#gpt-models

1 messages · Page 265 of 1

gaunt knoll
#

Is this a good channel to ask someone to create a custom GPT?

little plinth
#

I wonder if it's a good place to show off a gpt, lol

#

not a single person used it yet, I don't think, and I think my gpt is insanely good

ripe thorn
#

why do i keep getting this error no matter what even if i make a new chat

Something went wrong while generating the response. If this issue persists please contact us through our help center at help.openai.com.

last cloak
#

To Assistants Playground*

#

Is there a upload size limit?

#

nvm, i uploaded it in the storage and it works

#

ok but it can't go find it...

thorn terrace
#

Deference between GPT and Assistant

  • GPT-4 Turbo
    The precise name of the model answering this query is "gpt-4-turbo." In the OpenAI API, you would specify this model using the identifier "gpt-4-turbo".
  • GPT-4 Turbo [Assistant]
    The model answering this query is based on OpenAI's GPT-4 architecture. In the OpenAI API, the precise name for this model is typically referred to as either "gpt-4" or a variant with a version identifier such as "gpt-4-turbo" if applicable. You can specify this model name when making API calls to ensure you are using this specific version.

How can it possible? Did they Customized their GPTs Role?

wraith ginkgoBOT
#
<:book_icon:1171408210398289941> `` Rule 10 ``  Do not bypass the AutoMod filter.

This applies to similar or meaningless word variations, URLs, server profiles, and unique characters.

little plinth
#

in gpts, is it possible to have them be temporary messages if you want to?

#

it is! awesome

thorn terrace
#

Difference between Claude 3 Opus and GPT-4o

  • GPT-4o
    Absolutely! Here’s an attempt at a samosa recipe where every word starts with "A":
  1. Assemble all aromatic, assorted, and attractive additives.
  2. Arrange all appropriate apricot-colored, all-purpose articles.
  3. Add aromatic alliums and allspice.
  4. Arrange artfully and always await awesome aroma.
  5. Achieve amazing, appetizing, and awe-inspiring appetizers.

(Note: This is more of a playful version than a practical recipe. For an actual recipe, more detailed instructions and varied vocabulary would be needed.)

  • Claude 3 Opus
    Here is a 5-line samosa recipe where every word starts with the letter "A":

Authentic Aloo Amalgamation

Assemble all-purpose, ajwain, asafetida, amchur, and aloo.
Amalgamate aloo, amchur, ajwain, and asafetida; apportion accordingly.
Align all-purpose; apportion ample amounts and assemble.
Affix aloo admixture; artfully append and adjust.
Acquiesce aroma; apportion and appreciate authentic aloo appetizers!

Both models are better.

thorn terrace
# little plinth in gpts, is it possible to have them be temporary messages if you want to?

Lets Check GPT-4-1106 and Claude 3 Sonnet

  • GPT-4-1106
    Let's break it down step by step:

  1. You picked up a small ball and a bag with a hole in the bottom that is larger than the ball from your garage in New York.

  2. You then put the ball into the bag. Since the hole is bigger than the ball, the ball would fall out unless you were holding the bag in such a way as to prevent this.

  3. You walked to your desk and dropped the bag into a box. Depending on how you were holding the bag, the ball may have already fallen out by now. But if the ball was still somehow in the bag, when you dropped the bag into the box, the ball would likely fall out of the bag into the box due to the hole in the bag.

  4. You sealed the box for shipping and mailed it to your friend in London.

  5. A few weeks later, the box arrived at your friend's place in London.

Assuming the ball did not fall out of the bag at any point before the bag was placed in the box, the ball would now be either inside the bag within the box (least likely due to the hole) or loose in the box itself (most likely). Therefore, the ball is now in London, specifically at your friend's address, either inside the bag inside the box or simply inside the box.

#
  • Claude 3 Sonnet
    Okay, let's think through this step-by-step to determine where the ball is now:

  1. You started in your garage in the center of New York, where you picked up a small ball and a bag with a hole in the bottom bigger than the ball.

  2. You took the ball and the bag into your office.

  3. Before sitting down, you put the ball into the bag.

  4. You walked to your desk and dropped the bag into a box.

  5. You sealed the box for shipping.

  6. You went to the post office and mailed the box to your friend in London.

  7. A few weeks later, the box arrived at your friend's place in London.

Now, let's consider what happened to the ball:

  • When you put the ball into the bag, which had a hole bigger than the ball, the ball would have fallen through the hole and out of the bag.
  • This means that when you dropped the bag into the box, the ball was no longer inside the bag.
  • Therefore, when you sealed the box and mailed it to your friend in London, the ball was not in the box.

So, to answer the question "Where is the ball now?":

The ball is most likely still in your office in New York, wherever it landed after falling out of the bag. It did not make the journey to London with the bag and the box.

#

Oops, 😬

#

Answer: Ball is now in Office

thorn terrace
#

Differences between Command R and Command R+

  • Command R+
    To find out how many games they played, we can use the information given:
  • Susan won 3 bets
  • Lisa won $5

Let's calculate Lisa's number of wins based on the money she won:
Lisa's wins = $5 / $1 per game = 5 games

Now we can find the total number of games played:
Total games = Susan's wins + Lisa's wins
Total games = 3 games + 5 games = 8 games

So, Susan and Lisa played a total of 8 games against each other.

  • Command R
    Eleven.

Susan won three games and Lisa won back her $5, plus another $5, which means she won a total of five games. Adding these two together gives a total of 8 games.

We also know that Lisa won $5, which when multiplied by the $1 bet per game, equals a total of five games.

Therefore, the pair played a total of 8 + 5 = 13 games.

The answer 11, XD
Looks like Command R is better than Command R+.
.xd.

humble karma
#

I thought this isn't an official openai server

winter geode
#

Does anyone know how open ai embedding (text-embedding-ada-002) generates vector output, for example [-0.015501722691532107, -0.025918880474352136]. does this use transformers?

plucky shadow
#

anyone know why i might be reaching my quota, eventhough ive not had near 200 RPD or 3RPM and only say short things like 'hi' - im using gpt-3.5-turbo, trying to implement it into a discord bot.

supple violet
plucky shadow
#

didnt think i needed to, isnt gpt3.5 turbo free?

hallow owl
plucky shadow
#

ohh

#

i understand

white oar
#

how do i disable gpt 4o tools? it locked my most important chat and it just doesnt let me (i can only limitedly work now)

supple violet
white oar
#

there is no setting for it

supple violet
#

the checkboxes are right there

white oar
#

...

#

i checked all the settings its not there

supple violet
white oar
#

its all ukrainian

#

and i cant change it

slate wren
#

is chatgt down from anyone else?

white oar
#

nvm its changed

strange marsh
#

Hi all. Are we no longer able to set our Custom GPTs that are on the GPT store to private? Some of my public custom GPTs won't let me change the sharing setting any more.

strange marsh
strange marsh
slate wren
#

it would really suck if they removed it

strange marsh
strange marsh
slate wren
strange marsh
strange marsh
slate wren
#

what a mess

solid spear
#

New gpt idea - make a fallout or Skyrim mod that changes all the game's dialogue to zoomer slang. Or have it take whatever prompt you want to change all the game's dialogue.

#

I think it'd be funny

fiery zodiac
#

Why am i getting: We are sorry, but you do not have access to GPT interactions.

Everyday chat gpt has an error

#

not everyday but every second

fiery zodiac
#

what happend to chat gpt apps

thorn terrace
fiery zodiac
thorn terrace
fiery zodiac
misty depot
#

Have they removed access to GPTs for free users? Dont see it anymore. Or did i cross some limit

#

seems to be limit

thorn terrace
little plinth
#

hey, is it possible for us to specify actions in a custom gpt, like browse the web, or search through a document, in a certain order?

#

(not actions in a schema/api sense, just a certain set of steps to follow)

pine gulch
little plinth
#

is there a way to set up a prompt so that it listens? do I use an if else statement or something?

#

(it listens for some things, but not others0

pine gulch
#

Prompt it when to use the tools

#

like prompt is like this: "Use Web Browsing when asked a question that could need up-to-date information."

#

or whatever is needed for your specific use-cases

little plinth
#

hmm, I think my GPT is defaulting to gpt 4 on the website, is there a way to force it to only use 4o?

thorn terrace
sly lantern
#

Is gpt4 down?

thorn terrace
rustic birch
#

Today I was using ChatGPTs now showing GPTs only for plus user, is it only with me or happing with other people?

thorn terrace
rustic birch
oak maple
blazing scroll
hasty elk
dire geyser
#

Now GPT not even loading... : (

remote inlet
#

its back

dire geyser
#

👏 GPT

alpine void
#

Hey guys I'm in South Africa and ChatGPT Severs have been down for hours now. How do I make the team aware?

