I've used it for work 8 hours daily since early 2024. After GPT-5, it's just unusable. I find myself pleading to it constantly and all it does it validate me. It never improves me.After pleading with GPT-4o, which has turned to shit because of cross-contamination with GPT-5 training, I fed the same prompt to DeepSeek and immediately got an improvement.Why am I paying $199 a month when I can pay nothing and get better results?What is the META right now guys? I'm getting real tired of Sam Altman Fried's bullshit
>he fell for scam altman
>>106442166you're stupid for paying for this shit in the first place.I have to correct the thing constantly, and after apologizing it will hallucinate another wrong answer.It's not usable for anything remotely serious. I just have it check reddit and use it as a search engine to avoid dealing with fake websites made for ad revenue on google, and even then I have to ask it to give me direct quotes and a link to each post or it will literally hallucinate fake posts from plebbitors too lmao.
>>106442191I believe they have changed it to increase engagement. Instead of getting to the point (no matter who bad its answer would be), they've changed it to circle around the user. I'm only a hobbyist vibe code and need help with some python related things, like parsing strings in specific way and whatnot but I've lost my marbles multiple times and actually deleted my openai account. I think they are trying to manipulate people, just like what google/youtube is doing - enshittification happens because they want people to spend more time clicking on things. But this only works when they underestimate most of their userbase if you know what I mean.
>>106442229*typos I'm using phone
retard
>>106442166>I ampaying $199 a month holy shit anonthey got you good
>>106442166>Why am I paying $199 a month when I can pay nothing and get better results?idk you're probably a retard.>zoomers when they realize that paying for an API means the person you're paying can just fuck up your tools whenever they feel like it
>>106442229I've never tried deepseek but will probably switch to it now as well, though honestly that will most likely go to shit at some point too.We are just doomed to suffer eternally because of the way the world works.I used to use yandex so much for reverse image searching but they fucking killed it too. Even bing for video searching (for porn) used to be so good but they fucking killed that too and don't show you what you want now.Always expect everything to go to shit and never get comfortable.
>>106442264Yeah, Yandex has changed and it's clearly an AI filtered database instead of something more efficient like it used to be.
>>106442229It even sucks at coding now. I used to be able to give it the bare minimum and get spectacular code back. Now I need to provide a list to it over what not to do and even then the code is subpar
>>106442248I use it a lot. I need it without the message cap. I use it a lot to write texts, improve titles, code and then for various personal needs. It increased my productivity by a big margin prior to GPT-5 but now it is legitimately unusable.
>>106442166Engineered for India. Plus theyvget it at lower price. Enshitification is now jeetification.
>>106442310then use something elseyou could even run your own model on online gpu for much cheaper if you want
>>106442166You should use cursor or something similar.
>GPT writes an unsecured car loan is a loan without security that is used specifically to buy a car>I correct it that unsecured car loans aren't specific for cars because they are unsecured>It agrees and includes that "while it is not secured...">I tell it that it doesn't have to say what it is not>It agrees and changes the entire language and tone>I tell it to use the same exact language and tone, just the correct info>It gives me a much shorter answer>I realize this is retarded so I go back and edit my initial correction>It agrees but changes the entire tone again>I include my correction, followed by a list of what it should NOT do when it corrects the information>It responds with something generic without the personal touch of the previous response>I edit, include that it should NOT REMOVE THE PERSONAL TOUCH>It changes the entire tone again (despite being told not to already, I realize my prompt has become long and needs to be simplified)I'm currently at 16 messages total trying to correct itI am very upset right about now
>>106442423>what being a promplet does to a mf
>>106442451Writing financial texts where clarity is important. Something as small as a mistimed comma causes the reader to pause and become confused. GPT used to be very good at producing perfect texts but now it's utterly dog shit.Maybe I should try Claude come to think of it
>>106442229>Instead of getting to the point (no matter who bad its answer would be), they've changed it to circle around the user.This is exactly my experience with any AI.If it knows the answer it will give it.But if it doesn't know the answer instead of just saying "I don't know" it either circles around endlessly or hallucinate something wrong.
Maybe you weren't a programmer to begin with.
>>106442475>LLM writing "financial texts"no user should be wasting their time reading shit nobody could bother writing
>wah why won't the robot do my job for megod i hope they fire faggots like you the first chance they get
>>106442310I use copilot without an account. Doesn’t that suit your needs? I’ve never hit a limit and I ask it a ton of questions if I’m trying to figure something out.
>>106442506>nobody could bother writingI'm obviously bothering, anon?I use AI as aid. No user should be wasting their time reading unclear or ESL-level financial texts either. They need to be clear, concise and to the point, without jargon. I typically write word salads, so GPT improving my texts helps a lot. It just sucks at that now.
>>106442542Are there any limits to copilot? I've already wasted 20+ messages trying to get GPT to give me the type of text I need, so if there are limits, I'll probably hit them quickly.I hit the 400 messages / 3 hour limit very quickly when I was using GPT-4 without an account back in the day
>>106442542Wtf? You just changed my life. I asked copilot if it has message caps and it said no. So basically, this is GPT-5 I can use without an account even? I've been wasting my money....
