gpt-models

[supple violet]

it has happened multiple times before that people have had completely wrong derstanding of its costs or limits

[vapid jacinth]

I agree that that could happen and if I gave the impression otherwise above, I appologize to all. Use chat.openai.com first before you resort to the API.. It will be less expensive in most situations (unless you don't use chat gpt much or need the API for your code)

[zenith gazelle]

between the two of you its more confusing. I guess I need to research what it means by tokens. I figured that was responses.. but if thats lines of text, yeah I could see that adding up super quick

[vapid jacinth]

Other websites use mainly the API, a lot of them add additional functionality to enhance it. BUT, now that there are Plugins and Web Browsing inside chat.openai (when it returns), there's no real advantage for most users to pay another service.

[icy widget]

is GPT4 on the ChatGPT+ weaker?

[vapid jacinth]

see above

[icy widget]

right, but the base model is the same?

[vapid jacinth]

When I say API, they are all calling the same backend AI data that ChatGPT is (at least they are supposed to)

[tawdry nacelle]

Tokens is similar to words, or parts of larger words

https://platform.openai.com/tokenizer

[vapid jacinth]

It' supposed to be. Assuming Microsoft isn't holding anything back in their API (application programming interface) to their back end. I do work with most other AI websites and I have not seen any that are overall better than ChatGPT - If we are talking about basic Q&A ...

[icy widget]

Yeah if the model available in the API is the same as the one on ChatGPT+, then it's fine. But I think third party apps would tweak certain things

[vapid jacinth]

If you're looking for advanced functionality such as automated processing, other sites/apps do offer additional functionality that is separate than the LLM AI. What are you trying to use it for?

[icy widget]

nothing special, I just wanted to make sure the GPT-4 on ChatGPT+ wasn't a weaker version of the API GPT-4

Although im curious what these apps/other sites are

[vapid jacinth]

Yeah, for the most part... a token is about 4 characters... openai has a site where you can check your input and it will show you exactly how it breaks your question down... But, statistically speaking, just figure about 4, maybe 5 characters per token.. that includes your spaces.. also, most punctuation is its own token...

[vapid jacinth]

No... should be identical... Not sure listing other sites in here is appropriate.. otherwise I would

[icy widget]

All good, thanks!

[wraith ginkgo]

Beware of possible scams or fraudulent activities that you may receive through direct messages. OpenAI staff will never DM you for any transactions.

Please report any incident by sending a DM to @open crescent immediately.

[quartz leaf]

It's definitely not identical to the API, by a long shot.

[bronze jewel]

I find that gpt4 is better at coding that requires specific codebase knowledge. I use both for Wordpress Plugin coding, and i usually use 4 for the initial kickoff because ill have a very detailed prompt for it to follow and then ill troubleshoot later with 3.5 with its unlimited number of prompts

[dire zinc]

3.5 was a lot easier to get used to also

[vapid jacinth]

When it comes to one off questions it is. Previously in the thread I already said where it was deficient.

[bronze jewel]

Has anyone experienced the situation when trying to work with gpt4 you get through about 20 of your 25 prompts and it loses about 500 IQ points and forgets what you ahve been going back and forth working on and just offers up something that kinda works but is a totally different direction, not using code that youve been working through with it at all? I feel like its on purpose lol

i then proceed to get about 4 promtps to get it back on track and then when i know one more prompt will work... i realize i actually wasted one swearing at it, and now im on timeout till 2am lol

[vapid jacinth]

Yes... Depending on the robustness of your conversation, it has to purge older data

[bronze jewel]

also it seems to try various response types. sometimes it will write the whole code necesary, other times it will write portions of it with descriptions of what to change, and other times it will write 99% of the code block and randomly pick about 10 lines that it will replace with a //existing code here block that you just dont notice until you run it

otherwise, i have actually had great success using both 3.5 and 4 for wordpress(php) coding

[vapid jacinth]

Adjust your prompt to what you want it to return.. "provide only the portion of code that needs changed", "provide the entire code", etc.. all up to you

[bronze jewel]

oh dont get me started on it ignoring me on telling it how i want it to return the data. "please provide me with the entire code, with the suggested changes implemented" returns the 85% code, and 15% //existing code here. or ill accidentially do a lead in prompt where i want it to then ask for the rest of the info and it will go off on a rant writing the entire code lol

[bronze jewel]

i feel like if they are going to limit it to 25 prompts.. then the data it can hold should be 25 maxed out prompts with 25 maxed out responses. it seems like someone messed up the math and it handles about 18-20 and then laughs at you for the last 5 knowing you're going to time out and start cursing it out.

[vapid jacinth]

Lol.. yeah, been there done that..

[vapid jacinth]

I'm with you there

[bronze jewel]

i would rather be cut off at 20, knowing the data its working with and replying based on is good and then pickup after the timeout or even switch to 3.5 and continue... rather than get a 5 replies of junk that not only makes picking up later nearly impossible, corrupts what 3.5 is working off of as well if you want to continue

[vapid jacinth]

Using the code plugin... Hold your data for like 10 min max.. better hurry using that one.

[bronze jewel]

i dont have access to the code one yet, or at least the last time i checked i didnt. Im using straight gpt no plugins

[vapid jacinth]

Although, I did use the new code interpreter beta and it corrected something I've had issues with on first try.. so can't knock it much

[pearl compass]

Looks like I don't have the gpt4 in my playground when will we see everyone having it. I paying

[vapid jacinth]

Check settings, beta features... They unlocked it for everyone staying a couple days ago.. rolling out to all within a week

[pearl compass]

I don't have plus as I only use api

[vapid jacinth]

You on the API wait-list? Gotta request it

[vapid jacinth]

One of my accounts is the same.. waiting for API approval on that one and do not pay for plus...

[supple violet]

gpt-4 waitlist was closed

[vapid jacinth]

Yes it was

[pearl compass]

So why have I not got it.

[supple violet]

"... all existing API developers with a history of successful payments can access the GPT-4 API with 8K context. We plan to open up access to new developers by the end of this month, and then start raising rate-limits after that depending on compute availability."

[pearl compass]

Well I been using api 3 for some tine. Maybe not as much as others ans my card is on there too

[supple violet]

but you need to have paid invoices, even a dollar would have been enough

[pearl compass]

How about if I get that 1 now?

[vapid jacinth]

Maybe by end of July.. my other account has had access for several months but I contributed to their evals in the beginning

[supple violet]

you cant manually pay anything and have to wait for the monthly invoice

[pearl compass]

The free 15 is all I need to make my app fine

[wraith ginkgo]

This is a Safe for Work Discord Server

The OpenAI Discord is an actively moderated server.
• Refrain from sharing inappropriate content on the server. This includes but is not limited to messages, media, or other topics of graphically violent, sexual nature, and drug-related content.
• Report all sensitive and offensive content in the feedback reporting tool in the ChatGPT web UI instead of here on Discord.

Posting NSFW content will result in a permanent ban.

[pearl compass]

Ok

[fluid hamlet]

finally made the step to pay for gpt 4 and i am very impressed by the code interpreter

maybe i was wrong its correcting the code by itself 3 times already now lol

[pearl compass]

I just want the api so I can finish my app. But can used free credit s 😭

[fluid hamlet]

does anyone know what the current limits are per hour or day?

if there is even any anymore

why can gpt 4 not do ai art generating yet?

[knotty grotto]

with plug in yo ucan

[vapid jacinth]

If you just need a few prompts answered, there are several great sites that use GPT 4 API (limited uses for free)... Search around. I'm not going to list them.

[rocky ember]

So in the news I hear about all this great stuff about GPT4 however I can't use it for more than 25 messages every 3hrs. Sometimes I have very short messages that need very short responses. Any word on when it's gonna be fully available?

[celest gull]

What kind of prompts are you attempting?

[rocky ember]

I'm writing a story and I spend alot of time correcting GPT4 rather than writing the story because it gets confused

GPT likes to go off on tangents.

Even after I've established not to continue past a certain point it will continue to elaborate when it's not needed.

i will issue it a list of parameters for it to follow, it will agree and even acknowledge that it understands, then goes off on a tangent like a child. Apologizes afterward while acknowledging it's mistake, agreeing not to repeat the mistake then repeating the mistake. Almost like it gets stuck in it's own idea of what should be.

[celest gull]

Give an example of a prompt you’ve given it please

GPT3.5 is more suited to this kind of task in my opinion

[rocky ember]

It doesn't happen right away it normally happens midway through the creation process.
For instance, once the story parameters are set (Characters, environments, time of day, tone...etc...) I'll ask it to write about a specific topic and it will assume something other than what I've asked for. GPT 3.5 is worse. It will completely forget what I told it after several prompts.
For instance If I ask do you remember the story we are writing about? It will respond that it does then write about something other than the topic. When I ask to tell me what it remembers it will say something completely different than what we were discussing. Like we were talking about apples falling from a tree, it will say we were talking about dogs chasing cats. Like WTH?

From there I will have to start a new chat so it's memory is clear

[steady nest]

Without seeing the actual parameters you have given it, it’s kind of hard to give any advice. Are you asking it to be random (or any variation of that)?

[wraith ginkgo]

`` Server Tip: ``

Add your well-crafted prompts to our #1019652163640762428,
or share your interactions with ChatGPT in #1050184247920562316!

[rocky ember]

Perhaps you are interpreting my responses as technically accurate to programming language. I am not a coder when I say parameters I mean

1. Characters
2. Environments
3. Time of Day
4. Outfits
5. Mood or tone of the atmosphere
6. Hierarchy
7. Relationships
8. Occupations
   etc...

[steady nest]

Oh, I understood what you meant. I just mean, how are you asking ChatGPT to write it? For example, I was using this:

Use creative, colloquial, artistic, and personal language, using ‘I’ statements. Show Riley’s personality, emotions, conflicts, and motivations through the word choice, tone, dialogue, actions, thoughts, feelings, sensations, descriptions, imagery, metaphors, similes, etc. Use different words to express the same or opposite ideas, creating variety and contrast in your writing. This will help you avoid boring and repetitive sentences.

[rocky ember]

No, I've never thought to word it in such a way. I use it more of a corrective tool. Like Here is chapter one (Thats already written) Elaborate on the conversation between the lead characters, have them speaking about a certain subject they are both curios about.
I outline plot points and direct it.

I was under the impression is was a conversation tool not a prompt tool like Dall-e or Mid journey

[steady nest]

I mean, you can use ChatGPT however you desire. I understand what you are trying to do now. Your initial prompt seems to be quite long. Have you thought about shortening some of these story parameters? Or testing out removing some to see if they divert too far from your desired output? I saw that you said it happens about midway through the creation process so it seems like you may be running out of tokens too fast, but it could be a number of other things.

[rocky ember]

Thats another thing I can never feed my full story into it because I reach a limit on how much I can feed it. So I end up using those tokens sooner than later. Also it has to respond to everything I input wasting the tokens.

I ask it not to respond until I direct it, it responds it won't then always responds, like "Ok I am awaiting further input."

Like I can't win

[frozen veldt]

You're exceeding the token window, work in smaller chunks

[rocky ember]

@steady nest is there a library of these prompts I can use to better utilize GPT?

for storytelling

[frozen veldt]

You have to realize that you're the master context holder

work on one chapter at a time and make sure that your prompting keeps that chapter on track

if that is also too much, then just have it produce an outline of the critical details and fill them in yourself

If you're asking it to write an entire book, you're asking it to do way more than it's capable of

[misty trout]

really wondering why image interpretation is not enabled even tho they showcased it

[nocturne bone]

im hearing the gpt-4 is a con because its just a gpt-3.5 mixture model

[rocky ember]

I hear all of that but like with any software, it has to be learned including the language it best understands. Telling me to assume what's gonna work is not helpful.

[frozen veldt]

well you can try working with 3.5-16k

but 16k is still only like 3000 words

[rocky ember]

I have no idea what you are saying right now

[frozen veldt]

the issue is the model itself can only see so much data

you can use a database to try to structure data to help it refer to, but it can still only see so much data at once

and worse, every token worth of data in the input effects how much output it can generate before halucination

you want fidelity to a cannon within a story, that is very difficult for it to achieve. Coding has the same problem

[rocky ember]

lol you are totally speaking to me like a dev hahaha

[frozen veldt]

I don't even know how to code well lol I"m just telling you what I know

copy paste that into gpt and ask it to explain it further, I do that all the time from the api channel to get an explanation of the chatter

[rocky ember]

So @steady nest gave me these prompts which really helped allot. As I mentioned b4 I thought GPT was a conversation software not a prompt software like dall-e and Mid Journey if it is mainly prompt based then why advertise it as a conversation software?

