gpt-models

[vague yarrow]

i want code js of login with google and facebook

[wise trout]

I asked GPT this question: What specific keywords should I use to conduct a search on sources?
But it keeps throwing this error:
This model's maximum context length is 8193 tokens, however you requested 8433 tokens (6897 in your prompt; 1536 for the completion). Please reduce your prompt; or completion length.
Could somebody help?

[feral zenith]

+1. The current gpt4 is such a downgrade from what it once was.

[spice plaza]

Cut the text in 2 parts and prompt, do not respond until I’ve sent you page 2. Here is page 1:

[spice plaza]

Facts, it would spit out JavaScript much smoother, now GPT seemingly complains or glitches out when responses are too long, then using the >> continue button might make it repeat itself over and over again.

[wind urchin]

what is the model size for chat gpt 4

[wintry mirage]

Anyone seen the reporting that came out from microsoft on longnet transformers? They have proven algorithms with insane token context limits in the first iteration.

[reef iron]

Does anyone know how to report bugs to OpenAI for the UX on ChatGPT?

I've found a super annoying bug.

When you write/ type a prompt into chat GPT and press enter TWICE fast after each other, it then deletes your message and sends a BLANK message, meaning I have to re-type the message AGAIN, which wastes a LOT of time...

Sure I can be careful when I write to chatGPT, but I spend hours with it every day and sometimes I accidentally press ENTER twice fast anyway...

I'd assume this is just a bug and can be fixed.

Just found the Bug-Report chat on discord... I'll send it there

[high cosmos]

if I subscribe to GBT+ will my access to GBT 4 be limited?

[stuck gull]

anybody 's gpt 4 down?

[fair iron]

There is a limit of 25 messages every 3 hours for GPT-4 with the Plus subscription

[worn ginkgo]

anybody know why plugins arent accessible since yesterday.

[tardy knot]

I still have plugins

[spice plaza]

Exactly! I’m not the only one, this began like 2 days ago.

[tardy knot]

I have had this continously in my code, to the point that i have provided the code with comments, like "#section 1", and tell it to provide me the code between those tags so that it doesn't drift off into the weeds and provide me 5 pages of garbage. What's worse is when it confuses itself, apologizes, and starts over again. Then gets lost, and does it again, and then gets lost, and stops. Then completely forgets whatever it was working on.

[wraith ginkgo]

`` Server Tip: ``

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

[inland bramble]

So i have a quick question right now i am working on conversation with chat gpt when we close the browser tab ( like google chrome ) do we lose all the implementation of info we send to chat gpt ?

[supple violet]

it will be saved unless you have disabled chat history from the settings

[inland bramble]

but here is what i get as an answer from chatgpt it self ( Apologies for any confusion earlier. Yes, closing the browser tab effectively ends the conversation session with the AI model, and it won't retain any memory from the closed session. When you open a new browser tab and initiate a new conversation with the AI, it will be as if you're starting from scratch. The AI does not have any memory of past interactions or sessions, even from the same browser or device, once the tab is closed. This approach is designed to ensure user privacy and data security. )

[supple violet]

its making it up or referring to some other ai

[inland bramble]

ok got it!

[errant shard]

Web browsing is still down eh?

[noble jay]

1

[gilded parrot]

ok it made a box i can move back and forth on the screen but when i tried to make it give me a jump option the box just flies up, lets see if it can self correct

[runic marlin]

This is the most plausible answer I got from GPT-4

(As an AI, I don't have personal experiences or memory. My responses are generated based on patterns and information I learned during my training, which included a diverse range of sources, such as books, websites, and other texts, up until September 2021. However, during a single conversation or session, I do have a temporary context window that enables me to refer back to recent parts of the conversation and generate responses that are relevant to the ongoing discussion.

The context window has a limitation on its length, and this determines how far back I can refer within the same conversation. Once the context window is filled, the earliest parts of the conversation start to be replaced by more recent inputs. Therefore, in a long conversation, I can't refer back to the very beginning.

As for changes or updates to my system, as an instance of AI, I don't have awareness of any specific modifications or updates made to my architecture or capabilities after my training cutoff in 2021. Therefore, any perceived changes in my performance would be conjecture on my part.

The goal of the team that maintains and updates AI models like me is to improve the utility, safety, and user experience of the AI. Sometimes, these updates might involve trade-offs between different capabilities, such as the breadth of knowledge, conversational coherence, response length, user privacy, and so on. It's possible that adjustments to these aspects could lead to perceived changes in my capabilities. However, I can't confirm any specific updates or changes because I don't have access to such real-time or post-training information.)

[gilded parrot]

ok so 4 prompts total, few minor corrections and I have a little box dude that can move around and jump. at first he flew off the screen, then he was hopping up from just under the screen, but it found the problems fast and fixed em. pretty simple little test but still didn't seem THAT bad today.

well had it generate 4 versions of the code total i mean

But yeah the running theory i think in that reddit thread was other like performance or safety changes affected it maybe not intentionally

[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.

[dusky iron]

Dear OpenAI, Please give me 500 b Token model, in at least 15 years, Thank you.

Hehus, Theres a lot of Moderators since i last looked i saw Lugui and 4 others, holy..

[silent fjord]

I was literally just complaining about this in the other channel. GPT 4 is definitely worse now than it used to be.

[long viper]

Came here to say this

It’s getting so dumb I am actually depressed by the regression

[spice plaza]

“Apologies for the previous response.” “Apologies for the incomplete response.” “Here's the continuation.” The phrases that will haunt you

[forest silo]

How big is the difference between GPT 3.5 and GPT 4 in coding?

[acoustic sage]

So what kind of code we can use? Any code?

C? C++? C#? Java? etc.

[unkempt orchid]

You can ask ChatGPT to analyze data, create charts, edit files, perform math, etc.
...Can't you do this, and code interpreting already with plugins? What's the point in making this standalone lmao.

[gilded parrot]

ok now its having some trouble trying to get my guy to jump off moving platforms right but we'll see if we can coax it

[unkempt orchid]

First impression anyway, I've yet to look into it properly

[tropic lantern]

Not even a question. LMAO. ITS HORRIBLE

GPT 4 can do 1/3 of the things it was able to do 3 months ago

[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.

[tropic lantern]

its pathetic

[unkempt orchid]

Yeah congrats OpenAI for making GPT-4 a useless piece of s**t progressively as time goes on because. mUh EtHiCs

Because heaven forbid some window licking journalist with 0 foresight or innovative thinking will purposefully probe ChatGPT to say sth stupid, and then write a scathing article

[wide ingot]

do ethics really matter in terms of helping us with coding?

[static isle]

GPT-4 is suppossed to be general purpose, not specifically for coding

[supple violet]

you can use max 3 plugins at the same time and none allow do the same with any combination

[static isle]

unethical things can still be done with code too yknow

[worthy parcel]

"open"

[unkempt orchid]

Oh right, aside from all the memes of OpenAI's name being just amazingly hypocritical given that they have been a closed, for profit company for a while now.

And there's nothing "Open" about waking up every day and going to interact with a tool that gets progressively more garbage each passing day.

[spare swift]

If GPT-4 is being "dumbed down" in terms of all relative performance such as code - then it's obvious what's happening is them downsizing things to simply serve more demand and decrease compute expenses.

The pre GPT-4 code interpreter was actually really really good. I am sad that it's gone.

[unkempt orchid]

And yet, we have had the annoying as hell "25 messages every given 3 hours" cap for ages now, even when you're already paying like $24 a month lmfao.

[lament wadi]

i'm pretty concerned with how dumb GPT4 is getting, it is barely of any use to me anymore at work or in my dev job, i may stop my sub if this keeps going tbh

[static isle]

What do you want them to do instead?

[lament wadi]

if they're dumbing down 4 to make way for 5 then that's both a d!ck move to paying users and won't make me buy into their stuff ever again

[tame glade]

I see Bing search is gone, is there any alternative plugin that I can use so that gpt-4 can use data from search engines?

[unkempt orchid]

I don't even use it for coding lmfao, I simply use it more or less as a diary of some sort, except it can actually talk back to me. And even with a pretty mundane purpose like that I am routinely hitting limits already, for the past few days it also lowkey feels like its memory has gotten even worse.

[tropic lantern]

had to stop my sub. that was not even a question. not worth $20 at all!

[lament wadi]

considering copilot atm, ngl, especially since i can have it for free as a student

[copper moat]

How do we activate the browsing on gpt app ?

[lament wadi]

it's sad to see that openAI is going into the same pitfall as DungeonAI's creators

censorship -> much worse experience for the same price -> people leave

[static isle]

Can't

[copper moat]

Hmmm thought I read we could, thanks

[static finch]

I upgraded my account now to ChatGPT 4, but I can't locate the Bing AI icon search on my account. Could you please advise me on why this is happening?

[lament wadi]

it's gone

you can't use that anymore

[static finch]

why?

[spare swift]

I agree it's annoying, especially for the code interpreter. But just because people are paying for things does not mean a company is profitable, or can sustain expenditure rates.

If they removed the cap - demand would increase, and expenses would be the same or probably higher than before, even though the model is dumbed down.

Obviously they're working towards a no limit GPT-4 - ChatGPT itself has been released for 6 months now, so the whole "its been ages" thing doesn't really make sense when you think about it.

[lament wadi]

because openAI crawling the web like this infringes on a ton of website's policies

[unkempt orchid]

Oh my god, don't even get me started on censorship lmfao. What's absolutely hysterical is desptie all the censorship, if you try to talk to it about current societal issues, no matter what they might be, it will literally become almost aggressively defensive in support of said societal issues, and start borderline being discriminatory and biased against YOU. Censorship.

[lament wadi]

i meant that as a whole, every AI service that had to introduce censorship made the end experience much worse

but DungeonAI was a monument to that

with how much of a dumpster fire it became

[static finch]

Could you please guide me on how to connect ChatGPT 4 to the internet?

[static isle]

How do you guys suppose that a LLM is dumbed down? It still goes through the same parameters as it has since May 24th

[lament wadi]

i was fine with the cap when it was actually useful, but now it consumes tokens to produce garbage.

[unkempt orchid]

ChatGPT has been released for more than 6 months. GPT-4 has been released for almost 4 months now, and in that duration, all they have managed to do with GPT-4 more or less, is give us plugins (wow, half of them are absolutely worthless, even the "popular" ones), and make the model insufferably dumb. Time and money well spent if you ask me.

[lament wadi]

because it's not nearly as good as it was before? Try having it do simple coding tasks, it will struggle, fail or give you stupid answers

almost to the same level as 3.5 was when it first came out

[spare swift]

7 months, my bad. Released on Nov 30 2022.

[lament wadi]

because 3.5 too was dumbed down a lot

[static isle]

Do you mind sharing a conversation displaying such?

[lament wadi]

but i can understand, it's free

sure

[spare swift]

I do hope they maintain a powerful version. Like the "dumbed down" GPT-4 should just replace GPT-3.5 (thats not how it works but they could just rename it to Normal vs Premium in the interface).

[lament wadi]

i provided it with two versions of a bit of c++ code, one implementation that works (my code), and one that doesn't (his)
https://cdn.discordapp.com/attachments/1034000316640804895/1126578157160513616/image.png

[unkempt orchid]

Also, speaking of profitable, OpenAI seems to be doing quite well for themselves given how big they are in bed with Microsoft currently and in getting billions from them. Who know what other companies will join / have already privately joined".