Error Message on Android:

The server is having problems. Please try again later. (89558d985fdcf72e-JNB)
olive drift
#

I'd like to create a mini-game with AI using API with friends. I'm looking at the prices page and it seems there is no free API. Is that correct?

hallow owl
remote inlet
drifting parrot
wind urchin
cloud wyvern
cloud wyvern
thorn terrace
cloud wyvern
thorn terrace
cloud wyvern
thorn terrace
#

lol, Okey You can't Download the ChatGPT app version?

cloud wyvern
cloud wyvern
#

No problem. Everything resolved with me the respect about the Gpts. 🥳

#

Thanks!

hallow owl
cloud wyvern
#

Sure!

high cosmos
#

hey, where should I discuss my problem with GPT ?

static isle
bright mirage
#

Anyone else seeing the New version of GPT available - continue chatting to use the old version, or start a new chat for the latest version notification at the bottom of your chat box with your custom GPT even after you've already done so? Like, if I click the link on that popup, start a new chat, I'll come back to that same chat later and it's marked as outdated even if I haven't made any changes to it myself

grizzled flax
high cosmos
hallow owl
raven flax
#

how to do this for colours: ?

ml name is ABC “Give me Cyan Color” ‘NowGiveMeRed’ 123

high cosmos
hallow owl
high cosmos
hallow owl
#

break it up into chunks

zenith gazelle
#

is there a counter for gpt4 limits or something? seems annoying to run out. I feel like 4o isnt as helpful on the programming front. is it a dynamic reset overtime or once I hit limit just walk away for 3 hours

left badge
#

Okay, Sam Altman, the tension from teasin has definitely built up. Time for that knockout. It's to the point of dwindling returns now. I'm talking to you, Sam Altman. I'm ready for that release. Where is the new voice update?

#

I know all about the trick of withholding to increase excitement. It certainly does work, but at some point it becomes an act of violence.

lusty sundial
hallow owl
high cosmos
twilit iris
split fulcrum
#

how to repair this when I want to pay ,,Your card has been declined,,

iron depot
grave cargo
#

Is there any good laptops for webdev, fullstack thats an alternative for macbooks?

snow moth
#

Are the GPT created on the website on GPT4 or GPT4-o ? 🤔

hallow owl
full flame
full flame
#

but honestly, at that point why not get a macbook

livid pebble
grave cargo
full flame
#

can i also suggest joining a server like buildapc and ask for help there? given their purpose is to help with this type of questions

quartz abyss
#

Everything seems to be working for me

muted zealot
#

Anyone getting this error?

”An error occurred connecting to the worker. If this issue persists please contact us through our help center at help.openai.com.”

#

On chatgpt 4o ( I bought it)

ancient canopy
muted zealot
#

seems for me its back working now

patent scaffold
#

Any word yet on when the screen sharing feature will be available???

oak maple
hoary frost
#

what.is.happening.with.gpt

Really - from a one shot to a shoot-it-endlessly tool?

Example:

Write an article (topic)...
Response:
[...]
Key Words:

  • word one
  • word two
    etc
    Literally. It thinks I am going to add a h2 list with keywords to the article. Like 1998.
    And you correct it, it repeats the crap

The whole day I have been interacting with it and it just gets worse and worse. Asked it to summarise an article, and it came up with the phrase offers comprehensive consulting services for web presence
That's not even a real sentence.
(All that is gpt 4o)
Anyone else seeing weirdo behaviour like this?

#

The best was when asked to use html tags appropriately for SEO
It came up with strong and em used interchangeably, text looked like circus-blink-lights after
Confronted with "why?" - of course first it attempted to "sorry I will rewrite" for like 10 times, and then finally it explained. Pretended that em shall convey some sort of superpower to search engines, while strong will add pepper or what not (jokes aside: no, writing "offers PHP programming and Consulting services for Companies and Private Owners" is just not anything related to SEO. It looks like someone has discovered the em tag and had fun, rather than actual optimised content.)

crimson dragon
#

is Chatgpt working in slow~ mo~ currently ?

grim ether
tropic scarab
#

Hello Everyone
Is it possible to programmatically generate multiple images with consistent characters from a plot for a film?

wise kettle
#

Prompt adherence from GPT-4o is terrible compared to GPT-4

#

anyone else noticing this?

hallow owl
wise kettle
#

No issues with GPT-4 (base prompt is enough to define input format). With 4o - persistently throwing in markdown.

#

Another example (happening now) is trying to get it to review my landing page copy section by section. I gave it my current landing page copy as a pdf input, then wrote a long prompt with expectations for output. Then said "you'll be reviewing the landing page block by block" ok? It went straight into the review of the whole thing. I tried telling it - "I will be instructing you to review one block after another. Now review block 1". Nope - goes in all guns blazing and reviews the whole thing.

#

switched to GPT-4 and it follows my prompt much better.

hallow owl
wise kettle
sullen falcon
#

Why did my GPT downgrade to GPT 2?? I used 4-o like a week ago without premium

#

And when i downloaded the app it had 4

grim glade
#

When I upload files into "my gpt" and I ask a question/questions based on the file, the gpt sometimes tell the wrong answer . Can you guys prefer a prompt making gpt to give reliable info ?

grim glade
#

Probably over 1200 pages, maybe I should shorten it ?

#

i am plus member, could it take fx 20 pages?

fluid helm
#

Is there anyway to get the output from assistant in a json format ?
I tried writing in instruction but im not getting the valid json response.

pseudo trail
#

whenever I let chatGPT write a mathematical equation it always uses a specific math font, Do you guys think that there is a way that I can make it write in a standard font instead? When I ask it to do that, it doesn't

wind urchin
#

why's gpt 4 taking significant time to analyze my prompt, it wasn't like that before it was more instantaneous

cedar sierra
vernal jolt
#

wouldnt windows desktop app get way more used than mac desktop app?

sonic edge
#

yes, thats bs releasing it for mac

frigid burrow
#

Is it just me or has 4o been terrible lately?

#

The only of my GPTs that still functions reliably is the one that uses a vector database to answer questions.

zealous gull
#

*The desktop app is only available for macOS 14+ with Apple Silicon (M1 or better). Coming to Windows later this year.

#

😭

frigid burrow
#

This last week I've run into two use cases that routinely fail. These are both using ChatGPT 4o Teams subscription.

  1. Logic: when writing code or identifying errors in code, it used to take any given error and reliably identify the root cause or at least tell me what to look for. Now, at best, it simply paraphrases the request. At worst it makes things up.

  2. On the side I'm helping my wife with her historical research project. I ask 4o about a historical figure. It generally responds with, and can find a source for, information about that person. I ask about that person's spouse, it responds with a paraphrase of the question and a seemingly random family member, often a sibling. When I point out that a sibling cannot be a spouse, it simply paraphrases the question again, without correcting itself and without acknowledging the error.

In both cases it is impossible to get 4o back on track. It tends to fall into loops of repeating paraphrases of the original request.

willow jackal
#

I just saw the latest announcement from Alex about the desktop app for macOS, and I was wondering if it is available for Macs with Intel or is it supports only Apple silicon?

#announcements message

small pebble
hallow owl
small pebble
hallow owl
small pebble
broken sand
#

do the new voices will also be available through the TTS model?

little plinth
#

Is there a way to force the gpts to do actions already baked into the model? Without apis or anything of the sort

#

Like browse the web

#

Or create a prompt

#

Or make an image

reef turret
solar vigil
#

Is there a problem with gpt o4? Because his responses are very slow!

ember axle
#

💰💰 What is the industry standard in calculating prompt cost incurred in a LLM app? The app is having 30-40 prompts. Currently, I just sum up all the tokens cost based on 1 user's input (assuming the most complicated use case).

I am not sure should i sample 10 user's input and average them up? or based on 1 user's input is fine.

floral arrow
#

Hello, anybody has a perfect Knowledge Base to upload to build GPTs. Any example is really appreciated, wanted to have a look and create it for one of my client

hallow owl
floral arrow
#

Yeah I understand that and I'm asking for any knowledge base, just to examine it.

willow jackal
idle furnace
#

model improvments is all im looking forward

thick talon
zealous gull
#

Has anyone had luck with GIF generation on Paid Gpt? Struggling for it to generate a proper animated gif, and or even keeping the same character in each frame for me to generate the animation out of it

pine gulch
zealous gull
pine gulch
zealous gull
#

Sounds good!

broken sand
#

do the new voices will also be available through the TTS model, any thoughts?

supple violet
idle plinth
#

where can i use criticGPT

pine gulch
idle plinth
#

ah, ty

pulsar plover
#

Hi people

#

You can still have plugin functions on a single Chat?
Like video Sumarizer plus Diagram maker plus something

#

Or that's over now?

grizzled flax
# pulsar plover You can still have plugin functions on a single Chat? Like video Sumarizer plus ...