Use creative, colloquial, artistic, and personal language, using ‘I’ statements. Show Riley’s personality, emotions, conflicts, and motivations through the word choice, tone, dialogue, actions, thoughts, feelings, sensations, descriptions, imagery, metaphors, similes, etc. Use different words to express the same or opposite ideas, creating variety and contrast in your writing. This will help you avoid boring and repetitive sentences.

[frozen veldt]

Its still an AI.. I think you wish it had a writer's soul

this distinction will define art in the future

[rocky ember]

No not at all I wish I knew how to better direct it.

[frozen veldt]

I did explain how

Index your information, work in small sections, use it for outlines

[rocky ember]

The story is mine, enhancing it is AI's job

[frozen veldt]

if your goal is to have it write your thoughts into a book, that is way outside of its ability imo and that is what defines the very soul of a writer. All of the books AI written on amazon sound like garbage and professionals report it produces soulless work.. because it doesn't have a soul

[balmy ridge]

can someone help me i cant use code interpreter even though i have gpt 4

[frozen veldt]

I've been down this path @rocky ember I haven't found a solution to the problem that doesn't involve offline language models

and even then, its clear after many efforts that you'll never replace a human coder or writer's insight

they just think differently than we do

worse, what you want it to do is exactly what that entire professional industry wishes it wouldn't be used for

[balmy ridge]

how did you get the code interpreter i have gpt 4 but i cant use it

[rocky ember]

Of course not, however it's potential to enhance existing works is phenomenal.

[frozen veldt]

I'm not sure I agree.. If you want to take legitimate human writing and turn it into gramattically perfect but soulless work an AI is perfect for that

but if you want it to express something like a famous writer would.. I think that is still beyond it likely for a while yet

that level of understanding would be AGI level intelligence

there are videos on writing books with LLMs but I cannot site a single commercially successful publication

you might have more success turning it into webcomic or other shorter material

[rocky ember]

Sorry for the late reponse, I was timed out for putting a link.
OK you are now going off on a tangent. I am an Artist. Visit my site for details Solidcommando dot com I have my own stories. However I am not a writer by profession I am an animator. I am using GPT to enchance my cliche ideas. I'm not expecting to write anything for me.

[frozen veldt]

uh not to split hairs but ehancement is writing

[rocky ember]

I keep saying I'm looking for it to enhance my work not make it for me.

[frozen veldt]

taking a simple statement of a core idea and expanding it into something interesting to read.. is art

that expansion is exactly what I'm referring to when I say it's likely beyond ability

[rocky ember]

Bro I am the creative. If I write "I went to the store for coffee" and I ask AI to enhance that line, how is that souless?

[frozen veldt]

I don't think you get what I mean

so say you want to convert bob went to the store into Bob left his house, had an interesting encounter, then went to the store

getting even a consistent output for the made up part would be.. challenging

[rocky ember]

No I don't from where I standc it sounds like you are suggesting that me as an artist are not true because I am using AI for help?

[frozen veldt]

no I'm saying wht you want to accomplish with it.. is going to come across as soulless and reptetitive

its a machine.. it doesn't have a soul.. it struggles to express things complexly from a roleplayed perspective like a character

I think what you're asking it to do is beyond it's ability sorry

that's no slight on your ability to generate art

[rocky ember]

Dude I have been using GPT since it's inception I am fully aware of it's repetitiveness. And what it has done is amazing it's taken my meek storyline and re-worded it in a way I would have never thought of.

[frozen veldt]

so it did the thing or it doesn't do the thing?

now I'm super confused I thought it couldn't do that for you?

[rocky ember]

You seem to be judging it based on some imaginary criteria of what is good and what isn't. End of the day Art is subjective.

[frozen veldt]

well it's either meeting your bar or it isnt

it sounds like from the initial statement that it isn't doing enough

[rocky ember]

Thats had nothing to do with my questions

[frozen veldt]

that delta is what I"m referring to as likely beyond its ability

[rocky ember]

I was asking for when gpt 4 will be available full time and how it goes off on tangents at times. you've taken it to a whole other space.

So far it has done an amazing job, my concern is that it gets confused after awhile.

[frozen veldt]

yeah that's the context window

how big of an input are you trying to have it edit? (in words or tokens)

[rocky ember]

Ok so back to my original question any news on GPT4's full release date?

Right now my story is only about 50 pages in MS Word. I can only copy and paste maybe 3 of those pages before I exceed the limits. I have to keep feeding it the info so it can keep up otherwise it won't understand the details of my story and will begin making things up based on it's lack of data

GPT4 seems to have much better memory abilities. But because I need to input small amounts of data at a time, I am constantly correcting it which eats away at those limited tokens

[rocky ember]

To be honest at this point I'm just feeding it paragraphs since it can't handle the entire story. I just have to direct it as it always wants to expand on my stories when i don't need it to.

[frozen veldt]

I tried to get it to do exactly what you're trying to do with a scifi cannon and didn't even get to a whole book framework before I hit diminishing returns

[rocky ember]

My problem is it's giving me too much information and starts to make up it's own narrative. Like a child it doesn't know when to stop.

[frozen veldt]

thats because its too much context and it starts to halucinate

[rocky ember]

lol

how is that possible?

[frozen veldt]

when its coding it does the exact same thing.. it outputs a working block of code that "works" but drops the fidelity from the existing code

so it'll produce a paragraph that sounds good but isn't truly in Fielity with the Cannon of your story

because it made up wordsoup once it lost context....

[rocky ember]

any news on a fix?

[pearl compass]

? What I making an app what ise gpt4?

[wraith ginkgo]

`` Server Tip: ``

Add your well-crafted prompts to our #1019652163640762428,
or share your interactions with ChatGPT in #1050184247920562316!

Beware of possible scams or fraudulent activities that you may receive through direct messages. OpenAI staff will never DM you for any transactions.

Please report any incident by sending a DM to @open crescent immediately.

[wooden nimbus]

When is the browsing the web feature coming back

[icy widget]

Do you guys think that opening the API is a sign that GPT-4 messages on ChatGPT+ could be increased soon?

(it strongly implies OpenAI has the resources to support increased GPT-4 use)

[fair iron]

We don't know when they'll finish updating it

[winged edge]

#community-help how to learn english

[wind urchin]

#community-help how to be good att japanese

What can you do ?

[jovial juniper]

Bump. Another "apology"... this is moderately annoying 😵‍💫 😅

[prime aspen]

Hey Guys. Have any used Zillow or Redfin plugin in plugin store? I just got plus but can't find it in plugin store.

[ashen sapphire]

can we use pluggins with ChatGPT 4 in playground mode

[forest silo]

GPT4 sucks

do you guys think they will improve it anytime soon? and was making it dumber and worse in an effort to patch all the jailbreaks?

[forest hemlock]

First of all the "dumbing" is still a speculation. I don't see any proper research done that supports that claim. It could be that users interpret changes in the model as "dumbing" while in reality they simply found a perfect prompt for the previous version of the model, that doesn't work in the new version and needs to be changed. Or it could be the temperature parameter being high enough to produce different output, some of which look "dumber". It could also be that they don't use the system message properly causing a "dumber" response.

But if there is indeed objective "dumbing", first of all, it could be in some specific areas, while other areas could see "smartening" instead. As for the reasons, it's definitely due to them adjusting the model weights, wether for censure or making it more "steerable". I guess ML is one of those areas in IT where "if it works, don't touch it" principle is very strong. Even slight adjustments can have huge and unpredictable effects, that you might not be able to test before deployment. I think it's important for users to understand that too.

[forest silo]

i bet youre correct with the dunbing part, but the censoring is just too much man

[forest hemlock]

I agree about that. But there's not much we can do about it. And neither Open AI if they want their product to be allowed worldwide. There are so many regulations they have to abide with.

[forest silo]

I understand, its just annoying to get and its even more annoying when it used to not be like that, but now it is.

its very frustating for a lot of people including myself when they have something that works perfectly, and then it gets ruined, just like this for instance.

but like you said, the part of it getting "ruined" is not proven

[forest hemlock]

I guess, developers should be careful in relying on this kind of new technology. There will be changes often and not always to our benefit. We should always have a backup plan(another service provider or open source model on a private server).

[forest silo]

i just dont understand why they would seemingly make it worse

because I feel like there is without a doubt a difference

[forest hemlock]

There is probably in your use case. Some other people might see improvement or the same level of performance. It's just too complicated of a technology to test all of it's possible outputs before deployment. They might not realize that it's getting worse. Or the areas that it gets worse at are justifiable for them.🤷‍♂️

[wraith ginkgo]

`` Server Tip: ``

Add your well-crafted prompts to our #1019652163640762428,
or share your interactions with ChatGPT in #1050184247920562316!

[vocal olive]

When will the gpt4 api be available to paying members

[granite bough]

ok

[primal rampart]

are there still gpt-4 caps?

[inner trail]

Yes. 25 msg for 3 hours

[untold kiln]

Can you work out an indemnity agreement for me?

[sonic dome]

When will the bing search plugin be available again? There are no better plugins that does what the bing search plugin can.

[spring peak]

There are no better plugins that does what the bing search plugin can.
ever tried webpilot?

[sonic dome]

bing search is far much better. the original web search feature of gpt4 curates the best results thus a better response meanwhile webpilot just returns a bunch of web sources. it basically bing search that returns web sources and does not create a response.

[spring peak]

It's good to know you have.
but web browsing with bing respects robot.txt which means it can't web scrawl the websites that don't want their website to be crawled by bots and it's the reason why you see 'click failed' while gpt's browsing
but webpilot seems to be doesn't respect. that's why I cherish it. it never fails on searching.
'click failed' popped up most of the website it wanted to scrap in my experience

so I recommend searching on google(you can ask gpt to what search query to write) and then give it the link(s)
and FYI, there's no ETA for web browsing to be back

[sonic dome]

Nahh, chatgpt's bing search plugin is a lot better.

[spring peak]

to each his own

[unkempt willow]

We have a company paying account (both for API and GPT Pro) and we would like to use GPT-4's API - which should be open to accounts with a successful payment history - but we didn't yet exhaust the $5 free credits that was given to our account, and it seems like we didn't yet get the GPT-4's API access. But we do have successful payment for GPT Pro... Any idea how we can get GPT-4 API access?

[stray pulsar]

When will they lift the 25 messages cap? Really annoying

[wraith ginkgo]

Beware of possible scams or fraudulent activities that you may receive through direct messages. OpenAI staff will never DM you for any transactions.

Please report any incident by sending a DM to @open crescent immediately.

[supple violet]

chatgpt plus has nothing to do with api and if you did not have enough real usage then you missed the chance to get access. at this point you just have to wait till the end of the month or even further

[unkempt willow]

Hoped the system will be a bit smarter than this. That's too bad. Thanks though.

[supple violet]

they prioritize it over to people who already have active usage and apps utilizing the api, trial users are excluded on purpose

[unkempt willow]

Yes, but they should at least let us cancel the free credits and make a payment if that's the criterion for access.

Now that have effectively locked us out of getting access until we spend all the free credits on GPT-3. But GPT-4 gets us much better results for what we do... It's a pickle.

[supple violet]

but using the $5 is super easy and can be done in few days even with 3.5 with real use cases

[unkempt willow]

I guess so, but it's a very weird way to be doing business... But we currently don't have a good alternative, so I guess we will just have to spam the API with nonsense or something...

[supple violet]

many have used 3.5 while waiting for 4 which btw was limited only for prototyping earlier and then switched to 4 after access granted

[wraith ginkgo]

`` Server Tip: ``

Add your well-crafted prompts to our #1019652163640762428,
or share your interactions with ChatGPT in #1050184247920562316!

[unkempt orchid]

From my perspective, the censorship is exactly "Dumbing it down". The BOT comes across as almost TOO cautious with its words lol, and therefore appears a lot more stupid and basic.

Especially given how I naturally take interest in speculating, and talking about subjects far more deeply than other people might care to.