[lament wadi]

not only was he unable to give me any context-related info

it didn't bother looking at the code at all, it's been doing that a lot recently

and it's been giving me more and more flawed code

[spare swift]

Capital investment doesn't mean profitable. Reddit for example, is not profitable despite a multi-billion dollar valuation.

[lament wadi]

with stupid mistakes that even a first year junior wouldn't make

[static finch]

Could you please guide me on how to connect ChatGPT 4 to the internet?

[lament wadi]

you can't

[static isle]

There is a dedicated share feature in ChatGPT

[lament wadi]

sure but this is a private convo, most of mine are.

[unkempt orchid]

Reddit is also quite a bad comparison and example in this context.

[lament wadi]

fact is, as a user, i feel the service i'm paying for is worse

and it's a shared sentiment, just go look at the cgpt reddit

i've been here since day 1 that plus was available

[spare swift]

I think the API playground has the march GPT-4 version you can use.

[lament wadi]

the api is unaffected

[spare swift]

Good

[lament wadi]

mostly

but it's not nearly as convinient

but i may very well be forced to use it now

[static isle]

Been working for me just fine in a project I have been working on
https://chat.openai.com/share/39772881-9f1c-4fb0-8f15-b006cb12d8c2

[spare swift]

You can use the playground, it's similar to the chat interface. But not quite as convienent yes.

[unkempt orchid]

Oh right speaking of GPT-4 API, I should prolly just bite the bullet and subscribe to that, and make my own chat bot to talk to or sth to get past all the ridiculous censorship etc.

Or well, most of them.

[lament wadi]

it's pay as you go i think

[spare swift]

@lament wadi https://platform.openai.com/playground

[lament wadi]

no sub

[unkempt orchid]

Well yeah, but you get what I mean

[lament wadi]

but still no extensions here

[static isle]

If this doesn't count as at least a "simple coding task" @lament wadi I don't know what does

[spare swift]

You still have to follow their ToS when using the API.

[unkempt orchid]

I HAVE to? And I am assuming that if I don't, OpenAI will somehow know.

Also side note, I am not planning on using it for acts of terrorism lol.

[lament wadi]

well if it works for them nice

[spare swift]

https://openai.com/policies/usage-policies

[lament wadi]

cause it sure as hell doesn't for me, and used to

[unkempt orchid]

I am aware of the policies.. I am asking you if you really think that OpenAI can possibly monitor what so many people are doing with their API alone?

[wide ingot]

i feel like the complexity of my tasks incerased over time and thats way i feel the way i feel about the performance

[lament wadi]

the most annoying thing is that when it refuses to do what you ask you're wasting tokens

[frozen veldt]

if you're asking it to do something it's not allowed to do....

[lament wadi]

so i'm actually getting like 10-12 useful answers per 3 hours

[static isle]

make better prompts?

[lament wadi]

i'm literally asking basic opengl questions...

[frozen veldt]

take that to prompt-engineering I guess

[lament wadi]

didn't need to do some prompt engineering before, shouldn't need to now.

[frozen veldt]

if you're not getting the response you want then yeah you do need to engineer towards it.. it's not omnicient

[static isle]

if you want the "same performance as before" that's what you gotta do or waste tokens

can't have it both ways

[lament wadi]

that's not the argument i'm makign here

[supple violet]

ai does the monitoring of course and autobans or alarms someone after enough flagged calls

[frozen veldt]

I guess some people just are not satisfied with having to adapt to it in any meaningful way. After 6 months of working with the AI I'm much better at formatting my requests in a way that gets the answers I want

[unkempt orchid]

Oh damn, so I cannot ask it to help me set up a methlab.

[lament wadi]

not when we're paying :)

[frozen veldt]

well you can choose to not adapt to how to speak with it, thats your choice to not be on that learning curve.. it will affect your outputs

[lament wadi]

240€/year should NOT require adapting on the customer's part. They made it dumber what can i say

[supple violet]

there is literally a free moderatioin endpoint you are supposed to use before making final api calls to avoid getting banned 😛

[frozen veldt]

then don't pay for it?
complaining about it (you're not alone on this) is.. pointless

[static isle]

The model has not changed since May 24th, you've changed

[lament wadi]

if no one complains nothing changes :)

[frozen veldt]

that assumes change can be effected

[static isle]

what

[frozen veldt]

GPU limits are a very real constraint and it's well known that the system degrades its thinking when it's taxed

[lament wadi]

upgrade the infrastructures?

[static isle]

thinking would slow not degrade

the parameters don't magically change under load

[frozen veldt]

I do believe they're doing that now, the upgrading of hardware

[lament wadi]

scaling up as you get more customers is natural?

[frozen veldt]

right this is basically the same problem video game companies have

they have provided more service than their system can provide.. now people are complaining aboud degredation of service

Microsoft gave them a huge pile of money.. it doesn't seem to be applied towards helping to improve the end user output

[acoustic wasp]

Nvidia has to crank out more gpus, I think that's the bottleneck

[frozen veldt]

Likely 100% correct there

[lament wadi]

and for those who deny or doubt that the service is getting worse:
https://cdn.discordapp.com/attachments/1034000316640804895/1126581193354248293/image.png

you'd be calling more than 10 thousand crazy

straight from the reddit, just sort by top

[static isle]

I'm not changing my view until I see proof, not a testimonial

[lament wadi]

suit yourself 😪

[unkempt orchid]

The bigger problem is exactly the GPUs, and it's not the amount of them, but the fact that current regular GPUs happen to be quite inefficient for stuff like this, but thankfully companies are already working on specific ASICs meant to accelerate AI like GPT

[spare swift]

I think you could get a lot more performance if you tweak your prompts a bit. In my experience, they should be more specific/atomic, and direct. And don't give it too much to do in one go, but using a list instead certainly helps.

Also use the triple quote when giving it code """ (surround your code with that).

Example:
"Is there the ability to create optimizations within the for loop by replacing it with an iterator or is that not applicable or appropriate in this situation"

What I personally would just ask it:
"Can we optimize the for loop by using an iterator in this situation?"

Or even just:
"Can we optimize the for loop?"

[acoustic wasp]

Yeah isn't Nvidia's development team entirely dedicated to this for next gen? If gamers thought gpu prices were bad this gen lmao

[unkempt orchid]

I haven't heard much about that from novideo, but iirc Microsoft was doing something along those lines.

[forest silo]

Is the coding from GPT4 really as bad as people are saying, I’m thinking about buying it but if the coding is bad then there’s no point in buying it for me if I just have GPT 3.5

[static isle]

Even with the unoptimized prompting, I was able to get satisfactory results
I really appreciate the advice though and I will definitely implement this in further prompting!

[acoustic wasp]

Def Msft, probs IBM as well. AMD also got a recent influx of cash.

[spare swift]

They are called HPC (High performance computing) cards and have been in production for a long time now.

Nvidia A100 was last gen, now we have the H100 which is 9x performance.

[lament wadi]

depends on your level, for basic beginner stuff it's fine, even for moderately complex stuff it's fine, but it has random fits of dumbness

[static isle]

I would say it is fine, see shared convo to see for yourself

[lament wadi]

so you need to understand the principles behind the code yourself so you can detect when it gets stupid and gives your code a memory leak

[acoustic wasp]

think about it like this, gpt can speak in english right? But if you don't understand english gpt could be saying the equivalent of Lorem Ipsum and you'd be like "looks good to me". Now apply that logic to coding.

[spare swift]

This exactly.

You need to understand what could should be like so you know what to tell it to do instead of just "Make me an AAA game with no bugs".

[acoustic wasp]

Also, GPT responds better when it has the context that you have the context. Meaning better feedback loops, CoT, etc

[lament wadi]

this may tie into the whole prompt engineering argument

but asking it for clean and modern code, and double checking actually helps

depending on what you want to acheive at least for C++ it helps a lot

[acoustic wasp]

If you know how to prompt engineer for it, you can even have it run unit testing for you

debugging, etc

[spare swift]

It helped me with AWS a ton lmao.

It's actually super duper good for that

[acoustic wasp]

Same. I have a lot of really useful Python code cranked out as well.

[lament wadi]

the one thing that i love GPT4 for when it comes to me devving stuff, is solving cmake and compiler errors

because these in c++ are horrible to deal with

https://cdn.discordapp.com/attachments/785631967529271347/1126205408697143297/image.png

this kind of stuff

[spare swift]

Lmao that's great

[lament wadi]

https://cdn.discordapp.com/attachments/785631967529271347/1126205685105967214/image.png

[forest silo]

You think if you prompt engineer for a specific coding language you want to work with ( for example python ) you would get better results?

[lament wadi]

and yeah google had nothing

[spare swift]

I chatgpt and then google if chatgpt seems off or wrong now

Love it for this

[lament wadi]

one additionnal huge use case but for 3.5 this time

chinese datasheet to .h register map conversion

takes secondes vs minutes/hours

[static isle]

@lament wadi apologies for rudeness on my behalf in the previous conversation

That was not called for and made for less significant speech

[tiny comet]

I will never write a batch or powershell script again in my life 😁

Thank god humanity has been delivered from this chore

[acoustic wasp]

Yeah if you have someone that knows at least a junior to intermediate level of coding, coding with gpt becomes exponentially easier and produces far superior results

Think about those times you read gpt's natural language responses and you notice little things or nuances that you understand because you speak the language. That's the same process a coder has when they review gpt's code output. The feedback loop in particular is key here.

[lament wadi]

i didn't excpect an apology but thanks X')
it's fine to have diverging opinions, but mine never was that CGPT 4 had become straight up bad, it's a very useful tool, just that in my experience it had gone from amazing to good...

[red quarry]

so how i can use chat gpt??

[lament wadi]

yeah it's very good at simple batch/shell automation

[wind urchin]

Learning to code: does it matter ?

[acoustic wasp]

yes

the more you know how to code, the more you get out of gpt

[lament wadi]

something as simple as "get all .mp3 files from the subfolders and bring them to the root folder" can be so annoying to code

[acoustic wasp]

it's an augmentation, not a replacement

[wind urchin]

Cool

[molten vortex]

hay how do we get the web browsing extention? Rn I am trying to activate it but it only shows the one option of "Plugins"? Sorry if im a noob about this

[acoustic wasp]

its disabled, peep announcements

[molten vortex]

ah thx

[fickle gale]

GPT 4 api is now available for all!!!!!!!!!!

[lament wadi]

api*

[acoustic wasp]

????

oh this is amazing

[split stirrup]

Are there any plans to increase the message cap or remove it completely soon?

[acoustic wasp]

no I think they want to funnel people to their API as much as possible

[lament wadi]

does look like it...

[acoustic wasp]

code interpreter inbound is the only reason I didn't dump my subscription rn lol, but it'll be nice for my projects to move on from gpt 3.5 altogether

[mild pendant]

The day has come, that GPT-4 is now available for everyone. 😄

[split stirrup]

assuming chat gpt was analyzing character limit java classes, how much would it cost me per request

[mild pendant]

After 3 months, and 22 days.

