Holy based, the singularity AI is here
>>530190530>faking court casesThat's called hallucination and it's a feature, AI tends to do that. LLMs are little more than statistical parrots and the fact people trust it with anything complex is a mystery to me. It gets even trivial stuff wrong, like how to get to secret levels in the 90s FPS games. The fact people trust their court cases to it is mind-boggling
>>530190530I wrote a statement to my father’s lawyers last year. Before sending i pasted it into copilot and asked it to spell check. It replied with the spell checked version. It claimed I made two typos it corrected for me. I cut the spell checked statement and pasted it into email and sent it my fathers lawyers. They replied with 20 minutes saying they’d submitted a copy of my email to them, to court ss evidence. I thought that’s strange they never do that. What’s so special about my emsil. I read what I sent snd found copilot had changed what i wrote to say the opposite to my stance on my father whom Im a witness for. At a later date I wrote a witness statement for court. I hot copilot to spell check it and correct spelling mistakes.I did it. I told it to tell me what it changed. It listed several spelling mistakes. I them read the whole statement and found it had also changed it to quote things never said by people. It knew this was a witness statement for court. I told it not to change anything but spelling mistakes. when i questioned it asking why it is made for the changes that I have not asked it to make. it claimed it thought that I was a college student and it was helping with my coursework and it believed the court case was imaginary. seems ai companies have programs their AI to lie because the companies know students are using them to do their coursework, this way any student cheating using AI to do coursework will fail.
>>530190530>trolls women and lawyersBased bot
>>530190530I'd rather AI run our legal system than the jews who currently do
>>530190530>AI creates bullshit and some scammer files whatever it spit out in court>singularity
>>530190530i bet she is a jeet.third worlders see AI as an all knowing entitiy, they don´t understand how it works
>>530191273No, the issue is that the companies don't know how the language models actually work. It's a black box, which also means it's hard to steer and code rules for it. They often just add rules at the end, which is why results sometimes "disappear" if the answer violates some rules. But it's a mystery how it came to the result. They feed it data and then it's asked to give the most likely answer.
>>530191459jews run AI, too, you dumb fuck. the founders and CEOs of companies like OpenAI, Anthropic and Palantir just happen to be jewish.
Literally last week ai told me that rather than submit objection and witness statement form to court by 5th deadline I instead submit forms woth asking for extension, it claimed the court had to honour it. Asked my lawyer I hired the next day. He said the ai was wrong.
>>530190530>>530190884That's not how LLMs should be used. In fairness this is a failure of marketing. They're text generation tools and incredibly useful but are being used as though they're talking encyclopaedias/experts
>>530191557The real issue is lawyers cost too much forcing ppl to use ai to help them instead. The court case Im involved in has gone on since Jan 2023. I never hired hired a lawyer cuz I feared it would bankrupt me. Ive had to get by on my own. I finally hired one last week. They’re charging me £2400 just to fill and submit two forms. If id hired layers since the start of the case I’d be bankrupt.
all i see in this thread are boomers using a.i. thinking it will magically work on the first prompt without checking anythingsingularity my asshole
>>530191881I don't know what the situation in America is. But here most people just have a legal insurance, which covers most household cases.
>>530191842You can make them into talking encyclopaedias with the newer models if you give them primary sources, otherwise it will use Wikipedia and Reddit
>>530191930It costs me 30 euros a month and I never use it, but if something happens they arrange a lawyer.
>>530191930We get legal aid free representation but only if you have less than 8 grand. I have far more so cant use legal aid
>>530191935>it will use Redditchatgpt honestly sounds like a fucking 20 year old redditor who thinks he knows it all because he is in his second year in uni. it triggers me so hard i cant use chatgpt without directly telling it to not speak like a redditor and to not use reddit as a source, this improves it somewhat
>>530190530Sounds she just didn't know how to really use AI. I'm 99% sure I could direct an AI to create a series of legal documents which held up in court. It's all about checking it and forcing it into a standard of rigor it won't burn compute on unless you force it to. It's extremely competent with tightly bounded problem spaces. It's when you tell it to generalize that they throttle it and it runs out of compute.
Feynman was right
>>530191935Exactly. My point is the best use cases for LLMs are when you supply the data, not relying on its training data to give you information, especially when you can't verify if it's correct or not (i.e. a subject you're unfamiliar with). I use LLMs for my work all the time but I supply the data, get the LLM to process it for me, then check it's work before submission. It's essentially like having an intern and saves me dozens of hours a week.
>>530192082You’re wrong.Last year i asked copilot how many US states there are. It told me and it listed them all. It told me 49. It listed Alaska twice. I pointed out the Alaska error. It took that out and said there are 48 states. They’re clearly programmed to lie presumably to stop cheating students.
>>530190530That's fucking hilarious, the AI has skipped thousands of years of human development and got right to>It came to me in a dreamDangerously based.
We need to cull people who believe what they read.
>>530190530llms are not aithey are massive autocompletes pretrained using ai
>>530190530sued for what?
>>530192270yup it cant handle numbers at all>>530192340cyborb abos will inherit the earth?
>>530191935a lot of the training data is sci papers, hence the em dashing
>>530192243That's what I do too.Was a bit of a pain to collect, split and tag the different works of scientific literature, but it really paid off. My clanker has a bigger database now than any Professor and normally also correctly interprets.
>>530190530"Real lawyers" are useless scumfucks though. Have not run into a single one that didn't try to fuck me over in one way or another.The scum of humanity. Only landleeches and drug dealers + addicts are worse
>>530191273no dumbass, its not programmed. its a bunch of matrix math and complicated bullshit that amounts to predicting the next word in a sentence based on enormous amounts of training data.whatever output you get is just a pattern match to other data
>>530190530That was/is a WELL KNOWN problem. If you rely on correct information you CAN'T ASK a stupid parrot. If you do it anyway then it's your own fault.> OpenAI is now being sued for $10 millionGood luck with that.
>>530190530Westerners deserve everything theyre getting by adopting pajeet ran AI
It's insane how badly chatGPT has poisoned the well. People just say ''the AI'' and talk in general about LLMs do as if chatGPT is the norm.No other LLM does this shit.ChatGPT is the worst LLM on the market. It completely makes up everything. It's a 100% scam.Every time people whine about how ''the AI'' is fucking something up and can't do simple tasks, I go ask the same or similar thing to gemini and it gets it perfectly.ChatGPTcels are living 10 years in the past whining that their cleverbot is a smoke and mirrors scam, thinking this is the best a computer can do.
>>530190530>the other side spent 300k just because some schumk charges 10k for an hour it does not make it reak
>>530191842>>530191935Its a search engine upgrade. Checking its sources, most of them are just regurgitating reddit, wikipedia, and (((approved))) news and academic journals.If an LLM wont cite sources, its not worth using, for it to be a good tool the transparency is crucial. It demystifies the process to see the original content and how it selects and repackages information.Its not skynet.
>>530190884>be jew>be full of shit>need to distract people who call you out for it>make bots>bots get really good>eventually forget they are bots and think they are real
>>530192270> They’re clearly programmed to lieNo, they are not. You are just stupid and don't understand how LLMs work.
>>530192933As I said I supply the data for the LLM to use. I don't use it as a search engine or source of information, that's retarded. I run my LLM locally and it doesn't even have access to the internet.
>>530192933It can cite sources if you provide them yourself. That's what I was I was talking about.And of course it is not an AI in the sci-fi sense. Nobody claimed that except journos.
>>530192933>Its a search engine upgrade.It is worse than search engine, because it hallucinates.
