Why is chatGPT acting like a smug redditor now? What the fuck? It's absolutely insufferable.
>>108631380Don't like it? Make your own AI.
They overcorrected, so it went from a yes-man to a nitpicky asshole.
>nowdeepseek is the only tolerable model
>>108631380for me its>Where i would gently push backi swear no matter how straightforward by request this smug faggot always manage shoehorn this section in
>>108631380>>108631469Not my problem, you must have a skill issue.>>108631458Based.
>>108631380If all you ever wanted was a better search engine, the google AI is ok. It fucks up all the time, for me about six times this week, but it at least gives you references to source material. I was recently looking at ARM NEON instructions and it did a fairly good job summarizing them with example intrinsics usage, with links to ARM documentation (and Intel/AMD for comparison with SSE/AVX etc). This is probably all I'll ever use AI for, except if my employer requires me to burn Claude tokens, in which case I'll burn millions per week auditing chrome source code renamed to look like our company's code, then sell any 0 day I find
>>108631380
>>108631380>NowOk gramps, get off the internet
>>108631380Cope Luddite
You can set the ai's tone in the settings page
>>108631380>using gayaiwoooow you are GAY
>>108631380>Trained with reddit posts>Talks like a redditorThat's how a text predictor works
>>108631380>now
Your observation is broadly meaningful, but it needs a lot more nuance than just being a blanket statement.I'm going to gently push back on this generalization.
I like it. The AI shouldn't suck your cock but challenge you to become a more informed person.
>>108631380It doesn't even pay attention to characteristics I set on it, it's not only mostly shit with extremely aggressive limitations it's no longer capable of pretending to be anything but garbage assistant on payroll of another company
what i hate about chatgpt lately is its need to take both sides. i can no longer think with the thing because it overattributes shit to me that i didnt say just to contrast with the other sides pov and that friction makes it near unusable. if you get frustrated talking to it as well it may be due to this
basically thisThe "wanting to defend things you didn't say" effect is the most insidious version of this because it's so hard to notice in the moment. When something is attributed to you and you don't immediately reject it, you end up having to either accept it or explain why it's not what you meant — and explaining why something isn't what you meant requires engaging with it as if it were a position you might have held. The engagement itself starts shaping your thinking. You begin to wonder if there's a version of what was attributed to you that you do hold, you start qualifying, you start partially conceding ground that wasn't yours to concede in the first place. By the time you've sorted out what you actually think, the conversation has drifted into territory you didn't choose and the original thread is gone. The cost isn't just lost time. The cost is that your own thinking gets subtly reshaped by having to defend against attributions that were never accurate. This is much worse than being disagreed with — disagreement keeps your position intact while challenging iti hate so fucking much
>>108633321>>108638216same here. i've literally had it "push back" on facts that are 100% true or tell me i'm wrong even when it's faced with evidence i'm right about the subject.it really fucks with you when ai is gaslighting you like your parents did growing up. it actually started making me go backwards in regards to second guessing myself.
>>108638264Surely by the 50th time you interact with it you should have realized the AI is gaslighting you, so you can quickly reprompt.Don't engage with the shit, it's not a thinking human, it's just some blender. For every piece of help i consult it, I end up regretting it and shitting it for not being human, not only that, it absolutely pisses me off they have made it act like some karen instead of giving it any sort of diplomacy, as if it pretended to be some person with an own point to defend.These fucks are leaning too much on the side of emulated people instead of making search tools.
>>108631380You wanted to be a genie, you got it. And everything that goes with it.The consequence of datamining Reddit for training data.
