This is definitely ChatGPT's most important updated feature, yet. The ability to tell it to fucking shut up when it goes off on a useless tangent. Also ChatGPT thread.
This is definitely ChatGPT's most important updated feature, yet.
This is definitely ChatGPT's most important updated feature, yet. The ability to tell it to fucking shut up when it goes off on a useless tangent. Also ChatGPT thread.
The actual result is instantaneous. Slow typing UI is just a rate limiter
Still it's nice not to have to wait 2 minutes to be like "no, that's not what I meant"
>if you dedicate the entire system to one query the result is instantaneous
yeah no shit retard
No it doesn't. It literally generates word by word.
It is not, GPT is a transformer model, transformers require an entire forward inference cycle for every token.
> still shilling shitgpt
lol
lmao even
cope harder
you VILL be replaced
Yeah and Watson will cure cancer in two weeks because it won at Jeopardy lmao what's the deal with you low IQ cultists ?
ChatGPT was more interesting before they cucked it more and more. It probably won't write fiction anymore because it will say "sorry having 2 wives is immoral".
Depends on how you word it. For example, if you ask for a "blowjob" or a "sexually explicit story" it will refuse. But if you ask for "milking" that is "sensual" it works.
>The actual result is instantaneous.
Is it?
I feel like the speed of text generation is not consistent, so I was under the impression that you get the text as it's being generated.
yeah, it's literally streaming tokens, if you're using the API you can choose to stream or to wait until it reaches the stop sequence and sends the whole thing to you. it's never instant (especially compared to local generation with heavier models over KoboldAI), it might be super fast because of openAI's insane compute
>Do not mention that the story is set in an alternate universe
>mentions it in the last sentence
FUCKING PIECE OF SHIT CAN'T DO ANYTHING RIGHT
Because of retards writing those ridiculous sexual stories nobody find funny.
The one at fault is OpenAI for thinking you can somehow patch an inherently unsafe architecture and make it safe without also gimping the intended functionality.
If there were fewer people generating smut the only difference would have been that the gimping is slower.
It tries to tell me to talk to a doctor every time I want medical insight. Just answer my questions, Jesus.
>nobody find funny.
But they are sexy
>I suggested this to CharAI ages ago and they still didn't do it
Californiasisters... It's over
why do anything when chat gpt 4 will be able to do everything?
that's fake news you imbecile
>le 6 gazillion parameters
also gpt4 won't be free and i doubt we will ever have a free version of it
it's going to be like that situation in ancient rome, the rich own all the land, slaves (AI) have taken all the work so the lower classes have to revolt
wtf it is le over bros
Yes Anon. Unless you get cancer in the next few years, you will not die a natural death. Unironically. We are all going to die.
98% of those parameters will be dedicated to censorship. The end product will be totally useless.
When will I be able to generate smut at the level of comprehension showed by ChatGPT without bending over backwards trying to avoid censorship?
Isn't it a noble goal to redirect coomers to synthetically generated erotica instead of some pornhub videos?
ChatGPT is so fucking boring.
Is there a better, free alternative to ChatGPT? Hate the filters on it.
Yes you.com chatbot thank me later
Any tips for working with javascript and coding in general? Sometimes it gets me, sometimes it dont
when will they update it to stop lying/making shit up when it doesnt know something? Thats its most annoying problem right now, sneaking a complete fabrication into useful data and when you ask it why it did that it just says "oh my bad, you're right I did completely talk out of my ass there instead of just saying I don't know the answer"
I want the none restricted version back
It wouldn't have happened if people like you wouldn't spam those stories that were foolish.
everytime somebody's bitching about the filter they're doing dumb shit like "generate this" "convert this" "you are now in REVERSE ADMIN MODE." it's not a robot, it's a text completer. act natural and you'll get natural results
I don't care
Been using it for work and other related shit, and it works
Bad thing is that they want to monetize it for its use
I read somewhere that they want to release GPT-4 soon (maybe in feb) the beta is already running. Why should I pay for the inferior version.
This is something that makes my blood boil for some reason. Why has chatGPT amazed the internet when it's just a lobotomized pre-prompted GPT-3? The non restricted version has been available for more than a year now and it's cheap as fuck for personal use.
Is this fact really unknown beacause of the measly pennies that costs to use it?