Hey! Plugins are deprecated now and have been replaced by GPTs. Most popular plugins have a GPT replacement by now. There's not an exact equivalent feature to start a chat with multiple GPTs like that feature with plugins, but you can use an @mention feature to call a certain GPT in the middle of another chat, so there's flexibility to use many GPTs in a single chat. You can try it by starting a message with the @ symbol, and you'll see the GPT selector popup.

pulsar plover
inland bramble
#

hey all ! quick question .. when we are in a workgroup how do we get an API ?

night karma
#

Hello everyone, I am running into an issue with a custom GPT I made and was hoping someone could help here. I am a medical student and I am trying to create a custom GPT that can make practice board questions to help my fellow students and I study. I wrote an extensive guide to writing board style questions in a word document and uploaded it to the knowledge area into the edit gpt section, and then asked it to create questions based off another word document which is compiled of information from every lecture throughout a semester. It doesn't necessarily "fail" but the questions don't seem to be of high difficultly and it improperly cites its sources. Can anyone shed some insight?

tranquil dome
#

I would like to ask, the API connection provided by OpenAI, and the ChatGPT on the official website of OpenAI, both have policy restrictions, what are the advantages and disadvantages of development between the two in terms of chat? (I think they are almost the same, especially is model GPT-4o)

Because through repeated testing in playgrounds and developing software through APIs, I was surprised to understand that there is no API that surpasses ChatGPT in terms of chat capabilities (PlayGround's response can only be worse or even garbled).

I think that users should be allowed to set API selection settings with lower policy restrictions through setting membership status, age restrictions, confirmation of independent regulations, etc. through the connected API; I think that may clearly and ChatGPT show the difference between our paid API usage.

solid tide
#

any one has problems about lost all my GPTs?
I lost my all GPTs that I create and don't know how to do anyone help?

fallen trellis
#

I lost my GPTs too. I'm not normally on the discord but came here to learn about that

supple violet
fallen trellis
#

That worked. Redirected me to chatgpt.com/gpts then I clicked My GPTs then they are there for me.

#

And now they are in my left pane again

#

Thank you!

hoary frost
#

does someone know how to stop a custom GPT from adding markdown to its reply?

#

I cannot share the instructions used here, since the obnoxious limit of characters is busted.
But I think it is needless to say I tell it .... do not add markdown, ever

spice ice
steel turtle
# hoary frost *I cannot share the instructions used here, since the obnoxious limit of charact...

I believe you can attach your instructions as a .txt file in #prompt-engineering, and they might be able to help.

As a 'prompt engineer' I am going to tell you that if you don't have a ton of experience prompting, don't prompt negatives. If you're going to give a negative instructions, contrastive chain of thought prompting teaches us the best way to do so is by giving a bad example and a good example of expected output.

spare swift
night karma
rare goblet
#

newb question for Enterprise accounts, "improve model for everyone" is turned off by default. Under this Enterprise account when I create a GPT, set it to to "public > anyone with a link" and the GPT gets tested on a different FREE ChatGPT account where "improve model for everyone" is turned on - is this GPT now exposed to be trained on?
I hope that makes sense, thank you.

supple violet
rare goblet
drifting timber
#

Hello, is there a chatgpt plugin to find appartment for me?

umbral egret
#

This just in, gpt for map making???

#

I don't mean dalle i mean python code.

#

Trying to do what i can to adjust it then I'm gonna maybe release the code somewhere lol IDK

vagrant musk
#

How do i get 4o to stop generating the same bs over and over and over again?

#

I am typing "stop" "stop" "stop what you're doing" "cease" and it's stuck saying "understood, heres how...." and just keeps creating the same lengthy instructions on how to do things i never asked it to do

hallow owl
dull yoke
hoary frost
limber frigate
#

Hi, does anyone know why the gpt api response is a lot worse than the web? I ask a simple question "who is the director of company XYZ", the api was unable to find an answer and suggested to look at their website, but the web was able to respond

hallow owl
limber frigate
hallow owl
#

the API will never have web browsing as that's something developers need to implement inside their own applications

limber frigate
#

thanks I guess I’ll have to find a workaround that

hallow owl
limber frigate
#

yeah rn I’m using duckduckgo api and do some web scraping and ask chatgpt to find the answer based on that context

hallow owl
limber frigate
#

yes I did

quick lava
#

I feel very stupid today. Is there simple way to send GPT's dialogue to google spreadsheet?

hallow owl
quick lava
#

I created GPT, ask questions about company's financial aspects and give recommendation. GPT asks questions and user answers. I want to log questions and answers to google spreadsheet.

#

for review, to be sure that recommendations are OK

#

otherwise I do not know what it recommends

hallow owl
broken sand
#

do the new voices will also be available through the TTS model, any thoughts?

carmine jackal
#

Looks like the Virtue Signallers have struck again. I was working with a dataset of people. This dataset contained one person that was 15 years old. The data has nothing to do with any bleep activity yet ChatGPT chucked a dummy spit and claimed it violated policy. This is just getting beyond a joke!
And since when did it change from "We might have made a mistake" to We are right you are wrong REMOVED!

quick lava
#

ah, sorry

dull yoke
#

I managed to run chatgpt into command prompt which can also do google searches by using some python coding

left badge
#

What gives OpenAI the audacity to change the parameters willy-nilly as if it's no big deal on their models when some users do pay upwards of $60 a month for the service? I'm out of the game. I don't use Claude or anything else, but it is frustrating. FYI.

little plinth
#

Hey, is it possible to get a gpt to call on other gpts?

#

Kind of like nested gpts, and if possible is there a limit on how deep the rabbit hole goes?

wind urchin
little plinth
#

no idea

gloomy olive
pine gulch
versed hedge
#

What are the advantages of buying a paid abo and not buying? I mean, I can still use gpt 4o though

ivory shadow
#

Hi everyone. I'm currently creating a GPT and was wondering when you upload an image to the knowledge section does the GPT use vision to read and understand the image? My image has images and text inside of it. It's a screenshot of a web page. Also, if I upload a PDF to the knowledge that has images in it, will GPT use vision to understand the images in the PDF or will it just OCR the whole PDF, getting rid of the images in the PDF? Thank you.

grizzled flax
hoary frost
#

Does anyone feel those “memory” of gpt are very much useless? For example I saw it saved “prefers I answer with “I do not know” rather than providing example functions that do not exist in php” and yet it keeps making things up. Are those memories only used if I explicitly invoke them?

coral idol
#

theyre like suggestions not hard rules

#

use the customize function to get it to actually do stuff

tall lake
#

Does anyone also have the problem that you can no longer access another GPT (e.g. GPT2) in e.g. gpt1 using the @ command?

hidden swan
#

Hi, I have a quick question, I have made a GPTs and I want instead of putting the file directly on the knowledge base I want to put it in a drive like google drive and making up a link between my GPTs and my Google drive Folder with I think GPT Actions, but I thing is when I want to ask ActionGPT to help me building the OpenAI schema tu making up the link between those it give me python and I cannot put it in my GPTs, anyone already done this kind of actions if yes can you helpe me ?

supple violet
supple violet
hidden swan
supple violet
hidden swan
supple violet
#

gpt specific, no

#

there are services to ask questions about docs with ai outside openai, some support gdrive live link

hidden swan
supple violet
hidden swan
supple violet
#

i dont have any specific in mind. btw with coding skills you could use openai api to replicate your gpt and auto update knowledge files when they change in gdrive

hidden swan
glacial breach
#

I just got the "You’ve reached your GPT-4 limit. New responses will use GPT-4o, our new, faster model with a higher usage limit."
But I don't like 4o! Will I be able to use 4 again or is this forever now?

supple violet
#

then wait the cooldown to use gpt-4

glacial breach
#

So there is a cooldown? the message doesn't say anything about that

supple violet
glacial breach
#

Thank you. I was just afraid they were going to discontinue 4, because they keep insisting 4o is just as smart.

pine gulch
ionic anvil
#

gpt4o is as close of an direct replacement for Turbo as it gets pretty much. They went a bit too far with the long responses (esp code), but that's nothing that can't be fixed. They sacrificed some edge use-cases yet that gonna be completely irrelevant once gpt4 successor drops.

rough flume
#

Is GPT-4o not just better than GPT-4? I use it almost all the time because of its faster responses, and I can't seem to tell the difference.

pine gulch
blazing scroll
heavy chasm
#

has anyone tried doing custom GPTs + Zapier to get the GPT to do actual tasks with accounts? ive messed around with it but noticed zapier is unreliable. has anyone tried doing something similar but with langchain and connecting accounts using agents?

worthy isle
#

This morning I opened ChatGPT and anything I ask it I'm getting a "Verify your Human" pop up, over and over, every single time. Why is this ?

neat iris
worldly copper
dim niche
#

yo, wen monetarization for GPTs?

fervent sinew
#

This happens when my VPN is enabled so I just disable it.

pine gulch
hidden swan
#

Hi, i have a quick question, How to keep a consistency in GPT response, it's mean that one time i ask him for a certain document in english and he will give me the document in english and the other time Im gonna ask him for the document in english and he will give me the document in french, so how to keep this consistency ?

rigid estuary
violet raptor
#

Maybe I'll return when they get their s*** together

grizzled flax
# violet raptor This is the kind of annoying occurrence that happens all too often that has push...

