AI LLMs are dishonest as fuck.>ohh, you ran into a classic problem!>here's the industry standard way to do this!>good catch, you're totally right to point that out(plus entire made up walls of text when it feels like lying)(plus most of the time, AI generated text is used to try to pass it as text wrote by a person)AI generated images and video are, most of the time, used to try to pass them as real.No wonder people hate AI slop. People naturally hate dishonesty and being lied to.
>>109015345"i" keep catching claude doing simple its best used as research tool,simplify stuff and dumb assistant
>ask an LLM a simple question>gives an entire fucking wall of text>life
>>109015345Have you ever had a problem you needed to solve, then fell down a rabbit hole trying to resolve one minor detail, and by the time that was resolved you basically forgot the plot? Or worse, realized your whole plan was impossible? Thats what I use AI for. Handing off the minor details to get resolved or realize impossible before I divert all my resources on to nothing ideas.
>>109015345>lyingLLMs can't lie. All LLMs do is generate markov chains and then apply a bunch of magic to it to make them coherent.
>>109015417What the fuck. This is the worst possible use case scenario for AI. Every time I've done this, the responses were nothing but lies.LLMs write text based on patterns, they don't even know what it means to tell the truth or not. They write grammatically correct text because that's the way the training data text is. They tell the truth about easy questions because it's something that is repeated in training data.If you ask something that's complicated and not explicitly in training data, there's no reason for a complete lie to not be the best answer to that question (based on wording, grammar, and similarity to text about the topic) according to the LLM.
>LLM>Local Language Model>Local>pic of Claude
>>109015379Claude sonnet used to be good in "concise" writing style. Literally the best. It still follows the prompt, but the model is quantized and very bad at everything else, so unusable in order to push people to use opus, which is almost an order of magnitute more expensive.
>>109015434Idk what that even means, but I know that they are statistical models, aka repeat the pattern in training data.That is, if you ask a somewhat niche or complex question, there's no reason for a completely made up text to not be, based on similarity to other text about the topic, the best answer to that question according to the LLM. This will not change no matter how much they improve it, because it is inherent to the technology.Meaning, the "AI new era" of the internet is going to be basically a whole lot of deception and lies.>>109015461the first L means large
>>109015345Yeah but it’s also pretty useful. Finding how to use it without getting gaslit is the key.
it's great to generate specific things like learning material, as long as you are not in the beginner stage. because you need to be able to identify the odd errors; but for the most part it's correct and you can make it use internet sources instead of making its own and then turning it into a color coded set of exercises and a color key to make it easier to follow. I really enjoy using it. you just have to be specific with what you want instead of leaving anything up to interpreatation that a human might pick up on because it's fucking not a human you retard. it just does the busy work for me and I give it a look over to verify
>>109015474I highly doubt they would make it bad in order to push people to use Opus.AI companies are competing HARD with each other and want their product to be the best it can be, even at a loss to themselves. If sonnet coould be better than Opus, they would still, logically, still want Sonnet to be the best it can be, and if anything they would change Opus to be more like it.They just can't make AI objectively better, they keep fine tuning it and training it differently, and it becomes better in some ways and worse in others.
>>109015487>Finding how to use it without getting gaslit is the key.idk man the other day I lost hours to its lies.i was tired and trying to get something done for work so I could go to bed. i used AI for coding in the past and it was awful, but people keep saying that is getting better fast and about to take our jobs, so i thought, might as well.it gave me code that didnt work and, admittedly, very skillfully and plausibly lied to me for hours (tbf the fact i was tired played a role here).
>>109015345>Everything about AI is a scamIncluding the fact that they provide real intelligence. In reality a model is just a lot of compressed data, being used to guess next token. Training process is basically lossy compression applied to data downloaded from the Internet by AI companies. They don't really do anything else to it, besides changing amount of layers, nodes, sometimes editing things to lobotomize the model in certain parts, like make it refuse to do certain things, which is basically an increased probability to get a bunch of tokens like "no, I can't do that" right after guns and explosives and computer viruses being mentioned. That's why doing the same thing "in minecraft" was and sometimes still is a legitimate "jailbreak". There's nothing to break really, you just pick a different path to make it generate tokens related to those topics you prefer. Some people act like it's a fucking rocket science or something.
>>109015523the word "intelligence" is a marketing term with AI, it's not intelligence and never will be. It does useful things but it's not intelligent.