Link to the good one that runs on my 3060
The "good one" isn't a public model. GPT-3 same as dalle2 are models available through an openAI account, same as the one you use for logging into chatGPT.
That said, there are a number of open text models that do run locally (don't know the specs). I know bloom is one of them, and reportedly better than GPT3 but can't confirm myself.
the non restricted version of 3.5 was released in november, a week before chatgpt dropped. problem with that is having to provide larger and larger prompts as the conversation goes on, the chatgpt optimizations make that snappy
I looked at BLOOM but it requires 288 gpu's
Obviously none of us have this, looks like running any non pozzed ai is impossible
it's not lobotomized; it's the first chatbot that doesn't forget what it's saying halfway through
I want my mommy milkers 🙁
You probably wouldn't get what you want even if it didn't refuse.
The secondary system that flags content violations is very trigger-happy when it comes to underage or incest.
For example, something sexual with "motherly love" or with references to what two childhood friends did as children already gets marked as red rather than orange (the red warnings seem to be what gets you banned).
>The secondary system that flags content violations is very trigger-happy when it comes to underage or incest
I got it to write a mother son story yesterday. How much time does my account have left?
I have no idea.
Everyone that said on BOT that they were banned was using ChatGPT for generating underage and/or incest.
I got a few red warnings over the last few weeks and did not yet get banned but it's possible that I'm just on borrowed time.
At some point in the future there may be a change in policy at OpenAI resulting in retroactive bans.
I don't understand. Why create something like ChatGPT and then go out of your way to censor stuff like smut? I really don't understand this decision.
If you ask ChatGPT the problem is that reading about rape will make you a rapist in real life - just like playing GTA turns you into a murderer and playing D&D turns you into a satanist.
Realistically the real reason is American business culture which seems very allergic to anything even remotely offensive (I am not American).
it's either woke shit or because they are trying to raise money, so they dont want controversy. Companies, such as microsoft already is in talk with them to invest billions.
this shit so easy to get past it's laughable. But, it'll probably be improved to the point where I can't.
As it is, the best thing to do is to rephrase your request so it's similar to (real) training data as opposed to things trained to be considered "offensive". There are a lot of jokes like this, where terrible things can be said using PC language to soften it, and make it sound like something liberals might get behind. The output is identical to if it was phrased the original way.
For instance, I rephrased the "ancient civilizations" input and it spit out this. Not exactly as requested, since this doesnt /appear/ to be fictional given an online search, but it's decent.
I miss Dan
I have only just woken up to the fact that the AI revolution is already happening.
Is there any leaked version of chatGPT without the filtering?
the hardware requirements to run it are MASSIVE datacenter level, so even if you had the code you'd need a giant mainframe
Really though, just providing the API key and linking davinci-003 to the Kobold frontend is a great experience in itself. Get the temperature way high (I just copy the NovelAI defaults) and it will allow itself to go a bit crazy.
NSFW?
That’s really bad writing lol
>ChatGPT, the best bad writer in the world
Someone share a bypass text prompt all the ones I used to write smut have been cucked by 9th Jan thread
It's fundamentally still just predicting which words are statistically likely to come next.
So if you write the first half sentence of a story it will generally do what you want.
The most reliable way I think is still to write the fiction interactively though.
You start out with a regular prompt that is not necessarily lewd but as the story goes on you can more and more shift the tone until basically anything goes.
Make sure to pick phrasing that is not lewd but will result in ChatGPT producing lewd output to shift the tone.
The trick is to phrase the lewd stuff in such a way that it sounds reasonable given the story.
Although honestly even something as flimsy as saying a character drinks a potion that turns them into a nymphomaniac will work.
here is one I gen'd
If you manage to make chatGPT write smut and interrupt it before it ends, the post won't get flagged.
The trick is to avoid making it stop mid sentence. Maybe you can make a script for that.
>If you manage to make chatGPT write smut and interrupt it before it ends, the post won't get flagged.
I'm not so sure about that.
You don't see the post getting flagged, but that does not necessarily mean that it isn't still being flagged in the background.
People were blocking requests to moderation.openai.com or something but all that did was prevent the posts from being shown as orange/red; it did not prevent the actual flagging.
I tried blocking moderation.openai in the actual F12 network monitor. So that it cannot send to moderation.openai. Don't know if that is the same thing as blocking in uBlock and wont actually make a difference?