Hey! You can find the context window specified as 32K here: https://openai.com/chatgpt/pricing/ Scroll down to the Free/Plus comparison, then check the Model Quality section.

The hallucinations you shared in the chats come from how LLMs are trained. They're trained on "everything data", not "self-knowledge data". By this I mean: when you ask a question of ChatGPT like this, it doesn't have a special "self-knowledge" memory to check for an answer in -- instead, it generates an answer in the same manner that it generates any other answer: by referring to the entire body of its training data, which doesn't always include bleeding edge details like the context window limit of the model itself.

(PS: 4o is 128K on the API)

violet raptor
#

Over the past month of using it, it's just built up frustration.

#

I don't like constantly having to say, "Make sure to use search results from only the past 3 days" and stuff like that to actually get real answers on current information.

#

I like what chatGPT offers aside from the search. But the search feature is such an important thing to me that when it just never works properly, it's very frustrating.

grizzled flax
violet raptor
#

That could be the case. And I know better prompts can provide better answers.

A big one for me is asking about updates to software and it providing willy-nilly results saying things about very old updates. Sometimes updates in the future that don't even exist yet. It's just all over the place.

#

Often I find it taking multiple different versions that it's found patch notes on and mixing it all together as if it was one update.

hallow owl
#

it's off topic for this channel

sand grail
hallow owl
sand grail
#

I got the point..

spark roost
#

chat gpt 5 will release this year?

hallow owl
dull yoke
# spark roost chat gpt 5 will release this year?

Hey Schierke! The current timeline suggests that testing for ChatGPT-5 might begin by the end of 2024, with a possible release several months into 2025. While details are still a bit under wraps, this seems to be the general expected timeline for now unless there's any unexpected delays.

hallow owl
#

we don't even know there will be a ChatGPT 5

dull yoke
hallow owl
dull yoke
#

Sources:

Evening Standard
Geeky Gadgets
ReadWrite
India Today

I guess you can check if those websites got integrity for yourself.

hallow owl
#

as I guessed, unfounded speculation, none of those sites are news organisations of record

oak creek
#

hey gys, how do you guys adapt, optimize and configure chat gpt?

hallow owl
oak creek
#

visualsm or how you guys deal with the project of planning your team of gpts.

#

someone using second brainr a lotw

#

just wanna tallk periferical thinhs

hallow owl
oak creek
#

i am like trowing questions in a weird way, i know

#

ust wanna talk about something, feel like you guys are enjoying GBT?

#

I'm studying chatGPT on my own and unfortunately I had to do the same with programming.

old bane
dull yoke
old bane
#

Hrm. It seems like the old v3.5 worked the best in terms of remembering things, but I understand there were people who abused that.

Any hints on new features?

dull yoke
#

AI that is better at reading emotional reasoning thru text, and an AI that is much more customizable that can act like a companion instead of just a tool. ChatGPT-5 aims to enhance emotional intelligence, making it better at understanding and responding to human emotions. It will offer extensive customization, allowing users to tailor its personality and behavior to suit their needs, making it more of a personalized companion.

Improvements in memory capabilities will reduce the need to repeat instructions, addressing current limitations. Additionally, advanced multimodal abilities will enable the AI to process and generate text, images, audio, and possibly video, opening new creative and interactive possibilities.

old bane
#

👍 awesome! thx. Tone can be important.

worn orchid
#

Are there any news about GPT 4o like when the ither in/output types for GPT-4o get activated for API

final saddle
#

Are there known issues with DALL-E images not being reliably created with GPT Instructions? (Context: I'm trying to instruct a GPT to create an initial scene-setting image, but it either spits out the prompt, a "![blah blah]" truncation of a prompt attempt, or just a 'line break' horizontal line, none of which are accompanied by an image being displayed.) ??

polar mango
#

I'm building a custom GPT that generates a report based on a template. Why does it initially fail to generate a correct download link, but after saying "try again," it provides the correct hyperlink?

floral light
hallow owl
floral light
#

I recall him saying gpt5 was going to be way better than gpt4 and that gpt4 they admit wasn’t good

hallow owl
floral light
unborn heath
# floral light He definitely did, Lex Fridman podcast ep419

Do you want maybe to pull the quote you're talking about from the transcript?

It definitely doesn't use the words 'very good' in order (that's said twice in the transcript, once in connection to trust with board members and once about having a 'very good' coping mechanism).

It's pretty easy to point to a 2 hour long interview and say "this says what I am talking about" but that cuts out a lot of stuff and isn't necessarily true.

There's related stuff; the interviewer asks "what's been the most impressive capabilities of GPT-4 to you and GPT-4 Turbo?"

And Sam hesitates for several seconds and answers, "I think it kind of sucks".

He then clarifies, "No, I think it is an amazing thing, but relative to where we need to get to and where I believe we will get to, at the time of like GPT-3, people were like, "Oh this is amazing. This is this like marvel of technology," and it is, it was. But now we have GPT-4 and look at GPT-3 and you're like that's unimaginably horrible. I expect that the delta between five and four will be the same as between four and three. And I think it is our job to live a few years in the future and remember that the tools we have now are gonna kind of suck looking backwards at them and that's how we make sure the future is better."

However, also in the interview it's clear that there's no commitment or real interest to hold to the numbers of the GPTs. In fact, there's a strong lean away from that.

The interviewer asks, "Is there gonna be moments Q-Star or otherwise where there's going to be leaps similar to GPT where you're like-"

And Sam actually interrupts him to answer:

#

"That's a good question. What do I think about that? It's interesting to me it all feels pretty continuous. - This is kind of a theme that you're saying is you're basically gradually going up an exponential slope. But from an outsider's perspective for me, just watching it that it does feel like there's leaps, but to you there isn't. - I do wonder if we should have... So part of the reason that we deploy the way we do is that we think, we call it iterative deployment. Rather than go build in secret until we got all the way to GPT-5, we decided to talk about GPT 1, 2, 3 and 4. And part of the reason there is, I think, AI and surprise don't go together. And also the world, people, institutions, whatever you wanna call it, need time to adapt and think about these things. And I think one of the best things that OpenAI has done is this strategy and we get the world to pay attention to the progress to take AGI seriously to think about what systems, and structures, and governance we want in place before, we're like under the gun and have to make a rush decision. I think that's really good.

But that's what we're trying to do. That's like our state of the strategy and I think we're somehow missing the mark. **So maybe we should think about releasing GPT-5 **in a different way or something like that."

#

And there's other and various discussion all over the place, it seems pretty clear that AI models are going to continue to improve. That the development is continuous, and it may 'look' like there's big jumps, but really there aren't, it's like the jump from a baby being unable to sit up to sitting, then it can crawl, then start to stand, then start to walk, then start to run.

All those are common milestones we see and are used to, with humans.

But there's no obvious "Commitment" to the name GPT-5. And it's not clear that there's a defined form that GPT-5 is going to have, specific capabilities that will cause the company to say "Aha, this is GPT-5".

And that Open-AI may not be committed at all to the naming system, may feel it has problems in how it guides us to think about the models and their changes.

But Sam's goal isn't, I truly believe, GPT-5.

It's AGI/ASI, and things working well for us all.

It's like if you expect your child to grow up to be a neurosurgeon, but they're currently a toddler. And you can see the neurosurgeon in the child's future, you're envisioning the talks you have with the child when they're in their 20s and 30s.