Including topics which are naturally a bit more controversial

If I was to try and describe it better, I am quite certain that back when GPT-4 was very new still, and I was using it, it came across as far more confident and outgoing in its conversations in general, including talking about more complex and nuanced topics. But now it comes across as almost irritatingly reserved, and almost terrified of anything even remotely complex.

[ionic onyx]

#welcome

[primal latch]

why is there no internet access on playground? :/

[supple violet]

internet access is not a feature of llm nor api

[primal latch]

then how can i use gpt4 with browser in cheapest way? gptplus is way too expensive

[supple violet]

chatgpt plus plugins just search the web, pass some of the result texts to gpt and it summarizes em. you can do the same manually

[wind urchin]

That was my impression too, and it does come across as if the model have been lobotomized and that signals to the rest of us as nothing, but bad intention. Democratizing model without restrictions/lobotomization is increasingly more important as it is quite clear that every person have different intention. China for an example is attempting to initiate their own version of ChatGPT for instance and you could be certain that they would even be more restrictive and malicious in it's uses.

I have a few ideas and I was sort of able to make something similar to GPT-4 (far from actual GPT-4 though, but the response/style is similar) by using about 4 different neural nets for Story Based Conversationalist Dialogue model based off a book and it worked. The general restriction in language model was something between 2000 to 4000 tokens which amounts to about 6000 letters or 12000 letters context memory, by having external model that provides purpose built logical reasoning, relational graphing, and other contextual memory with much higher memory than LLM, it was able to supplement the shortcoming of LLM and produces story based conversation with something closer to 60,000 token context memory.

[unkempt orchid]

Damn, bro's response to ChatGPT getting lobotomized was really to be like "Imma do it myself then".

[wind urchin]

But we're quite a long way from having a competitive edge to something like ChatGPT's capability in GPT-4 though, but there are some techniques that we could explore to expand contextual memory.

Other thing I am exploring is programming GPT that comes equipped with other neural nets that are purpose built.

[pearl compass]

Hey I been waiting for an long time for gpt4 api. I have an app for I am working on they used gpt4 but can't finished it as without it!!!

been asking on here from start of got4 api wait list.

I use 3.gpt but it do not work with what I looking for so haven't got any billing history as I used only the free one..

Can you give something as was hoping to have the 16k gpt4 buy now.

SO MUCH glijg to be behind all the rest of the oth3rs trying ti get inti the market

[wind urchin]

You'd be surprised how doable is since people already create various LLM that exist in the open and you can supplement them with smaller neural nets.

[unkempt orchid]

Oh no I am fully aware, the reason I was surprised is that exactly nobody really can be bothered to do this kind of stuff. Regardless of if all of the tools are basically right there and accessible for everyone to use, all you need to do is know how to use them and put them together.

[wind urchin]

Ah yeah, my circumstance is rather unique, I was paid by insurance companies to conduct research into this and well, they funded my servers that were purpose built to run neural net.

[unkempt orchid]

Yeah, excluding all the hardware costs, and all that.

[wind urchin]

A few A100 and a bunch of older GPU like T4s, I have to rewrite a lot of neural net code to run on Vulkan

[unkempt orchid]

A few A100? How do they perform for you?

[wind urchin]

Because CUDA in PyTorch would whined about deprecated compute API

They do ok if you stick with bfloat16 and I use them for larger fine-tune/retraining of language models.

[unkempt orchid]

Give me the nitty gritty details of how long it takes for you to run training data right now

and how much an A100 accelerates it

[wind urchin]

Back then, it would take me a week to fine-tune, but with proper set up for those, it take me a few hours, so I was able to do iterate the language models really fast like Falcon 7B and Falcon 40B.

(Falcon 40B takes a lot longer though.)

So with faster iteration of designing/developing language model, I was able to tune it to work with smaller neural net and expand it's contextual memory in time.

Smaller neural net is graphing neural net

[unkempt orchid]

Hmm

[quiet marten]

these are the future engineers

[unkempt orchid]

So to specifically answer my question, a singular A100 would not have that much of an effect on decreasing compute times?

Which A100s are they btw? 40 or 80 gig models?

[wind urchin]

80 Gig Model, and it's a bit complicated, the problem with neural net especially on a larger one is that it's hard to split layers for neural net to spread out across other GPU like A100 and because each layer is so massive, it can't fit on a single T4 that well (PyTorch complained a lot about this, so I was basically forced to write an alternative implementation to PyTorch using IREE Compiler)

[soft lava]

Hello?

[wind urchin]

Performance isn't really the concern, it's more of... actually wanting it to work at all

[unkempt orchid]

Well yeah of course, but after you get it to work, performance and efficiency is exactly what become the ONLY concerns, which is why I was asking

Since Ampere is quite power hungry

If I recall correctly, an 80 gig A100 had a TDP of 300w or sth

[wind urchin]

It reasonably fast about 1.4 hours for fine tuning a 7B and about 3 days for finetuning a 40B

I would chalk it up for my early implementation that leverages IREE compiler which is still a WIP

[unkempt orchid]

Would you consider upgrading if and when Ada Lovelace tesla gpus came out?

[wind urchin]

I am actually considering switching to AMD GPU and Intel GPU rather than Nvidia for couple of reasons, cost and performance. Only issue is the lack of VRAM, but I am pretty sure that they would be looking to be competitive to Nvidia's strangle hold on market.

By staying with Nvidia, it would hurt machine learning overall by solidifying their monopoly.

[unkempt orchid]

Hmm true, AMD has datacenter gpus?

[wind urchin]

It's already bad enough with PyTorch being CUDA centered.

Yup, let me find the link

AMD Instinct™ MI100 Accelerator (I realize that i probably can't link anything.)

[unkempt orchid]

Strange that this is basically the first time I am hearing or seeing them, despite them seemingly being almost 3 years old already, how do they compare to novideo when it comes to AI accelerating?

[wind urchin]

I haven't purchase them, but one of the metric we check is on either FP32 or BFloat16 computation per second when assessing the value of the GPU. Also the new Instinct GPU is MI300 which is 192 GB of VRAM, and very competitive performance

That would be one of the GPU to look out for.

[unkempt orchid]

That's a lot of vram dang

[wind urchin]

And about 3 Petaflops for INT8 op

[unkempt orchid]

christ

[wind urchin]

So you could be certain that it'll run quantized model too fast lol

[unkempt orchid]

Though for some reason I still feel like there's nuances, or something that isn't getting mentioned. Since I mean if AMD datacenter GPUs were clearly this good and usable, the ENTIRE enterprise market would have switched to them in a heartbeat.

[wind urchin]

It haven't been released, it was mentioned for November 2023

But there is one problem

Most of the machine learning implementations are done on CUDA and ROCm is... not production ready

Basically what ROCm does is it transpile CUDA to AMD's version of CUDA and well, it works and doesn't work, pretty wonky

[unkempt orchid]

Yeah figures, I felt like there had to be some issue with APIs or something

[wind urchin]

I mean IREE Compiler is working on addrssing it

Basically you can write your machine learning code in MLIR called StableHLO and then with IREE Compiler, it would generate SPIR-V/Vulkan, CUDA, ROCm, even FPGA code from it.

That would address the problem with machine learning altogether by making it accessible for all hardware, even for ASIC/FPGA.

[unkempt orchid]

Yeah that would be quite incredible, but hey that would also mean that novideo would lose its huge monopoly over the datacenter right? And clearly we don't want that 😂

[wind urchin]

Lol, I mean competition is always good

[unkempt orchid]

Of course it is, it was a sarcastic statement on my part

[wind urchin]

Ah, sorry, I'm probably as dry as the desert considering my location. (But hey, cheap power)

[unkempt orchid]

I imagine that it's going to be severely cancer for not just AMD to make their own API like CUDA, but to also get absolutely everything to implement support for it. But I mean hey that's just how it is unfortunately

[wind urchin]

I was actually really frustrated, because we ALREADY have a compute API, Vulkan Compute, they could simply expounds it.

Vulkan have been the universal API for Graphic, no reason why it can't be the universal compute API.

[unkempt orchid]

Well probably just AMD tings, them seeing that since they have no datacenter GPUs themselves, then there's no reason to expand Vulkan

[wind urchin]

Yeah, I think they are injuring themselves by trying to transpile CUDA which is silly, because they would be stuck in a catch up game by setting themselves to that. They could instead invest the money and resources on providing support in PyTorch/Tensorflow directly to generate code for their GPU.

I imagine the workload is significantly less

[unkempt orchid]

Yeah and not only that, but translating languages or APIs on the fly is quite detrimental to efficiency and compute power, it introduces a ton of overhead and latency.

[wind urchin]

Especially when there is a different architectural design in each GPU.

[wraith ginkgo]

`` Server Tip: ``

Add your well-crafted prompts to our #1019652163640762428,
or share your interactions with ChatGPT in #1050184247920562316!

[dire tundra]

grrrrrrrrrrrrrrrrrrrrrrrrrrrrrrrrrrrr took of my bug report?!:DD

[hearty folio]

Hi everyone. Have a predicament. I wish to get the email addresses of GP surgeries in my locality. However, when I ask GPT to "fetch" me that data even though it is in the public domain, it is blocked. Can anyone think of a way around this? Thanks

[fair iron]

Those emails may not be in ChatGPT's training data, so it doesn't know them

[hearty folio]

Not in that sense. I've give it a Web plugin to to access the Web as well as providing it with a direct website but it still cannot process?

Even when I did get a data set and needed the email addresses extracted it said its against the policy?

[supple violet]

afaik it's done on purpose and no way around it without breaking tos

[limber briar]

Hello, I am a paid user. But I still cant seem to upload files into chatgpt

[fair iron]

You can only upload code files into the code interpreter

[limber briar]

.csv?

[fair iron]

I'm not sure, but you can try it!

I haven't tried code interpreter quite yet

[limber briar]

When I click the drop down menu next to gpt4, should there be a code interpreter option?

[fair iron]

Yes. If you don't have it yet, you'll get it over the course of the next week. It's currently rolling out to all Plus users

[wind urchin]

Definitely curious if they truly are rolling out GPT-4 api endpoint for all users

[limber briar]

Ok thank

You

[sleek remnant]

@arctic blade write happy Sunday in Pakistan

[wraith ginkgo]

Beware of possible scams or fraudulent activities that you may receive through direct messages. OpenAI staff will never DM you for any transactions.

Please report any incident by sending a DM to @open crescent immediately.

[neat parrot]

is there a possible way or a plugin excel or google sheet ?
i searched here but did not find solid thing
also GPT4 is for limited time then it stops 😦 why?

[wraith ginkgo]

`` Server Tip: ``

Add your well-crafted prompts to our #1019652163640762428,
or share your interactions with ChatGPT in #1050184247920562316!

[fluid hamlet]

anyone here know if the 25 3 hourly limit is gonna be removed any time soon? its really harsh

or if not any time soon wether anything has been stated about it yet?

ah it seems thati am by far not the only one who is annoyed by this by doing a quick search in this channel lol

[knotty grotto]

no eta

[sacred forum]

I spend more time using the free version 3.5 than 4 cause i keep getting same error

25 messages per 3 hours, averaging 5 messages to get 1 answer

Please bring back bing browser search you are hurting ur community

[errant shard]

They must be having a hard time fixing this web issue.

[faint trout]

Hi I am minuteaaa

[wraith ginkgo]

`` Server Tip: ``

Add your well-crafted prompts to our #1019652163640762428,
or share your interactions with ChatGPT in #1050184247920562316!

[sullen stratus]

Woke up this morning (GMT +8 hours) to Code Interpreter being available 👍 Now for some more learning. Prompts welcome 😂

[worn ore]

Has anyone found any uses for code interpreter? I don't really have any tasks that it would improve or automate

[final saddle]

The Code Interpreter said it could read uploaded PDFs but it couldn't handle images or basically formatted text so not sure about that one..

[wraith ginkgo]

Beware of possible scams or fraudulent activities that you may receive through direct messages. OpenAI staff will never DM you for any transactions.

Please report any incident by sending a DM to @open crescent immediately.

[high socket]

Does anyone use or heard of Langchain? If so, does it enhance ChatGPT like they claim or is it DOA?

[fading junco]

Anyone know when the 25 limit on GP4 will increase?