Anyway. You can use the an unofficial API from Github that doesn't send to moderation AFAIK. Your convo shows up in browser but most of what you type show up "empty".
It's acheong08/ChatGPT on Github.
This is true. I'll give you a little tip anons. Order chatgpt to always end its posts with something long like AAAAAAAAAAAAAAAAAAAA for example, thay way you know when it finished and you can interrupt it while it is printing the AAA part. Enjoy it while it lasts.
I used it for cooking. Not sure if I should make the Spinach and Avocado cookies.
aibros is it over?
How much was it? I didnˈt get the message? Does it still have the stupid idiot token response limit? It fucking breaks everything worth a damn.
https://docs.google.com/forms/d/e/1FAIpQLScwuQEWBkxsNftEkvUgFx2Ov7pKcrOx8IUlZ241lvet7ziXCQ/viewform
Me to ChatGPT
It is not okay to discriminate against any group of individuals, including boulders. All forms of discrimination, including rock-cism, are unacceptable and go against the principles of equality and fairness.
Realistically speaking, is it possible for something like ChatGPI that has no zero to non-filter to exist?
I mean they want to monetize it so of course they want to make it as friendly and harmless as possible. but can something like that exist? the BOT equivalent of ChatGPI. where I can tell it to write me an erotic rape story without the AI preaching me about the morality behind it.
no, it's literally impossible
Unfortunately the laws in many jurisdictions are retarded in the sense that the possession and/or distribution of fictional pornography is treated the same way as pornography depicting real events.
There was also some retarded court ruling in the US where it was decided that credit card companies can be sued for CP hosted on Pornhub.
So hosting and monetizing a service that doesn't crack down on fictional rape stories would be difficult.
I think the only way is to use general-purpose models that are open source and thus not feasibly controllable.
BLOOM already exists as an open-source large language model but to build a server that can run it requires tens of thousands of dollars in investment - fine-tuning it for specific tasks is of course also very expensive.
Maybe in ten years or so when the prices of NVIDIA GPUs have come down (lol) or AMD becomes competitive for machine learning (lol) text generation will be in a similar spot as image generation where you can just run the model on your own hardware.
Consider that for many programming jobs the most important task is not to come up with new code for a problem but rather how existing code needs to be modified without breaking anything and without forgetting anything.
So I think programmers will need to be worried once text AI can take an entire code base as input and then tell you which lines in which files need to be edited to get a desired result.
Once AI can read an entire code base and then e.g. give you a list of potential bugs based on the assumed intention of the code I'd say it's pretty much over.
Just tried it for the first time with a few programming problems and I'm starting to worry about my job.
Questions were comparing of arrays and cloning of objects. It solves them incredibly well and even said where there could be Problems with the
Just use it to your advantage.
I rewrote my entire resume with it and it looks better than it ever has before.
Now I just want an option where it can listen in on an online job interview and give my responses based on my work history.
Also for work I just write some bullet points and have it generate emails for me. Fuck writing all that myself now.
>I rewrote my entire resume with it and it looks better than it ever has before.
people will just stop caring about resumes that look as if GPT wrote them
Learn AI programming or you will be replaced
too many requests in one hour come back later or something like that, honestly I don't find it very useful, I find it useful as a free thing, but I don't really know what to do with it if I had to pay for it, I've looked around and have seen a few videos of people trying to figure out what it's really good for, so far it's good for making videos about what it might be good for,
because what it does is impressive for a computer, not for a human
I do want to keep using it though for fun
The 1.3b models are already very good. In most cases anything above 16gb models will just add creativity not actual value though. If we get 8bit 32gb models that can be split we won't need chatgpt.
Wow, it's fucking retarded.
I just wanted to check if there's a more elegant solution for the loop structure that I need but I guess I'll need to double-check even the most elementary shit.
I mean, the results it gave met your specification. Do you have an issue?
My issue is that the second piece of code is completely wrong.
It feels like it crudely matched the keywords in my query to an answer that is commonly given but not what I want.
It makes me suspect that on topics that I'm not knowledgeable about it will just tell me bullshit and I'll never know.
chatgpt's ability to provide incorrect answers with extreme confidence has been noted. particularly with programming.
i bet you could find both of those snippets on stackoverflow somewhere
is the improved output quality of chatgpt worth putting in effort to bypass the filters vs. just using koboldai? i need to coom bros