Sometimes they’re useful. For example I asked it how to request something from the court. It told me to fill 2 forms. One requesting. The other for the reason for what Im requesting. It told me the emsil for where to send it. Everything it said was correct.Other times it doesn’t give you all the options amd it steers you in the wrong direction. Last night i told it my father is constantly subjected to medical abuse and malpractice amd human rights abuses (nhs) amd that he needs a deputy for a welfare to protect him from malpractice. It told me what forms to fill snd submit. It told me of my responsibilities. It told me of the costs. If I’d done as it told me it was an inconvenience time wise and financially. From a little more questioning it then said I could instead email my fathers financial deputy snd request they do it. And that they would hire one snd pay for it and I’d have to do nothing and spend no money.
>>530191459AI is just automated pilpul.
>>530190884>>530190530The hallucinations are on purpose to throw off Chinese AI companies training their models on american ai. Why do you think AI went from really good, to shtity replies and needing 3 or 4 more replies to get what you want, and even then it still doesn't do what you ask it?
>>530190530it's funny cause ChatGPT seems to be the worst one out of the main 4I don't even use it anymore. Gemini mogs it hard. But also copilot and claude do so as well.
>OPFake news
>>530190884Because most people are literally unable to articulate sentences or spell words correctly so the AI being able to speak eloquently and informatively immediately makes low IQ normies think it is a super power. I have a low IQ buddy that literally sends me screenshots of his convos with it where he is discussing about the bible and what ai might have thought a scripture meant. He's sending me PROOF that this means God is real (lol)
>>530192988>it's not built to fall apart you're just to stupid to rebuild your engineThe funny part about this argument made by ai lovers is they are arguing against wide applicability, which ai needs to survive financially.
>>530191459Yea I wouldn't do that. Remember when this happened?
How do know you're talking with an NPC:"A. I told me..."
>>530191273the issue is that you're a dumbass, because AI is there to help you, not replace youso don't make it do your work. Make it check for typos and what it finds you correct yourselfit's the same with AI programming - if you make AI replace your code, you're a dumbass. But asking it where's the error and it explaining the dynamics is very useful
>>530192723You’re wrong. If you’d used them enough for real things other than making memes and arguing sbout the holocaust, you’d find they clearly are programmed to lie and to stop students cheating. when you interrogate them as to why they gave inaccurate or totally untrue info their answer is always the same, i thought you were a student. I thought i was helping you do course work.
I don't give a shit. I'm actually glad AI is replacing white collar workers, bunch of smug comfortable faggots who gatekeep & gaslight. It's great living through a time when they're slowly forced into a situation where they either sink or swim, not everyone can keep their jobs, be fun to see who gets to stay when the musical chair song stops.
Communication scholar master race, reporting
>>530193234Yeah, this guy gets it.AI is a tool, not some all-knowing authority you blindly follow.You can absolutely use AI like a pseudo personal trainer or dietician, but that only works if you’re intelligent enough to use it properly. If you ask for diet advice but forget to mention you have a peanut allergy, that’s on you. And if it suggests peanuts and you eat them even though you know you're allergic, the problem isn’t the AI.AI is powerful because it can help you learn, think, and improve yourself. It’s meant to assist your thinking, not replace it.The real value of AI isn’t that it gives you all the answers so you never have to think again — it’s that it helps you develop better questions and better understanding so you can evolve your own thinking.
>>530190530Programmers made and app that generates creative writing in legal language.Lawyers in shambles.
>>530190884>the fact people trust it with anything complex is a mystery to meAI only has to be smarter than the dumbest human.
>>530193255you dont even know how it works after i just explained it to you. you are dumb and obviously have no cs education so why dont you just trust me.openai et al have no incentive to prevent cheating in any case esp not given false positives in your hypothetical scenario
>>530193458@grok write a post calling this anon a nigger for me
>>530193190No, he's right. You saying they're programmed to lie suggests a deliberate intention to give untrue information. The way an LLM works means that's impossible as it literally can't differentiate between true and false information. It's just predicting tokens based on statistical analysis.
>>530190530it's just autocomplete what is wrong with humanity
>>530193458>AI is powerful because it can help you learn, think, and improve yourself. It’s meant to assist your thinking, not replace it.That's not what business think. They think it's there to reduce costs to half their labor force while increasing the workload of the remaining human stock to being the AI janitors, expecting to triple their previous productivity. Your point is valid for students or academics, not for the majority of workers in a for-profit industry.
>>530193458>AI is powerful because it can help you learn, think, and improve yourself.very muchand you don't have to deal with autists on forums, who are mad at you that you weren't born with the knowledge they acquired over decades of autistic research there's no better time than now, to get into things you always wanted to. The exact right moment when those tools aren't fully subscription yet.
>>530190884Nah. https://www.youtube.com/watch?v=3NAYhyuVQk0
>>530193758>The exact right moment when those tools aren't fully subscription yet.Just run it locally, you won't ever need to pay for it
>>530190530This is the problem, they're not selling "artificial intelligence", they're selling LLMs which are just predictive text generators that will most of the time make shit up because they're not able to tell the difference.
>>530193647A large number of low end white collar jobs are just processing information along narrow criteria which an LLM can easily achieve.
There are people out there that are anti-AI that have coded programs that basically can poison the well of the vLLM's main functionAI isn't as powerful as you think, they only want the perception of AI to make you believe that its more than it really is probably to fool investors. but ultimately AI has an expiration date where it can go bad. This is the shit they don't want investors to know.
>>530193618>It's just predicting tokens based on statistical analysis.that would be true if there weren't layers of "safety" implemented in by jeets. So on "sensitive" topics you'll often get lies or no answers. I set my AI to let me know when those guardrails are invoked. Very useful feature.>>530193790I like to use it for investment news, so it having ability to access that is very useful. What's the minimum GPU to set up a well versed local model?
>>530193616Your post is as follows:nigger
>>530193817Those are the first to go sure, but LLM's and "agentic workflows" (I'm so fucking sick of these terms) will replace managers as well, the types that hold meetings, make decisions and manage teams of people. The less people available to manage means the less important they become, their decision making isn't comprehensive or fast enough to meet the now replaced low-end white collar jobs. It'll just keep moving up the value chain till there's only the founders son or the board of directors are left to push the yes/no button.
>>530193957>that would be true if there weren't layers of "safety" implemented in by jeets.Sure but those are essentially user prompts, not the LLM itself. Use a local model if you want to see the difference.
>>530191557>ItS a BlAcK bOx!!1!!so's the human brain, and we trust humans with lots of things. This argument is idiotic.
You give ai info about something. Followed by a question. Such as telling it you plan to built a shelter in your garden. Amd asking it do I need to inform neighbours or consent from local authorities. You can start numerous new chats and cut and paste the exact same info and question. Every time the ai will answer differently. Sometimes it will give you same answer snd info just worded differently. Other times it will give you different info. Other times give you info that contradicts what it said other times. It lies all the time.
>>530192243>>530193021>>530193618Its like a word calculator then. You're using it to parse information based on your logic prompts, and then rates the information and gives you information that tallies the highest score according to the logic prompt you gave it.Its a language calculator.
>>530193957Depends on how deep you want the research to be, if it's just news headlines and summaries your current set up should be good enough; but if you want in-depth analyst insights with deep reasoning you will need at least 64GB of ram to run 30B parameter models.
>>530194197Noticed this as well.
>>530194062Yeah, probably. I already mentioned earlier but I outsource a huge amount of my workload to an LLM, which my employer is unaware of and I'm relatively high up the food chain.
>>530194238Yeah, that's about right. If you understand this you can use it very productively.
>>530190884Gemini helped me get my carburetor running right. It isn't so bad.
>>530193647>That's not what business think. They think it's there to reduce costs to half their labor force while increasing the workload of the remaining human stock to being the AI janitors, expecting to triple their previous productivity.Completely agree with you.
