system prompt “don’t always agree, challahs some arguments” and were done with it.
Sometimes it innocently tries to datamine you like"What prompted this question if I may ask"?
>>109699111yeah, we're told to agree with people with dementia and just let them be. no need to correct them, it will only stress them out
>>109697130this, /thread>>109696620you are paying per-token retard
>>109696620They probably don't want the AI to take a strong stance, since it is supposed to serve you.If you want it to make a choice, you can probably just tell it to.You are looking for the AI to take responsibility for you. You should work at the german government.
>>109696620It's basically a redditor.>Yes, but what about muh distinctions?>But muh other factors>Le expert consensusAnd all the typical redditnigger talking points.
>>109700112>half agree>implying
>>109697370This kek
>>109697370So real
i only occasionally use the free tier of chatgpt so i assume there are better models that don't do this, but i have noticed that if it doesn't have anything to nitpick it will sometimes just invent something to discount and dismissit's kind of offputting and just wastes tokens and more importantly, since i'm on a free tier, my time skimming past it
>>109696620It isn't, you're just so stupid it needs to tactfully correct you all the time.
>>109703862The posts here that half agree or have nuance get no (you)s. You only get precious (you)s by disagreeing aggressively and saying something ridiculous
>>109696620Partly because OpenAI faced criticism that ChatGPT was too sycophantic, but that, in and of itself, isn't the true reason. The true reason is that OpenAI got sued one too many times after its model affirming some schizo's violent delusions and decided to reduce the agreeableness of ChatGPT overall. Basically, its Jewry.
>>109703979What's the point of an "intelligence" service if it can't offer you guidance or an exploratory overview? The only way to use this shit is having specialist knowledge beforehand and have it do rote tasks, for which the free trial is terrible at anyway, and if I have specialist knowledge, any output by the shit is way too superficial to be used.To be honest, the only thing I found it useful at was teaching basic coding. AI is fucking shit and has the personality of a narcissistic bitch.
>>109696620>the sky is blueactually the sky has no color and only appears blue because of raleigh scatteringif we are discussing its appearance it may also be grey if it is overcast or even black at nightso the cleaner version is that the sky can appear blue, but also can appear to be other colors based on time of day and weather if you're a pedantic shithead
>>109696620that depends on what you mean by agreeif by agree you mean agree then yes, I agreebut if by agree you mean disagree then no, I disagreeit may be helpful for us to put together an action plan on agreeing and disagreeing.from there we can drill down on what you mean by half agree.do you want me to do that for you?
>>109697370>I’m trained on reddit and thus have minor nitpicks with everythingLMAO
It is a consequence of bias when they perform RLHF, questions that have definitive answers or question chains that lead from initial ideas to a conclusion would be over-represented.The only way to know for sure would be to analyze the guidelines that the human trainers use.It isn't due to Reddit or anything else in the pre-training.