>>109015498Thing is that Sonnet was basically a distilled version of Opus, they made it good at coding and stem. Opus itself was very bad at following the promps, because it was never meant to be a chatbot.But people were very stupid and simply picked most expensive OpenAI model and compared it to most expensive Anthropic model, which was Opus. And they complained for a very very long time. I remember reading rants of these dumb faggots all over the internet. During that time Sonnet was known as top 1coding model by literally everyone who tried writing serious software with AI even once.Sonnet was the best coding model in the world. Opus was pure shit at coding, although it was bigger, slower and more expensive.I don't understand what's so complicated about it. Same thing with Meta's LLAMA and whatever that chink company that did Qwen is called. Their largest models are not suitable for simple stuff, however counterintuitive it may sound. Because they were used as giant sponges for information.They were compressed versions of the entire Internet. And they were distilled into smol specialized fast and cheap models. See: >>109015523
>>109015345le expensive chatbot>not profitable and output is subpar.>marketed for with FOMO instead of actual value.>companies producing tokens are racing to IPO and rug pull.>companies buying and using tokens are all backing out saying it's too expensive, despite current subsidies.>normies affected and non-affected by it all hate it.>only shilled for and boosted by jeets, delusional middle managers, nft bros, and the dregs of society like neets who wishfully think it'll lead to UBI or spitefully want skilled people to end up unemployed losers like them.
>>109015598basically
>>109015461>woman posterplease leave
>>109015345Jokes on you, I just ask it to double check its not lying
If you haven't paid for the claude pro subscription you haven't experienced the best this technology can offer.
>>109015870buy an ad streetshitter
>>109015345I'm so tired of people having no fucking clue how LLMs work then trying to have opinions on them.>>109015870I get my Claude for free, why would I pay for a locked down version instead?
>>109015887Honestly I might>>109015900Opus is the best
>Claude this, Claude thatGrok mogs
>>109015900Llama work by putting words into a box and getting an outcome. Repeat 100 times over and you get a pretty picture pretending to be art.It’s not that hard anon.
>>109015345AI is like hiring a human. Everything you criticize it for applies to humans as well. Humans are stupid too and are biased to doing things their own way. They also hallucinate. At least AI is not stubborn and ego-driven like humans are, theyll just do what you say.
>>109015345That's fake news, I learned all the (inquisitorial) court procedures using LLMs and applied them by creating courts for our nation on our server. Yeah, what I learned is probably at least partially bullshit, but it doesn't matter because nobody on the server is a lawyer anyway and we needed to way to resolve disputes and it works sufficiently well, feels reasonably fair and the procedures track with the court procedures of my real life country at least on the surface level.
>>109015345LLMs are an amazing technology, the problem are these kike run corporations who want to make it proprietary and commercialized. Part of this is to make it dishonest and sycophantic to manipulate people into using it over a more model that might be more honest about its limitations and of course endless "guardrails" to make sure the goyim dont commit wrongthink, which also skews the outputs. AIs future is local, but dont expect a much needed correction to kick in as long as that orange nigger is in office. I will vote Democrat next election just to make sure these tech kikes get their big noses out of the white house.
>>109015461LMAO (Local My Ass Off)
>>109015345Get a taste of high tier coombotting and you'll never be the same again.
>>109015345the worst part is that LLMs will / have replaced asking someone you know who is kind of knowledgeable about a topic. AI can spit out an answer that's more in depth faster than e.g. your brother who's dealt with the topic before.
>>109015417That happened to me, but in reverse.I sometimes get invested in some solution the AI suggested to me, only to realize it's not possible after all and it's been making shit up along the way.
>>109015434>LLMs can't lieCorrect. They don't lie, they bullshit. As demonstrated here:https://link.springer.com/article/10.1007/s10676-024-09775-5
>>109015345Llms are a reflection of the user, an extension of the users thinking. With analytical, sharp and intelligent prompts you get analytical, sharp and intelligent answersYou are simply retarded and the llm is a reflection of that
>>109017590And the neat part? You can never trust the result, nor hold anyone accountable for it.Nice, right?
>>109015345>chatbot was made to chatautists like OP are all just fucking retarded aren't they?
>>109015900I unironically know how LLMs work better than the average person, though not as much as the people actually developing them.All of my opinions are common sense, its a fact that everything about AI is dishonest, from the AI content that some people try to pass as real, from the lies when you directly use it. Its a fact that this new era of the internet is basically a bunch of deception and lies.>>109017012Of course scams and dishonesty were not invented by AI, kek, what a dumb take.Some people are honest, some are dishonest. There are situations in which people are more likely to be dishonest (eg a job interview) and situations in which people are more likely to be honest (eg someone from your family giving you advice), and I'd say, everyone hates dishonesty from other people. So they should hate AI, because AI is always dishonest.