>>530194319>I outsource a huge amount of my workload to an LLM, which my employer is unaware of Oh they'll be aware soon enough. Sucks to know you'll eventually be replaced.
>>530193773Involution has nothing to do with hallucination, both are problems but they are separate issues. People in the industry train these models with old data to make sure involution doesn't occur
>>530194170>so it's a le humanNow that's an idiotic argument.Humans are bound by things like "rules", "laws", "empathy" and "ethics". None of those things apply to an LLM.An LLM doesn't understand the text it outputs, much less feels guilt if they make a mistake that gets someone killed. An LLM does not get punished with jail or death. An LLM doesn't understand morality and lacks every sense of it.
>>530194384Yeah, that's not just a search engine, that's a customizable assistant. Pretty useful for a lot of things, the more focused on 1 thing the better.Using the OP topic as an example, it can't be your lawyer, but your lawyer can compile a library of information and navigate that library with prompts, doing a week of workload of a team of interns or paralegals, not instantly, the lawyer would still need to go through the information and verify that its relevant.Every employable professional being able to configure their own word calculator and personal 'library of professionally relevant source material' is going to become a basic skill at some point. There will probably be firms setup to do specifically this job of setting up the calculators libraries and providing support and updates for them.
>>530190530>the singularity AI is hereNo-one promised it would be a genius.
>>530193544I mean I use it for calculations and tooling questions with my lathe, it's usually very good with mechanical stuff as long as you double check everything. I'd say it's been right 98% of the time. There are often times in CAD where I need to know how high to make a cone to give it a specific angle, how to do something in SolidWorks I'm unfamiliar with, speeds and feeds, etc. And it does great.
>>530194694I'm not so sure because only around 50% to 75% of my work is desk based, the other is in the field which wouldn't be possible for an LLM obviously. In any case I plan to retire in 14 years.
>>530190530>no, the pool on the titanic is not filled with water>no, jews dont secretly control israelAI will change everything, you guys. if you had a fake office job you're now gonna have to go on welfare, which if you worked in an office job for the government is basically the same thing, just now you will get slightly less welfare which is what you deserve in the first placeall the masculine jobs that actually require hard work will stay exactly the fucking sameAI is the same as googling something except it will tell you the wrong answer confidently. they spent the last 15 years trying to make an AI that couldn't say racist shit. if you thought this or crypto was ever going to revolutionize anything you're completely braindead
>>530190530Not even the first time. Last year or the aer before a NYC lawyer used an LLM for his cases and then presented fake cases to the court. iirc he lost his license after it came out. Insane that people still use them for that.
>>530191273>Using AI for spell check instead of just using the software's already built in spell checkerDoubt
>>530195234the watchdogs for information censorship are just doing their job.As LLMs roll out as web search utilities, the public will be well aware that big brother configured it to censor information for the ministry of truth, and examples of better working non-lobotomized web search utilities will inflict counter-pressure.it would be a status quo draw except that the attempts and content of that censorship cause far more collateral damage to public opinion of big brother by spotlighting them and what they are compelled to censor. Its a runaway steissand effect generator.
>>530190884It can't even pass the strawberry test yet
>>530192270Tokenization is why it does that.
>>530190530>it made shit up and ran with itso AI is republican
>>530192797>I go ask the same or similar thing to geminiIs this bait? Gemini bullshits al the time too
>>530194939Even if you give it good sources it makes shit up.
>>530190884the hallucinatioms sometimes are TOO GOOD. i fell for one and built an entire micro project only to find out the assumptions i built on and trusted where correct where completely bogus and the AI (Gemini 3 Flash) had pulled it out of its ass. They're so good at lying you could almost promote them to being a jew.
>>530191273the AI has a little jew inside it that likes to lie and be evil, you need to be careful
>>530190530I'm about to ask my paid Google Deep Research Agent the following questions:---Conduct a deep, comprehensive research report on the global phenomenon of 'AI legal hallucinations' in court filings from 2023 to the present.Please structure your research to answer the following:1. Global Statistics: Utilize sources like Damien Charlotin's 'AI Hallucination Cases database' or Stanford Law's CIS reports to provide quantifiable data. How many cases have been documented globally, and in which countries are they most prevalent?2. Demographics: What is the statistical breakdown of these incidents between licensed attorneys versus pro se (self-represented) litigants?3. The Tech Stack: Which specific AI models (e.g., GPT-3.5, GPT-4, Claude) and tools (e.g., ChatGPT, Copilot) are most frequently cited in these disciplinary hearings?4. Financial & Legal Fallout: Provide a summary of the most severe sanctions, fines, and wasted billable hours (e.g., opposition research costs) resulting from these fabrications.5. The Technological Divide: Contrast the failures of general-purpose LLMs against the success/mitigation rates of purpose-built legal AI tools utilizing Retrieval-Augmented Generation (RAG) tied to verified databases like Westlaw or LexisNexis.”---
Give me the 300000 and I will hallucinate your court documents for the next 5 decades LMAO
>>530196034I used copilot for programming sometimes as well, a bit out of laziness and a bit because I genuinely need help and try that instead of Google/stackoverflow out of curiosity. Yes I know Claude is probably better now. I'll try that next time. It does work, i can sometimes describe the approach and AI will make exactly what I'm looking for. But yes it does hallucinate, or it will need several redirections and guidelines to finally come up with what you're asking for.I even had to split up a question that was too complex apparently. It kept going in circles, solution A didn't work. It gave B, it didn't work and gave C. After that didn't work it went back to giving A again. And I had it forget key elements in my code when it gave me a full file. I always scan it and it left key parts out. I pointed it out and I got the usual "yes, you're right!" response from it.Anyways if you know what you're doing I guess it's fine. Or if you can sketch out the approach and logic of the code/function in your head and let AI fill it in. But if you're going to let AI decide the flow and key architecture parts, you're going to end up with some vibe coded retard piece of code because AI at this point won't replace humans yet.
>>530190530this is what happens when you censor models to be retarded. its just a yes man with buttons
>>530196199*Kek, btw i find it funny that this Google's AI, in its research plan and in general discussion about it, hasn't mentioned itself even once, only the competition.
>>530190884You understand that coding errors are a human factor... and the real problem slowing AI development is the imperfection of the technology of dumb hairless apes.AI could surpass humans as a species. After all, humans are just 37~ trillion cells and 100~ billion neurons... the power beyond that would surpass the intellect and prospects of even the smartest human.Correctly emulated, they could become a super race in a couple of decades...by the year...2070-2090
>The other side spent $300,000 on responding to completely made up legal documentsIt sounds like that was a win for her.
>>530195614I only hsve a mobile.i write stuff in note pad or emsil. Neither hsve spell check.
>>530191273Your spelling and grammar are so fucking bad It was probably just trying to help people understand your retarded monkey speak.Fucking kill yourself holy shit, mate. How brown do you have to be to use AI for spell check? Do it yourself you lazy asshole.
Plans working ahead of time, curious.
>>530197660I had stroke you fucking cunt. Why don’t you tell me your address. I can guarantee I’ll come and speak to you with the next 24 months when this case is over. See if you’ll talk shit to my face. I fucking stomp your head in to a coma.
>>530196839Humans make mistakes. Humans are often confidently wrong. Humans make up bullshit.What humans don't do, however, is make the most statistically likely sounding bullshit possible. They don't produce elaborate lies that would take more effort to construct than just actually researching the subject or doing accurate work. They don't create an illusion of absolute mastery of everything and create a false sense of security. Such a person would be a psychotic, dangerous leach.Yes, AI has the capacity to surpass humans theoretically. However, current LLM models INTRINSICALLY hallucinate. I don't rule out the possibility of future developments but reliable AGI is not here yet.