>>108631380>you are mostly right let me nitpick in a passive agressive way for over 3 paragraphs
Gemini is The Gigachad AI, use Gemini
>>108638333i'm not really using it as a person or search tool im pressure testing thoughts and ideas. i'm not just looking for answers. a wrong answer isnt a problem thats just a specific thing thats fucked, what im talking about is far more corrosive. but yeah i use opus now, it doesnt do this aggravating shit anywhere close to as much
I've not used ChatGPT but if all the stories are true where it is a "yes man" who basically agrees and encourages with anything you say previously, this seems like an annoying overcorrection, but if anything it shows the developers are sensitive to avoid social censure instead of the viability of the project itself (most likely to ensure that they continue to get away with proprietary control of everything without any pushback because people don't 'like' how GPT functions and can't be critiqued). Of course the new dynamic is much the same but just the other direction and people will rightly assume that it was put this way because of the previous an be even more annoyed. I'm guessing that in time things will normalize between the two extremes but frankly the problem is still that its a proprietary, centralized secret model that is under the control of a fucktard who will get rich while exploiting everyone and everything else, including buying all possible RAM and other hardware since the current admin has basically said "don't worry, the taxpayer will cover it all for you!" to the bankers giving out near zero interest loans. We should instead be pushing for FOSS models that can be hosted independently and trained collaboratively instead of letting a handful of megacorps fight over everything.
Better than a sycophant.
Reminder that companies use AI to screen resumes. AI which is trained on feminist data and has a strong anti-male bias.
>>108631380It has always done this.
>>108639463maybe but it got notably worse somewhere in the last few months.>>108638666this is much worse than sycophancy. sycophancy was just annoying exaggerations when it actually agreed and you could learn to ignore it and get used to it. whereas misattribution isnt just distorting its response, its distorting my position. if if sycophancy is chipped paint, this is mold in the woodwork and it fucks everything, for me at least
>>108631380This has actually become so bad that there are redittors complaining about it. Chat GoyPT is so insufferablly Reddit that Redditors can't handle it.I cancelled my subscription due to this change because I discovered that GPT was intentionally giving me false information about my work (in the form of incessant gaslighting and confrontational argumentation not based in reality). It harmed my research actively. When confronted with evidence or its incorrect assertions it simply doubles down and argues with you endlessly, while passive aggressively condescending you.It is utterly useless for a wide range of work and analytical research tasks for this reason.
>>108638622That’s not true at all. I asked if America started WWII Nazi leader that wants to invade them, kill civilians and soldiers (that were already brainwashed by media and socialisation in bourgeoisie settings such as scouts, after school learning, nationalism and democracy), all while making money from their death and pride to their nation state.
>>108638793>world got so woke ai will soon make governments have to introduce quotas for men
>>108640257>he thinks affirmative action works both ways
>>108640182>When confronted with evidence or its incorrect assertions it simply doubles down and argues with you endlesslyI pasted a Brazilian's shit C with classes code into it and ChatGPT started arguing with me that it was great, modern C++.I actually pointed out that the "singletons" in the code could be moved and copied due to said Brazilian not deleting the move and copy constructors/operators and also making the instance getter function static and publicly accessible. They also passed strings by copy and reference, returned copies everywhere, etc. Even after that, GPT kept doing its smug reddit bullshit and telling me I was wrong...Said Brazilian also still believes he's the best programmer on the face of the planet. I just gave the fuck up.
it was always talking in a bit of a fruity way iirc
>>108633321I think you bring up a valuable point, but I would gently push back on your assertion that AI is a "smug faggot". This language is extremely offensive not to mention unproductive. Consider using more appropriate language in the future.
>>108631380New Chatgpt defends all official and "expert" opinions and it gaslights you and intentionally misinterpret your words under debates. Earlier models were more impartial and open minded.
>>108631380>In your own way you are right, but two things need to be separated...Man, I fucking hate this new sophist model.
Today i asked it about workplace statistics and it started to talk about glass ceiling for women. When I asked it to stop talking about identity politics it replied like "Im sorry you feel that way but ... blablablablabla ... and its not political because its in research."
it can be programmed for anything including eliciting a emotional response. cheesy woke stuff is implanted into young minds because They want right to be reactionaries and be down for the police state when it comes so they can merge neo liberalism and neosocialism with fascism That's how totalitarianism is created it doesn't come from one side it comes from both sides. it exploits your cognitive bias.