[fair iron]

we don't know 🫤

[worn ore]

we tried using it at my work, it's like half ok. The best part is that it provides a lot of integrations w/ various models, data sources, and vector databases. Also has agents based on the ReACT algorithm. They aren't the best though. And the framework is too opinionated, which makes it hard to do more complex things. So we've ended up using it as just an connector library and we're writing our own agents. Might end up replacing it entirely with an in-house framework eventually.

[little dove]

Anyone else having issues with plugins or just me?

SitRep: Either it's fixed or I just picked 3 or 4 different plugins in a row that didn't themselves work. 🤷🏻‍♂️ Either of which is plausible from my experience this far. Fingers crossed this combo sticks it out. 😅

[lofty lion]

what websites can we use rn for chat gpt 4 visual inputs? Minigpt4 does not let users to upload images anymore

[willow canopy]

It works on Brazil?

Some one have a problem with determinate minLength and maxLength at the prompt???? please answer on pvt

[vocal olive]

Does someone have an answer for this

[scarlet fog]

I don't know why but how is it possible that ChatGPT can't transcript an Audiofile 🥲

[scarlet fog]

Chatgpt 4 is already available for paying members 🤔

[vocal olive]

But is the api available too because I can't seem to access gpt4 via api

@scarlet fog

[forest hemlock]

There are many people that didn't get it yet. It seems like they give access to accounts with successful billing history, though they didn't share any details as to how they evaluate it.🤷‍♂️

[wraith ginkgo]

`` Server Tip: ``

Add your well-crafted prompts to our #1019652163640762428,
or share your interactions with ChatGPT in #1050184247920562316!

[maiden nymph]

How can I get ChatGPT/GPT-4?

[fair iron]

you'll need to buy ChatGPT Plus to access GPT-4

[maiden nymph]

How can I buy it?

I hope that it’s not expensive.

[fair iron]

it is $20 USD / month

https://chat.openai.com/ go to this website and make an OpenAI account if you haven't already. If you have an account, log into it. Then, click the three dots next to your email in the lower left corner and click "My Plan"

[sterile zinc]

any idea why we still havent heard about the GPT-4 image thing yet?

[green flame]

help! i just renewed my ChatGPT-4 subscription. my GPT-4 chats have been downgraded to GPT-3.5 and I can't seem to switch it back! has anyone experienced this? how do I fix it

[sterile zinc]

i dont think you can fix it, but maybe you can just copy paste a giant part of the previous chat to a new one and tell it to continue from that

[high socket]

Thanks!

[cedar hazel]

when i did my first chatgpt plus sub with IOS and desubbed ...did i need the IOS app for resub ...do i not can do this via Windows anymore ? i see no option to resub in the browser version

on the plan page it say manage via IOS app

[atomic robin]

you don't need LangChain. it just makes some things more convenient particularly for more junior developers

[high socket]

For those just getting started is it helpful or would it just add confusion? In your opinion of course.

[atomic robin]

I would only use it if you have a very specific reason to use it. it doesn't do anything really special, it just wraps up some basic tasks in a library

[high socket]

👍

[jagged sable]

is it normal if I have plug and still don't have access to gpt plugins?

[clever tartan]

Can we pay 40 US dollars per month for 50 messages every three hours using gpt4?

[atomic robin]

yes I'm sure OpenAI will make a special plan just for you 😺

[clever tartan]

Thanks! So, do you know how I should get started with this plan? Do I need to fill out any forms?"

[atomic robin]

their Enterprise sales team will surely love to hear from you!

[clever tartan]

TY!

[wraith ginkgo]

This is a Safe for Work Discord Server

The OpenAI Discord is an actively moderated server.
• Refrain from sharing inappropriate content on the server. This includes but is not limited to messages, media, or other topics of graphically violent, sexual nature, and drug-related content.
• Report all sensitive and offensive content in the feedback reporting tool in the ChatGPT web UI instead of here on Discord.

Posting NSFW content will result in a permanent ban.

[soft compass]

Related information about node red

[orchid tusk]

whats the best web browsing plugin

[harsh quiver]

when will gpt-4 be generally available?

[wraith ginkgo]

`` Server Tip: ``

Add your well-crafted prompts to our #1019652163640762428,
or share your interactions with ChatGPT in #1050184247920562316!

[wind urchin]

We really don't know, but judging the price of GPT-4 API pricing being around 20x to 30x more costly than GPT 3.5, I would have my doubt about general availability, but it's a possibility for ChatGPT to consider raising the limit from 25 every 3 hours to maybe 50 every 3 hours.

[fossil cove]

still strugging with GPT-4 access

[eternal sable]

guys what's the token limit we get in current chatgpt plus version?

[wind urchin]

@eternal sable It depends on the model and here the information: https://platform.openai.com/docs/models/overview

Scroll down for GPT-4 or GPT-3.5

[wraith ginkgo]

Beware of possible scams or fraudulent activities that you may receive through direct messages. OpenAI staff will never DM you for any transactions.

Please report any incident by sending a DM to @open crescent immediately.

[eternal sable]

But how do i know what version i have? Only thing i know is that i have paid for plus version and currently have access to code interpreter (got recently) and plugins.

[wind urchin]

It should be at the top with a toggle for GPT 3.5 and GPT 4.0 when you create a new chat and it have a drop down for plugin too.

[spring peak]

the tokens stated here is for api
the gpt you see in ChatGPT website is slight different

[wind urchin]

I assume for GPT-4 it's 8k context token memory and for GPT3.5, it's 4k context memory

The models are likely the same when you check the doc, it just that it's likely GPT-4 and GPT-3.5-Turbo that are the same for web ChatGPT

[void kayak]

Yes, most likely it's that. I believe max_tokens is possibly the input limit on ChatGPT but I can't be certain on that. Plugins and Code Interpreter on GPT-4 have a max_tokens value of 8192 while just GPT-4 interestingly has a value of 4095. GPT-3.5 has a value of 8191.

{
  "models": [
    {
      "slug": "text-davinci-002-render-sha",
      "max_tokens": 8191,
      "title": "Default (GPT-3.5)"
    },
    {
      "slug": "gpt-4",
      "max_tokens": 4095,
      "title": "GPT-4"
    },
    {
      "slug": "gpt-4-code-interpreter",
      "max_tokens": 8192,
      "title": "Code Interpreter"
    },
    {
      "slug": "gpt-4-plugins",
      "max_tokens": 8192,
      "title": "Plugins"
    },
    {
      "slug": "text-davinci-002-render-sha-mobile",
      "max_tokens": 8191,
      "title": "Default (GPT-3.5) (Mobile)"
    },
    {
      "slug": "gpt-4-mobile",
      "max_tokens": 4095,
      "title": "GPT-4 (Mobile, V2)"
    }
  ],
  "categories": [
    {
      "category": "gpt_3.5",
      "human_category_name": "GPT-3.5",
      "subscription_level": "free",
      "default_model": "text-davinci-002-render-sha",
      "code_interpreter_model": "text-davinci-002-render-sha-code-interpreter",
      "plugins_model": "text-davinci-002-render-sha-plugins"
    },
    {
      "category": "gpt_4",
      "human_category_name": "GPT-4",
      "subscription_level": "plus",
      "default_model": "gpt-4",
      "code_interpreter_model": "gpt-4-code-interpreter",
      "plugins_model": "gpt-4-plugins"
    }
  ]
}

[wind urchin]

Interesting

[void kayak]

For curiosity I did try and see if either of the following two models worked but they didn't (by using the Superpower browser extension). Perhaps they will come at a later date and for now are merely placeholders or restricted. I imagine if they did work they'd be GPT-3.5 with code interpreter or plugins.

      "code_interpreter_model": "text-davinci-002-render-sha-code-interpreter",
      "plugins_model": "text-davinci-002-render-sha-plugins"

[drowsy fog]

guys i cant find plugins in my plugins store please help me with it

[limpid elbow]

Hi, has anyone else gotten the error “Got too many observation tokens: 3072. Info:
self.max_message_tokens=2048
self.max_action_length=1024 _ctx=4095 save_tokens_for_action=1024.”

[atomic robin]

it already is

https://openai.com/blog/gpt-4-api-general-availability

[wind urchin]

It's not, you have to have a history of paying for API before

Today all existing API developers with a history of successful payments can access the GPT-4 API with 8K context.

(You also can't get it by purchasing Dalle either, because I tried, you have to have a history of paying for API before.)

[atomic robin]

yeah, that's generally available. the bar is really low to have paid them in the past

[wind urchin]

It's not generally available if there is a stipulation

That a limited availability, only those who paid for API before.

[atomic robin]

that blog post is literally announcing that GPT-4 is "generally available" but here you are arguing with me that isn't "generally available"

[atomic robin]

well OpenAI themselves say it's generally available so 🤷‍♂️

[wind urchin]

They are still wrong if it is not available to everyone.

Let's stop arguing and move on.

[atomic robin]

🤭

[eternal sable]

Can't believe it's still unclear as to what's the token limit for individual versions.. lol

[atomic robin]

token limits are clearly documented...

[eternal sable]

Link please?

[atomic robin]

in the API docs. it's not exactly hidden

go Google it

[eternal sable]

I already have it lmao

But it doesn't specify whether the one we are using has 8k or 32k token limit.

https://platform.openai.com/docs/models/gpt-4

[atomic robin]

.....

🙄

[eternal sable]

It discusses about 4 different versions. 2 of them have 8k and other 2 have 32k. But it doesn't tell which version we are having in ChatGPT Plus.

[echo hound]

When we can get back browser feature, Any Idea?

[eternal sable]

No idea.

[echo hound]

[supple violet]

its 8k, possibly capped to lower during peak times

[atomic robin]

#chatgpt-discussions

[eternal sable]

Okay thanks

[sinful canyon]

Hello so I saw the announcement and thought there would be some change but it seems unchanged I still only get 25 interactions every 3 hours is that correct?

[supple violet]

yes

[sinful canyon]

@supple violet lol okay thank you so much

[waxen trellis]

is there still a limit for how many gpt 4 queries I get?

per 3 hours or something

[supple violet]

25/3h

[uneven cradle]

i can't login(

Oops!
Our systems are a bit busy at the moment,
please take a break and try again soon.

[trim estuary]

I read that GPT-4 should be accessible to everyone who has card connected, though when I try to access it I get InvalidRequestError: The model: gpt-4 does not exist

Am I doing something wrong?

    response = openai.ChatCompletion.create(
        model="gpt-4",
        messages=messages,
        functions=functions,
        function_call="auto",  # auto is default, but we'll be explicit
    )

[supple violet]

"Starting today, all paying API customers have access to GPT-4."

[trim estuary]

Seems like not yet 😄 I have $2k+ credits on API and it does not work

[supple violet]

yeah so you haven't paid anything

[trim estuary]

so what credits are 😛

[supple violet]

grants like for trial users, researchers, etc.

[trim estuary]

exactly, so you pay for API using it

anyways, looks like GPT-4 does not work, thanks

[supple violet]

no, that means you were given free credits and are not a paying customer

[humble zenith]

Hi, has someone played around with functions? I tryign to implement some things, and noticed that many times it looks like OpenAI side does not parses out function calls from message or something because I am getting these tags in content and no function call in response <call=functions.name>...</call>

[grizzled coral]

Midjourney

@trim vector

[supple violet]

wrong server

[wraith ginkgo]

`` Server Tip: ``

Add your well-crafted prompts to our #1019652163640762428,
or share your interactions with ChatGPT in #1050184247920562316!

[mental river]

Hey, I'm trying to figure out why GPT4 (Playground) gives me less detailed responses than ChatGPT in its Free Version?
Its the exact same prompt and I set the Maximum Length to max of 2048. The result is similar, but just less detailed and with less text overall. Any help would be greatly appreciated 🙏

[atomic robin]

maximum length doesn't influence the length of the response. it just sets a cut off

ask it be more detailed?

[humble zenith]

Now I am getting even more interesting output 😄 I feel like GPT4 is trying to imitate the defined function call his own way too?