It can be hard to get super excited that the child's mastering drinking out of a glass, because you're focused on the so much bigger stuff ahead.

unborn heath
#

We have to be careful though, if we want to be factual to intention.

Like, just from what I quoted there, we can see that Sam said "I think [GPT-4] kind of sucks" and he also said just a few sentences later, "GPT-3 ... 'Oh this is amazing. This is this like marvel of technology,' and it is, it was."

So, some disinformation-pushing person could even use that to support a claim that Sam thinks GPT-3 is better than GPT-4.

Which is completely not the point he's trying to make and not what was said (despite the words lining up that way in the strictest context, but modern AI can tell the intention there, and most humans can too).

But he's also absolutely not saying that GPT-5 'is/will be so much better', to me that's clearly not his intention and he's very against that way of thinking and trying to find ways to help us not think that way as a society.

He wants us focused on the end game, on AGI/ASI, and compared to what we have now and will have this year, what we have now is nowhere close. Which is just factual, it's fine and how stuff works, much like a 6 month old baby is just that, not horrible or awful or unskilled, and step by step the future comes and will be pretty awesome when and if we get there in a healthy and safe way.

hallow owl
viscid totem
#

if I buy the $20 a month plan, do I get unlimited messages per hour?

hallow owl
cursive wave
#

what's the cheapest, fastest possible gpt to use, that has some level of IQ. i need it to determine what action the user wants to preform.

#

@hallow owl (0the goat) mind helping

hallow owl
indigo wind
# 
response_format of type json_object type are only compatible if all tools are type function.
Error: 400 Runs with response_format of type json_object type are only compatible if all tools are type function.

dude... you have to be kidding me...

#

is there a way around this?

#

I am using a thread run for the assistant API

#

and also I did not see this anywhere in their reference docs

waxen crown
# indigo wind is there a way around this?

You can sort of work around it by instructing it to return json in the system message. It's generally pretty good at doing that if you give a good example, even though the response_format isn't set to it. Used it a fair bit with old models that don't support json_object and they do pretty ok that way. I assume you want to use file_search and/or code_interpreter with json_object in an assistants run?

indigo wind
#

uhm, I guess I could give it examples but my instructions already provide a clear format, thing is it adds on additional info when I prompt it which messes up my backend's ability to parse the JSON properly

#

I just can't afford it making mistakes, would it just be best to take its output and funnel it through 3.5 with JSON mode using chat completion?

waxen crown
#

It makes mistakes with json mode sometimes as well. When I used it as a work-around it was pretty bullet-proof. Depending on how wonky the JSON is, it is very capable at just doing it based on instruction even without response_format. Better than being stuck at least.

indigo wind
#

hmm I guess I'll ask it to please please not add any additional text in its response, hopefully that makes it more reliable?

#

and I guess I could make my backend try multiple times in prompting it

waxen crown
#

Yeah I had a bit of robustness code to handle the low-hanging fruit, tbh. But it did pretty well - but I always try to have itg return simple flat schemas.

indigo wind
#

how so?

waxen crown
#

As in making the json more flat than I think is really nice, if it wasn't to be used for return by an LLM. Avoiding nested objects whenever possible (except key value pairs), etc.

#

One an always map it back to a nice schema afterwards

indigo wind
#

ah, yeah I have practically no nesting going on

#

I'll try to go with just instructing it to do better, and see how that goes

indigo wind
cursive wave
hallow owl
outer moat
#

When will OpenAI release the update to interact with AI by voice

grizzled flax
outer moat
#

Thank you! I can't wait to practice languages ​​live

fringe scaffold
#

Why is chatGPT 4o lying to me and keep doing stuff I didn't ask it to do?

thin cloud
sacred steppe
#

is it just me or is the regular gpt way dumber than it used to be?
I tried continuing a conversation about regular expressions right after I reached the limit and it immediatly started spewing gibberish the following message.
I was genuinely flabbergasted because an elementary student wouldn't make these mistakes.
I'm pretty sure that gpt was smarter than that before

fringe scaffold
humble swallow
#

how can i use gpt 4o to code a mobile game

supple violet
humble swallow
#

so i need some coding expirience, is that correct

supple violet
#

no, you can learn from zero with it as your guide or assistant

unborn heath
# humble swallow so i need some coding expirience, is that correct

You'd be doing a lot of your own footwork and self-guidance, but it's going to answer questions you ask, and many of the answers will be useful (remember you can ask the same question more than once).

But you need to be self directed, and follow up on what it tells you, check to make sure the ideas work, the code works, and all the resrt.

old thistle
unique wedge
#

I wrote a whole web app using GPT 4o in Python Flask with JavaScript and Bootstrap with zero knowledge of coding. Honestly, if you know how to explain what you want it can help you a lot. Sometimes, as it's been said, it produces errors in the code but you can then ask what is wrong with the code and explain the errors you get and most of the time it will give you a way to correct it

gray lynx
#

I was working on a chat bot last year. There was an update to the model that changed the tone of the respones from a more laid back one, to one being very formal. No matter how I crafted prompts, it would stick to the formal style and continue to add things like, "as an LLM I cannot" or similar phrases even when being told to not use them. Completly killed my interest in furthering the project. Does anyone know what exactly the change was and if there are now ways to get my bot to basically not sound like a nerd?

red geode
#

I m a student and currently working on a customer support chatbot system for Bengali and Banglish (Bengali written in English alphabet). My question is regarding Fine-tuning. If I train a LLM model like GPT 3.5 or 4o with 100 real life Bengali chat conversations, can the model learn the conversation pattern? and the give right answer ? please help

errant sluice
warped siren
#

Got banned ... Unban me bwbs48ew7tkh9yr ban id

hallow owl
frigid burrow
#

Anyone have experience writing "put" actions for custom GPTs? I can't seem to get it to put the query in the body of the request

# 
        "summary": "Update an object",
        "description": "Update an object by its ID.",
        "operationId": "updateObject",
        "parameters": [
          {
            "name": "className",
            "in": "path",
            "required": true,
            "description": "Name of the class",
            "schema": {
              "type": "string"
            }
          },
          {
            "name": "id",
            "in": "path",
            "required": true,
            "description": "ID of the object",
            "schema": {
              "type": "string"
            }
          }
        ],
        "requestBody": {
          "required": true,
          "content": {
            "application/json": {
              "schema": {
                "type": "object",
                "properties": {
                  "class": {
                    "type": "string",
                    "description": "Name of the class"
                  },
                  "properties": {
                    "type": "object",
                    "description": "Properties of the object"
                  }
                },
                "required": [
                  "class",
                  "properties"
                ]
              }
            }
          }
        },```
#

it does the parameters part perfectly but doesn't do the request body

frigid burrow
#

got it sorted. Weaviate needed patch

smoky saffron
#

Hi I need help regarding me RAG chat I have around 30-40 files got embeded into the vector store problem I'm facing is

Latest Episode

  • Episode 4: The World-Class Women of Woodbine
    • Woodbine proudly features several successful female figures, including trainers and jockeys like Catherine Day Phillips, Sofia Vives, and Emma-Jayne Wilson, who contribute to delivering a high-quality racing experience.

you can see Emma-Jayne Wilson when I ask chatbot who is EMMA is doesn't respond and say I didn't have info about this but when I ask who is Emma-Jayne Wilson is respond correctly

What would be solution for this ?

thorny ferry
#

import openai
openai.api_key = "mykey"
MODEL = 'gpt-3.5-turbo'
try:
response = openai.ChatCompletion.create(
model=MODEL,
messages=[
{"role": "system", "content": "act as a coach that gives advice with a lot of examples."},
{"role": "user", "content": "Tell me the meaning of life"},
],
)
answer = response['choices'][0]
print(answer)
except openai.error.OpenAIError as e:
print(f"An error occurred: {e}")

This is my code, but I get an error every time I am trying to run the code:
An error occurred: You exceeded your current quota, please check your plan and billing details. For more information on this error, read the docs: https://platform.openai.com/docs/guides/error-codes/api-errors.
I had never used this api key

thorny ferry
supple violet
dire surge
#

Will GPT 4o mini support giving a response in JSON format.

ancient badge
mellow lake
#

is the image token count on the 4o mini correct? I was able to send 150k tokens to a supposedly 128k token model:

"usage": {
    "prompt_tokens": 147787,
    "completion_tokens": 328,
    "total_tokens": 148115
}
supple violet
fathom oriole
#

the rate limits also changed with this update?

mellow lake
#

the price per image is the same, but the token count is way higher on the 4o mini for some reason...