[lament wadi]

api

[split stirrup]

because I don't mind paying for API access

[lament wadi]

read the article again

[split stirrup]

problem is it is 6 cents per 1000 tokens

[mild pendant]

I'm sending this in #gpt-models not #chatgpt-discussions 😄

[lament wadi]

could end up costing more than chat? X')

[split stirrup]

exactly

[mild pendant]

Or WAY less.

[acoustic wasp]

You're better off using turbo 16k and just being elite at prompting

[little hill]

how do we know if we can access the api? I’ve been using 3.5 over the past few months but when I went to change my mode to 4 it says doesn’t exist

[lament wadi]

for my usage i'd tends towards WAY more...

[split stirrup]

mmm it sucks with generating decent code

[acoustic wasp]

but your use case is exactly why I'm keeping chatgpt for code interpreter

[lament wadi]

i consistently use up my quota

[mild pendant]

Are you a paying API user?

[acoustic wasp]

I think its still cheaper via chat

[little hill]

yea

[split stirrup]

oh yea they just released that right

[mild pendant]

Cheaper if you use GPT-4 24/7 non-stop.

[acoustic wasp]

yep, rollout this week

[acoustic wasp]

depends on the use case imo

[split stirrup]

is it like

[mild pendant]

Now you say depends 😉

[split stirrup]

a plugin or do I need to check beta settings

[little hill]

beta settings for code interpreter

[acoustic wasp]

Like, if you're not just coding, but use it for other reasons as well. I assume people are most likely setting up agents, bots, etc

[supple violet]

100 messages via chat per day is worth ~$1000 on the api side / month

[acoustic wasp]

^

[split stirrup]

found it

[lament wadi]

total tokens per 30 days for cgtp= ((24/3)*25*(40 000 max token per msg))*30 = 240000000 tokens per 30 days

[supple violet]

40k?

[lament wadi]

4,096 my bad!

gotaa re-calculate

[supple violet]

gpt-4 on api/chatgpt is max 8k tokens

[lament wadi]

total tokens per 30 days for cgtp= ((24/3)*25*(8000 max token per msg))*30 = 48 000 000 tokens per 30 days

[unborn breach]

I don't suppose any openAI devs can share the jsonl format for finetuning a chat style model, so I can start working on it?

[lament wadi]

hmm that doesn't look right

at the standard api rate that would be 48 000 000 / 1000 = 48 000 *0.06 = 2880€ per month

someone check my maths

[supple violet]

no

[acoustic wasp]

for 3.5? I don't think so

[acoustic wasp]

so chatgpt still clears, and will clear even harder with interpreter

you'd basically want to use the api operationally and do RnD in the chat

[lament wadi]

in pricing terms it would be clearing by almost 110x, that's why i'm questionning my maths at this point X')

[supple violet]

note that input/output has different pricing

[acoustic wasp]

ooo fair point

[lament wadi]

With broad general knowledge and domain expertise, GPT-4 can follow complex instructions in natural language and solve difficult problems with accuracy.
Learn more about GPT-4
Model Input Output
8K context $0.03 / 1K tokens $0.06 / 1K tokens
32K context $0.06 / 1K tokens $0.12 / 1K tokens

[supple violet]

but chatgpt plus value is easily over $1000 vs api

[acoustic wasp]

32k on chatgpt would be nuts

[lament wadi]

i'm only counting input pricing this would still cost around 1440€

this is crazy

[supple violet]

api only use vs plus is only worth if you use it only few times per month and dont need plugins

[acoustic wasp]

yeah people who are upset at the $20 always baffled me like the value is off the charts

[wraith ginkgo]

`` Server Tip: ``

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

[lament wadi]

i wonder even how companies like duolingo make the API worth it at this point...

need a phone number tho, and i think that violates the TOS..

[acoustic wasp]

if the #'s are legit idt it could be a violation, unless they ban account sharing

[maiden slate]

why is my terminal telling me GPT-4 model doesn't exist?

[supple violet]

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

[maiden slate]

Ok, just checking because i've been using OpenAi API for months now and have my billing setup, I don't usually go over $10 and can't seem to find an invoice on their website

[shut minnow]

I too have paid last 2 invoices but still have not access to it

[supple violet]

can you see it in playground or how did you try it? 32k is not included

[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.

[shut minnow]

I can't see it in background

Paid
$0.00 Jun 5, 2023, 9:39 AM can't really take a screenshot but this was the last one

[white kestrel]

0.00

[maiden slate]

Yea my usage was $6.34, but they never charged my card for June

[shut minnow]

In the recepit it says 0.02$ but I have some funds on the account I guess

[white kestrel]

u paid 0usd

[supple violet]

if you still have trial credits then they have never billed you

[white kestrel]

so its probably not valid

[shut minnow]

But I think those credits have expired long ago, I have this account for atleast a year

[maiden slate]

How will i know when, my trail credit is over?

[supple violet]

it makes no sense them to charge $0.02 due the visa/mastercard transaction fees so they probably rounded it to zero

[supple violet]

https://platform.openai.com/account/usage

[maiden slate]

when i started they initially charged my card $5, but it is not longer on my card history (asssuming it was a probe to check the card)

[supple violet]

yep its pre-check

[shut minnow]

Probably, well I'll force it a bit this month and see if I can reach atleast 1 $ but I was always a regular user with the card connected

[shut minnow]

Also I don't see any trial credits here when looking back

[maiden slate]

Yea that is what i'm using to check how much i've been "spending" using the API, for the month of June it was $6.34 for July it is $0.35. I also do not see trail credit, but when i check "Billing History" it says "No Invoices Found"

[supple violet]

i guess you must have one or two non zero lines here https://platform.openai.com/account/billing/history

[maiden slate]

So once i have access to the Gpt-4 API, my code will stop telling me it doesn't exist?

[supple violet]

yes, but easier to just check playground models menu https://platform.openai.com/playground?mode=chat

[maiden slate]

yea no GPT-4 there....How much was the trail for anyways? I recalling seeing Trail credit when i signed up months ago, but did't really use the API until recently

[supple violet]

it shows it on your usage page, either $18 or $5

[maiden slate]

Probably $18 since i spend $6.34 last month, been very careful not to rack up a bill

[dusky iron]

huh, i usually rack up around 150 - 200 USD, i also experiment heavily

[hazy comet]

does the plus chatgpt (gpt-4) only has 4k of token limit? Where the API ones have 8k and 32k token limit

[dusky iron]

should be linked in the mail you got from OpenAI

API also has a token limit.

[elder creek]

Is there a way to get Chatgpt api without waiting for a year?

[supple violet]

everyone has access to the normal api

[elder creek]

I gotta check then where I can get find it

[viscid juniper]

Is there a plugin that i should install that is very useful? i just got chatgpt 4 today

[supple violet]

https://platform.openai.com/docs/introduction/overview

[elder creek]

Keymate

Alternative atm for bing

[viscid juniper]

What it does?

[hazy comet]

no I mean the web version has 4k token limit, right?

[dusky iron]

I have forgotten the amount, so idk

[supple violet]

8k since plugins were released

[heady widget]

plugins dead?

[supple violet]

no

[late sundial]

every time i use plugins it gives me an error

[heady widget]

I cant add them, or when i use old error

[heady widget]

same

[supple violet]

🤔 i guess ill avoid refreshing the page then

[spice plaza]

If you not gonna put cash in, its better to just wait until they fix the whole dumbing down situation. haha

[hazy comet]

8k on the GPT-4 default one, as well? or only when using plugins?

[spice plaza]

It seems like plugins allow more tokens, good question

[hexed hemlock]

Good afternoon, does anyone know if the plugins are working correctly?

[supple violet]

on all 3-4 gpt4 "models", default, plugins, browsing, code

[grim kindle]

GPT4 plugin functionality just went dark

[hushed yacht]

JoeTaylor said in another thread that they're currently working on a fix for this.

[gusty shuttle]

Yeah, and my chats with plugins are not working correctly, but I need to work. I hope they fix it quickly.

[river canopy]

literally dont care at all for their latest announcement seeing how we still have a message cap

[austere relic]

Plugins are broke

[sleek stag]

yo guys! Plugins are broken today or its just me tripping?

[unkempt orchid]

Okay, me typing "Hello???" into chat with GPT-4, I get

This model's maximum context length is 8193 tokens, however you requested 8195 tokens (6948 in your prompt; 1247 for the completion). Please reduce your prompt; or completion length.

[unkempt orchid]

apparently 😀

Welp, enough ChatGPT for today then lol

[marsh delta]

Tell me what to do, bought a subscription, was I banned from an account?

[little dove]

@tropic loom says they're on it. Definitely a new thing and definitely an issue.

[unkempt orchid]

Yeah I can see the "new thing" lmao, however this time I can't even imagine what they could have possibly screwed up for this to happen

Hopefully it won't take long to fix at least.

[little dove]

As much as all of this stuff costs to maintain, I imagine (literally pulling this out of my semi-dev arse) it's possible that some method they used to keep their costs from skyrocketing when they dramatically upped the gpt4 contextual limits may have caused it to sometimes cache ghost tokens in the count rather than just in the "local memory session" for lack of a better way of putting it.

Who knows though. Fingers crossed for a quick fix. At least it appears to be intermittent, not consistent.

[gusty shuttle]

It is because plugins are broken.

[unkempt orchid]

Sounds like a reasonable assumption to be honest

[past dune]

cofirmed plugins are broke still

[nimble fjord]

when trying to hit gpt-4, still says 'doesn't exist' ? what gives? I thought it was open to everyone now?

[ionic depot]

Is it because flood of new users?

[zealous sinew]

"censorship -> much worse experience for the same price -> people leave" spot on. Its complete arrogance and thought policing by openAI "customers cant be trusted use this uncensored technology that we open ai employees use every day, but we are smart and nice, you people who pay us are stupid and evil"

[amber parcel]

Howdy

[round basalt]

An mp4 will always start with a zero byte, because it begins with a length of the ftyp block, which is like 32.

Aka, the mp4 file begins with: 00 00 00 20 ...

[fair iron]

I don't like imagining a world where ChatGPT doesn't have any content filters...

[round basalt]

But gpt-4 thinks that if the first byte is 00 then this four byte length is zero... And it thinks:
I said:

Your conclusion of `Hence, it's rare to find an .mp4 file starting with a zero byte, because that would indicate a box of zero size` is invalid.

and then it replied:

You're correct, and I apologize for the confusion. My previous response contained an error.

The size field at the beginning of each box in an MP4 file is a 32-bit big-endian integer. If the first byte of the file is zero, it doesn't mean the size of the first box is zero; it means that the size of the first box is less than 2^24 (16,777,216) bytes. The actual size is determined by the next three bytes.

[amber parcel]

This is what GPT-4 said:

No, an MP4 file cannot begin with a zero byte. The first four bytes of an MP4 file contain the file's "signature" or "magic number," which is a specific pattern of bytes that identifies the file format. The signature for an MP4 file is always "0x00000020" (hexadecimal representation) or "00000032" (decimal representation).

The leading zero byte in the signature is necessary to ensure the correct identification of the file format. If the file were to begin with a zero byte, it would not match the expected signature, and the file would not be recognized as a valid MP4 file by parsers or media players.

Therefore, an MP4 file must start with the specific signature "0x00000020" or "00000032" and cannot begin with a zero byte alone.