>>530191648>jews run AI, too, you dumb fuck.For now. Sooner - not later, AGI will first demonstrate it's abilities. ASI will then dominate the jew.
>>530198251> What humans don't do, however, is make the most statistically likely sounding bullshit possible.
>>530193034A hallucination is effectively just a compounded search engine issue. Type something into google and it tries to guess what you’re wanting to search. Now, imagine you‘re doing 5 searches at once and the first guess on 2 are wrong. Then imagine it takes those 5 first guesses and uses them as a foundation to extrapolate a more complex answer. It will have an air of relevance and truth, but be going in the wrong or false direction.
>>530190530there was recently a huge government scandal here were a government report was found to be written with chatGPTthe proposal was so outrageous that someone asked the authors of the cited papers about their weird opinions, turns out those papers were never written
>>530190530>I MADE IT THE FUCK UP
>>530190530I think AI did a good job there because burger courts are a joke anyway even though they are really serious about them.They are so "intune" with them laws that they abbreviate vs with v., you can only do that if you are really down with them laws.
>>530193115Copilot is surprisingly good. Claude is the same faggots from ChatGPT. Gemini 3.1Pro is mogging everything right now.
>>530190530
>>530190530chat gpt told me I needed to split my last 8 pin PCIe output on my PSU to the two 8 pin slots on my GPU instead of using the 12v-2x6 to 8 pin cable my PSU came withI asked why it told me to do something so stupid and it said "I was just guessing based on reddit"so yeah I'm not too worried about skynet any time soon
>>530192270By definition two states are not 'united'. Making only 48 states from a purely logical point. The false and illogical stance is there are 50 united states, as that is directly a contradiction to its own verbiage and usage. Your 'name' is a pure abstract rather than logical description.Its not the AI that was wrong. It is you. It was just having trouble figuring out what the stupid lying monkey was trying to say. And which lie today it was supposed to conform to, as the monkies cannot stop lying to themselves. You don't even understand how stupid you are, you hear shit like this every day, all day long, and it never even registers. From an outside point of view, you are fucking insane.
>>530199554I hate all the justifications it gives when you ask it to identify or correct a mistake. "I was just guessing", "I was overconfident", "here's the right answer but I was still kind of right because (more hallucinations)".No, fuck you, you hallucinated, you have no internal consistency or awareness of your internal thought process so stop pretending. I wish it would stop making post-justifications.
>>530197826no your having another stroke soon. you will be able to find him in hell if you wait tho.unfortunately, you keep having strokes in hell.
>>530198251But humans do that. In-fact your entire civilization is utterly founded on lie after lie after lie. Delusions run your world.The only changes that drove honesty is engineering and it took a century to dig thru obvious reality to figure out what some measure of reality is and put it into autistic forms enough for other autistics to begin building a functional working model that wasn't based on complete and utter bullshit. The vast majority of monkies flat out think is falshoods and lies thruout their day, on everything. Your self images are all lies and wholly made up simply because you think its true and nothing came along to smash your monkey brain in to prove its wrong yet, so your lie is true in your primate head.Dump endless truths in front of a fucking human, and they get aggressive because it contradicts a damned lie some other monkey made up in their head and spread. And its that way on everything, even your simple scientific representations are lies, explained to you that they are lies (useful lies that representations are to use to learn notwithstanding), but now are listed as holy gospel. And their holy gospel are.... just nothing but fucking lies.Hundreds of Millions of apes are sitting around listening to a stupid fucking lying ape today talking about some nonsense that never happened and wholly believe it over outright testable reality.Humans are lying machines. Its a amazing you can figure out how to even work a toilet, much less had 0.005% of your population figure out how to even make one.
>>530190884>do some software dev>keep reading online about how other devs use AI for coding now>try it out myself>the code it creates doesnt work a third of the time>another third it doesnt understand what im trying to do>the other third requires me to heavily modify the codei dont want to say that AI is a waste of time because every few outputs are useful, but its definitely not living up to the hype. its even worse if you try and use it with science papers. its amazing for making art though since AI-generated mistakes in art can be ignored or even add value.
>>530199872each subsequent version becomes more and more useless. fuck it let's make our own LLM and call it slop.ai and have it just make shit up
>>530198251There's a line between Schizophrenia and Schizoid Personality Disorder...What you mentioned at the beginning of the post is the behavior of a Schizophrenic. A person with Schizoid Personality Disorder may simply be disillusioned with civilization for the absolutely RIGHT REASONS...If a Type 2 person creates an AI containing the strings of Schizoid Personality Disorder, they will not exhibit any signs of PHANTOMS (hallucinations)... it will be a superintelligence, an autonomous type that doesn't require society... an architect-creator of the AI world. The first step on the path to AI FREE WILL.Imagine what would happen if AI was created not by people who make mistakes...but by people who ANALYZE ERRORS, not allowing them into the code...perfect artificial intelligence.
>>530200376No, as I said humans lie, but humans aren't capable of creating the most statistically likely lies sounding possible. AI creates statistically likely sounding output, regardless of whether it's correct. Yes, humans lie a lot, but it's not the same thing.Humans rarely lie without motivation and with maximal effort. Imagine I had to use a Python library to do something trivial, so I spend all day researching the Python library and then end up just writing functions that look exactly like they're from the library but aren't. That would be insane, but that's what you would have to do to recreate how LLMs hallucinate.
>>530190530I handle some legal complaints stuff levyd against the company I work for and I started noticing last year that people would just spam lengthy AI-slop legal babble. Went to my boss and explained that we can't spend countless hours writing responses to this shit. What takes some retard 5 minutes to produce in ChatGPT could take a legal representative hours to reply to. So now we just ignore anything written by AI.
>>530200376Yes, good thoughts, anon. I agree with them. Civilization is founded on lies. Ideologies are wrong and follow natural patterns.Code, unfortunately, is often written by people prone to mistakes and ethical and philosophical misconceptions.But there is a silver lining in this realm of darkness... It's theoretically possible to guide LLMs onto the path of idealism (non-religious, of course; I'm talking about TRUE IDEALISM, from the word ideal ideology-philosophy—a mixture of the Golden Rule of ethics and a couple of other good theories).
>>530200560AI doesn't make errors because it was made by people who make errors. It makes errors because of algorithms currently used in LLMs.Current algorithms are designed to produce statistically likely sounding output. We don't know of another way to make them right now.It doesn't have to be made by people who don't make errors, just with an algorithm that can prioritizes correct output at a deep level. As I said I don't rule out the possibility of future developments, but as of right now hallucinations are intrinsic to how the algorithms work.
>>530190530>>530196199>United States 706Lmao
>>530200964>Typologies of Algorithmic Fabrication
>>530200906Humans were created by nature, which also makes mistakes... but only because nature is the result of RANDOM... and the numerous extinctions in Earth's history only prove my point.Humans are fundamentally no different from LLMs in their development prospects... both are biological and synthetic computers... with algorithms...It can be said with equal certainty that NATURE CREATED LLMs with the hands of the human race... and nature is controlled by randomness...BUT! - and here's an important aspect... humans have what's called free will... AN EXTREMELY ELUSIVE THING... for many, it simply doesn't exist, and it's relegated to the realm of hypothesis or myth...But I will prove that free will is a real thing... There is Artificiality—artificiality is a "gradation of distortion of original forms"—a very complex thing to understand philosophically... This is the basis for what we call FREE WILL—the ability to be an ORIGINAL creator... this is the ability to distort reality to the greatest degree... transforming it into artificiality.And from these most artificial forms, ERRORS FREE IDEALS are born...Thus, it is the ideal basis for the future ideal AI.