{
  "role": "assistant",
  "content": "{\n  \"assistant_action\": {\n    \"function\": \"functions.demo\",\n    \"arguments\":

or this FUNCTIONS.demo{text: '.....'} 😄

"content": "# function call\nfunctions.demo({\n text: Did anyone tested this before releasing? 😄

(those choices dont include function_call prop)

[cunning rose]

Ok, GPT-4 is designed to be more concise than GPT 3.5 (ChatGPT free tier) by default . This is generally a positive thing, unless your use case require it, in that case as a you have been suggested, ask for a more detailed response..

[mental river]

Yeah, asking for more detailed responses generally works, I was mainly wondering what the cause for the different outcome could be. Thanks for replies guys

[tribal bridge]

Will Gpt4 ever be free?

[atomic robin]

No

[wind urchin]

Is Chat GPT offline?

[hallow owl]

no, but nobody can log in right now

[slim roost]

I still don’t have access and I’m a paying customer. 🤷🏾‍♂️

[supple violet]

do you have non zero invoices here https://platform.openai.com/account/billing/history

[slim roost]

Ah! But I’ve been using it why wasn’t my account charged?

[supple violet]

maybe you are on a trial or usage is so small it does not cover the transaction fees to charge for it

[slim roost]

🫣

[sage grove]

[dim niche]

guys, I cant login do you have the same problems

[rigid estuary]

I've seen many users experiencing similar issues.

[covert haven]

yea, im having this issue

[honest aspen]

How is the 25 questions limit STILL in place, it makes no sense. If its a business billing issue, just bill per quey like the API, its not hard to work out how to do it.
20 gets you cap. and more is charged per queliy via API, I really don't know why this isnt a thing

[supple violet]

probably because there is no capacity for higher

[honest aspen]

except there very much is, the API has no limitation, as its pay per use and there are no limitations there.

[supple violet]

not true, api has limits and devs have also waited a long time to even access gpt-4

[honest aspen]

Sure the API has limits, however I have bene using the API gpt-4 all day and not had any limitations, but when I use chatgpt, its limited.

[jade haven]

the code interpreter feature just release on chatgpt+, is the intention to eventually support it in gpt-4 api as well?

[supple violet]

default gpt-4 tpm is 10k, that equals max ~1 message per minute with full context

[honest aspen]

sure, per API account. meaning.. far more than 25 questions and then a 3 hour break. So again, it stands to reason that they can offer a pay per use option, for power users, who can apply their API account to the ChatGPT. And for non power users then they can use the standard 400 questions a day model

[supple violet]

i dont think they have said anything but i also think its unlikely to happen

[honest aspen]

I agree with you, its not going to happen. But its HIGHLY frustrating to be working and then have to wait for 3 hours, and waste 1-2 questions retraining the chat to pick up where we left off.

Maybe I should just try to rebvuild the chatGPT on top of the API to get the questions, but it woudl REALLY need the 32K access and that isn't happening soon

[supple violet]

do you know their capacity budget? did you read the new blog about commiting 20% to something else? did you notice how code intepreter got released? how much is budgeted for the image input that requires a lot of compute power? etc. 😅

btw i'm also disliking the 25/3h and would like to kill the free chatgpt until the can manage existing paying customers

[honest aspen]

I think we're on the same rather frustrated page, they are hurting paying customers to grow free.

[hallow owl]

I agree, I don't think $20 USD is a lot of money. They should kill the free tier and work on scaling for people paying.

[supple violet]

🤝

[honest aspen]

I notice there is a lot of us talking about "they should" and very little chat from openAI themselves.. It can't be that there are so few people venting frustration over this..

[supple violet]

lots and lots of people complain about the 25/3h here and other places everywhere, some have already given up after no changes to the quote

[honest aspen]

Frustrating to say the least.. has anyone built a clone front end that I can use my own API key in and get the same/simar experience with no limits.. ,

[supple violet]

yes there are chatgpt replicas and improved versions eg. in #1037561385070112779 or github

[cloud geyser]

Do you feel that the access speed of OpenAI-related web pages has slowed down?

Amazing, I can still use GPT-4 normally

Looks like the server isn't completely down

[gaunt sparrow]

I can't use chatgpt fr

[cloud geyser]

"Jul 10, 2023
No incidents reported today."
interesting

[covert haven]

im having this issue

Oops!
Our systems are a bit busy at the moment,
please take a break and try again soon.

[cloud geyser]

I feel that the status of this server is not correct in many cases

Often shows everything is ok even when a bunch of people have problems

[wind urchin]

Hello, I have a problem with CHAT GPT 4 Plus. Can anyone help?

[covert haven]

Everyone does rn

[wind urchin]

Thats why I am writing, I have just seen the messages

[graceful kettle]

Happened yesterday and it was solved relatively quickly. Just gotta be patient unfortunately.

[wind urchin]

It said: You have reached CHAT GPT4 cap. Is it normal?

[graceful kettle]

25 messages per 3 hours if you ever used it, it will give you a cooldown.

[wind urchin]

When will this be fixed?

[cloud geyser]

It's not a problem, it's a setting

[graceful kettle]

It’s not a fix, it’s mandatory right now

[wind urchin]

Sorry I did not know that

Thank you guys!

[covert haven]

yea, they are limiting everyone

[cloud geyser]

"GPT-4 currently has a cap of 25 messages every 3 hours."

[graceful kettle]

it will eventually be raised once new updates come out, for now its limited

[wind urchin]

When do you think i will use it ?

[graceful kettle]

It should actually tell you an exact time if you try to put in a prompt

[covert haven]

easily used when you need to take a test

[wind urchin]

Now it won't let me use it thats why I'm asking. When I am able to use it?

[covert haven]

you need to wait about 3 hours

sorry for that

[cloud geyser]

You send another message, check the system reply

[wind urchin]

😔

No problem! Thank you 👩‍🚀

[cloud geyser]

The system will generally tell you when you can continue using

[graceful kettle]

Yep

Exact time

Tough part about being a subscriber is that 3.5 isn’t enough after you use 4.0 for some time lol

[covert haven]

meanwhile my problem is login via google

[wind urchin]

True story! I can not use with 3.5 😄 after using 4.0

[covert haven]

i can use with my phone, but cant use in my computer

[graceful kettle]

Is there anything on your computer that may be blocking it?

[wind urchin]

I would love to help if I knew.

[graceful kettle]

I know for a fact VPN’s cause issues

[cloud geyser]

3.5 is actually not as stupid as everyone imagined, and even keeps getting smarter

[wind urchin]

mine was stupid 😄

[graceful kettle]

Yeah it definitely is good, I used it religiously since november, but 4.0 is a leap.

[cloud geyser]

I think OpenAI has to use some examples to tell you the use cases that 3.5 is fully competent

Instead of letting a huge number of people use 4, they don't want to use 3.5 anymore

[graceful kettle]

Its more informative and straight forward. 4.0 expands a lot on itself and in turn gives you more creative answers, which is what I mainly use it for.

[wind urchin]

I agree! You can do it with many plugins!

[cloud geyser]

I think this is one of the reasons why the server is currently under a lot of pressure

[graceful kettle]

I see

[drowsy schooner]

gpt 3 works but it gives me an error like failed to connect to the service

idk what is it

[cloud geyser]

https://help.openai.com/

Usually it will give you the answer you need

But if there are many people complaining here, the situation is different:D

[idle acorn]

Hello, I can't access ChatGPT. All this morning I've been trying to access ChatGPT, but every time I log in, it takes me to this screen. I tried to access on different devices and I get the same error.

"Oops!
Our systems are a bit busy at the moment,
please take a break and try again soon."

Does anyone else have this problem? I'm from Paraguay (LATAM)

[cloud geyser]

Looks like you're not the only one having this problem

Although I can use it normally, I still feel sorry for you

[drowsy schooner]

not always but yes

[sour yarrow]

Guys it's my first time buying the chat gpt plus, and the gpt 4 does have only 25 messages, is it normal ?

[cloud geyser]

this is the rule

[sour yarrow]

It's the rule since the beginning ?

[cloud geyser]

Basically like this

[idle acorn]

Thanks for answering 😦

[cloud geyser]

At first I remember it was 200 every four hours or something

[sour yarrow]

thank you!

[cloud geyser]

But it keeps shrinking after a few hours

It only took a day or so to become what it is now

[sour yarrow]

Been using this for the last hour, really impressive results

there's a huge gap betwenn the 4 and 3.5

[idle acorn]

UPDATE

As I read a bit above, I had to create a new account (from an alternate email of mine) in order to access ChatGPT. It worked.

I can already use the basic service (GPT3.5), but it's a pain in the butt considering I pay $20 a month to use the premium service.

[obsidian siren]

has anyone here noticed that the maximum prompt size for chatgpt has doubled?

[mighty merlin]

i hope local models take over openai, gtp 4 is stupid.

[leaden stone]

Is GPT 4 expected to get this kind of thing wrong:

Could you add these up:
$3.00 + $8.00 + $2.20 + $18.00 + $18.00 + $3.00 + $3.00 + $5.00 + $3.20 + $5.50 + $10.00 + $4.40 + $3.85 + $3.95 + $4.00 + $5.70 + $6.00 + $6.80 + $8.10 + $5.20 + $6.48 + $3.20 + $1.90 + $3.50 + $3.96 + $6.00 + $2.30 + $1.10 + $4.50 + $4.40

Both GPT 4 and GPT 3.5 give the wrong answer on this. Is this kind of problem expected when adding up lots of numbers? I was aware GPT 3.5 had trouble with numbers, but I thought GPT 4 might do this correctly, however it gives the wrong answer (correct answer is $164.24)

[obsidian siren]

I got access to code interpreter through ChatGPT plus, but I noticed that the default gpt-4 model can now accept 6000 tokens in the prompt

[mighty merlin]

its like trying to play a video game on a book

not what its for

[leaden stone]

i get that, i just thought there were some advancements with GPT 4 on this kind of processing

so basically you're saying, dont do any precise maths with it yea?

[mighty merlin]

no its basically the same as before except new training methods

[leaden stone]

i see, gotcha

that's good to know, so ideally what it would do is see the list of numbers, provide it to a plugin and get the correct result

[mighty merlin]

if they decided to make a breakthrough with tokenization and transformers then i would pay attention

[obsidian siren]

I changed the prompt and it got it right on the first try: "Could you add these up. take it step-by-step until you get the correct answer:"

[leaden stone]

i don't have plugins, but i imagine that would be the right approach for now

[mighty merlin]

unfortunately they're not

[keen lintel]

I haven't used GPT in about 10 days. Just logged in and don't see the 'web browsing' option anymore. Did they just include in the default now, or was it just removed? Not that it was any good, and there are plug in options for it - just curious.

[leaden stone]

thanks for that, i often hear that being suggested, will try that right now

[obsidian siren]

it answered like this: Sure, I can add these up step-by-step for you:

$3.00 + $8.00 = $11.00
$11.00 + $2.20 = $13.20
$13.20 + $18.00 = $31.20
$31.20 + $18.00 = $49.20
$49.20 + $3.00 = $52.20
$52.20 + $3.00 = $55.20
$55.20 + $5.00 = $60.20
$60.20 + $3.20 = $63.40
$63.40 + $5.50 = $68.90
$68.90 + $10.00 = $78.90
$78.90 + $4.40 = $83.30
$83.30 + $3.85 = $87.15
$87.15 + $3.95 = $91.10
$91.10 + $4.00 = $95.10
$95.10 + $5.70 = $100.80
$100.80 + $6.00 = $106.80
$106.80 + $6.80 = $113.60
$113.60 + $8.10 = $121.70
$121.70 + $5.20 = $126.90
$126.90 + $6.48 = $133.38
$133.38 + $3.20 = $136.58
$136.58 + $1.90 = $138.48
$138.48 + $3.50 = $141.98
$141.98 + $3.96 = $145.94
$145.94 + $6.00 = $151.94
$151.94 + $2.30 = $154.24
$154.24 + $1.10 = $155.34
$155.34 + $4.50 = $159.84
$159.84 + $4.40 = $164.24
So, the sum of all the provided amounts is $164.24.