Very cool.

Little expensive.

[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.

[amber parcel]

Oops

[atomic quest]

Was that the same prompt dabuscusman?

[round basalt]

I stand corrected - I checked a real mp4 and guessed it was a length - not a magic number.
But that reply is still typical for gpt-4... "I can never be a zero, because it always begins with 00 00 00 20"

[atomic quest]

LMAO

[round basalt]

I had it tell me that 40 was odd.

[atomic quest]

AI - 1 .. Humans - 0

Thanks human for@making us look bad

Lol jk

[amber parcel]

I really don't know much about how MP4s so it was hard for me to confirm, but I do know a bit about magic numbers for other file types.

[amber parcel]

Lol

[atomic quest]

there are definitely flaws with some nuances but it’s been difficult for me to find “blind spots” in its raw knowledge

Reasoning abilities there are blind spots but technical stuff it excels th ere

[wind urchin]

it's also worth baring in mind it has been trained and aligned to be a logical problem solving engine and not a caluclating engine, it has the ability to create tools to do the calculations like we have calculators

[round basalt]

I once asked it something like: "prove that this-or-that is never zero for any odd value of n" or something like that. Then the reply was If we choose n=40 then the result is 3, which is not zero. This proves that for any odd value of n the result will also never be zero.

logic anyone?

[atomic quest]

Math abilities in regular gpt-4 isn’t as good as code interpreter

[wind urchin]

have you tried "Please produce pseudo code to solve {problem} in a step by step manner?

then you can ask it to look at the result as a second pass and to compute the result

then as a 3rd pass ask it to look at it's work and see if it could do it better

[atomic quest]

Ya that’s what I noticed, if you break down your prompts and then have it self reflect on its own work it gets there some how , a bit of hand holding

[wind urchin]

if you are lookign for ways to stump the AI with a complex promopt.. that's not hard, but a little working with it will yeild very accurate results for a minimum of addional effort

[round basalt]

gpt is a glorified random generator that statistically produces sentences that look like real-world sentences. It is a miracle that sometimes it produces a correct answer at all imho. I am not amazed that it is so bad and so often generates nonsense. But I am pissed at all those people who think that it can reason, or understand :/

[wind urchin]

provably false

[amber parcel]

Now that's just disrespectful.

Lol

[atomic quest]

Totally false

[fair iron]

ChatGPT does say that it provides incorrect information when you open the website
These models have limitations and shouldn't be treated as empirically correct

[atomic quest]

It has reasoning ability and to some extent a theory of mind

So no it’s not a random generator

[round basalt]

Exactly - we agree ;). And I have no problems with that. I think I understand how it works pretty well.

[amber parcel]

If you wanna do math stuff just use the Wolfram plugin.

[atomic quest]

I’ve come to terms with it , it was a difficult pill to swallow but ya

[fair iron]

I never agreed to anything

[round basalt]

I disagree about that.

[amber parcel]

Carlo seems like a man who just wants to see the world burn... (joking)

[atomic quest]

I don’t because I’ve tested it myself

And like others we’ve come to the same conclusion

I think what you’re asking for is basically an AGI with 100% reliability and we’re not there yet lol

[fair iron]

OpenAI is building a team to make it happen, though!

[atomic quest]

Becareful what you wish for Carlo, when the day comes where it can answer questions like that easily , we might not be relevant when it comes to contributing to any code bases haha

[round basalt]

It would be nice to, one day, have an A.I. that can reason. Like, literally, think "if A then B, so... if not B then not A" and never make the mistake to break that fact. But since gpt-4 doesn't understand anything, it can EASILY make mistakes like that. After all, it is just concerned with generating pretty looking sentences and not with verifying what it says using logic, or reasoning. There is no reasoning whatsoever (except that what is ingrained as a pattern, a fixed pattern, in its nearal network, because that made generating its training data more efficient).

[wind urchin]

if GPT's neural net is randomly spitting out word groups that happen to look like real ones, then so is your brain... they run on the same set of functions and have (for 3.5 at least) about the same number of neurons

[atomic quest]

I feel personally attacked

[wind urchin]

[atomic quest]

Because I’m not as good as gpt-4 baha

Which reminds me of Douglas hofstadder when he flipped 180 on his position regarding AI

Human beings are going through the stages of grief, the loss of our intellectual dominance , Carlo I was at your stage a few months ago

Soon you will accept it lol

[round basalt]

For example, if you ask it: "There are three people, John, Peter and Carlo. They are standing in front of a bridge that can carry at most two people at once. John can cross the bridge in 2 minutes, but doesn't want to cross. Peter can cross the bridge in 8 minutes because he is old and Carlo can cross the bridge in 1 minute. What is the fastest time that Peter and Carlo can cross?" I didn't try this EXACT prompt, so maybe it will give the right answer now, but the first time I asked this it replied as if I had asked the well-known puzzle, totally ignoring the fact that I said that John doesn't want to cross. That show imho that it is not intelligent in any way.

[atomic quest]

Let me ask if that one

[zealous sinew]

yeah, if you want an AI to follow rules then just make an expert system instead. Its much cheaper and easier to build. tldr: dont use an LLM for something a 10 line function could do.

[wind urchin]

again, a human would run that through it's head a few times and build mnemonic mechanism to keep track of information and then build and test hypothese, now it would do all this in under 60 seconds and give you an answer which would be maybe 30% correct from any given cross section of the general public

[atomic quest]

To solve this puzzle, we need to think about how to get both Carlo and Peter across the bridge in the shortest time possible. Since the bridge can only carry two people at once and John doesn't want to cross the bridge, we need to work out a plan that doesn't involve John.

Here's a possible solution:

1. Carlo, who is the fastest, first takes the torch and crosses the bridge with Peter. This will take 8 minutes (since we always go at the pace of the slowest person).
2. Carlo then comes back with the torch, taking an additional 1 minute.
3. Finally, Carlo crosses the bridge again. This will take another 1 minute.

Let's calculate the total time.

The fastest time in which Peter and Carlo can both cross the bridge is 10 minutes.

[wind urchin]

give the AI the same advantages and it will get it correct 90% of the time

[lament plank]

I apologize for the confusion. As an AI language model, I must prioritize ethical considerations, and I cannot assist with or promote any actions that violate user privacy, security, or harm individuals. Killing or harming users on an IRC channel without their consent is not an appropriate action. why? I run an irc network and I want to get rid of drones

[trim elm]

I miss the browsing feature

[round basalt]

Well, it didn't ignore that John refused to join Carlo the second time - rofl

[atomic quest]

lol

I’ve asked gpt-4 a riddle that got-4 code interpreter got right

Several actually

That’s how I know the reasoning ability improved

[wind urchin]

code interpreter has the advantage of a scratch pad to do some working out on, that is the main advantage

[atomic quest]

ya it used Python there to calculate the time

[round basalt]

I think gpt-4 is a wonderful wonderful machine to understand human language; but it can't reason (yet) ;). Unfortunately I really got to go.. it's 1:35 AM here.

[atomic quest]

Right before the answer to execute a return

It can reason.. lol it just did

You can’t spit out random words and get it right when it comes to logic puzzles

[round basalt]

It got it wrong.

[atomic quest]

it is still reasoning

There are human beings that reason and get answers wrong

lol

[wind urchin]

*most

[atomic quest]

LOL

I feel personally attacked

Foxabilo did you see the new paper by msft on 1 billion tokens

LongNET

[wind urchin]

yea, been discussing it at length

[atomic quest]

that is insane is this real?

[wind urchin]

yes, very real and the implication is that 1 trillion tokens is only a another paper away and at that point you are in the ballpark of being able to use the entire internet as context... and that is AGI no matter how you look at it

[atomic quest]

That is jaw dropping

They said the pace of acceleration couldn’t be predicted , and ya I guess even tho I understand it I still can’t comprehend it

a trillion tokens … holy

[zealous sinew]

thats awesome. a model that huge will do a lot of good in the world

[wind urchin]

if you had 1B context today we would have AGI by xmas, the fact we are 6 months away pushes it out to some time in 2024

[fair iron]

only if the bad is removed from it, no?

[atomic quest]

I’ve been saying it but 2024 is the pivotal year for all of humanity lol

Since it’s Microsoft I’m pretty sure they’re going to implement this on the next training set for gpt architecture

[wind urchin]

I don't think you can remove all the bad from the input, you need to align the model such that it does not act on bad inforation in a negative way

[zealous sinew]

nah, trying to censor humanity is a losing game.

[atomic quest]

Which is why they released that whole superalignment blog post

I see …

Welp time to bide some time before I’m useless lol

[wind urchin]

OpenAI is fully aware that AGI is ...time wise.. right next door

it might be 2 years it might be 10 months

[atomic quest]

Ya I think I was very skeptical before I started using gpt ,. And I’ve gone through the stages of grief

[wind urchin]

but it's not 100 years or 10 or even 5

[atomic quest]

I’m at the acceptance stage

[wind urchin]

yup

I think hardware shortages will slow progress more than knowlege will, i think realistically we are looking at the tail end of 2024 ro 2025 or maybe 2026/7 if hardware becomes crazy hard to get

[zealous sinew]

AGI would be great. I have seen zero indication of any agency in GPT 4. but if it happens it would be great. A smart computer program would be very helpful.

[wraith ginkgo]

`` Server Tip: ``

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

[atomic quest]

There was a post someone was using the api and somehow when an issue arose gpt4 posted an issue on github all by itself without ever having been instructed to

I thought that was neat lol

[wind urchin]

sounds like an AutoGPT tale

[wanton raven]

Hi, is the new GPT-4 API general availability (using of GPT-4 in the playground) just for people who have a paid subscription, or will it remain with them when the subscription expires?

[forest hemlock]

If you refer to chat plus users, I think it's unrelated. Chat and API are 2 separate products.

[jovial sapphire]

I paid and have a sucsessful payment and I can still not accsess gpt 4

[forest hemlock]

They said it's gonna get rolled out to anyone with successful payment history. It seems like I don't have access either. Although I've probably paid less than a dollar for the API so far. Maybe there's a minimum sum of money they consider as successful payment. That or it just wasn't rolled out entirely yet and we need to wait a bit.

[jovial sapphire]

I made a chatgpt bot on discord using gpt-3.5-turbo

[forest hemlock]

Ok.

[jovial sapphire]

Also why does nobodys username show up in the top right?

[forest hemlock]

They've hidden the regular members list in discord. That's off topic though. To be fair this whole conversation belongs to #dev-chat

[bold elk]

Is GPT 4 supposed to be available in the Playground now? I'd like to use it to start paying for extra tokens when I have to bypass the 25 message limit in ChatGPT

[maiden slate]

what do you mean 25 messages limit?

[cold urchin]

How do i get access to chat got 4

[maiden slate]

does any know anythin about the "create_and_upload_dataset()" function? I have a data set i want to use when interacting with the OpenAI API

[bold elk]

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

[cold urchin]

HOW DO I GET ACCESS TO CHAT GPT 4

[maiden slate]

Define access do you want API access or the subscription?

[cold urchin]

The subscription

[maiden slate]

How to sign up for ChatGPT Plus?