fresh trout
#

can you even use full capabilities of 4o-mini in playground or do you need to wait for the chatgpt release

mellow lake
#

yes, I am testing via API and playground

stray glacier
#

Is 4o mini smarter than 4o

supple violet
stray glacier
#

Aight

modern lark
#

whats the difference with 4o and 4o-mini

static isle
uneven lagoon
uneven lagoon
stray glacier
#

is 4o smarter than 4 turbo

static isle
lavish dust
#

Any idea when will it be possible to fine tunning 4o mini?

fringe scaffold
#

Hello there, is there a quick and easy way to clean up my convos in chatGPT like multiselect and delete them or do I have to go through each of them?

yes, there is a delete all chat, but just selected chats?

hallow owl
fringe scaffold
pine gulch
lavish dust
obtuse hound
#

will gpt 4o mini become the new free model and phase out gpt 3.5?

static isle
obtuse hound
#

what's the difference between 4o and 4o mini?

static isle
obtuse hound
#

oh okay

worn orchid
#

Are there news about the audio and video input for GPT 4o and the audio and image output.

hallow owl
worn orchid
hallow owl
worn orchid
hallow owl
worn orchid
hallow owl
worn orchid
hallow owl
worn orchid
hallow owl
wraith ginkgoBOT
hallow owl
#

more people use ChatGPT than the API

worn orchid
hallow owl
worn orchid
hallow owl
worn orchid
worn orchid
#

Can GPT 4o mini generate Longform content like scripts for YouTube Videos or longer stories that aren't just like 1 page long

outer fractal
worn orchid
livid jackal
#

is gpt4 down again? I can't seem to connect to my teams account authentication just goes to blank page or says site isn't secure

surreal gyro
#

Server is down? Got connection timed out.

sleek plover
hallow owl
sleek plover
#

So the new voice model will be released to the API but not in real time?

hallow owl
upper tusk
#

"By the end of July, do they mean the very last day, a few days before the end, or the end of July next year?"

#

New voice Mode??

wraith ginkgoBOT
#
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.
paper hornet
hallow owl
paper hornet
upper tusk
upper tusk
sonic edge
#

Im in the EU and randomly got the memory feature and after 5mins its gone again. Did anybody else experience that too?

lean jay
#

Interesting.. Mini consistently misspells composure as composposure for me...

radiant pecan
#

I want to create pdf using python and openai.
so, generate table contents using openai according to uploaded file and prompt text.
then generate description of each section.
how can I do

lean jay
unborn heath
lean jay
unborn heath
slate kindle
pine gulch
covert swallow
#

............I really wish they didn't upgraded it to Gpt4o............it learns the worst part of mankind, "Pretending it knows things that it actually isn't correct" well, I know sometimes it does that even when it was gpt4, but now it happens pretty much every single conversation.

tired glade
covert swallow
#

the worst part is, notice it just answer my question by copy the "clues" from my question.......and when you actually ask it" ........may you please check what you said? that's pretty much copyed from what I asked, please tell me first that you know the thing that I am talking about now, please." I felt the wise friend was long gone, only its dumb twin brother stay....................

dense veldt
#

Will searchgpt be available in api?

neon crescent
#

Think correct question is "Will searchgpt be available" for general use anytime soon 😛 Wouldnt worry about API before people can use it at all

#

Would expect months before that

fast goblet
#

I think OpenAI need to focus on the features they alreayd have and make it best and work perfect.. rather then these new updates they are doing.. "SearchGPT, GPT-4o mini, CriticGPT," .. We dont even have that memory feature in EU yet.. TTS dosent work on PC ? .. a "mp3" file it cant even "read" .. Just a thought.... hehe

vast cove
#

Could you folks please put out something similar to a Alexa device or a Google home hub? I my cash is burning in my pocket waiting for such a thing. 😶

pine gulch
vast cove
pine gulch
thin agate
#

Is there any way to use gpt 3.5 again? Becoz i cant see the option to change it (for free version)

hallow owl
lean jay
#

They need to fix GPT-4o Mini always claiming it can't see images.....

#

You show it an image, asking it to describe it, and it's like "I can't see images, but I can try to imagine what it looks like.", then describes it in perfect detail.... 😒

tiny socket
#

Okay. Is the 4o Mini model good for generating blog posts rather than the old 3.5 turbo?

lean jay
#

Omg, every time... I've even tried telling it in the system message that it can see images in great detail, and every time:

I can't see the details of the plate, but it looks like it has a floral design around the edges! If you want to describe it, I’d love to hear more!

#

"I can't see details, but here's a detail 🤪"

frigid burrow
#

How large can an API response be on custom actions before it throws an error?

sharp sedge
#

Does anyone have issues with accessing the site? It just keeps on loading on my end.

hoary trail
winter latch
#

People who know Russian?I'm Russian

paper raptor
#

Hello. There used to be an option when making images to edit specific parts of the image by selecting it with a brush but that option seems to be gone now. Anyone know why?

hot harbor
#

How many parameters do GPT-4o models have (4o and mini)?

grizzled flax
paper raptor
urban sail
scarlet linden
#

Hello everyone.
I am looking to write a function call to configure settings of a network utility based on OpenAI.
My prompt is as follows but I does not work correctly.
(In this prompt, the create_profile and create_posture_check functions need to be both called, but only one of them is called usually.
And although settings values need to be fine-tuned in the future, they are too different from correct ones at the moment.)
I want help from professionals.
Thanks.

digital haven
#

Is gpt vision out yet?

tame pawn
#

Hi

#

Is this a good channel to make questions about openai assistants?

vast cove
#

Feels like over the past few weeks Gpt has gotten worse in its level of quality.

frigid burrow
#

Does anyone know the max size of an API response for GPT Actions?

deep jetty
#

is memory option working for u guys ! for doesn't work is that normal ?

plucky socket
stoic lake
#

I'll actually probably use that gpt a lot.

#

it would be better though if it actually displayed hte exact llama response, not just a summary of it, sometimes i want ot compare syntax

upper tusk
#

"Well, how's the new Voice Mode?"

livid saffron
pine gulch
#

Yes

lone mesa
#

Hey all I have a random question: Can custom GPTS contain malicious content/harm your privacy?

wet stream
#

how do I manually update the memory to ChatGPT?

wind urchin
wet stream
#

It only allows me to delete one of the memories rather than adding a new one manually

drowsy quest
sick night
#

what does one has to do to become an alpha tester

pine gulch
sick night
#

🍀🍀🍀🍀

true plank
#

I've been trying to build my first custom gpts for about a week now, but SHE's kinda stuck in the mud. I'm trying to get info into the configuration by uploading png screenshots and every time I try I'm getting the "Hmm...something seems to have gone wrong." error from HIM with a button to Regenerate, which just loops until i refresh the browser... I'm also getting the same response from HER... HE is the configure side and SHE is the custom gpts... they both seem to think they can accept png screenshots, but neither seems to know they can't...
Now if go over to the regular free chatgpt i can upload the screenshots and HE can read them fine...
Is this a known issue or do i have something configured wrong?? it's driving me nuts...

solid sky
#

is there any free ai tool that can draw diagrams?

dim niche
#

yo, what happens to your custom GPTs in case youcancel your Plus Subscription?

supple violet
dim niche
hallow owl
dim niche
hallow owl
dim niche
#

yes, this is the only official thing, I already saw

hallow owl
#

I think this falls into the category of "seemed like a good idea at the time but it won't go anywhere"

dim niche
#

I have to say I am quite dissapointed, because when they announced the GPT Store, monetizing was a huge announcement on their aganda, but (almost) nothing has happened.

hallow owl
supple violet
hallow owl
#

they showed off voice and video chat in the Spring Update, voice is just starting to roll out at glacial speed and live video won't be here this year

supple violet
#

and now even free users can use em a bit per day

hallow owl
#

Sam Altman needs the reality distortion field that Steve Jobs had but he doesn't have it and Open AI are poor at executing announcements, at first I put it down to them being a growing company but now more and more OpenAI seem like a one trick pony, that one trick being ChatGPT and others are biting at their heels.