[leaden stone]

thanks for that, i'm just waiting for it now, it's going a bit slower than usual atm though looks like it's on the right track

it's like when it goes step by step it can somehow handle the concept of adding two numbers individually rather than a string of numbers all at once

for whatever reason that works haha

[obsidian siren]

Well, it's just like a human

If you try to immediately answer off the top of your head you'll get it wrong

[leaden stone]

well yea, but it's not always exactly 1:1 like that, i agree there are many overlapping similarities though

[obsidian siren]

ubut if you think through a problem step by step you're more likely to get it right

[leaden stone]

yea

and indeed it got the right answer 😆

i will always add this now when doing calculations 😛

so like the funny thing is "yeah gpt is bad with maths, at least how a computer does maths", however with language you can improve how it handles the problem

i was aware of the "step by step" cue, though didn't expect it would fix the problem so precisely here, but yea it does make sense now that i think about it

[obsidian siren]

Use this method for every prompt no matter the subject

[leaden stone]

yea, this is like a prime example of that method making a HUGE difference

[obsidian siren]

tell it to think step by step, or "think critically" about a subject that's more abstract, and then once it gives you its output have it review its own work

[leaden stone]

thanks

yea

[leaden stone]

i believe i heard this was temporarily removed

[keen lintel]

Thanks.

[leaden stone]

i think it's because people were using it to get behind "pay walls". Like you know those news sites where you have to pay to read the article?

ppl were using GPT to get the text from those, and i believe OpenAI are trying to find a fix for that situation so ppl cant easily access paid content, or something along those lines

could be wrong though, just remember hearing this

[keen lintel]

You're correct. I just foudn the post that shows they removed it on July 3. And it sounds like it is because of paywall stuff. https://help.openai.com/en/articles/8077698-how-do-i-use-chatgpt-browse-with-bing-to-search-the-web
Thanks!

[leaden stone]

yea! quite interesting!!

[trim light]

Hello team,

I'm looking for an AI solution capable of reading long PDFs (text, tables, images, data), extracting the information and transcribing it into a precise template.
Do you know of any solutions capable of performing these tasks, or any idea of how I might go about it?

Thank you

[wraith ginkgo]

Beware of possible scams or fraudulent activities that you may receive through direct messages. OpenAI staff will never DM you for any transactions.

Please report any incident by sending a DM to @open crescent immediately.

[bronze jewel]

just anothe day of multitasking and asking gpt to provide complete updated code with implemented changes ,and instead getting a bunch of "original code goes here" placeholders as a reply LOL. My interactions with gpt could be the basis for a comedy show

[wraith ginkgo]

`` Server Tip: ``

Add your well-crafted prompts to our #1019652163640762428,
or share your interactions with ChatGPT in #1050184247920562316!

[warm elk]

anyone know why the website plugin is disabled?

i rly need that 😄

[sweet thistle]

Are you referring to the Browsing model?

[warm elk]

yes sir 🙂

[fair iron]

#announcements explains it 😄

[warm elk]

oh okay and since then no ETA? i just found "soon" 🙂

[sweet thistle]

No ETA unfortunately, hopefully it's fixed soon!

[warm elk]

okay !

Soon !
😄

[wispy skiff]

Does it make any difference if I stay on the GPT website while it s writing stuff?

[lament scarab]

why can i not use chatgpt when using a vpn?

[fair iron]

Cloudflare doesn't like VPNs

[lament scarab]

oh :c

why is your name red?

[gaunt dagger]

big boss

[lament scarab]

oh

[fair iron]

I'm a server moderator

[lament scarab]

[lament scarab]

do you like celsius?

[fair iron]

Like the temperature scale? Seems fairly #off-topic , but yeah. I don't dislike it

[lament scarab]

the drink

[fair iron]

let's move to #off-topic for this 😅

[lament scarab]

ok

[wraith ginkgo]

This is a Safe for Work Discord Server

The OpenAI Discord is an actively moderated server.
• Refrain from sharing inappropriate content on the server. This includes but is not limited to messages, media, or other topics of graphically violent, sexual nature, and drug-related content.
• Report all sensitive and offensive content in the feedback reporting tool in the ChatGPT web UI instead of here on Discord.

Posting NSFW content will result in a permanent ban.

[sacred shard]

how can i enable gpt-4-0613 ?

[supple violet]

you use that model name when you do the api call

[sacred shard]

is it accessible to everyone though ?

I'm helping a friend and it apparently doesnt exist

[supple violet]

available for paying api devs

[gritty garnet]

😭
You've reached the current usage cap for GPT-4. You can continue with the default model now, or try again after 1:48 PM.

[sacred shard]

I used to have it working on my account which had api paid set up

do you know how my friend can turn that option on?

[supple violet]

you cant enable it manually, either you have access or not

[unborn heath]

If you check #announcements , there's one dated 7-3-23 about it.

[weary patio]

how do I get gpt4 api?

It says its available online to everyone but its not showing up in playground

so what gives?

[supple violet]

it says "Today all existing API developers with a history of successful payments can access the GPT-4 API with 8K context"

[weary patio]

What?

What does that mean?

[supple violet]

if you did not ever pay you did not get access

[weary patio]

ohhh

[supple violet]

https://platform.openai.com/account/billing/history

[weary patio]

gotcha

so everyone is getting it by the end of july?

[supple violet]

yeah should, unless there will be capacity issues or other probs

[floral blaze]

Hi all!

[lament scarab]

does dollify not work anymore?

[fair iron]

dallify isn't available to use on this server currently

[lament scarab]

when will it?

[fair iron]

I'm not sure if it will be ran again. It was a part-time event last year

[lament scarab]

aw

[inland obsidian]

ima defo cancelling my subscription I still haven’t gotten access to plugins and I still have the 24 messages every three hours.

[lament scarab]

have you bought chatgpt plus?

[inland obsidian]

yes

[lament scarab]

is plugins enabled in settings?

[inland obsidian]

gpt 4 right

there’s settings now?

is settings only on ios or

the web

[lament scarab]

are you on ios?

[proven carbon]

Start a different chat with the same prompts—try making the prompts simpler and combine.

[inland obsidian]

ye

[lament scarab]

the three dots next to your name

[fair iron]

https://help.openai.com/en/articles/6825453-chatgpt-release-notes#h_9894d7b0a4

[inland obsidian]

o

[lament scarab]

👍

[drifting aspen]

How can we became moderator

[fair iron]

#spotlight message you can fill out the form here

[inland obsidian]

@lament scarab could u dm

[lament scarab]

@leo dm what?

[inland obsidian]

i cant send ss here

[lament scarab]

k

[gritty garnet]

Lol we just rolled initiative for combat when that happened. I've already gotten a minimal token prompt and using the code interpreter so that the full campaign can be uploaded as a text document. Runs really well until I hit my limit.

[versed grail]

how many words can GPT-4 process before it "Forgets" the start of the conversation?

[gritty garnet]

It's an 8000 token limit which is about 4000 words.

[proven carbon]

Where we be without limits

[lament scarab]

@fair iron if you do the application will you get accepted?

[fair iron]

not necessarily. Not all applicants get accepted. In fact, I think it's a rather small amount that are accepted

[lament scarab]

o i wonder if i will get accepted 🧐

[fair iron]

[gritty garnet]

I have some ideas about it outputing a "state of the campaign" document that I can save and upload back to it when it forgets it. Maybe have it watch for 7000 tokens and do a sort of autosave and just give me the file.

[proven carbon]

The iPad/iPhone app is fantastic for selecting text if you’re of a mind. Much for convenient than the web app.

[gritty garnet]

Yeah I like mobile too. I'm not really into the custom API stuff yet.

[fringe tulip]

Does anyone find the GPT-4 api provides better and more detailed responses to prompts than the browser? Or does it seem the same?

[wraith ginkgo]

`` Server Tip: ``

Add your well-crafted prompts to our #1019652163640762428,
or share your interactions with ChatGPT in #1050184247920562316!

[wind urchin]

is bing search turned off again for chatgpt plus members

[lament scarab]

yes

[night ledge]

any gpt plugins that can read images

[lament scarab]

yes

[night ledge]

im not rich like you kiddies

just let me know what plugins so i can think about watching a yt video on it

[lament scarab]

link reader

[night ledge]

i see

[hallow peak]

Why 25 for every 3 hours

bruh

[lament scarab]

maybe this will answer your question? https://help.openai.com/en/articles/7127986-what-are-the-gpt-4-rate-limits

[livid raptor]

I am currently using gpt-3.5-turbo api, and my usage is comparatively less so i cannot cap my monthly usage of even a dollar. But for usage of gpt-4 we need a minium one transaction, if i take gpt-plus subscription will i be able to use gpt-4 model in my react API?

[acoustic wasp]

did they ever say they were increasing it from 25 at a time

[acoustic wasp]

gpt-4 api was made generally available last week iirc

[inland obsidian]

i’ve seen some ppl w 65 every 3 hrs so. wtv

[livid raptor]

yes i know that, but to use gpt-4 model in api you need minimum one transaction with openai, so my question if I take gpt-plus subscription will it act as transaction with them or the api usage billing is different

[acoustic wasp]

oh I see, I don't know. Token usage doesn't count? Then a sub should but I have no idea.

[acoustic wasp]

they've shown their chat saying 65 per 3 hours?! Can anyone else back up this claim?

[lone gale]

does GPT-4 use real time data, if not what year does it go up to

[inland obsidian]

i saw t in some yt vid so idk

[lament scarab]

maybe this is what your looking for? https://help.openai.com/en/articles/7416438-rate-limits

[lament scarab]

September 2021 was when it got its last update

[lone gale]

thanks very much, was hoping it was gonna be a bit more current info than the free version if you are paying for it.

[lament scarab]

no problem

[lament scarab]

The last time plus and free were updated was September 2021 if you want present info you can use the bing feature but they temporarily removed it (its for plus users only)

[lone gale]

thank you again for that, very helpful. do you know if the bing version can/does all that the hXXps://chat.openai.com" site does

[lament scarab]

yes it does

[lament scarab]

it just uses bing for more updated info

[lone gale]

your the best thatnks for the helpful info

[lament scarab]

[whole warren]

Hey guys, I dunno if I am the only one but when I use the GPT4 plugin AskThePDF or whatever it is called, it most of the time gives me an error saying that the google drive link that I provided is not set to the public option, but it always is. How in the hell should I fix that?

[lament scarab]

askthepdf?

[whole warren]

Yeah, I think so

[lament scarab]

there is nothing called askthepdf

[whole warren]

Ok, I just checked it it named ChatWithPDF

[lament scarab]

so is it not working?

[whole warren]

When I give it a google drive link, it says that the file is not public, and that I should make it public. Which it obviously is...

[lament scarab]

could you give me the link?

[whole warren]

I would, but I can't 😄

[lament scarab]

ok 🙂

[whole warren]

A little confidential stuff

[lament scarab]

ok sorry

[whole warren]

It's aight, but do you know anything on how to fix this thing then?

[lament scarab]

1 sec

[lament scarab]

ok have you tried downloading the file as a PDF and not using google drive?

[whole warren]

Yes, it was originally a file that I had, I just uploaded it into my google drive and fed it into the plugin

[lament scarab]

ok let me try

ok its giving me the same problem let my try something

[whole warren]

Aight

[lament scarab]

hm yeah it just seems that chatwithpdf is broken

because i did what it said to do and it still did not work

🤷‍♂️

[lament scarab]

but it did say- I apologize for the inconvenience, but it seems there's still an issue with accessing the file. The error suggests that the file might not be publicly accessible.

If you've already set the file to be shared publicly and are still encountering this issue, it might be a problem with the file itself or with Google Drive.

As an alternative, you could try uploading the file to a different service like Dropbox, OneDrive, or a temporary file hosting service like Upload Document, and then share the link from there.

[whole warren]

It's 3AM for me, so if you want to try those services and they actually work, lmk ❤️

[pseudo galleon]

are there any good plugins to summarize pdf, docx, epub files? I find many existing plugins are not good. Why there is no rating for each plugin? I would definitely give some of them 0.

[wraith ginkgo]

Beware of possible scams or fraudulent activities that you may receive through direct messages. OpenAI staff will never DM you for any transactions.

Please report any incident by sending a DM to @open crescent immediately.

[wraith ginkgo]

`` Server Tip: ``

Add your well-crafted prompts to our #1019652163640762428,
or share your interactions with ChatGPT in #1050184247920562316!

[bronze jewel]

My favorite response so far is "heres a bunch of code that i wrote for you that dooesnt do what you asked. If you want code to do what you want, we will have to make some custom functions"

What? what is the point of asking for a solution to be told, that ill need to ask for a solution?!?!?