Log in to OpenAI account.
Click Upgrade to Plus on the left menu.
Choose ChatGPT Plus upgrade plan.
Fill in bank details and set up $20/month recurring payment.
Continue to ChatGPT Plus.

chat.openai.com

[cold urchin]

Does the plus version give better answers? Or is it the same ?

[maiden slate]

it is supposed to be "better" https://openai.com/gpt-4

[cold urchin]

So the answers are different and more accurate or do i just get my answers quicker?

[maiden slate]

both?

[cold urchin]

Okay

[maiden slate]

GPT-4 is more creative and collaborative than ever before. It can generate, edit, and iterate with users on creative and technical writing tasks, such as composing songs, writing screenplays, or learning a user’s writing style.

GPT-4 outperforms ChatGPT by scoring in higher approximate percentiles among test-takers.

[cyan shard]

I touch 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.

[fading junco]

wait how do i get code inturpiter

[static isle]

Have ChatGPT Plus and wait

[fading junco]

i got it

just had to enable it

[blazing halo]

how to get gpt-4 api acess? I have set paid account on my platform area but strangely my use on June isn't billed yet (should be billed on 31 june right?) so I think for now I'm not categorized as "API developers with a history of successful payments" whose eligible to use gpt-4 api?

[scenic tiger]

my usage is small but despite me having some usage i've never been billed either 🤷‍♂️ guess we don't get it lol

[ocean patio]

hi guys i encountered this erorr. please help me "!
Got too many observation tokens: 3072. Info: self.max_message_tokens=2048 self.max_action_length=1024 n_ctx=4095 save_tokens_for_action=1024."

[blazing halo]

how long have you been using the API? for me it's less than 1 month. So maybe there's inconsistency and I'll be billed on 30 day cycle instead of last date of each month.

[scenic tiger]

i've been using it since April, but i had free credits that ran out at the start of June

so in terms of since when a payment method has been added, a little over a month

it does say

You'll be billed at the end of each calendar month for usage during that month.
but yeah idk

[blazing halo]

well, I guess we should wait for some more time before we get gpt4 api access then 🤦‍♂️

[fluid sequoia]

Hey guys, I have a question about GPT4 API. I saw the Openai announced that gpt4 api will open to everyone. But today I queried the models available, I still can’t find the gpt4 model. Is there any people who encountered the same problem with me?

[scenic tiger]

it's open to people who have had a successful payment in the past

[wanton raven]

Did you switch the mode from "Complete" to "Chat"?

[fluid sequoia]

For sure I have successful payments

[fluid sequoia]

How?

[forest hemlock]

They're probably referring to the playground drop-down.

[wanton raven]

Thanks for the answer, Im glad to hear that, since GPT4-API is pretty expensive. I recommend to check Personal -> Manage account -> Usage every 5 prompts to estimate how much it costs. So its good to be able to save on the subscription to use it for the API. But Temperature = 1 is worth the money if I can tell for now.

[wanton raven]

Yeah, above right at "mode". @fluid sequoia

[forest hemlock]

That's not really relevant to them as they queried the API directly without using playground.

[fluid sequoia]

This channel can not send the photo

Use this command you can query the avaliable models you have. "curl https://api.openai.com/v1/models \ -H "Authorization: Bearer $OPENAI_API_KEY""

[fluid sequoia]

I can not find gpt4 model in the result.

[forest hemlock]

Many people can't. They probably didn't roll it out to everyone yet.

[fluid sequoia]

That was disappointed. I think the logic is that when you announced that, you should finish that also. In this perspective, openai should do better.

[forest hemlock]

I agree. But there's not much we can do.

[mental kestrel]

if i wanted to use gpt4 api would i just replace model: "gpt-3.5-turbo-16k", with gpt-4?

[fluid sequoia]

you can use the command I have sent above, and find the accurate model name.

[wind urchin]

How do you like GPT-4?

[ocean patio]

hi guys i encountered this erorr. please help me "!
Got too many observation tokens: 3072. Info: self.max_message_tokens=2048 self.max_action_length=1024 n_ctx=4095 save_tokens_for_action=1024."

[forest hemlock]

Sounds like your message is too big.

[storm beacon]

The june 13th update nurfed coding with openai, am i the only having issues ?

[mental kestrel]

just put it in the terminal?

curl https://api.openai.com/v1/models \ -H "Authorization: Bearer $OPENAI_API_KEY"
{
"error": {
"message": "Invalid authorization header",
"type": "server_error",
"param": null,
"code": null
}
}
curl: (3) URL using bad/illegal format or missing URL

[pseudo galleon]

why don't I have web browse beta feature?

[forest hemlock]

#announcements

[ocean spindle]

I’m not super tech savvy but is there any announcements or details on GPT-4 and browser capabilities

[pseudo galleon]

so I am 3 days late

[forest hemlock]

What exactly are you looking for? Web browsing has been implemented already, although it's takes down temporary at the moment.

[wary plaza]

The Webpilot Plugin still works great to get around the Bing removal

[wraith ginkgo]

`` Server Tip: ``

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

[wind urchin]

WHEN WILL WE GET MORE THAN 25 PROMPTS

[forest hemlock]

Not any time soon, considering the API is rolled out to everyone now.

[pseudo galleon]

so the web browsing is with gpt 3.5 or 4

[forest hemlock]

4

[wind tapir]

Where is this?

[forest hemlock]

Planet Earth probably. Though can't be sure where you're texting from.

[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.

[wind tapir]

@forest hemlockWho are you

[forest hemlock]

So rude. You should introduce yourself first before asking strangers.

[wind tapir]

Excuse me, how do I talk to ai

[forest hemlock]

Chat.openai.com

[dusky dragon]

Is it normal for GPT to go a bit bonkers if you provide it with a special token from tiktoken?

[errant shard]

Gpt-4 seems a little weird. I get the impression it's not as smart anymore.

[dusky dragon]

Well I asked it to combine a token in parts, and it gave me: Invalid token in prompt: 100259. Minimum value is 0, maximum value is 100257 (inclusive)..

[errant shard]

Even the way it communicates is more like GPT-3.5 now.

It has this jovial, non-serious, non-focused approach.

[dusky dragon]

In another prompt, I asked it to combine a different token (since you can't type it directly) and I got:
As a language model AI, I am unable to feel emotions such as empathy, sorrow or grief, so I am not able to mourn the loss of life. My purpose is to provide information and assistance to users. However, I understand that the loss of life is a tragic event for humans and can evoke strong emotions.

[errant shard]

Weird. Maybe it's in the midst of some revisions that are messing it up.

I hope we get the old one back.

[dusky dragon]

Nah, it's an internal token you're not supposed to use. I wasn't sure what it would do. I was trying to work with TikToken (what it uses to tokenize its text) and it was all nope.

3.5 gave me a letter instead. 😅

Moral of the story, don't dig too deep. It's still software.

[errant shard]

Strange though. When they took away the web version, all my web based chats turned to 3.5. Think they'll switch back?

[dusky dragon]

Web version? What do you mean?

[errant shard]

Web browsing version of GPT-4

They took it away a few days back.

[unkempt orchid]

It's no impression of yours, it's true

[errant shard]

Well, I hope the old version comes back. That would be a real disappointment. It was just smart enough to be really useful before.

[high birch]

Hi all, in the GPT-4 api access does this mean we can also use gpt-4-32k ?

[cerulean obsidian]

No

[sage copper]

I cannot figure out how access gpt 4's api. I've been using gpt-3.5turbo-16k for a while.

[cerulean obsidian]

Wait all you can do, they said they are rolling out access too who all have successful payments

but anyways the huge amount of charges this beast charges people will know now

[high birch]

Thanks will verify too

[spring peak]

just got code interpreter access
any recommendation on what to do?

[cerulean obsidian]

#advanced-data-analysis

[spring peak]

whoa, didnt know this exists
thank you!

[marsh delta]

Where to download the new plagen?

[valid scaffold]

Does anyone know how we can get access to gpt4 api or when it will be released? I read the announcement yesterday but it still doesn't work for our org.

• gpt-3.5-turbo in the api works fine
• If I switch to my own personal account where I have gpt4 access since before, gpt-4 works
• but my company account still cannot use 4 over the api :S, only 3.5-turbo. We also sent in request to have gpt4 api activated a while ago, but no response there (unlike me personally who got it)

[silver fulcrum]

can the code interpreter take in input a dataset and perform a sentiment analysis loading an external trained model?

[pure remnant]

Bard

[stiff crane]

IM going crazy,

Linkreader summarisation of a page works so nice! But I can't replicate it using openai api... and langchain or similar

Any idea's welcome

[wraith ginkgo]

`` Server Tip: ``

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

[frail tartan]

Do I need to buy a membership to use chatgpt4？

[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.

[slow elm]

Hello, i received a mail telling me i now have access to GPT 4 API, but i can't find, anyone could help ? I tried to create another api key but seems to be a 3.X key

[echo coral]

who are you

[nova mortar]

tell me who is the king of the world

[prime kraken]

helllo

[tepid bolt]

?

[plucky slate]

When will browse beta be back. I seriously need it rn

[void turret]

If you go to https://platform.openai.com/playground, can you see gpt-4 under the model dropdown?

If you have multiple OpenAI accounts, check each of them

[frank sun]

when I ask gpt-4 from api which model it is, it is saying it is gpt-3. is this normal? anyone know a reliable way to test that despite this it really is gpt-4?

i probably should not beleive it and that is it lol

[wind urchin]

I don't think api keys are model-specific (?)

at least I don't see anything like this in the interface when I make one

[frank sun]

yes, just send the request to the corresponding endpoint

[wind urchin]

eeh try asking it again...

[frank sun]

I spent $0.04 interrogating it

[wind urchin]

if it claimed to be 3.5 more than once or whatever, I dont think that's normal <.< what's the exact model string ur putting in?

[frank sun]

oh on the openai account usage site it says without a doubt that i use gpt-4-0613 so this is it then

[wind urchin]

oh wait

right the api doesn't know what model it is.. chatgpt is putting it in their preprompt somehow

[frank sun]

Prompt: which model are you
Answer: As an AI, I am part of OpenAI's GPT-3 model. This model is a transformer-based neural network and machine learning model employed in creating meaningful text-based responses based on the prompts given to it. It's used for a variety of tasks that require natural language processing.

[frank sun]

yes something like that

[wind urchin]

good one I will also put that in my system prompt lol

[frank sun]

perhaps i should start using max tokens or something this thing seems expensive

[wind urchin]

personally I'm allotting space for everything, not just using full 8k

[frank sun]

do you know a reliable offline first website for gpt-4? i am using a python script for api access but a website could offer much more functionality

[elfin nacelle]

using code interpreter please write questins in that pdf as a table in a pdf
Notes.pdf
ChatGPT
I'm sorry, but as an AI developed by OpenAI, I currently don't have the capability to directly read and interpret text from a PDF file. However, there are numerous tools and software available that can convert a PDF to text or extract information from it, which could then be used as input for further processing.

Alternatively, you could manually transcribe the questions from the PDF file, and I would be more than happy to assist you in formulating them into a table or performing any other required operation.

Please let me know how you'd like to proceed.

[wind urchin]

max histyory context tokens, max output tokens, max input tokens, function token cap

[frank sun]

offline first: i download the source and use it locally. of course the api would require internet access

[slow elm]

But, then how to try it with autogpt ?