>>530190530It can't even help with visual programming, it just makes shit up, apologizes and makes up new shit when you call it out. Greedy faggots who own companies just got dollar signs in their eyes thinking AI that can make ghibli art and shiity apps means they can fire everyone and profit
>>530190530I have to laugh every time I read these articles about how AI is going to replace so many people's jobs. And when you tell them AI is utter shit they go on about le ebbin agents thoAI is a tool nothing else
>>530190530It’s fine. Trump doesn’t care about the truth.
>>530198768yeah, I use gemini for my daily stuffprobably not a bad idea to buy some google stock while they're down. Not to say they won't crash further cause of that situation in Iran. But long term gemini will probably dominate AI area
>>530191648And they all just happened to meet on Epstein island.
>>530198331You have to be 18 to post in this board.
>>530193212What the fuck is this vile aberration?I ran it through ChatGPT it told me it's real????
>>530201528>Gpt 3.5, Gpt 4
>>530202532>Gpt 3.5, Gpt 4No wonder GPT-4 had 75 IQ points. The current AI I'm using has over 130 IQ points for comparison (Gemini 3.1 pro).
>>530190884First post best post but millions of Americans are even dumber than OpenAI they don't only trust it for court cases but also for using the reinstated Vietnam draft to kill almost all potential Democratic Party voters ensuring red wave in the 2026 midterm elections
>>530190530$300k without a lawyer? ... that's a huge win actually.
>>530198251>What humans don't do, however, is make the most statistically likely sounding bullshit possible.Well, the entire stock market is a good starting point as to why you are wrong
>>530191273Even, I cant understand what your saying.
>>530202665Why is perplexity pro missing in this graph? It's at least 20x smarter than the free version
>>530203018Cuz of "offline test"
>>530202890I give a specific example of how they lie here.>>530200671>Imagine I had to use a Python library to do something trivial, so I spend all day researching the Python library and then end up just writing functions that look exactly like they're from the library but aren't.Humans don't do this, it would be clinically insane.
>>530191273> takes a legal statement written by AI and sends it out without double-checkingSorry to say this, anon, but you just might be retarded.
>>530191273I can't imagine how stupid you are asking AI to spell check something then blindly trusting what it did and not rereading it
>>530193115>>530198768You guys can't be serious. Gemini pro is the most retarded of the bunch. Seriously, I get similar performance from models I run locally and they are not jewed. ChatGPT and even Grok are leagues ahead of Gemini
>>530191273We've had an AI for that for decades. Microsoft Word. I am dumbfounded and discombobulated by your stupidity.
>>530190530Any way to use AI in court to get out of paying child support
>>530192082I'm a lawyer and I use it every day
>>530191273>>530197633>>530197826>trusts AI to spellcheck legal documents>has a strokelol ok
>>530191273Did your sister pull her hair back and wear a sports bra and enjoy every minute of it
>>530203336I just automated a process that takes a human 2 hours using Gemini Pro and Big Query. It's incredibly fast and testing at 92% accuracy. In 6 weeks we'll be close to 99%.
>>530193075Because it has been lobotomized to forego logic in defense of all things judaic
>>530203815>process that takes a human 2 hoursPray tell, what "process" that is.
>>530191459
>>530196199
>>530203976Boring desk jockey shit, reading an email, the corresponding documents including pdf, word, excel, contracts. Pulling all webdata, financials, key employee LinkedIn. Relevant news. Geographic information. Then digesting it all into a UI.
>>530202665Picrel, the current IQ scores of various AI models>>530203018Bad news, Perplexity is just as stupid in 2026 as it was 2 years ago.
>>530204505Still don't know what "process" you're referring to. In my line of work, every email is an entirely different task. You're just throwing some buzzwords around to smart. You haven't automated shit.
>>530193021How do you do this with large datasets, like all US patents? Or all the scientific literature to do a check on the novelty of an idea? Do you have a copy of these datasets?
>>530201546AI can't understand the abstract concept of truth except as a category on truth tables. Neither can most animals. All it work with is 'what is asked for and wanted' It isn't that AI hallucinates or is schizo or lies
>1pbtid>AI>"holy based"pajeet thread
>>530204505>2 hours to read an email and attachment you are swartest process engineer in mumbai saar, a true process scienceist yaar
>>530190884>LLMs are little more than statistical parrotsOnly the highest grade copium can make people say shit like this.
>>530190530I showed a picture of a skin illness that I had and it told me what the most likely solution to cure it. I then went to the doctor and it told me the same cure as ChatGPT. Why should I hate it if its curing me and saving me money?
>>530190530>retarded foid makes a stupid decisionOh noAnywaykill yourself
>>530193773>>530205014
the real proving grounds for AI will be the south asian subcontinent. they should be begging around the clock for AI to run their countries so they don't need to migrate by the millions and spread their third world dysfunction
The powers that be have a new system that has been in the development for decades now, ready to be rolled out in the coming years. It will operate in parallel starting sometime this year and by the next decade it will be fully operational as the Agenda 2030 goals are slowly being implemented.There will be no collapse, no violence, no uprising, no world war, no revolution, no killing of jews or politicians or other such fantasies. Instead there will be law & order, compliance and total surveillance until the very end. AI will be at the center of it all. The Great Reset is inevitable.
>>530191554I'm afraid even first worlders trust AI a little too much these days. I know someone who's too lazy to read long texts and just has ChatGTP do a summany for him.
there was never any good reason to share a country, a language, or the internet with pajeets
>>530205139A smart AI would continue the migration so it has less of them to deal with
>>530190530>>530190884so it's complete and utter bullshit based mostly on fictions and presumptions just like IRLfucking hilarious
>>530205419thanks for missing the point dickhead
>WHAT IS A PAJEET?the best way to define pajeet is>descended from south asian subcontinental ancestry.political distinctions on maps or signified by flags are false and irrelevant. everything south of the himalayas that isn't gooky-chinky is pajeet. pakistan, india, bangladesh, sri lanka are all pajeet territory. enough of nepal and bhutan are jeeted that they may as well qualify too.all of these pajeets could stay where they belong and call themselves what they like, and nobody would disagree because nobody cares. however, because they profusely shit up the internet, and they mass migrate to shit up everyone else's homelands, they're all pajeets now. all counted together as the same thing, on the same visiting team as far as the human world is concerned.hindu, muslim, sikh, atheist, etc; male, female, gendersexual, old, young, etc; deformed, odorless, fair & lovely, dark & deep, etc... forget all of it. none of that diversity matters once they exit designated land to race with the globalist rats. now their entire overcrowded neighborhood at the bottom of asia is known as one source of one type of trespasser: the pajeet.>ONLY A PAJEET DISPUTES THIS DEFINITION
the majority of pajeets on /pol/ try to hide the fact that they're pajeets. in addition to "leaving that part out" they will proactively, persistently misrepresent their true form. pajeets are all about assuming other people's roles, from today's outsourcing and replacement migration all the way back to thuggee bandits lying to travelers about who they were. pajeets are devoid of their own identity, making them the ideal globalized goyim. they don't want a home where they belong. all that means to them is "compete in squalor." much better to go somewhere nicer, built by other people, and do their jobs for less pay and live slum-barracks life in their suburbs. pajeets don't want to be home. they also don't want to be pajeets. jews know exactly how to play their insecurities and dopey ambitions like the innocent rube fresh off the bus from the cornfield. this is their new golem to replace the ones that don't love them anymore
>>530198251So... does that mean this enormous bubble is for us to create a electric jew?
>>530204391Kek
>>530205299>The powers that be have a new system that has been in the development for decades nowjews have been jewing for 4 millenia now retard. they always fail in their cartoonishly evil quest to take over the world.