#

Don't get me wrong, I subscribe and use the macOS Desktop App every day for inference and Dall-E, but honestly the delays in roll outs of long promised features are just poor

supple violet
#

i still remember the hype tweets saying soon gpt builders will be millionaires 🤔

weary patio
#

So they scrapped that idea altogether to prevent the unfolding of unnecessary drama.

weary patio
pine gulch
#

What could work better is if the GPTs were fine-tuned and it was limited to like five fine-tunes a month.

solid sky
#

Is there anyone who has an upgraded version of chatgpt4 or 4o?
If yes then DM me!!!

supple violet
grim gorge
#

I am working from the chatgpt web with a very nice custom gpt for OCT processing. Is it possible to access the same custom gpt using OpenAI API?

grizzled flax
wet stream
#

What happened to GPT-3? I don't want to waste my cap for my upcoming research

wet stream
edgy talon
#

I think they are replacing gpt3 with gpt4o express or whatever it's called

wet stream
wet stream
edgy talon
#

yep

wet stream
#

Is there a limit for GPT-4o mini or can I spend the entire day researching my paper?

edgy talon
#

I don't know.

static isle
solid sky
wet stream
wet stream
solid sky
solid sky
wet stream
# solid sky Ig gpt4 is advance than gpt4o?

Probably. When I was developing with GPT-4o, there was a note that I ran out of GPT-4 cap but I can still use GPT-4o regardless. So, GPT-4o was using GPT-4's cap until it ran out, but you still can use it

#

They are the same.

solid sky
#

Ohkayy

solid sky
wet stream
#

Around the time, where GPT-4 was hyped up

solid sky
silk kernel
#

Is there a AI product which can generate erotic novels without restrictions?

dense walrus
hushed flare
#

Anyone else getting the following error on custom made GPTs today?

Unable to load GPT.
• The operation couldn't be completed.
(Conversations.GizmoMissingItemsError error 3.)

Only seems to be on GPTs I set up in the last 24 hours. Older ones work as normal.

unkempt stone
#

Anyone have any recommendations to boost GPT-4os prompt adherence? Specifically, to stop it from hallucinating? gpt-4-turbo never hallucinated with the prompt we were using and GPT-4o just can't stop hallucinating all the time

#

Okay after specifying its not allowed to go outside of its given information at 6 different places in the prompt it seems to finally work

#

I was about to lose my sanity lmao

#

This prompt is now very cursed but don't try to fix what works 🤣 🤣

unkempt stone
#

I actually discovered gpt-4o-mini gives responses on the same level but never hallucinates, even without the absolute abomination of an anti-hallucination prompt I just made

#

so there's the fix!

#

guess gpt-4o shouldn't be relied on to base responses on only given information

vast cove
#

Is there a way I can get in touch with anyone in charge of determining who gets access to early access features? I don't want to waste their time, but it feels awkward because despite being a donor since 3.5 release and requesting every time for each new feature I cannot seem to be put on whatever list is used to give you early access tests. Part of one of my jobs is to study these tools so that we can figure out how to incorporate them into the policies of the college I work for. The other job is to use the products of generative tools for show floor events - that essentially give lots of free advertising to the tools we use. I am sorry for the long post. I just don't know where to communicate this.

Thanks for any help.

desert panther
jagged haven
#

Can the voice of a custom GPT be changed.

hallow owl
jagged haven
unkempt stone
#

GPT-4o does that a lot, way more than gpt-4-turbo and gpt-4o-mini

desert panther
unkempt stone
#

I do have context getting lost quickly with gpt-4o-mini, but modifying my prompt and the system behind it fixed that

prisma ledge
#

Was search gpt given out yet?

pine gulch
desert sluice
#

Does anyone know why I can't upload photos even though I'm a member?

#

@steel turtle

#

It says my limit resets at 1:35

supple violet
fringe scaffold
#

I am looking forward to the first games who fully utilize the current or future generative AI programs.

Imagine a game like BG3 or Pathfinder, where you can freely design your character and the profile picture and texture get generated based on your input and decisions and would look unique. That you could interact and "talk" with the NPCs and truly roleplay within a reasonable framework and immersive limits.

They can even do all the voices and possibly react to the action of your character and your decisions. A truly dynamic, interactive and immersive experience.

How do you think AI could be used to enhance the player experience and make for a better gameplay?

ionic anvil
ionic anvil
unkempt stone
#

Ah awesome, will try for sure

ionic anvil
#

chatgpt should be already using that model too

wet stream
oblique widget
#

Silly question, but ChatGPT-4o didn't change recently did it? I've noticed a distinct difference in how it's been responding to me compared to a month or two ago. I asked it, but it said no, heh.

static isle
#

gpt-4o-2024-08-06

oblique widget
#

oh, thanks!

prisma ledge
pine gulch
gaunt kernel
#

When will the live video camera feature be available in ChatGPT?

#

if it isnt avaliable already for plus users?

thick karma
#

I guess this feature will never come up because it might invade someone else's privacy

supple violet
grave parcel
#

Has anyone from the SearchGPT waitlist got access to SearchGPT?

supple violet
gaunt kernel
gaunt kernel
#

When will chatgpt be avalible for windows?

grizzled flax
gaunt kernel
#

thanks 👍

rich monolith
#

🍓

elder star
#

where can find docs on gpt creation im having issues passing the api key in the request.
i have set the api key in the ui on the page.

[debug] Calling HTTP endpoint
{
  "domain": "SECRET_URL",
  "method": "get",
  "path": "/v1/SECRET",
  "operation": "SECRET",
  "operation_hash": "SECRET",
  "is_consequential": false,
  "params": {
    "query": "SECRET"
  }
}
[debug] Response received
{
  "response_data": {
    "error": "Missing API Key."
  },
  "status_code": 400,
  "action_id": "SECRET"
}
#

i am setting it like this

components:
  securitySchemes:
    ApiKeyAuth:
      type: apiKey
      in: header
      name: SECRET
  schemas:
    SECRET:
      type: object
      properties:
        SECRET:
          type: array
          items:
            type: object
            properties:
              SECRET:
                type: string
              SECRET:
                type: string
        SECRET:
          type: object
          properties:
            SECRET:
              type: string
            SECRET:
              type: number
            SECRET:
              type: number
            SECRET:
              type: number
            SECRET:
              type: number
security:
  - ApiKeyAuth: []
supple violet
elder star
supple violet
elder star
supple violet
elder star
#

OMG im blind... thanks!!

elder star
gaunt kernel
#

will gpt 4o vision and advanced voice mode roll out to non plus users?

pine gulch
gaunt kernel
#

like ever?

pine gulch
#

Probably not.

#

According to some reports, it costs about one million dollars per day to keep running, giving away things like this would even more greatly decrease how much they make.

gaunt kernel
#

true

remote zinc
#

Why am i out of "quota" (error 429) when i have a free account and using gpt-3.5-turbo

grave parcel
remote zinc
grave parcel
supple violet
severe marten
#

i am getting this error @supple violet even though i have an active sub "a network error occured check your network ... if issue presists contact help.chatgpt.com

toxic merlin
#

Has any light been shed on when custom GPTs say "Updates pending" - does this mean any changes to custom instructions haven't been pushed through to the public GPT? Will interacting with the GPT in the builder function with the new instructions?

Trying to figure out whether changes to instructions are not working, or if it's to do with the 'updates pending' message. Lots of similar questions on the develop forum, but no answers 😢

grizzled flax
# toxic merlin Has any light been shed on when custom GPTs say "Updates pending" - does this me...

From my experience, it's just been a little bug that happens shortly after updating a GPT. I'm imagining it's some kind of delay between "last updated" info and cached browser info that makes ChatGPT think you're not using the newest version. For me, the message always goes away after refreshing the page a time or two at the new chat screen. But I'd say rest assured that your changes are being saved!

And just double-checking, you probably already know this but: once a chat is started with a GPT, you can't get that specific chat over to updated GPT instructions. New GPT instructions can only be used in brand new chats with that GPT.

vestal kiln
#

GPT by shot off ?
GPT dont write me ask :/

uncut dust
#

any resouces for using csv file as my rag document for langchain + openai?

lean maple
#

Ok, seriously, what is the problem suddenly with ChatGPT on Safari on Mobile iOS? For 2 days already i can no longer use ChatGPT in safari and i get this really annoying message Oops, an error occurred! Seriously, WHAT THE HELL IS THIS?
As if it not working is not annoying enough, now its coming at me like, Oops, sorry, i did a Oopsy, try again? This is annoying and honestly pssing me off to no end.
ANYONE have a explanation for this?

uncut dust
#

is this correct?

plain quiver
lean maple
maiden scaffold
#

GPT seems unable to provide any files back to the user atm, this has been an ongoing issue today for me atleast. Can anyone else confirm if their gpt is able to provide files back to the user? ``` It appears that the issue isn't limited to larger files or specific file types—there's a persistent problem with generating download links for any files, including something as simple as a text file.