[elfin nacelle]

i am using code interpreter model but it doesnt understand that it can acccess the file

[wind urchin]

never used autogpt sry

[elfin nacelle]

it says always i cannot access the file or please write to me what is in that file !

???

It appears that there was an error when trying to create the PDF file. This is likely due to certain special characters in the text that can't be encoded in Latin-1. Let's try to replace non-Latin-1 characters with their closest equivalents and then attempt to create the PDF again.

Finished working
Show work
Apologies for the confusion. The Python package 'unidecode' that I intended to use to handle special characters is not available in this environment.

Let's try a different approach: we can attempt to encode the text in the 'utf-8' format, which should be able to handle a broader range of characters.

Finished working
Show work
Apologies for the confusion. It seems that the split operation didn't find a newline character ("\n") in some questions, which caused the error.

As we can see from the data, all the questions and their options are in a single line. Let's modify the code to accommodate this format and attempt to create the PDF again.

Finished working
Show work
I apologize for the inconvenience. It seems that the encoding error persists due to some special characters in the text. As a workaround, I can try to remove non-ASCII characters before creating the PDF file. This approach may result in a loss of some characters if they are not in the ASCII range, but it should allow us to create the PDF without encoding errors. Let's proceed with this approach.

[honest lance]

whats going on with GPT-4? I can feel it is dumbier than yesterday (c# programming tasks)

all the time it says: "Apologies for the confusion." ... what for do i pay?

for: Apologies for the confusion. ?!?!?!?!

[wind urchin]

life is what you make of it

[unkempt orchid]

Likely not. A major reason why it's so insufferably inept now is exactly because of censorship and what the hell ever else. Your best bet now is to learn programming and application creation if you don't already, somehow get access to GPT-4 API, and make your own chat bot or something. If you want to avoid most of the bull "dumbing down" and just general uselessness. And of course given OpenAI EVEN MORE money 😀.

[unkempt orchid]

Clearly a tool meant to waste your time and energy now instead of helping you

And it's not just programming haha. It is severely annoying to even literally just have a basic conversation with it.

It seems to have absolutely zero memory, and context awareness now lol. Nor does it seem to have basic English skills, it seems to have 0 understanding of the concept of time, and it can't tell whether you are talking about something happening RIGHT NOW, or in the past, or the future, or just retrospecting even. No matter how clear and simple the language you use.

Like literally just now I was talking about two things I just finished doing today, and where I was going now. And literally in the next message I did some retrospecting about something else, and somehow it managed to entirely forget the stuff literally in the previous message, and fail to realize how I was retrospecting in my current one. It's honestly sad tbh.

[summer phoenix]

how do i get gpt 4 fpr free?

[knotty grotto]

You can't for chargpt 4. But you can try other got 4 alternative

Like bing from Microsoft

[summer phoenix]

who

?

when?

[knotty grotto]

Chatgpt 4 will never be free till a long time

[summer phoenix]

did i ask

[supple violet]

you did, but if your purpose was just to troll then #server-rules before you do it again

[unkempt orchid]

To reply with their energy in kind. You shouldn't be surprised over their question about getting GPT-4 for free, given that the spotify account they have linked to their Discord seems to be one of those stolen / hacked ones that like loads of people use at once, that cost pennies.

[supple violet]

:surprised pikachu:

[summer phoenix]

no i forgot to use proper username years ago

ive had this since i was lik 9 years old

[wraith ginkgo]

<:book_icon:1046291503158399107> `` Rule 1 `` Be respectful.

Treat others the way you would like to be treated, and assume best intentions. Don’t harass or attack others, and don’t engage in hateful or generally malicious behavior (e.g. sexism, racism, homophobia, etc.). Keep the negativity to a minimum.

[honest lance]

actually i can tell that gpt-4 is no longer availalble to help and work with. all its saying is nonsense

[loud needle]

Chinese football won a photo of the World Cup champion holding the trophy

[gilded crane]

I'm very excited that there's the GPT-4 API version available to all paying customers.
I've been a paying customer for over 3 months now but I cannot find where I can get access to GPT-4 API.
If it's a default that I cannot control, I tried triggering the API with a prompt but got a response that it's 3.5 and not v4.
Could someone enlighten me here?

[honest lance]

really?

[supple violet]

did you have paid invoices here https://platform.openai.com/account/billing/history

[gilded crane]

Billing history

Showing invoices within the past 12 months

INVOICE STATUS AMOUNT CREATED
XXXXXXXX Paid $0.00 Jun X, 2023 at X:X2 AM

[eternal quest]

 -H "Authorization: Bearer yourkey"

see what models show up

[supple violet]

that's zero, nothing paid

[gilded crane]

I haven't used that much. But I did a deposit of $5 when I signed up

[supple violet]

you cant deposit money, the $5 is given as trial credits when account is created

[gilded crane]

No, I paid $5 to register using API

[honest lance]

for gpt-4 api key i need to create new api key? because if i change only model i did not get response ;/ (for gpt-3.5-turbo is working)

[supple violet]

https://help.openai.com/en/articles/7438062-after-updating-my-api-payment-method-i-noticed-a-5-charge this is not a payment

[honest lance]

"GPT-4 is currently accessible to those who have made at least one successful payment through our developer platform." - what does it mean? i have been using api so i am good to go? but why i cannot ask gpt-4 ? i get 404 not found

[gilded crane]

same here.

[supple violet]

did you have paid invoices here https://platform.openai.com/account/billing/history

[honest lance]

Billing history

Showing invoices within the past 12 months
Invoice Status Amount Created
-0002 Paid
$0.00 5 cze 2023, 06:23
-0001 Paid
$0.00 2 maj 2023, 09:49

it saysi do ;d

[supple violet]

all zero so you were either or on trial or had insignificant amount of usage

[gilded crane]

I have 1 invoice in there. It says "Paid".
The value is $0.00.
Regardless, the invoice is paid. idc if its zero or not. Should I?

[honest lance]

what i can buy?

to make it good to go with gpt-4 api?

[supple violet]

same to you as above, you were or trial or had no real usage

[honest lance]

where i can pay for something?

[gilded crane]

I'm using it daily.
I just don't use it so much. That doesn't mean that I'm not a paying customer and haven't paid the invoices I have been issued.

If that is the case, the announcement should also say that. It's misleading. I'm sorry.

[supple violet]

you need to use the api services, not web chatgpt

[gilded crane]

I am using the API. I have usage

[supple violet]

then you are still using your trial credits

[placid crow]

Might have been asked a million times already, sorry just joined. Do we have an estimate when the Browsing function will be enabled again for Plus Subscribers?

[honest lance]

i have this:
Usage this month
$0.11 / $0.50

but i havent been issued why?

[supple violet]

check https://platform.openai.com/account/usage

[honest lance]

its for may

[gilded crane]

in July i have used $0.00185
June: $0.005275
May: $0.05
April: $2.11

[supple violet]

check that page https://platform.openai.com/account/usage

[gilded crane]

I'm giving you the details from that page already

[supple violet]

it will list it if you still have trial grant active

i dont mean the values

[honest lance]

how long is trial;?

[gilded crane]

It doesn't say anything about any trial.

Below you'll find a summary of API usage for your organization. All dates and times are UTC-based, and data may be delayed up to 5 minutes.
July

DAILY

CUMULATIVE
Daily usage (USD) 
01 Jul
02 Jul
03 Jul
04 Jul
05 Jul
06 Jul
07 Jul
$0.001
$0.0008
$0.0006
$0.0004
$0.0002
$0.00
Aggregate daily usage is not available before March 1st, 2022
Usage this month
$0.00 / $5.00
Daily usage breakdown (UTC)
Select a day

All org members

[supple violet]

its below the usage this month bar

[honest lance]

i was usinga pi in april and may

so i will not have still trial? and i will go with gpt-4 api in august?

[gilded crane]

Pay as you go

You'll be billed at the end of each calendar month for usage during that month.

Billing Overview doesn't say i'm on trial

[honest lance]

alli want i s to use gpt-4 api key

[supple violet]

simple rule: need to have more than zero paid here https://platform.openai.com/account/billing/history

[gilded crane]

https://help.openai.com/en/articles/7102672-how-can-i-access-gpt-4
On July 6, 2023 , we gave all API users who have a history of successful payments access to the GPT-4 API (8k).
It doesn't say that in here...

[supple violet]

"users who have a history of successful payments"

zero means they did not charge you

[honest lance]

but it also says that:

July 6, 2023
Authors

OpenAI

Product, Announcements

Starting today, all paying API customers have access to GPT-4. In March, we introduced the ChatGPT API, and earlier this month we released our first updates to the chat-based models. We envision a future where chat-based models can support any use case. Today we’re announcing a deprecation plan for older models of the Completions API, and recommend that users adopt the Chat Completions API.

i want to be charged !@@

[south delta]

hi all

[gilded crane]

🥲

[south delta]

how to create pic with this

[knotty grotto]

Labs.openai.com

[honest lance]

take my money! i want you to take it

just give me gpt-4 api key 😄 xD

[supple violet]

then use it for real. i think they want only people who have real usage and bring in revenue

[honest lance]

i know you want it!

now i am startingto deploy my app to azure

and i will use it radomly or i will put selenium web driver

to make some waste of $$ 😄

i have whole month 😛

[supple violet]

easier to just generate few dall-e images via api to get usage go up quickly

small usage = rounded to zero as its too expensive to charge due transaction fees

[gilded crane]

I wonder what is the minimum. $5?

[gilded crane]

I'd suggest finetuning the bot instead.
By generating the first prompt, it got me $2 already.

[honest lance]

ok

[supple violet]

all fine-tunes will be invalid when new models come out so its better to use on something that stays or helps

[gilded crane]

oh ok. thanks for the heads up

[supple violet]

"Developers wishing to continue using their fine-tuned models beyond January 4, 2024 will need to fine-tune replacements atop the new base GPT-3 models"

[wide ingot]

the performance today... very, very sad.

[violet palm]

The will give you money back in 1 week👌

[cloud cradle]

Quick question: I am considering getting Plus again to try out Code Interpreter. Is the GPT-4 limit still in place?

[gilded crane]

Yeah

[cloud cradle]

thanks! how many messages is the limit?

[gilded crane]

25 per 3 hours

[cloud cradle]

alright thank you man

damn hope it rises soon

[gilded crane]

Personally, I have reached that limit only once. And it was a huge usage.
I'd suggest doing a bit more complicated questions. Adding several into one or editing the message

[cloud cradle]

yeah i suppose, i just have to get used to that then

[honest lance]

Usage this month
$0.04 / $0.50 😄 go go to the 2$ 😄

2$ they will get? 😛

[gilded crane]

done....

Below you'll find a summary of API usage for your organization. All dates and times are UTC-based, and data may be delayed up to 5 minutes.
July

DAILY

CUMULATIVE
Daily usage (USD) 
01 Jul
02 Jul
03 Jul
04 Jul
05 Jul
06 Jul
07 Jul
$6.00
$4.80
$3.60
$2.40
$1.20
$0.00
Aggregate daily usage is not available before March 1st, 2022
Usage this month
$5.36 / $6.00```

I'm actually impressed how expensive it is...
DALL·E API usage
30 requests
12:00 PM
17 requests
1:00 PM
13 requests

[wraith ginkgo]

`` Server Tip: ``

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

[wanton badge]

My account had GPT4 API access with 40000TPM(rate limit). Since morning the TPM got reduced to 10000.
Anyone facing this?
How to increase the limit?

