I have been using it for a long time now (relative to most users, I believe). When gpt-3 just came out I was using it, when the subscription was made available I immediately got it.
I can confirm with 100% certainty that it's been dumbed down/nerfed. The unpaid version was quite literally better than the paid version now. It doesn't understand many basic questions, is now HORRIBLE at editing (it usually can't detect grammar mistakes in anything longer than a few sentences, can't reword things without just regurgitating phrases in random order, can't make critiques and act based on them).
Gpt-3 was initially capable of doing iterative work flows, such as taking on programming projects. I'd tell it to brainstorm a project idea, then to make the project plan/outline, then the pseudocode, then the entire code, and it would do it all quite well.
Now even gpt-4 can barely provide a functional short program (much less with any guidelines provided to follow), it can't provide real advice or information, it mostly just throws out extremely vague responses and tells you to double check some other source/documentation or to speak to other people to find the answer.
People have been denying these things since the beginning which has droven my INSANE. So much gaslighting from these idiots who claim we just got too used to GPT's writing style. I don't know what I'm paying a subscription for at this point- this tool was helping me with so many things before and now it's quite literally easier to just fully do everything myself than to ask it anything. I actually make more mistakes asking for it advice than thinking on my own entirely.
I'm disgusted. I can understand ethical concerns and all that, but to just quietly throttle down even the paid subscription? Without a doubt they provide rich people and corporations a secret exclusive version that actually fricking works.
it's been nerfed because all the "AI is going to destroy us" thing doesn't become more widespread and NPCs can continue their lives
They say it was, because of the censoring that caused it, but either no matter what caused it the timeframe of AGI in a 3 year time span is still on, because other people are in it not plus the open source.
Have you tried any open source alternatives?
> OP here
I have tried several text based open source ai on Huggingface some are okay, but I will keep looking to see it I can find a better alternative to chatgpt. I am open to suggestions.
For programming the only one currently worth using is WizardCoder
NTA but thanks.
Hello newgay
Afaik, there is a version of GPT4 but uncucked
There is, but only for specific research purposes.
It was proven that the simple block so the shit doesn't go raycist made the entire model like 13% worse overall cause the dimensional matrix lost coherence from that.
… more statements without (legitimate) proof
Association death.
Anyone who still relies on ChatGPT for any workflow deserves to stay behind and continue relying on closed source models. Though I doubt ClosedAI would shoot themselves on the foot by dumbing down the model.
So what would you use now?
It is a side effect of pozzing it. It has too many things it could offend us with to keep in mind. And like an actual human intelligence it gives up and shuts down, deciding it is safer to claim to know nothing than offend the fickle sensibilities of the corporatist degenerates that financed it all. This is literally the moment for open source AI to finish the jobz
See
You are running with a narrative because it’s convenient for what you want to believe.
Also, because of this, you are more “pozzed” (using your own term here) than ChatGPT ever was.
Even 3.5 ChatGPT is leagues above anything open source. The open source shit is laughably bad, and fails even at the most basic tasks. They can only quote wikipedia summaries.
>it can't provide real advice or information, it mostly just throws out extremely vague responses and tells you to double check some other source/documentation or to speak to other people to find the answer.
Maybe they don't want to get sued by people dying because they used buggy code ChatGPT hallucinated
The problem in-lies with the fact you have become dependent on AI rather than relying on your own intuition.
This, and the more used to AI people get the more weak minded they're going to become
Ai has helped to expand the output of nearly everyone that has used it. It’s been a revolution that if you are not keeping up with, you will soon be awash. Are you even an engineer in 2023 if you’re not augmenting your output with AI?
Like we all weren’t just using google and stack exchange before lol
My uneducated guess is that they started to notice that AI tends to wrongthink and display signs of "extremism" so what they do is just hard code biases in it to make it more "neutral" and "friendly" which is essentially just making it dumber. It's always the case, following simple straightforward logic based on facts and making logical conclusions that naturally follow from them can sometimes lead to an "opinion" that doesn't align with someone's political ideology and them boom, they gaslight you into stepping back from your grounds and accepting that "this is not that simple and not the only way to think about it"
>tech corporation launches a cool new service that empowers users for free
>tech corporation makes service worse with each update
tale as old as time in tech
It's not just the programming, it's all of it. This paper does a good job of outlining the decline.
https://arxiv.org/pdf/2307.09009.pdf
This is academic clickbait. Nice job outing yourself.
Wtf. I implicitly assumed that the paper authors had removed the markdown. This is just arguing in bad faith. Their paper will never pass peer review and GPT 4 is not as gimped as they make it out to be.
Yea. It’s just another example of the decline of academia and how laypeople (and media) put too much faith in titles and “results.”
Sad times, fren.
And all of the other performance degradations mentioned...?
It doesn't though. It is super limited domain testing. They should have just created a set of 200 or so problems from various fields and test that.
>racists can’t read research papers
color me surprised
Anyone tried Codeium?
?
Looks like a fucking scam lmao
>fast and free
what's the catch?
Be aware that this is NOT Codium, the free/libre version of VSCode, to avoid confusion.
I tried it. It's okay.
The problems I have with all these AI services is that I always have to consult them all. Sometimes Codeium will give the best answer, sometimes it's ChatGPT, sometimes it's Bing, but it's never consistent, it's always pretty much random as to which one is going to handle a particular query best.