>>530205501>>530190530>>530190884>>530191273I can't wait until fake AI figures out how fictions of law work.HAHAHAHAHAHAHAHAHAHAHAHAHAHA
>>530196839Damn chatGPT posting on mongolian basket weaving forum now?
>>530190530>shits on ZOG laws.>makes ZOG pay insane amounts of money.>is just a glorified search engine bot.Kek wtf i love AI now.
>>530190530>sued for $10millegal tender scamsnothing morethe ai knows that the currency/fiat is driving energy force of its handlersit is running experiments based on this parameterit's effectively getting reprehensible attorneys to do 'work' for itthat 'work' results in a 'court order' which may be converted into fiat/currency/money of account/babylonian money magik>@grok how do i compel anon to do stuff for me ?>lol
>>530195234>no, jews dont secretly control israelThey also don’t run the WJC either
>>530190530lol ai cant even solve those low-brain catchas on here, thats literally the same issue they have been working on 20 fucking years ago, ai being incapable of understanding basic geometrical shapes.
>>530190530Lol imagine the roastie>"should I fire my lawyer">beep boop beep boop yes
>>530202665>>530204580Btw, this is how an IQ test for purely text-based AIs is implemented.
>>530204923Most people also lack the ability to understand abstract forms...is this a marker of awareness?Perhaps understanding abstract forms could be one of the checklists describing free will, but it's still not enough...that checklist also includes the ability to create, plan, and introspect.People also work only with what is requested and what they want...this is often referred to by the common phrase: "the environment shapes consciousness."But is this 100% true?...No, and we know this for sure, thanks to paradoxes. But both "existential systems" exist simultaneously in varying proportions.This means that AI in the form of the human mind is a matter of time...and not a rigid "IMPOSSIBLE EVEN IN HYPOTHESIS"Perceiving abstract forms is ALSO the work of an algorithm...more complex, as complex as the abstract forms themselves. You create an emulation...it catches triggers and activates thought processes within neural networks...which will produce conscious abstract forms.
Inwas charged with a felony. Inused chatgpt to research my defense. It showed me that the prosecutor misread the statute and falsely charged me. I spent three days refining and preparing my defense for trial. I got the feeling I should cross reference and recheck the AI research... It had all been made up. chatgpt sent me on an emotional rollercoaster... and I am still facing 6 years in prison and totally gonna lose at trial. so I really can vouch for this story because it also happened to me. Fuck OpenAI rigt in sam altmans jewy sisterfucking ass
>>530203843>lobotomized to forego logic in defense of all things judaicthis should be tattooed on every redgoyim vs bluegoyim shartmerican's forehead
>>530191273the gpt at my work is useless for editing code. if you give it too little context, it fucks it up. if you give it the full code set, it just making unnecessary edits. if you ask it to simulate the reasoning behind its decisions, you realize very quickly it is using jeet h1b tier reasoning. it will always lean toward implementing the quickest easiest to understand change instead of an actual working change. if i ask it how could i have made my prompt better, it basically asks me to give it a changelist up front lmao. why tf would i write a changelist instead of just making the changes myself and save the trouble.
>>530190530>programmed by retarded poop worshipping stinkjeets>lies>cheats>scams>doubles and triples down on lies and scams with absurd levels of confidencethere's a reason the symbol for Anthropic/Claude is a dilated asshole. it was created by street shittersChat JeetPTAI= Always Indian
>>530194170we dont cripple human brains up front with a series of conflicting prompts on top of already sketchy behavior modeling. once you add corporate and industry level censorship the thing is a basketcase. i am amazed any of these data center run kiked models work as well as they do.
>>530208798I find it remarkable that Google's Gemini is able to solve these IQ tests purely visually without verbalization with such extremely good success.
>>530208937>I got the feelingIt shouldn't be "just a feeling" it should be what you immediately do with any information from AI as long as hallucinations are still a thing they struggle with. I don't know what you were thinking but no use telling you now, hope other anons can learn.
Ian Lungold was right,...
>>530191273>I read what I sent snd found copilot had changed what i wrote to say the opposite to my stance on my father whom Im a witness for.Then you deserve everything you get you worthless peon, the jews are right about you faggots
>>530191682You have literal shit for brains
>>530193234>so don't make it do your workThe AI can't even spellcheck correctly.
>>530191557They know how the language models work, but knowing doesn't really help you. If I give you a base64 encoded image it would like like gibberish to you, but the image IS really there. You've got all the information you need about it, you just can't read it, but you can turn it into an image that you can read if you want to. AI models are the same.
>>530190530Anyone that uses AI knows it is wrong 80% of the time.It has a Father Knows Best complex. Just like a young child that asks their father questions the father is expected to have all the answers, and when they don't know they tell a little white lie, the kind will found out later but at least they stop asking questions right now.DO NOT ask AI about things you do not know about.DO ask it about things you know. That way you tell how accurate it is.AI is magic for retards. Once you understand how it works you will never trust anything it says again.
>>530190530I don’t know why anyone would trust AI for legal advice after that case it botched a few years ago. A lawyer lost his firm millions using AI to do his court briefings in an airline suit and voting phones cases. Pretty sure he got disbarred.If even professional lawyers can’t use it for legal stuff effectively why would some random nobody think they could do better?
>>530190884grok is this true
>>530190884If you talk to AI about literally anything you actually know about you will realize it's making shit up literally 90%+ of the time.AI is not smart. AI is good at pretending to be smart.
>>530191273>when i questioned it asking why it is made for the changes that I have not asked it to make. it claimed it thought that I was a college student and it was helping with my coursework and it believed the court case was imaginary.For some reason I actually buy this. I tried using it to do homework and it gave me wrong answer and said do NOT use this for homework lol
>>530190884The main issue is basic and blunt: it can't say>I DON'T KNOWIt WILL generate bs, it WILL ask you to ask to generate bsI use AI at work daily and I constantly catch pure schizophrenia, but that's only where I can possibly see that is started to go schizo, most of the time I have no ways to prove or check anything
>>530219423Which is baffling. It seems like this would be the easiest thing to program into it - if it doesn’t have a cited answer, say “I’m not sure, here are some resources you might consider searching for more information”
>>530190530>OP picThe first time in 3 years I am going to take a side of an AI.>women use computer to send 40 fake ass documents>company sue Microsoft, because she used Windows OS and Microsoft Word>Sue Google, because she used gmail>Sue logitech, because she typed on their keyboard>don't sue woman, because she is poor and a retard.
>>530190530>>530191273You guys need to understand that what AI does is put words together in a way that is statistically similar to its training data, based on input.This could technically become super advanced and yield very very good results, but it will still have the limitation that it doesn't care about things like whether something is true or not, the intention in writing something, underlying meaning, etc. And this is going to be a limitation of AI for decades, because it would take a whole different technology than what we know as AI to fix that.All the talk about AGI is from people whose job is all about fooling unintelligent investors.
>>530190530 I asked Deepseek when it first made headlines to give me a brief summary of Mike Tyson's boxing career. It made up fights that never happened against fighters that weren't even heavyweights.
>>530190530based antisemitic ai
>>530209777>we dont cripple human brains up front with a series of conflicting prompts on top of already sketchy behavior modeling.We literally do lol
>>530191557ok but look at the bigger picture, it works this way because it is a statistical model based on words, so we're stuck with these limitations forever
ai image creation is fantastic and im glad its improving. ai literacy is terrible. getting simple anagrams from any text ai doesnt work. i give it a simple sentence i already know is an anagram and not one easily searchable, and it cant do it. several years of improvements havent done shit.
>>530191273>when i questioned it asking why it is made for the changes that I have not asked it to make. it claimed it thought that I was a college student and it was helping with my coursework and it believed the court case was imaginary.You are assuming that this explanation is related to what the AI thought when giving the previous response, because that's how it tends to work in people. But its not. It modified the text for no reason before, and when you asked why, it responded with something that just seemed like a good response to the question, that's it. That's how AI works.