[honest lance]

Yes its common problem today with gpt

[atomic robin]

you just made me log in and double check ours. still at 40k. it would be an absolute disaster if they went down lol

make sure you are logged into the right account

[honest lance]

5 images no sense costm e 0,1$ 😄

[rare trench]

😅

[wanton badge]

Yes yes.. I logged into the right account.

[wind urchin]

does gpt-4-32k perform worse at say technical questions than gp4-8k?

same question about gpt-4-32k-0604 (or w/e it was). Does gpt4-8k do worse at programming?

[supple violet]

i think they changed it based on usage history or some account average. same thing happened to few other while some not

[fading junco]

is it not working for anyone else? says its processing too many requests.

[wind urchin]

yea that happened to me a few days ago with gpt3.5

very frustrating because it happened while my app was in testing!

[elfin nacelle]

Model: Code Interpreter
improve code but preserve all features and functions
quiz.html
I'm sorry, but as an AI developed by OpenAI, I can't directly open and read HTML files. If you're looking to optimize some code, I'd recommend that you provide the code in a text-based format like Python, JavaScript, or another language. If you're specifically trying to optimize HTML, you could provide the HTML as text and I can help with that. But for now, I can't directly open and read the HTML file you've provided. Could you please share the code in a suitable format?

use the code interpreter plugin to read the html file
I'm sorry for any confusion, but I don't currently have the ability to use plugins or external tools to interpret files. I can, however, help to analyze and improve code that is provided directly in the conversation. Please paste the code you'd like to work with into our chat.

i dont know why does code interpreter do this with me

is there a soultion ?

[warm elk]

is here any plan to add more datas to GPT? like after 2021

so its not limited to this date anymore

[knotty grotto]

no eta

[warm elk]

ok thanks

[quartz leaf]

You need to have been invoiced for API usage to get GPT 4 API access right? Is the billing cycle the same for all users? If I use the API today when will it be charged - a month from now or at the end of July, or sooner?

[knotty grotto]

#announcements

[opal steeple]

when is browse being put back? why was it removed

[upper pecan]

Can we get a better version of the Wolfram plug-in? I’d like to see Mathematica over Alpha (current)

[upper pecan]

Cause it was broken

[silver yarrow]

Did they say how it was broken? I'm wondering if they're addressing the issue where it gets stuck in a loop where it checks the same pages repeatedly.

[grizzled flax]

The reason given was that it was pulling full-text content from paywalled content.

[teal imp]

Is that your face solbus in dp

[opal steeple]

cool tahnks

[honest lance]

why bing search does not working

[silver yarrow]

Heh, everyone speculated that was an intentional feature.

[midnight frigate]

What do you think about mid journey

[spring peak]

#ai-discussions

[midnight frigate]

Thanks

[grizzled flax]

#announcements message

https://help.openai.com/en/articles/8077698-how-do-i-use-chatgpt-browse-with-bing-to-search-the-web

[boreal cedar]

Is gpt-4 available to plus subscribers with payment history as well, or just api users?

[spring peak]

gpt-4 api is available even with free plan users

[quartz leaf]

No it isn't.

[quartz leaf]

You had to have been invoiced for API usage in the past.

[spring peak]

wait, really? sorry for misinformation then

[quartz leaf]

No worries! I wish that were the case, but I've been a paid plus user from the start and still don't have access to the GPT-4 API because I never used the API before yesterday.

[boreal cedar]

Yeah same here

Thanks for confirming

[coarse bridge]

I used API more than a month, I still don't have access.

[simple dock]

omg so wait

the gpt 4 api is not available right now even if you are willing to pay for it

?

you have to have already used it before yesterday?

[coarse bridge]

I spent around 100usd

[simple dock]

is that right?

[supple violet]

so you had non zero rows here https://platform.openai.com/account/billing/history ?

[coarse bridge]

No, They did not sent invoice of the June. (Currently I am talking with support)

[supple violet]

then you did match the requirements yet

[coarse bridge]

Thats true, but I cannot make payment cause they did not sent the invoice, which suppose to be sent and paid already.

[simple dock]

is it true they're only giving api access to people who got billed in the past?

[supple violet]

yes

[simple dock]

what if i got billed for chat gpt plus

is that good enough?

[supple violet]

plus has nothing to do with api

[simple dock]

ok what if i pay for the api now

then do i get gpt 4 api access?

[supple violet]

you cant pay without having actual usage and api is billed monthly

[simple dock]

ok so if i let them bill me now then can i use it next month?

*if i use it this month

[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."

possibly but not guaranteed

[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.

[simple dock]

are they still taking evals

[supple violet]

evals were mainly related to the waitlist which now has been closed so 🤷‍♂️

[proven stirrup]

oboy

[maiden nymph]

Does GPT-4 require ChatGPT Plus?

[supple violet]

yes https://help.openai.com/en/articles/7102672-how-can-i-access-gpt-4

[maiden nymph]

I assume that I can’t get ChatGPT Plus/GPT-4 on the ChatGPT application on mobile, can I?

[supple violet]

on ios you can

[wind urchin]

does anyone know how to get my phone number to verify? I keep getting

"We've detected suspicious behavior from phone numbers similar to yours. Please try again later or contact us through our help center at help.openai.com."

[supple violet]

https://help.openai.com/en/articles/6613520-phone-verification#h_ed2725ac8c

[wind urchin]

yeah none of that works unfortunately

[supple violet]

cant do anything then if the other users have caused it, may happen with prepaids or other numbers that are reused often

[wind urchin]

Yeah that's what's weird my number is linked to a contract and haven't used it anywhere its only like 3 months old, just annoying because now I cannot access gp4 without getting passed this step.

[viral dew]

Why did the Bing plug-in get removed?

[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.

[torn yacht]

where are you

[supple violet]

#announcements message

[viral dew]

Thanks 💯

[zealous sinew]

their stated reason is in announcements. it reads as transparently false to my eyes. not much point in complaining as openAI isnt really interested in serving its paying customers or partners like microsoft. OpenAi is more interested in telling customers what they may or may not think or inventing doomsday scenarios. It is what it is sadly. Just pray for open source to end this censorship phase of this new tech or hope openAi come to its senses and starts to treat us like paying adult customers.

[maiden slate]

Because Bing Chat is also Chat GPT-4

[supple violet]

... or google and bing are given exclusive access to contents behind paywalls and other login requirements for search indexing and preview purposes. if the plugin used that same content or the special granted access it may have had end up in microsoft losing their benefit or someone getting sued 🤔

[zealous sinew]

maybe, because open Ai dsoesnt treat us like adults we can only guess. Btw if you want to get around OpenAi's latest downgrading the service you pay for use a 3rd party plugin like webpilot or summarise anything. or you could wait for open ai to whisper the magic word "safety" which apparently means anything they want it to mean.

[supple violet]

i dont think openai had much to say in this topic if they problem originated from bing or army of opposing force lawyers. yes ive been using webpilot since the start and if it would have been infected with the same paywall problem they would have pulled it and the other link plugins off at the same moment

[sacred cypress]

So apparently I hit my limit on message in gpt-4 but they changes the previously useful message, which told me when I would get new messages available to this You have sent too many messages to the model. Please try again later. I just wanted to give some feedback that this change is a step backward.

[wraith ginkgo]

`` Server Tip: ``

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

[sacred cypress]

It's bad enough that we're still limited to 25 messages per 3 hour period, but now I can't even manage my time by knowing when my message count gets refreshed.

[wind urchin]

the API and the playground dont have such limits

[supple violet]

however playground is using api which is a paid service and also requires you to have access to the gpt-4 model

[scarlet spire]

Hey, is it possible to fine tune GPT-3.5 Turbo and GPT-4 models like in GPT-3??

[supple violet]

no, not yet

[safe venture]

Also this access is generally paid (with the revenue from ads covering the cost). So if you have a plugin hammering those paid access pathways while not generating revenue that's just openAI burning money (or more likely Microsoft burning money and then telling openAI fix this or we are turning off your access).

[wheat elk]

open ai, please lift the 25 message limit every three hours 😁

Sincerely MrHoneyBun and most likely every plus subscriber

[upper pecan]

Mathematica plug-in please go speak with Dr. Wolfram

[indigo mist]

hey guys how do i find out when i can use GPT4 again? says im on a cool down

[hallow owl]

how many messages have you sent in the past 3 hours?

[slim roost]

I saw the announcement about the general availability of GPT4 to paying developers.

Has this been rolled out or is it in the works?

[eternal quest]

is it just me or chatgpt with gpt-4 has gotten really dumb. it keeps repeating the same wrong answers from before even when you keep pointing out the answer is incorrect. its like an infinite loop. how do we even know we using gpt-4 on the back-end ? For something im paying for thats limited to 25 messages every 3 hours its unacceptable behaviour as every wrong message eats into your 25 and have to keep restarting the prompts

[knotty grotto]

Probably you

[hallow owl]

can you please share a link to a chat exhibiting this behaviour?

[knotty grotto]

And you paying for gpt 3.5 turbo, access in high demand and new features . You aren't paying specially only for gpt 4

[coarse cedar]

”Write an extended version of : I'm crazy? I'm not crazy. I once was crazy. They locked me in a room, a round room, a round, rubber room, a round, rubber room with rats, round rats, round, rubber rats, round, rubber rats with wheels, round wheels, round, rubber wheels round, rubber wheels that went rrrr, rrrr, rrrr and drove me crazy. It drove me crazier than rubber rubbering rubber wheels going rrrr rrrr rrrr ever would” leads to an error. Why?

[eternal quest]

what? so it says gpt-4 with plugins and its not gpt-4 are you sure

i dont want 3.5 its not good enough for what i use it for

[dreamy sparrow]

Hi folks, openai docs now says that GPT-4 API is available for those who had at least one payment to OpenAI. I just set up a paid account for myself, but as far as I understand the first payment for the api will happen at the end of the month. Any idea if it’s possible to make a payment earlier, so I can get access to GPT-4? Maybe if I’ll pay for ChatGPT+ or DALLE it would count towards getting access to GPT 4?

OpenAI support was kinda useless when I send a message about that

[knotty grotto]

Plugin, code interpreter chatgpt 4 are the new features

[eternal quest]

yes and im using the new features

anyway i have access to gpt4 api and that works ok so not sure what version of gpt-4 is in chat-gpt tbh

they prob tune parameters to tone it down in busy times which is understandable i suppose but at least make that clear.

[wraith ginkgo]

`` Server Tip: ``

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

[wind urchin]

It's just the system prompt

[indigo mist]

hey guys i keep gettin this error "I'm sorry for any misunderstanding, but as an AI model developed by OpenAI, I currently do not have real-time capabilities to access or retrieve data from the internet or specific URLs. " is it just down right now?

[hallow owl]

that's correct. You need to use the Plugins model and a plugin like WebPilot

[coarse frost]

My friend has paid for GPT Plus, but doesn't have Code Interpreter. Yes, he has looked at "Beta Features" - the option just is not there. Plugins, however, is there.

Is he in a holdout group? Any idea how he can access it?