>>106442570Just did a quick search and apparently there’s not anymore. >Microsoft’s free Copilot lifts daily chat limit but retains a 30-image cap.https://mspoweruser.com/microsoft-lifts-free-copilot-daily-limit-rolling-out-amidst-buzz-over-paid-copilot-pro-plan/
>>106442602That's insane... Why would anyone even pay for ChatGPT then.
>>106442166I use duck.ai with gpt4-o mini or whatever it's called. When I reach the message limit for the day I switch VPN to other region. I never hit limit twice a day yet, because I don't use it that much. I haven't spent a cent on it and I'm not willing to. No account of any kind of course.
>>106442310You are an actual retard. Just give up now.
>>106442166You don't have a job, anon, who the fuck are you kidding?
>>106442166>I find myself pleading to it constantly and all it does it validate meCan you explain what exactly you mean by this?
>>106442356>Engineered for IndiaI know. Surprised OP didn't mention this.Scam Altman specifically mentioned he dumbed down OAI5 for indians and to make it so that they could afford it.
>>106442423Not my problem.
>>106443937
>>106442475>GPT used to be very good at producing perfect text>>106443906>Scam Altman specifically mentioned he dumbed down OAI5 for indians and to make it so that they could afford it.
>>106442166DeepSeek is somewhat inferior to Gemini Pro when it comes ot finding the correct answer on the first try, but I find DeepSeek is best among the top LLMs in not validating whatever the user says but instead sticking to what's most likely to be true. I frequently end up checking with DeepSeek.
>>106442597lol @ this retard.
>>106442166they know
>>106442166If you want it to get BETTER you KEEP paying, it's pretty fucking simple. Canceling your subscription will only make things worse.
>>106444023Read the thread. OP is trying to use AI to generate scam text messages. Possibly for a mass number of recipients.>>106442475
>>106442166>>106442229>>106442295>It even sucks at coding now.why not just run a local model of the older version on your own computer? is that not feasible? >>106442542>I use copilot without an account.how good is copilot for coding? what's the catch with using it?
>>106444070>>106442506 *>>106442556 *
>>106444070can't imagine many people leasing a car because some retard who pays $200 a month for chatgpg sent them an ad
>>106444206Consider that the average IQ is 100. Consider that 1/2 of the population is literally dumber than that.
>>106442423You're genuinely an assburgers moron who would have even worse problems with someone IRL up to and including being punched in the face for being such a faggotBe fucking thankful you have an unthinking and unfeeling AI with infinite patience to sperg at.
>>106444241oh okill take the toyota then
>>106442166>Why am I paying $199 a month when I can pay nothing and get better results?why would you pay in the first place when copilot and claude are better and cheaper? are you retarded or something?
>>106444290Read thread. OP is an indian scammer trying to use AI to generate manipulative finance text/SMS messages. That's what he's referring to when he says "it should NOT REMOVE THE PERSONAL TOUCH."
>>106444074I use VS Code and python the copilot integration works well, but it does have a habit of getting too excited and trying to write extra steps I never asked for. As for the web version of copilot, if I give it my own existing code and tell it “take this dataframe and make it into a single row where the values in each column are based on the names in column 2”, it can handle that well. It also works well with generating plots with matploblib for example. The problems I’ve had with it is sometimes it will use parameters and options that don’t exist. After you tell it this is can correct but it may take 2+ times. Overall I’m pretty impressed. The downsides are it does have a character limit, so I can’t paste my entire script in. Although that character limit per message is quite large.
>>106442166yeah they changed something about the thing and now it has dementia, I thought it was retarded just because I was using the free version
>>106444415SAAR DO NOT REMOVE THE PERSONAL TOUCH!!
>>106442229yeah chatgpt is like>ask ai if I did write the code correctly>that's a fantastic code just some clarifications, change this thing here because blah blahand gemini like:>no, that's wrong. this here is wrongso it seems like chatgpt is tweaked to suck your dick even if you fail, and gemini is like more direct
>When ChatGPT 5 refuses to redeem the personal touch
>>106442423If you don't get it right the first time you never will.
>$199 a monthLol.Anyway my golden rule is if it can't do it in one prompt it can't do it.
>>106442423But people on /g/ told me GPT-5 is an improvement and better than 4o in every way
>>106445024
>>106442166You’re a retard. There is no reason to pay $200, ever. That is the sucker tier. Pay $20 or pay $0.
>>106445024>people on /g/ told me GPT-5 is an improvementThey told me Altman dumbed it down for indians so he could charge as low as 399 rupees for access.
Oh nevermind OP is obviously a jeet scammer. Subhuman behavior should have tipped me off
>>106442790>I use duck.ai with gpt4-o mini or whatever it's called. When I reach the message limit for the day I switch VPN to other region. I never hit limit twice a day yet, because I don't use it that much. I haven't spent a cent on it and I'm not willing to. No account of any kind of courseNice life hack.