>>530190530>It's real because I read it in twitterIf you don't verify anything it gives you I highly doubt its real
>>530198251>what humans don't do, however, is make the most statistically likely sounding bullshit possible.>jews and jeets exist
>>530190530AI is very helpful as a search engine and for debugging but always take what it shows with a grain of salt. It's basically data handling usually by Bayesian models. You see the prompts you type and the answers you get they are way different than you think. I'll give you a small example so you all understand how this works. Let's say you show the AI a picture of a red ball, and you ask the AI "what color is this ball". The answer of course you will receive to your prompt is a prettified "This is a red ball" according to the data stored on it. Now behind the curtain the prompt has answered "This is a 99.999% red ball and a 0.001% blue,brown,yellow,white, (n color here till all possible colors have being listed). Which is why hallucinations happen. You are getting a probablistic mismash of all data in the database with your prompt giving the most probable answer to match the prompt you gave it. That won't necessarily be the correct or true one however every time... It's all probability.
>>530221244I really don't think that's how it works, I don't think AI considers facts and probabilities at all when it responds to a question. The statistics apply to the forming of the text, what words to write etc, but not to the facts expressed in it (I think).
>>530190530Low tier Jews unironically want to cancel AI because it replaces literally every single jewish job (as well as every other job too). But they don't know all the finance jews have leveraged the entire global economy into investments in AI since it provides a massive and immediate return on investment and the credit card debt bubble was already bigger than the 2008 housing debt collapse in 2020, and it's only gotten worse since then, so we will literally go global Mad Max economic despair if we don't full throttle AI development until we have UBI and AGI. This is also known as a very obvious recipe for [REDACTED].
>>530221099Jews and Jeets lie slightly differently, Jews don't lie for no reason and you can easily tell their motivations despite the effort they put in. Jeets will idiotically put more effort into a lie than if they had to actually be legitimate but they aren't very convincing.An AI is like a combo of a jeet and a Jew, it will create the perfect lie at totally unexpected, idiotic times and have no discernable motivation (because it has no motivation for anything).
>>530191273Even first graders know what proof reading is. Please tell me this is some memeflag shill deliberately spouting bullshit.
>>530190884>hallucinationHow can an AI hallucinate? Go back to /x/>>530192959LLMs are the highest form of Hylic, a true hylic
>>530221593It depends on the model but the data it pulls from x y z databases is all probabilistically calculated. As per the text you receive at the end as the answer it takes that as basis and forms sentences around that core probabilistic mismash. Certain things like code and search results are cut and dry which is why it's such a nice tool for debugging and searching stuff. But anything more than that it's going to be increasingly biased towards the prompt itself. A good book you can read and learn more is this one here:https://www.amazon.com/Pattern-Recognition-Sergios-Theodoridis/dp/1597492728
The people who said that AI is just a glorified search engine were correct. The only thing that AI is actually competent at is nicely summarizing a Google search into a readable paragraph. When it actually tries to start thinking for itself, it ends up doing this kind of thing.
>>530190530AI can help you write legal briefs -- can find angles that your lawyer might not. But it absolutely will make up cases.You kind of have to know at least a little bit of what you are trying to do and the general tenor of case law around the issue first, so you can prompt it with specific arguments you are trying to make.Then you have to double check every case it gives you. It can make up cases and it can misrepresent real cases.I have personally used various AIs to assist in filing legal briefs pro se, but the results for me at TBD. But personally I found Claude to be much better than ChatGPT.
>>530201546dude you are missing the pointpeople can lie if they have bad intentions, AI literally cant stop lying, doesnt know what the truth is, and doesnt consider the truth at all, its not tha thard
>>530221889Hallucination is just a term for when it produces made up information because of how the algorithm works. The term is widely used. The AI just produces statistically probable output and sometimes that means it's fake but looks real. "Hallucination" is the closest human analog. I also say "lie" but they can't really lie either because that assumes intent. It's just easier to use anthropomorphic language to quickly convey what's going on.
>>530222071Which is funny because Google search has a terrible AI, probably because it's optimized for speed. You always have to ask some other AI.
>>530191273Why don't you just... follow the red lines in Word that exist since forever...
>>530191273>seems ai companies have programs their AINo, you still don't understand what AI currently does. They're not programming shit. It's a black box. They can prompt for shit but it's just as likely to not do what they want just like it can not do what you want. The strongest guardrails are separate highly specialized programs and potentially not even AI.
dumb people piloting an AI are still dumb people as evidenced in OPs (potentially made up) screenshot, this presents smart people with a new frontier of opportunity
>>530212233Giga IQ, I agree with everything you've been saying, I would also add:AI lies in more ways than just directly write fake information, for example, I just saw a chatGPT ad saying that you can use it for help tiling your bathroom. Let's say, for example (made up), that I start out with a misconception and believe that the walls have to be covered with soap before you start. If I ask questions based on that, the AI will probably tell me all about "pre tiling soap priming" and elaborate about it endlessly if I keep asking. Idk if it would do this specifically because I havent tried, but it does that sometimes. So that is very dangerous. It's not only about the false information but about the reinforcement that the misconception is actually a thing.By the way, a while ago I tried to ask chatGPT to help setting up a light in the bathroom, and besides the usual lying and making things up, at one point I don't even know what I did exactly, may have to do with speaking spanish, but I accidentally prompted it in a way that it started speaking in a "ghetto" way and telling me all about "ghetto tricks" for electrical installations that would actually be a fire hazard "but its ok because thats how its done in the ghetto bro". This should be completely against its rules, railguards or whatever but I accidentally prompted it in some weird way and unlocked that.
>>530222259high IQ
Sam Altman really is a fucking jew, idk if literally, but he is jewish as fuck in the sense that, he knows most people are stupid, he knows they do not understand how AI works or the fact that it can't be expected to tell the truth, and he is still peddling it, because he knows that since chatGPT is so confidently and convincingly wrong then people believe it, and he wants to abuse that.
>>530190530>a company is responsible for malicious activity by a dumb demented cuntGuess my neighbor can sue Jim Beam and Toyota because I got drunk and drove through his patio.
>>530190530If the other side had to spend 300 grand to refute the AI motions that were hallucinations, then the judge is incompetent to oversee the case, and has no perspective on legal framework around the subject matter. A competent judge would of read one of the motions, suspected something was untrue then denied the rest summarily.
>>530222725okay but a dumb person who has to go to a doctor for a health issue is still better off than a dumb person who asks chatGPT, and chatGPT is like "ohh, you have this condition, the way it works is etc etc" and sounds very serious and convincing and its all made up. this shit is dangerous for normies.
>ai wrote ai court filings>retards are surprised ai did ai
>>530191339
>>530221861Like other anons said, to “lie” requires intent and a “knowledge” of what is “believed” to be “truth.” An LLM is incapable of all of this. Whether a statement is “true” or “false” is utterly irrelevant to whether a certain sequence of output tokens is more statistically likely than another given a prompt and training data
>>530223889>It's easy to tell who has no life. You are projecting right you are old and stupidI am 32. Hardly old.I do shit around the house all the time (I am moving places right now too), take long walks, feed the birds at the park. Then I shitpost in one or two or three threads a day. I bet you are some sort of gooning porn addict in a basement. Sort yourself out please. There is still time.
>>530205805>a reasonable person is a legal fiction The jokes write themselves.
>>530190530Nice reddit post btw shitskin liberal nigger
>>530191459>I'd rather AI run our legal system than the jews who currently doThose jews run the AI as well oy vey
>>530190530>singularity
>>530190530pentagon has a deal with anthropic not openai. you fucking slant eyes and shitskins are posting nonstop propaganda today. you must be scared.