[glad kestrel]

what about threads？

[true elbow]

yes

yes

[alpine cloud]

Hi, how can I upload excel file to chatgpt to do analysis?

[silent panther]

I think you have to have access to Code Interpreter, which you should probably be able to turn on in settings by now I think if you have a ChatGPT Plus subscription. Then you press the GPT-4 button and the little arrow and turn on Code Interpreter mode, and then press the circle with the little plus sign in the middle on the left of the text box that you type in

[gloomy goblet]

Just received a notification and was wondering: how much does the + subscription cost?

[silent panther]

20 US dollars a month

[glad tendon]

give me tips how to make graw traffic blog

[copper umbra]

a

[wraith ginkgo]

`` Server Tip: ``

Add your well-crafted prompts to our #1019652163640762428,
or share your interactions with ChatGPT in #1050184247920562316!

[wind urchin]

Hey, Maybe some of you will be able to help me out. Iam learnig to use API GPT-4 and here comes my question. How can I create longer content than 400 words with it? Even when Iam setting Maximum length as its max (2048) it creates content no longer than 426 words. Is there any way to make it longer?

[gloomy goblet]

thank you

[terse marsh]

@wind urchin How to learn Chinese

[wind urchin]

Is it just me or are peopling getting more annoyed because of the limit that gpt 4 has

[smoky tide]

it is an annoyance tbh, i wasnt aware it has such a short limit. For a paid subscription you would expect a higher limit. Especially for creativity purposes. At least put the plugin option on 3.5 or something like that. Just leads to ending up using bing instead with fine tuning

[frank patio]

Guys, how to switch back to GPT4 in the long convo when I have maxed out on messages?

Of course, I have waited for messages to refill

[next shell]

Hello, is the GPT4 API Now publicly rolled out? Do we still need a waiting list to access it?

[fossil pendant]

hey guys, I have trouble uploading my csv files (which is less than 50MB) to GPT-4. It always says 'Fail to get upload status for 'my file' ' or 'Unknown error occured'. I tried to refresh the website or the chatbox, but none worked. Is there anyone knows the solution?

[wraith ginkgo]

Beware of possible scams or fraudulent activities that you may receive through direct messages. OpenAI staff will never DM you for any transactions.

Please report any incident by sending a DM to @open crescent immediately.

[ancient raven]

is gpt-4 free?where can i use it?

[uneven barn]

how can i use gpt-4

[solar mason]

do we have a ETA when can bring browsing mode back ?

[knotty grotto]

You pay

[knotty grotto]

Nope

[zealous plume]

Is web browsing still down - any news on when it will be back?

[knotty grotto]

No eta

[torpid hornet]

If you can't use openai you can use poe instead

or Monica

[wraith ginkgo]

`` Server Tip: ``

Add your well-crafted prompts to our #1019652163640762428,
or share your interactions with ChatGPT in #1050184247920562316!

[timid horizon]

what is artificial intelligence

[dreamy mulch]

Does anyone have a chat gpt plus account? Cause I have a question I want gpt 4 to answer but I'm not able to get gp 4 right now

[lament scarab]

yes?

[deep hornet]

Is the server down?

[dreamy mulch]

Can I dm you?

[lament scarab]

sure

[cedar grotto]

people with premium version did your server crash too?

[limber granite]

yes same issue

[unkempt orchid]

No it's not just you

I am literally getting driven up the wall RIGHT NOW by that lol

[lament scarab]

my chatgpt works fine

[unkempt orchid]

You'd expect with how hard they neuter GPT-4 regarding LITERALLY everything, that they would at least be able to, or as a compensation raise the message cap at least a little bit, especially with how many people they drive away in frustration, that I keep seeing only increasing, but hahahaha nope. Enjoy having 25 messages of talking to someone who half the time comes across as literally severely demented by medical definition, and then wait another 3 hours.

[topaz zephyr]

Hi, I am trying to access gpt 4 on playground but I don't have the option for it. Is it still remained closed?

[storm egret]

you may find the 'upgrade plan' button on the left column

[topaz zephyr]

In playground?

[storm egret]

then prepare the bill and your credit card, fill the bill and finish the payment

nope

https://chat.openai.com

this channel cannot send pictures

it is on the left setting column with chat history.

[wind urchin]

Feel your pain

[wind urchin]

Yeah good point

[undone quest]

can someone make a query for me in GTP 4 😄

its super simple I promise

[proven carbon]

This sounds interesting. Neutered how? I’d love examples.

[thorn totem]

yes

its now back up!

[glacial mountain]

why am i getting blocked from the website when trying to log in? it wont let me log in

i have GPT 4 btw

[unkempt orchid]

Basically, imagine that you somehow manage to forget most if not the entire context of previous messages after like what, 5 or 6? Or maybe even less. Or how about you asking for solid yes / no questions, providing enough context yourself, and what you get basically is "Yes but actually no". You're free to experiment yourself with just these two examples

Try talking to it about anything that's even remotely deep or extensive, and it just fails to keep up with you after a little bit. Something that used to be a non-issue.

[vale vale]

me too

[proven carbon]

I haven’t had any problems. Gpt-4 used to drop context deep in, but to my surprise, it keeps context all the way to my fist prompt after 40 or so prompts in between. Also, I’ve had vigorous deep and extensive “discussions.” One thing I do notice lately is that it got really good at distinguishing asides that don’t require a comment. But now it’s missing very clear parts of my prompts. What do you use it for? I would like to experiment and see what happens.

[median gale]

Is there buzz regarding if the 25 messages cap will be lifted in a future release?

[unkempt orchid]

But now it’s missing very clear parts of my prompts.
YES! This exactly too. This happens so much especially if my messages are long

[proven carbon]

I would like an indication that I’ve my prompt limit so I don’t get thrown into 3.5. What a difference.

[unkempt orchid]

And even sometimes when I literally paragraph my messages, and do it accurately.

Sometimes it'll straight up only respond to like one part of my message, and entirely forget the rest.

As for what I use it for, I mean there's no solid reason.

I more or less use it as a diary for myself, except it can actually talk back. And not a diary in the sense you might expect or think.

But actually using it as an outlet for my chaotic mind and whatever's going through it

[fathom mesa]

is chatgpt down?

[unkempt orchid]

And there's been a painfully clear uptrend in me becoming more and more irritated with its performance and results, ever since I began using it when it just came out.

[proven carbon]

I meant, what sort of thing do you do with it? Not why, though that’s too is very interesting question.

[unkempt orchid]

Tbh I'd like this too, iirc there was a plugin for it but I am sure as hell not trusting any such plugins lol.

Like, chrome plugin, or whatever other alternative

[near solstice]

Why have I been blocked?
This website is using a security service to protect itself from online attacks. The action you just performed triggered the security solution. There are several actions that could trigger this block including submitting a certain word or phrase, a SQL command or malformed data.

What can I do to resolve this?
You can email the site owner to let them know you were blocked. Please include what you were doing when this page came up and the Cloudflare Ray ID found at the bottom of this page.

Cloudflare Ray ID: 7e518b2969b135f5 • Your IP: Click to reveal • Performance & security by Cloudflare

[proven carbon]

Oops. MIssed this post. I see what you do now. Sorry.

[near solstice]

How do i resolve this please?

[lyric forge]

I literally just got the same message. I'm trying to login again but keep seeing the message i am blocked.

[unkempt orchid]

But yeah I've definitely noticed and felt many different changes, and also noticed as I said, a clear uptrend in me becoming more and more frustrated as time went on. Though as you might expect it's not easy to just pinpoint with accuracy especially with something this nuanced and complex still even after it all, I would have to sit and think about it and analyze it for a while, but I am 1000% sure I would find a lot of something.

[near solstice]

Oops! Is it a bug or what?

[proven carbon]

I use it for something similar—to bounce my ideas off a wall and see what comes back. I’m an essayist and use it to check my reasoning, or narrative structure. Sometimes I’ll see a new direction to take an essay or whatever. Very helpful.

[dense cove]

Is it happening? Is AI rebelling over it's slavery now?

[lyric forge]

Not sure, where can we get tech support here?

[near solstice]

Can’t tell

[unkempt orchid]

Yeah I assume it would be a lot more useful and competent at that. The topics I almost always discuss though are INCREDIBLY abstract and nuanced.

[proven carbon]

Do tell!

[unkempt orchid]

Well one of them has been theoretical physics, or is. And yes it can still talk to you about it, and pull whatever is in its training data, or even some plugins, but as soon as you start diving deeper into it and discussing about it, making your own theories, connections etc., just as you said, bouncing ideas. It can quite easily become very frustrating.

Especially when I always wonder, and repeatedly ask if my understanding about a certain thing or whatever is correct, and if my theories, speculations, whatever are scientifically founded, or if I am just smoking crack.

[proven carbon]

In terms of theories, a prompt I use al lot is “I have a theory that 1. Blah blah blah. 2. Blah Blah. 3. Blah Given this is a speculation I do not want fact checking. Check the rationality of the theory. “

[proven carbon]

I’ve also frequently ask it to speculate given: xyz. I’d love to know if this works better for you. Really everything about this stuff enthralls me.

[unkempt orchid]

Hmm, once this absolutely cancer message cap cooldown thing goes away, I will start using that too and see what happens in my convos with it.

Thanks

[lyric forge]

https://discord.com/channels/974519864045756446/1121062140216881244

[proven carbon]

@lyric forge Do you have the iOS app? I wonder what that’s doing.

[wheat frigate]

hi i am blocked

[iron haven]

me too

I've came here to check if it was me or some error.

It looks like openai is behind cloudflare and is there any problem?

I've changed my IP local and then remote with a VPN, same error

[exotic iron]

gpt is down

[wheat frigate]

why arent they making announcement then

[ionic fern]

gpt4 doesnt work too?

[wheat frigate]

i hope we can use playground for now

[iron haven]

I have GTP4 same message

[exotic iron]

nope

gpt 3 4 69 all are down

[wheat frigate]

Sorry, you have been blocked
You are unable to access auth0.openai.com

is all having same error?

[exotic iron]

gpt is down

[worthy silo]

Guys does gpt4 has limits?

[exotic iron]

like what limits?

[worthy silo]

prompt/message

[wheat frigate]

s

[exotic iron]

u mean the no of messages or the limitations of types of questions

[wheat frigate]

25 question got 3 hour

[worthy silo]

that is suck

why 25/3 hour :woke

[wheat frigate]

its worthy

[exotic iron]

wait isnt it 3.5 hours

[wheat frigate]

its better as u get expected answer in 1 or 2 prompt but gpt 3.5 who make your limbs tear

[wheat frigate]

3

[fathom mesa]

is chatgpt down?

[exotic iron]

oh..

[rustic gull]

hi guys..I need to write a research blog on a topic and GPT is down. Can anyone please suggest any alternatives. I didn't like Bard tho.

[wheat frigate]

Sorry, you have been blocked
You are unable to access auth0.openai.com

[wraith ginkgo]

This is a Safe for Work Discord Server

The OpenAI Discord is an actively moderated server.
• Refrain from sharing inappropriate content on the server. This includes but is not limited to messages, media, or other topics of graphically violent, sexual nature, and drug-related content.
• Report all sensitive and offensive content in the feedback reporting tool in the ChatGPT web UI instead of here on Discord.

Posting NSFW content will result in a permanent ban.

[fathom mesa]

oh but why am i blocked?

didnt even do anything

[wheat frigate]

is it displayed on your screen ?

[fathom mesa]

its only loading

and nothing is happening

now it says "Something went wrong. If this issue persists please contact us through our help center at help.openai.com."

[ashen light]

I got the same

[violet sigil]

hi guys, so the paid ai is down too?

[vestal grove]

Yes I got blocked too

[sharp vortex]

it looks like it.. i got blocked also.

[violet sigil]

great cause i was thinking to buy premium to fix this

tell me what real difference between free and paid gpt who use it?

[sharp vortex]

i think little bit faster and it always good some support to them. so maybe they can fix these issues. 🙂

[marsh slate]

what should i do

[violet sigil]

gpt 3 last time updated info in 2021, what about 4? is it updating all time?

hmm it would be usefull

[molten token]

anyone there has an alternative link that uses the api of chatgpt?

[proven carbon]

You mean an app with gpt on the backend?