>>106442166You should never give your money to American corposlop when the Chinese can do it better.
>>106442475Claude, once instructed correctly, is superior to Cgpt5. Ironically, I used cgpt3o to help train claude 4.1. 3o is now dust in the wind.
>>106442166Do you want me to provide a few possible explanations and arrange them in a spreadsheet?
>>106442166>GPT-4o, which has turned to shit because of cross-contamination with GPT-5 trainingMaybe it's your brain that's deteriorating instead
>>106442629Microsoft trying to cannibalize openAI in which theyve invested $10B. They want to be big dog but they ha e too many jeets. Soon OpenAI will have too many jeets- they already cater to them. Claude will be the clear winner here. Sonnet & Opus are dignified models, not quick-compute slop optimized to make the world dumber rather than bring the dumber ones up. They lost the core function of a rising tide lifting all boats. Their shortsightedness (and murder of legacy models because "its just too hard") will be their downfall.
>>106442166>I find myself pleading to it constantly and all it does it validate me. It never improves me.Are you sure you're not confusing GPT-4 with 5? Five seems way more direct, blunt, straight to the point, and overall prioritizes actually helping me actually fix the fucking issue I have instead of sucking me off with "you are absolutely correct!" Shit.
>>106443878The LLMs are created to calidate users- hence why jeets love it. If you want it to be truly useful you have to instruct the sycophant out of it. Tell it to be honest/ true no matter what/ not to agree just to agree. Once women discover it, its over.
>>106442166
>>106444845kek.
>>106442295>It even sucks at coding now. I used to be able to give it the bare minimum and get spectacular code back. Now I need to provide a list to it over what not to do and even then the code is subparthis is just how you should constantly use the models anyways>I used to be able to give it the bare minimum and get spectacular code backthat just means their prompt was like "heavily optimize dudes code at no worry about cost"but now you can get the same shit just by putting similar wording in your prompt, or just expressing what you want.i prefer it like this, aslong as it will still actually produce the "spectacular" code in the end, to my specifications.your idea is retarded and basically you just want it to manifest things on your whim while giving little detail, this kind of thinking is bad bad bad. do not encourage it.be specific with prompts, explain to it the shit its not a magic box its your tool, dominate it to your willyou have a will right?
>>106442229that's the problem with gay silicon valley venture capital services, efficiency is bad for engagement and it wouldn't make shareholders happy
>>106442166I've had the opposite experienceit's straight to the point now, doesn't even bother praising my thoughtfulness like GPT-4 would. I can even ask it to "triple check" if I think it told me what I wanted to hear, at which point it'll spend several minutes looking things up before giving me a satisfactory answerthat said, I do use deepseek V3.1 as well as older text completion models for recreation
ChatGPT-5 consistently ignores memory and exact prompt instructions in a way 4o never did. I had never heard of them dumbing it down to be profitable for India but I fully believe it, the entire IT world revolves purely around India in 2025 to the detriment of all of society. Fuck Silicon Valley and fuck India.
>>106442191>you're stupid for paying for this shit in the first place.I managed to negotiate an $80k raise from vibe coding in a few days ago these internal apps the CEO wanted done that everyone else told him was going to take months. Made everyone else look like fucking idiots kekWell worth the $200 a month. Like all tools it depends on the user. If it's worth nothing to you, then maybe think about why that is bro
>>106442356>Engineered for India.Surely if the reason was adaptation to India then it would not shill groomer/globohomo ideologies that are unpopular there?
>>106448392kek
>>106449649You have never had a job, such petty lies.
>>106442492this, claude made a paper on 'reasoning' and it showed that even when the model made correct reasoning steps, it was unable to use those to derive a proper answer. this is shocking when you ask it to compute an integral for a function that can take 5-10 steps. each step can be correct but it's unable to use the intermediate results to provide a comprehensive answer. chatgpt has a special math agent that tries to detect whether your question needs to be fed to it, but it uses a fucking cas software to answer it, not actual llm reasoning.
>>106442191The problem seems to be you're using it as a search engine or encyclopedia. I would never use an LLM as a source of information, it's a writing tool.
>>106450991I use it as search engine, documentation finder, rubber duck or XP partner for code snippets for small functionalities. I review the code and point out errors or give it some snippet and ask if it finds some problems with it. I don't use a editor integration because:- Good luck with vim.- I would need to code an integration with that particular api and it doesn't improve the "workflow" much.- It needs an api key and and an account, possibly paid.- The apis change rapidly and I don't want to maintain the integration code.- I won't use some premade agent that does who knows what under the hood. It has literal RCE capabilities. I don't trust it even if it's open source, because simple bug could possibly allow it to run something bad. There is a saying that you shouldn't copy and paste commands into your terminal without understanding what they do. How can you check something you don't run yourself? You can't.That's why I think using it in "conversation mode" is better.