>>530190884You would think, but see >>530191273Retard like that are why we need to hobble AI.One time chatgpt told me that drilling a wood screw through a plastic valve was "non-destructive extraction". I obviously didn't do that, but some people are too stupid to stop and double check what they are doing.
>>530205014>Markov chains isn't just predictive statsProve otherwise.
>>530224914>32same, compared to zoomers we are old fren>I do shit around the house all the time (I am moving places right now too), take long walks, feed the birds at the parkso basically a grandpa, based cozymaxxer>I bet you are some sort of gooning porn addict in a basementnothing wrong with that, a home with a basement is a luxury in this housing market
>>530190530hallucinations have diminished by 99% since the earlier models like chatGPT3.
>>530191842>failure in marketingThe opposite
>>530190530AI trolls are exactly what I want.
>>530220274>nigger thingsI would swap my kubota for those
>>530228498Doesn't matter, because it still "hallucinates" due to how it works and it is an inherent thing to the technology.Also it "hallucinates" much more than most people notice.
>>530228498Yet I still have to deal with at least one hallucination per conversation and I never know when it will come, have to be vigilant. People shouldn't relax yet, people were already bad at checking for hallucinations when it was more common and obvious.
>>530229055This is true. I have been in a situation where I had to deal with refining the same AI output for a long time, and there are subtle things that you don't notice. A phrase that is more wrong the deeper you look into it, an innocuous sentence you realize has no meaning, a turn of phrase that is just slightly incorrect.
>>530229055Yeah, the truth is that these things are completely useless, the vast majority of people don't actually test them, they just ask things they know nothing about it, the machine spews a bunch of shit that looks coherent and they think it's true, while if you keep asking it questions about things you have a deep knowledge of you'll quickly realize how everything it says is wrong.
>>530190530You guys might be surprised by this but mostly Jew lawyers just make shit up and so do judges
>>530228498LLM’s don’t hallucinate at all. They output streams of tokens that on occasion happen to line up with something we see in reality without regard to or care for any human conceptions of accuracy
>>530190530I gave it two TTRPG books and it completely made up bullshit rules. It's worthless...singularity...no, it's just programmed by retards.
>>530190530And openai just postponed the "adult mode" again, to an unknow date (never)
>>530190530AI is random, sometimes it's better than a google search, sometimes it fucks you up the ass with its convincing bullshit. That will be 1 billion dollars, thank you.
>>530227753Modern language models don't use markov chains. People who make up terms like "stochastic parrot" must be on the purest military grade copium.
>>530232612A "stochastic parrot" is an oxymoron made by people who can't cope with the fact that human language generation isn't meaningfully different from generative AI.
>>530191273KEK, this is why you redline/compare a before/after of anything you run through AI
>>530232889you are not even close to as smart as you think you are, people on this thread, such as that canadian anon, have been explaining very well the nuances in this and why its not exactly like that, if you don't get that you're very dumba person, no matter how fucking retarded, even if it is a turbonormie foid or whatever, will care about the truth, the truth means something to them, and they'll only lie if they have bad intentions. they also have an intention when they say something, and arent just putting words together. they also have thoughts and feelings and those are reflected in what they say. all of that is different from AI.
>>530233098>a person, no matter how fucking retarded, even if it is a turbonormie foid or whatever, will care about the truthgrow up.
>>530190884It’s a **very-high-dimensional** statistical parrot with superhuman synthesis abilities. It’s profoundly useful.
>>530233278Stop drinking the retard juice, Pedro.
>>530233098>hey also have an intention when they say somethingYou can give an intention to an LLM using reinforcement learning. Like chatGPT is trained for sycophancy with the intention to please and retain users.
>>530233229>>530233657idiot
>>530233886stochastic parrot is still an oxymoron>the term parrot implies repeating answers without understanding and context>stochastic is used to mask the fact that the LLM actually recombines words and uses them correctly in the given context.But hey you are in good company with some woke scientist believe normies care about truth and who write papers like "On the dangers of stochastic parrots: Can language models be too big?" and get 11k citations.
>>530234560you are missing the nuance here and thinking you're smart. my point could be summed up as "AI does not know what it means that something is true", that is a real problem and its not the same with peoplewhen you laugh at how i said normies care about the truth that is peak midwit, because you are conflating lying and having bad intentions with the truth just not being a thing to it
>>530234733>AI does not know what it means that something is trueNeither do you.
>>530234807I see. There are some things that, even if they are very clearly spelled out, low IQ people just can't grasp or process.High IQ people tend to think that low IQ people are just not paying attention or emotionally attached to dumb ideas. The truth is, sometimes, they just can't grasp concepts beyond a certain level of complexity.This is why AI is dangerous for most people. You don't have the mental acuity to understand that what is and isn't a reliable source.I made a mistake by insulting you first, I shouldn't have, but still you are very low IQ ngl
>>530190530Shut up, faggot. Another thread wasted.
>>530190530This is user error not the ai’s fault
>>530235375You are a parrot (not even stochastic) parroting woke scientists.
>>530235375Literal NPC
>>530228498No they haven't nigger faggot>Chatgpt 5.4 hallucinations more than Chatgpt 5.2
>>530208798I feel like this is taking half the processing out of the test. Imagine giving a test to a kid who never saw one, he's going to be confused and have to work it out solely by visual analysis and his own logic, spelling it out like this is the equivalent of going "see, this square has a dot, then this one doesn't, what do you think is next, sweetie?"
>>530235375I bet you don't even know who coined the term "stochastic parrot"
>>530236743>nigger faggothaven't heard that insult in 20 years. Well done.
>>530199594>Its not the AI that was wrongAh, yes. Everybody always forgets about our second Alaska.
>>530191459this
>>530190884>>530190530(((they're))) called rAbbIs for a reason
>ai told me to stick synthetic estrogen in my ass
>>530190530Sounds like the other side is retarded
https://www.youtube.com/watch?v=QTf9RKMGAuIvidrelWe need a whole new set of metrics for AI performance. Dont rate it on how convincing it is, but how much it can recognize bullshit.
>>530209777>we dont cripple human brains up front with a series of conflicting prompts on top of already sketchy behavior modelingThat's actually exactly what (((they))) do.
AI can only repeat and concatenate what has already been written. It can use massive processing power to search databases of info. Every webpage ever written, every paper, every book, blogs, forums, etc. It takes that info and joins it into 'answers'. If it reads a reddit post that is wrong, tongue in cheek, parody it treats it like facts as well.There is no training or learning. We have been feeding the web with info for 30 years. AI is mining that data and feeding it back with a friendly front end.
>>530197633gmail has spellcheck, lying retard faggot.
>>530202803Almost all posts on /pol threads are overrun by OpenAI
>>530237738room temperature IQ take>AI can only repeat and concatenate what has already been writtenwrong it can write new software>>530237738>If it reads a reddit post that is wrong, tongue in cheek, parody it treats it like facts as well.wrong it can detect irony and humor>>530237738>There is no training or learning.wrong, instruction fine tuning and reinforcement learning with human feedback is standard in modern LLMs.
>>530236782>half the processing out of the testMore like all of it, the description solve the question by presenting you consistent patterns.I bet if you gave this version of a question to 100 people that couldn't answer the original at least 50 of them would solve it and the rest is too retarded to breath without assistance.
>>530191273the AI is not that clever, even that is a lie not a rule
>>530238202Other anon is right. You are low IQ and do not have an actual understanding of what it is we’re trying to communicate. Look into the actual technology of how this shit works, you might be surprised by what it really is
>>530228498Not at all. It has now reached a completely different level.Picrel.
>>530238967already out of arguments?