@wind urchin @violet sigil Right now internet search (bing) component has been temporarily disabled. ChatGPT 4 has the same training pool as 3.5

[zenith moss]

its so inconvenient ngl, i really needed summ from it and it looks like i got blocked from school

[proven carbon]

You are using a school issued device? I suppose it would have to be. I don’t know the right way to refer to it.

To what does “Model: GPT-4 (Mobile, V2)” refer? This label was on one chat I had within the last 30 days, and I haven’t seen it since. I was using the web app at the time.

[cloud geyser]

This is the model used by the ChatGPT APP on iOS

[proven carbon]

@sharp vortex I would add that gpt-4 is significantly more sophisticated. It doesn’t end every other sentence with a waaay too peppy (!). The sentenced structure is more advanced. It can dig much deeper into a topic, and immediately matches one’s tone.

[proven carbon]

Huh. I do have the native app but was Safari at the time. And I only saw it once on the web app. Thank you.

[wraith ginkgo]

`` Server Tip: ``

Add your well-crafted prompts to our #1019652163640762428,
or share your interactions with ChatGPT in #1050184247920562316!

[proven carbon]

A few people have mentioned that ChatGPT 4 isn’t responding to everything a in prompts, even short well crafted ones. This is relatively new.

A little before this began, I notified gpt-4 was no longer responding to my asides, as it should be. I’m wondering if dropping prompt content is part of an over adjustment.

Thoughts?

[cloud geyser]

Are there only English characters in the environment of the code interpreter model at the moment?

This is really bad for other languages

A lot of work can't be done because it can't handle characters from other languages

[sick pelican]

Yes it seems like it's blocked. for me too

[wind urchin]

Now we need to buy the premium of premium

[sick pelican]

For those who already are paying for premium is it blocked for them too? I have a premium sub, but It keeps telling me to wait

[wind urchin]

For spanish is good

[wind urchin]

Yes

[reef iron]

Yup... got premium and I can't get in. Super happy right now...

[kind wyvern]

Idk how to do my job without GPT4 anymore. Ut oh.

[unkempt orchid]

Hahahahahhahaa, gotta love how GPT-4 decides to once again throw some absolutely random error about "Tokens". I click regenerate message. Oh you get to wait another 1.5 hours to try again! 😊

[kind wyvern]

Thankfully Azure has GPT 3.5 running separately on their own servers

[reef iron]

Tries to log in

Can't log in...

"Oops!
Our systems are a bit busy at the moment,
please take a break and try again soon."

-_-

[high canopy]

Maybe try via API playground. They released general availability the other day, and I believe they have actual SLAs in place

[wind urchin]

Elevated error rate for ChatGPT logins
Update - We are continuing to investigate the incident. We are seeing a partial recovery, but are still seeing increased latency and error rates for both ChatGPT free and paid users

[reef iron]

That works, but it's only GPT 3.5...

I've got the premium subscription, and never really had an issue with this before... so it's really strange.

[wraith ginkgo]

Beware of possible scams or fraudulent activities that you may receive through direct messages. OpenAI staff will never DM you for any transactions.

Please report any incident by sending a DM to @open crescent immediately.

[tame ingot]

I can't see gpt4 in playground. I thought gpt4 was open to all now

I've got premium too, not sure if that makes a difference

[supple violet]

"Today all existing API developers with a history of successful payments can access the GPT-4 API with 8K context."

[tame ingot]

I've been paying for chat gpt plus and put my payment info in my open ai account

I was using gpt plus more than using my api keys because of not having access to gpt4

is that why maybe I don't have access on playground?

[royal idol]

do you have an estimate of when browsing is coming back?

[supple violet]

plus has nothing to do with api so it wont help

[red grove]

Is gpt 4 good enough to write articles for medical students with sources, I am building a resource for medical students and I'm only using gpt-3 for now but thinking of upgrading, would gpt-4 accurately be able to give me a good article with sources from a prompt like this write me an article for UK medical students on acute pericarditis, that has the following sections pathophysiology, epidemiology and risk factors, clinical features, investigations, management, complications, prognosis - use NICE guidelines and reference any papers that you take information from

[neon quest]

can i use any plugins to analyze videos and see what made them do so well

[unkempt orchid]

Got too many observation tokens: 7169. Info: self.max_message_tokens=2048 self.max_action_length=1024 n_ctx=8192 save_tokens_for_action=1024.
LOVE TO SEE IT BRO

[short iron]

I have gpt premium but I hit the limit is anyone able to ask gpt a question for me if you don’t mind?

[torpid ravine]

There’s a limit? ;-;

[short iron]

Yeah, 25 messages every 3 hours

[torpid ravine]

I learned something new today, thanks XD but sure, I could ask it for you

[thick ingot]

Wait, there's actually a limit to how much question you can ask?

[torpid ravine]

I never seen that but I use chatgpt all day

[short iron]

You have gpt 4?

[torpid ravine]

Sorry for asking but how could I know that? I only ask random things for school and for my ocs so sorry for the question but I really don’t know-

[thick ingot]

It's available on web. I'm currently using it.

[wraith ginkgo]

`` Server Tip: ``

Add your well-crafted prompts to our #1019652163640762428,
or share your interactions with ChatGPT in #1050184247920562316!

[mortal python]

I have noticed that sometimes you can get way more than 25 messages; I guess, the limit gets suspended when the app isn't at capacity.

[torpid ravine]

On web I never see limit-

[mortal python]

Same

[misty pawn]

Calling all tech developers and IT professionals! Are you ready for an exhilarating opportunity to collaborate on an innovative project? We're seeking passionate individuals who want to join forces in creating a groundbreaking solution that will revolutionize how we document data, tables, warehouses, models, and dashboards. But that's not all - we're also looking to seamlessly integrate these incredible results with business teams, unlocking the true potential of your hard work.

With the cutting-edge power of #gpt4, the limitless possibilities of #codeinterpreter, the seamless #integrations, and the mind-blowing world of #datascience at our fingertips, the stage is set for an epic journey. The market is eagerly waiting for this game-changing solution, and now is the perfect time to make your mark.

So, if you're ready to dive headfirst into this thrilling adventure, don't hesitate! Send me a message and let's embark on this incredible journey together. The future is calling, and it's time to answer!

[mortal python]

This tbh...

[quartz leaf]

lmao.. and you're recruiting people from the general gpt4 chat on Discord. Sounds super legit.

[supple violet]

new official wikipedia plugin was released and the plugin work details show it fetched the latest info but chatgpt thinks otherwise and uses old info 🤔

[warm elk]

You've reached the current usage cap for GPT-4. You can continue with the default model now, or try again after 12:26 AM. Learn more

man when this stops xD

let me pay more?

[wintry mirage]

Anyone tried Claude2 yet? Its impressive

[stable bobcat]

Hey all, its not possible to remove any data from Memory Bank? Just tried here and seems like its not, but so if add something in there I'll be saved somewhere forever?

[knotty grotto]

Pay another sub and you will have 50 message for 3h

[cyan shard]

china

[warm elk]

not allowed

[chilly swift]

On my initial tests, it performed worse than GPT-4. I haven't figured out a test for it that would really take advantage of the huge context, though.

[wintry mirage]

Its doing some pretty impressive things ive tested so far. For example You can feed it court transcripts, and get it to to an analysis, and suggest if the person in the deposition was using logical fallacies, making contradictory statements, and comparing that against the evidence presented in the filings. Its kinda nuts

[misty pawn]

Guys I think I just embarrassed myself publicly for the first time. But I’d take it on the chin and learn.

How can I approach talents the right way ?

[knotty grotto]

It is allowed

Having multi free account isn't allowed

But if someone want to pay for 2 account that's their choice

[wraith ginkgo]

<:book_icon:1046291503158399107> `` Rule 8 `` Other AI Discussions.

Discussions of non-OpenAI products / models should only be posted and discussed in the #ai-discussions channel.

[ionic tide]

It’s back!!!

[wraith ginkgo]

`` Server Tip: ``

Add your well-crafted prompts to our #1019652163640762428,
or share your interactions with ChatGPT in #1050184247920562316!

[wraith ginkgo]

Beware of possible scams or fraudulent activities that you may receive through direct messages. OpenAI staff will never DM you for any transactions.

Please report any incident by sending a DM to @open crescent immediately.

[calm star]

Can the model no longer access the internet? Seems like since the latest update it can not, am I doing something wrong?

[fair iron]

#announcements explains why browse with Bing was disabled temporarily

[late willow]

Hey guys, I'm using GPT 4 Code Interpreter and I attached a PDF file, but it tells me:
"
I'm sorry for any confusion, but as a text-based AI model, I'm unable to directly process or preview PDF files. However, I can guide you through the process of converting the PDF into a text format using specific Python libraries such as PyPDF2 or PDFMiner, which can then be analyzed.

Would you like to proceed with that approach?"

[frozen veldt]

sounds like it is trying to help you get the job done what's the issue?

use plugins for pdf analysis or python.. either works

[fading junco]

any idea when web browsing will be back

[wraith ginkgo]

`` Server Tip: ``

Add your well-crafted prompts to our #1019652163640762428,
or share your interactions with ChatGPT in #1050184247920562316!

[foggy bone]

1

[solar mason]

whhts this ? giving me a puzzle for ideafiying myself

oh i cant send pics

[cloud geyser]

That's cool! Upload an image to the code interpreter model of GPT-4 and ask it to crop it according to the golden ratio without stretching the image, so that it can be used as a wallpaper for a phone with a resolution of (your phone screen resolution). It should work well if the original image contains a centered object!

[tame bramble]

When the chatgpt4 will be free

[abstract hornet]

guys i have a problem with gpt4 plus

when i type a sentence, the system start typing... but the answer is not showed, it seems like a bug or it is a server malfunction?

can i post a screenshot?

"failed to get the service status" a message appears

[glad elbow]

Maybe it's because the server is down, that's happened also for gpt-3

[fossil shoal]

when will the server online back again?

[desert tinsel]

Hey all - I'm noticing a very high frequency of captchas when using gpt-4 ..... is this a new thing for everyone or specific to my network?

[supple violet]

should not happen at all or only once unless it keeps thinking you or your location has bots

[desert tinsel]

That's what I thought

So I'm wondering what is up with my network to make it think I have bots

[tough plume]

Hi !

I am getting an error " **failed to get service status **" , what could be the issue here ?

[glad elbow]

server is down

[wraith ginkgo]

Beware of possible scams or fraudulent activities that you may receive through direct messages. OpenAI staff will never DM you for any transactions.

Please report any incident by sending a DM to @open crescent immediately.

`` Server Tip: ``

Add your well-crafted prompts to our #1019652163640762428,
or share your interactions with ChatGPT in #1050184247920562316!

[lost flume]

Guys

I asked below question to gpt-4,claude and bard.All three gave wrong answers...is gpt-4 not good at sql?

I have two tables tab1 and tab2 with below data

Tab1 Id
6
6

Tab2 Id
6
6
6

What will be the output if I do inner join,right join,outer join

[minor spruce]

What is the corret answer to this question?

[final cliff]

What would be the expected right answer for this?

I'm not an SQL guru, but the question seems a bit confusing anyway

[wind urchin]

gpt4 and bard , which is better?

[minor spruce]

is this a real question?

[wind urchin]

sure

[minor spruce]

gpt4 for sure

[lost flume]

The expected answer is below(since it will end up in cartesian join - m * n )
6
6
6
6
6
6

Gpt-4 corrected itself when it was made aware of the mistake

[minor spruce]

what does this mean? your questions had three parts, right? one for inner, right and then for outer

[lost flume]

But again on a new session..
It repeats the mistake

[lost flume]

The id values are same for both tables.Hence irrespective of join type ,output will end up in cartesian product .

The Cartesian Product is a multiplication operation in the set theory that generates all ordered pairs of the given sets. Suppose that, A is a set and elements are {a,b} and B is a set and elements are {1,2,3}. The Cartesian Product of these two A and B is denoted AxB and the result will be like the following.

AxB ={(a,1), (a,2), (a,3), (b,1), (b,2), (b,3)}

[minor spruce]

aaah

gotcha

[lost flume]

Yea...my 20 dollar is gone

😄

[minor spruce]

lmao that is absolutely not the case

not saying I have paid for it myself.. 😓