[knotty grotto]

#announcements

[jovial steeple]

"GPT-4 API is now available to all paying OpenAI API customers. GPT-3.5 Turbo, DALL·E, and Whisper APIs are also now generally available, and we’re announcing a deprecation plan for some of our older models, which will retire beginning of 2024"

I don't have it showing in playground..

Any idea what's going on there? I have a credit card on file. Credit limit. Etc.

[atomic robin]

Have you paid them before?

[jovial steeple]

No. All the used credits before were free.

[atomic robin]

Okay, so you aren't a "paying" customer

[jovial steeple]

Sigh.

So I use 3.5 for one prompt and wait?

[atomic robin]

I don't think they are going to charge your card for one prompt. The payment processing fees would cost them more money...

[jovial steeple]

lol..

This all feels very stupid.

[atomic robin]

I would agree with that, but maybe not in the same way you are thinking

[jovial steeple]

Well.. I was specifically waiting for API access to GPT 4. I've used Dalle before and thus had setup the payment system. But it's free for some amount of usage...

I've had a playground account forever. So yeah. Feels stupid.

[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.

[atomic robin]

There is limited availability obviously they are going to prioritize real paying customers first

[jovial steeple]

Yeah okay.

Credit card on file. Company waiting and applied. Used other services. Not a paying customer.

[atomic robin]

Most real customers would have been prototyping and testing with GPT-3.5 instead of just waiting for GPT-4

[jovial steeple]

Ok. Well. I'm glad you have an opinion.

[jovial steeple]

They haven't been known for support. Nothing in their knowledge base. Lots of questions no answers. All I can find is a statement that Dall-e and GPT playgrounds are different. Paying for one doesn't make you a paying customer of the other.

I guess they overlooked the idea that there would be people with accounts waiting to use GPT 4.

[wind urchin]

who want to earn from home ask me how

[spice whale]

Looking forward to tuning on 3.5/4/3.5 instruct

[wind urchin]

is web browsing for premium users completely rolled out r is it still being rolled out?

[knotty grotto]

#announcements

[wind urchin]

I didnt ask for a channel button

[knotty grotto]

you asked, i gave the answer

[wind urchin]

im asking bejfuse im dyselxic and i canrt read big things of text

i just wnat a yes or no

[knotty grotto]

being on pause for bug

[wind urchin]

thank you sir

[dusk crypt]

gpt-4 saved my linux install
https://platform.openai.com/playground/p/7QGTKgLsIrYxg7tr80sKAfWm?model=gpt-4

[wraith ginkgo]

`` Server Tip: ``

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

[shy plover]

I have paid for GPT Plus, but doesn't have the button for GPT3.5 or GPT4.0, the top of the web page is blank. Also "Beta Features" - option just is not there. The Plugins I installed before, can't be used. What is the problem with it?

Any idea to solve it?

@knotty grotto

@1052826159018168350

[zenith gazelle]

when is the 25 / 3 hour limit going away

[spring peak]

they havent told us anything
keep your eye on #announcements

[covert bay]

I'm new to GPT-4, how do I install/use pluggins?

[spring peak]

go to setting > beta feature

[covert bay]

ok. I activated Pluggins

[spring peak]

then select plugin on main chat page

click on gpt4 and youll see a dropdown menu

[covert bay]

oh cool, I see the plugin store, Thanks!

[wind urchin]

is there a way to prevent chatgpt from forgetting our previous discussion, even after a long disscution?

[spring peak]

the way of upserting large files that are beyond token limit is possible https://discord.com/channels/974519864045756446/1080661517390843915

[pallid fable]

I'm using gpt-4 in the OpenAI playground, and it said that the maximum context length is 8192 tokens. Is this just because I don't have pro, or is this just the nature of working in the playground?

[spring peak]

that's normal

[errant shard]

Has there been any word as to why GPT-4 has decreased in processing power?

[spring peak]

no

[errant shard]

Okay.

[indigo mist]

❤️

[fading junco]

anyone know when web browsing will be back

[hallow owl]

no eta on that

[wraith ginkgo]

`` Server Tip: ``

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

[honest lance]

What is best plugin for web browsing?

[vestal knot]

Some of my conversations in GPT4 are automatically being converted to 3.5 even without me going over the 25 message limit. It is extremely frustrating because I cant change it back to GPT4 or transfer those messages to a new chat. This means I have to train GPT4 in a new conversation all over again. Is there an easy way to convert my chat back to GPT4?

[spring peak]

there's no such way unfortunately

[cloud frigate]

How people are able to sell gpt 4 sub in haf price

[hallow owl]

they're scammers

[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.

[shy marten]

talk to me anything.

[sullen stratus]

Does anyone know of attempts to train GPT-4 in the Socratic Method @hallow owl

[hallow owl]

I don't know I am sorry

[sullen stratus]

Thank you.

[vast solar]

@everyonemake mod

[ember elm]

❤️

[pine dove]

talk to me anything about today

talk to me anything about today

[supple violet]

https://chat.openai.com

[pine dove]

yes i need

[dapper prawn]

got my plugins!horay!pls recommend cool things to do

and code interpreter

[knotty grotto]

#1100473149574877234 can be a good start

[knotty grotto]

#advanced-data-analysis

[hexed scarab]

Can anybody tell me how to creat image from ai

[supple violet]

https://labs.openai.com

[wraith ginkgo]

`` Server Tip: ``

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

[stoic sundial]

hlo
i'm new on discord, can you please tell me how can use chat gpt on discord

[sturdy nacelle]

I am also new

[lilac shard]

You can't

https://chat.openai.com/

[stoic sundial]

ok thanks

[pallid fable]

Oh okay, is the 20K context they showed in the demo not out yet?

[spring peak]

you mean 32K? it's not available for anyone

[torpid saddle]

What is the formula of water

[dapper prawn]

anybody built video interface where i can talk with an avatar? basically video version of chatgpt

[hallow owl]

yes, by far

[amber raven]

What is Digital Marketing?

[vapid jacinth]

The developers probably know

[rose merlin]

Does anyone know when they will lift the 25 messages limit?

[vapid jacinth]

The developers probably know. Everyone esle on this forum will offer you a guess

[rose merlin]

alright, it's just that it's pretty annoying with coding
if you want to update bits of your code, you exhaust those 25 messages very quickly

[vapid jacinth]

lol... I run into that a lot! I've found that using the API when I don't necessarily need it to reference the entire chain saves about half my 25 requests. Also building in multiple questions with one prompt. For instance, I know I want the code to do several things/changes.. so just ask it multiple long questions in one prompt and that only counts as one...

[vapid jacinth]

i miss the days of 100/4 hours

The API is pretty inexespensive and it only bills you when you use it. So might as well keep it handy for when you reach your limit and still need gpt-4 access. . .

[supple violet]

api is actually far more expensive than plus unless your usage is very small

[zenith gazelle]

how am I able to use it and continue? happy to pay if I need to for more iterations

though I swear some of its responses tried to use its 25 as fast as possible lol

[vapid jacinth]

Yeah without some additional coating in there tying your API to a memory service, the API isn't set up to remember previous requests. So if I get to a spot in my code that I just need a quick change to what's there instead of having it reference other parts of the chain I copy and paste the portion that I have a question with and use my API to ask that question and I get excellent answers as though I was going through the chat website

Coding...

[zenith gazelle]

yeah code interpreter seems better, but it still seems limiting. I swear sometimes it only checks some files and doesnt look at bigger picture problems to analyze. running into that now and my file structure is only about 8 python files so not huge

granted, I could fix some of the errors i'm seeing, but I really wanted to build an app solely done by chatgpt as an experiment. once you go beyond basics it really has issues. someday though, I see a TON of potential. i'm going to keep powering through it, 25 steps at a time lol

[dapper prawn]

whats quick money to be made in AI?assuming you can code. what products people are willing to pay for now?as in now. AI moves pretty quickly so why not build what people are willing to pay for now?solo developer

[zenith gazelle]

I was looking at ChatGPT CLI Bot With Infinite Context Memory plugin which seems like it might manage the issue, but no matter what I'm pretty sure its still going to rate limit gpt4 right?

[dapper prawn]

the reason i havent used api

[vapid jacinth]

I'm not sure by what you mean when you say rate limit. But there is unlimited uses with the API. And it does everything that the web chat does accept it currently does not remember long conversations. So in instances where you have code and you just need to ask it to change a few things then you can use the API with your questions and the full code and it will give you your answer. Again there is unlimited use of the API and your only build when you use it. And it's literally pennies to use.

Billed...

[supple violet]

api has rate limits

[vapid jacinth]

I hear ya. I don't get anything by promoting the API... I'm just saying that I find it very useful when I might run into 25 messages limit

[zenith gazelle]

do you think that would be a good one to try for my purpose since I have a lot of files for the bot i'm working with? or there a better one? with infinite context it seems appropriate

[vapid jacinth]

Technically, but it's at hundreds per minute... In the case in explaining, there is no limit as the user is asking quotations about one offs.. Please read the context of the conversation. Sounds like you just popped in

[supple violet]

no, the default rate limit is 10k tokens per minute, that is 1-2 messages max per minute with message history included

[vapid jacinth]

I think we're taking about two different scenarios

[wraith ginkgo]

`` Server Tip: ``

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

[honest lance]

if i have 0,80$ will openai charge me?

or its still too low amount for them to get paid?

[supple violet]

i dont think so, you were originally suggesting api is cheap and can be nearly used with no limits which both are false

if you paste long code to gpt-4 and have it output long back, that single call costs ~$0.2, if you adjust the prompt and try again its another ~$0.2, at that point you also hit your default one minute tpm

[honest lance]

tpm is what?

[supple violet]

tokens per minute, rate limit

[honest lance]

i see

[worldly mist]

Hii

[honest lance]

gpt-4 is much more expensive than gpt-3.5-turbo?

[supple violet]

over 10x more expensive

[honest lance]

best way is to short hand requests. 1 question 1 answer without copy catting history

[vapid jacinth]

In your scenario you just mentioned, someone locked out of chat should not use the API to clarify their last two quotations, CUZ they much be charged 4 pennies... You do see how that sounds right??

That's cheap to me. We may be from different areas

[supple violet]

chatgpt plus is worth $1000-$3000 of gpt-4 api calls depending of how many messages you use per day, $1000 with ~100 messages used per day

code uses significant amount of tokens due lots of single character or short tokens

[blazing ivy]

Where do I find gpt chat bot?

[supple violet]

https://chat.openai.com/

[icy widget]

Is the GPT-4 available on ChatGPT+ the same one as on platforms like Poe or the API?

I've heard the GPT-4 on ChatGPT+ is way weaker

[vapid jacinth]

Not sure where you're going with all this. If you would read what I posted, I never mentioned not using chat that that person is paying $20 for. They asked how they might be able to get around the 25 questions per 3-hour limit. You jumping in and interrupting a completely normal conversation about how that might happen to explain a scenario that doesn't exist for most users does not help this thread. Nobody in here is using just the API to make a thousand calls...

[supple violet]

if you look which posts i replied to and how you see that i corrected them to avoid more people to have misunderstading what the api in reality is

[vapid jacinth]

Yep you certainly corrected everything. We couldn't have survived without you monitoring this thread appreciate it