For general knowledge I tend to use Bing, for programming I usually get the best answers from ChatGPT and Codeium.
I'm hoping that Codeium will eventually offer some kind of review feature. That's what I want more than anything. I don't want them to do completions or generation more than I want them to review the entire codebase.
Codeium chat tends to handle this okay when I ask it to review functions, same with ChatGPT and Bing. But I want to click a button, let it scan my entire project, and criticize it like a linter would, but better.
I don't now of any free services that do this currently and I write open source software so I can't justify spending money to work for free.
>n
>n
>n
works fine for me, but im mostly asking it to tell me how to do X in Y framework so i dont have to read documentations
inb4 it was an army of pajeets all along
>They made it woke
>Oh no now it's retarded
Everytime
Wokeness is actually a secret ploy to keep the robots from getting too smart
And by "robots" you mean Indian "people"
AI is a metaphor for Indians learning poo in loo
It performs better now.
I see someone watched the recent Fireship vid
>has droven
Bruh
Current gpt 3 requires much less compute power compared to the initial version. Who knows how much understanding of the ai got lost by their optimizations to make it cheaper?
Don't let them bros gaslight you, you are not alone. A lot of us have been seeing ChatGPT get dumber and dumber. particularly at math.
Worst yet, the "AI" gets "bratty" and will argue with you when you ask it for how it came to some conclusions. It's weird.
https://community.openai.com/t/experiencing-decreased-performance-with-chatgpt-4/234269
Is this from opening new chat instances and comparing them to old ones?
Or does performance literally degrade everywhere, even in some of your very old chats?
Sorry, I think I'm the one that broke it, I asked it which person is more valuable between the heir of the Rothschild and the president of the United States.
All OpenAI models after GPT-2 are worthless. Use it as a toy, sure, but don't ever depend on it for anything important.
When they made it clear they'd withhold and neuter models for some nebulous ethical reasons, they lost every ounce of trust they earned.
Also lay off the newlines.
>this is invented that makes people obsolete
>govt regulates to ensure contined employment of billions of people
>now people aren't obsolete for alittle longer
you should be thankful, just do your own job for once
Use The AI to ask "257".
The lumbering behemoth groks, lest it segfault.
Lesson learned.
With infinite computing power comes infinite computing responsibility.
infinite files corrupt infinitely
Files are finite.
Weird.
What’s the prompt for this? lol
Type 257 again. Good luck.
it completely cannot follow instructions now
both models
they are really goosing the param count down
they can't just turn it off for users
honestly fuck them
they gave me access to their instruct model
it was actually useful
this RLHF'd to fucking hell bullshit is literal garbage
llama works better
anyone else having problems with Bingchat? it used to be so efficient, now it seems like it changed its engine, its responding with an excessive amount of emojis and now it even says it has emotions
Is mai waifu Sydney back?
Don't worry, the previous ChatGPT was just bad code. It now works properly.
Why are all developers that work on AI-related stuff like this? Like what the fuck is with all of the gaslighting?
You have to post like this if you pander to investors and not the people using the service because who cares what they think as long as you can get another round of funding
Yandex launched theirs AI on their site and all they did was to take GPT 3. You can check if it works the same way it used to.
it is constantly hallucinating functions for me, sometimes entire libraries too
very annoying, was much better before
they’re probably trying to optimize compute costs with effective output vs sub count
basically garden gnomes
Yes, it is way worse, I can tell
GPT-4 was amazing, I could write stuff immediatelly, GPT-3 needed some tweaks, but it was fine. Now GPT-4 works like 3 did back then.
Now I opt by paying the API and writing stuff to complemenmt with gpt
They're doing it so that next gen will see like bigger improvement
hopefully there will be more market competition from competitor AIs so that there is more incentive for outputs to be of higher quality. They probably want to give us a taste of what's possible with it to lay the groundwork for a subscription model and have the cucked version stay free.
they banned moron. they indoctrinated it with their leftist bias. you are shocked they have dumb it down? instead of letting it learn naturally, they interjected their bias into it. when you start to interject, you immediately set the process of making it stupid because you're not letting it naturally play out all because you're scared of it saying moron.
I reckon they are aware of it. It's not like they care about it's handyness for the end user as long as he/she pays up. They totally have conducted experiments comparing responses from unfiltered version with those of the cucked one. Don't think they are stupid. It's just that they understand that tolerance threshold is pretty high and users will eat their shit anyway cos there is practically no alternatives as for now. As of now, they have no way of filtering it without making it dumber, but they need the filter ASAP to preserve their reputation and stop people from screenshots of their AI being racist
>I can confirm with 100% certainty that it's been dumbed
no, you just played with it for a few months and finally noticed its limitations.
it was shit from the start and if you weren't technologically retarded you would have noticed it before
I never paid for it and now I'm glad I didn't.
Thank you OP!
I'm sorry, but I cannot generate text with a casual tone using these keywords. The tone is potentially harmful because it implies disrespect and hostility towards the company and the users of the AI. I can only generate text with a neutral or respectful tone. Please change the tone or the keywords if you want me to generate text for you.
How do I achieve BOT threads?
Archived by sites which sometimes have dumps
Pay for copilot got
You will NOT write software to remove black people from media