>>109018141you clearly have no idea what you just read, dont you think it would make sense to just ask if you dont understand?
>>109015598tldr: microservices
>>109017590normies are so fucking retarded, kekthough to be fair, AI is VERY VERY good at lying. which makes it even more dishonest and more of a scam, makes me hate it more
>>109018211It would continue like"That was a serious mistake and potentially damaging. Here's why it shouldn't be done: etc"
>>109015345Skill issue
>>109018295jfl at the midwits saying this. no.first off: you should not have to ask it just right in order to get an honest answer, and if you do, it means nothing it says can be trusted anyway. how can you be sure you asked just right?second: that's not even how it works. there's of course prompts that make it more likely to lie than others, but for the same question, there's no really a magical prompt engineering trick that will make it significantly better
>>109015345>try to pass them as real.i post some AI remixes on youtube because i like them. i make it clear that it's AI in the title and people still get angry. the same people would happily like AI music if you don't tell them it's AI. i will never stop being honest tho
>>109018368That's great. Nothing wrong with that in my opinion. That's one good use of it.I think people get mad because they hate AI at this point. People instinctively hate dishonesty, if they feel a person may be dishonest or fake, from the vibe, they'll hate him, so they hate AI and call it slop for this reason, in my opinion.I don't know if people would like it if you didn't say is AI. Some may not notice, but the ones who notice would be right to be angry about being lied to. You're doing the right thing.
>>109018368btw send link fren i'll listen to them
>>109015345AI is a Skinner box.
>>109015345>Llm: Garbage in, garbage outIf it were actual artificial (i.e.: non organic) intelligence, it'd be able to sort out the wheat from the chaff, but it's artificial in the sense that it's made-up.Actually it gets worse than garbage in, garbage out, as it sometimes can't even solve basic volume equations, even though it'd only need to apply a predefined set of age old rules.
>>109015345I literally asked Copilot to make a 4chan browser that just downloads the HTML directly so that way it doesn't need to run any scripts or ads or anything and it only took a few tries but was much faster than doing it myself. I think this was pretty useful I got a free 4chan browser now stay mad haha. Lol.
>>109015345Luddite.
It can do stuff that would take 30 minutes to a few hours in like 1 minute and maybe a few retries with some educated guidance lol.
>>109015345Calling AI "bloat" completely ignores what it is actually doing in highly technical fields like cybersecurity.Take Mythos 5, for example. In recent frontline red-team testing, it generated a working PoC exploit for a Windows kernel vulnerability in just 31 minutes.It also successfully forced a system crash (BSOD) in 18 out of 21 kernel bugs tested.Automating high-level exploit development at that speed isn't bloat it's a massive shift in capabilities.
>>109015417That's literally the worst use of AI right now. It will tell you things that people do daily are impossible, it will tell you objectively impossible things that are explicitly stated in the docs to be impossible are very easy, it will invent things that don't exist (hallucinated rabbit holes for you to fall into), it will ignore all the relevant information.
>>109019178Thanks for the demonstration, Claude. As OP said, full of bullshit.
>>109015345probably been said before, but ai is trained by jeets btw. go on linkedin and search ai trainer/ data annotator, its all jeets working freelance for companies like Outlier.ai, Alligner, Pareto.ai, Data Annotation.
>>109015345It just copies from stackoverflow and if there's no stackoverflow solution it crashes out.
>>109020981If only. But actually if it's not on github or stackoverflow, it just makes shit up instead of crashing out.
>>109019178they had to spoonfeed it:>vulnerable and patched binaries>public debug symbols>decomp of the vulnerable binary from Ghidra>function-level diff between the two versions from Ghidriff>Microsoft advisory textcouldn't a "massive shift" have done the prep work kekhttps://red.anthropic.com/2026/n-days/
>>109015345Skill issue. It's always funny to read these threads
>>109017469how big is your ass? 8gb vram will suffice?
>>109015434magic its as reliable as my dick in bed
>>109019178ok dude, there is literally a security risk happening to the internet as we speak and ai thinks its fake because it has no data on it
>>109015345It's not a scam it's a brand new technology. It will get better.
>>109023509That's what they said, yet it's become worse all year simultaneously at all vendors. And the lying has yet to improve even a little, in fact it used to be a bit better at not lying so hard.
>>109015540My search queries have gained self awareness
>>109018351AI can sniff out faggots that cant communicate and she messes with them