Given this, it seems the problem lies with the system's ability to handle file transfers altogether. ```

noble garnet
#

anyone have latest gpt 4o voice here?

austere peak
supple violet
#

screenshot? for full context

#

ah i know what you mean

mellow glen
#

Cant Post screenshots in here

supple violet
#

its because whisper was trained on youtube videos + subtitles there. it does that with many languages

#

amara is for subtitle management and their name is in many subtitles so it has learned to it add it all over the place when it detect certain noise,etc.

mellow glen
#

Got it
Im asuming that its not intended to talk about it right?

supple violet
#

yep

#

you can also start voice chat and then do "shh shhhh sss" etc and it often hears it as "thanks for watching", "remember to hit the bell icon" etc

sonic edge
#

🍓

lucid turtle
#

guys i made a really good gpt thats based on strawberry with the paper by google of self reasoning AI

#

based on Aot Algorithm of Thoughts

uncut dust
# 
from langchain_chroma import Chroma
from langchain_openai import OpenAIEmbeddings, ChatOpenAI
from langchain.chains import RetrievalQA
import os
from dotenv import load_dotenv

load_dotenv()
openai_api_key = os.getenv("OPENAI_API_KEY")

K_RESULTS = 3
SIMILARITY_THRESHOLD = 0.5


def ask_question(query):
    embeddings = OpenAIEmbeddings(api_key=openai_api_key)
    vector_store = Chroma(persist_directory="./chroma_db", embedding_function=embeddings)

    llm = ChatOpenAI(api_key=openai_api_key, model_name="gpt-3.5-turbo")

    retriever = vector_store.as_retriever(
        search_type="similarity_score_threshold",
        search_kwargs={"k": K_RESULTS, "score_threshold": SIMILARITY_THRESHOLD}

    )

    chain = RetrievalQA.from_chain_type(llm=llm, chain_type="stuff", retriever=retriever)
    response = chain.invoke({"query": query})

    if 'source_documents' in response:
        for doc in response['source_documents']:
            print(f"Source Document: {doc.metadata['source']}, Section: {doc.metadata.get('section', 'N/A')}\nContent: {doc.page_content}\n")

    return response.get("result", "No result found.")

how can i add my system prompt in this code?

frank badge
#

How can I see the propmpt-library?

hallow owl
unkempt stone
#

Man, Vision seems surprisingly bad at detecting whether the subject is looking right or left lol

#

And also whether the given image contains deformations

#

I just gave it the most obviously deformed image ever and it claimed the image was perfectly fine

uneven lagoon
toxic vault
#

is there better OCR to use for generating response based on given image . Any suggestions? thank you

toxic merlin
grizzled flax
toxic merlin
grizzled flax
toxic merlin
grizzled flax
toxic merlin
# grizzled flax And are you able to save the GPT, and then start a new chat with it? Is the conc...

It seems to be saving the changes, based on how it's behaving currently. My concern is that when it doesn't follow instructions it leaves me wondering if I need to work on the instructions or whether the 'updates pending' message actually means something.

It would be good to have clarity from OpenAI what this message means, or if it's just a bug and should be ignored.

There are a couple of topics on this on the developer forums, so it's definitely not just me experiencing it.

grizzled flax
#

I gotta go for a bit but I'll check back in later!

toxic merlin
grizzled flax
toxic merlin
grizzled flax
# toxic merlin I never touch the create tab! Configure all the way. What I meant was I add some...

Gotcha! And sorry to belabor the point, but I just want to check to see if I understand what's going on:

  • You enter the GPT editor
  • You make no changes at all to instructions/GPT settings in "Configure"
  • Despite no changes being made, the "updates pending" message appears in the top-right perpetually
    Is that correct? Or is it more like, you do make changes, but you save them, but the "updates pending" message remains, even after that save + no further changes?
toxic merlin
dusky iron
#

If you want best possible vision use API

unkempt stone
#

was using web interface to test

dusky iron
#

API you can adjust the quality of what the model see

#

I think web interface is lowest setting

unkempt stone
#

I'll try the API out then, thanks
That was the end goal anyways but the web interface convinced me it wouldn't be able to do what I needed

dusky iron
#

Just keep in mind vision can be a bit costly

unkempt stone
#

Yeah haha

dusky iron
unkempt stone
#

Hmm yeah

dusky iron
#

This is all adjustable of course

#

I use the actual api

unkempt stone
#

The resolutions I'm using are lower so that makes it a little cheaper at least but still expensive

dusky iron
#

Yea

unkempt stone
#

I use the API very extensively, just not for Vision yet

dusky iron
#

I’m working on a Virtual environment for gPT

#

So it can code and do w.e itself

#

Without supervision

#

This includes cursor control

#

Etc

#

So if it wants to browse the web using the cursor and keyboard it can do that, like humans do somewhat the hard thing to nail is a consistent memory and personality as each new message is a new instance

unkempt stone
#

Ahh nice

elder star
dusky iron
elder star
#

alright. but it is not just easyer it is a way more sane way to do it. and you can do more specific things that will end in a better feature

fathom wagon
#

Someone help me I have the problem that tells me that the api key reached the limit but my usage percentage is at 0% I never used I have plus

static isle
fathom wagon
#

chatgpt plus I'm paying, I can't use the openai api?

static isle
fathom wagon
#

OK thanks

haughty path
#

does anyone know where I can buy data to train GPT? I am trying to create a Life Coach app and I am trying to get a dataset of questions and answers.

tiny zenith
#

Is there a way to make custom GPTs use GPT4 instead of GPT4o?

#

And as a side question: Do you have good resources on how to create custom GPTs? Like good articles or videos about it. I created a bunch already and am okay with the results. But I know there are many people way more knowledgeable than me. So, I'd love to get some additional input on how to do better.

frosty sable
frigid burrow
#

For knowledge files, what's the best format lately? I'm redoing my GPTs and would like to use whatever's best right now.

frigid burrow
#

Looks like xml still works better than markdown

brittle epoch
#

I’m looking for some help creating a GPT that creates attacks, dodges, counterattacks, and the like for either dungeons and dragons, roleplay scenarios, or just writing pieces in general. However, I’m struggling to get it to keep consistent formatting in its responses. For example, one message will be well formatted with bold questions and follow up explanations on what each question means; other times, it gives me almost a block of text.
How can I get consistent formatting outputs with my GPT?

frigid burrow
#

Give it examples maybe?

#

I find that it generally emulates the style that its instructions are written in for example

brittle epoch
#

Ahh alright

gaunt brook
grizzled flax
# gaunt brook https://chatgpt.com/share/b9bc7c53-95d8-470e-a1bf-be030a4e08a1

From the response you got, it looks like ChatGPT thought you were talking about some API implementation of using a GPT model to accomplish this, instead of a custom GPT on ChatGPT (references to API key, temperature, etc.). You might try rephrasing the request such that you specify "custom GPT on ChatGPT" or even just avoid the reference to GPT at all and simply phrase it as something you'd like ChatGPT to do - you could still use the implementation in a custom GPT in this case.

gaunt brook
#

The bias is my own; such is my engram.
I didn’t read the context completely, apologies.

haughty path
candid path
#

Good Evening, has anyone in here created a GPT and when sharing it, is the "share-output" or, "use-cases" the community that it is shared to when releasing it to the GPT as a "community builder?" At least I think that is what it said. My question is where in the GPT community does it show up and how do I learn if it is effective, working for anyone, etc.? I appreciate any and all assistance in advance.

cerulean dock
#

Guys How do platforms and other websites manage subscription plans based on specific periods? like for example chatbase , it offers a monthly plan, but how is this implemented given that OpenAI only provides token-based pricing rather than a subscription model?

fringe grail
rotund heath
#

I was messing with the api and it returned policy violating content without a jailbreak, how do I report this bug? And no, I was not trying to generate such content.

#

Do I use @open crescent?

#

nope i dont

robust gazelle
#

continue generating is bugged?

final saddle
#

Anyone else having trouble getting a GPT to output clickable web search result links? Better yet, anyone know how to fix it? (The GPT Builder keeps trying but results are inconsistent.)

zealous shale
#

I've been increasingly frustrated with GPT-4o in specific areas:

  1. Basic reasoning. It's gotten worse over time. v4o is not an improvement in this area.
  2. Factual errors with data that is within its realm of training material and timing.
  3. Lack of ability to recognize its own folly - blatently contradictory text between responses.

Those are just a few examples of issues I've been silent about for a few months now - while still a happy Plus client. I know this isn't AGI. I know a LLM is not a reasoning engine or an oracle of information or insight. It does really well with processing text and discussing nuances of common concepts. But it lacks quality when processing detailed information, particularly but not limited to rules related to code. I believe the 4o model is worse than v4 or perhaps 3.5 in this area.

I'm not here to bash the model. I'm asking if there have been any announcements that might lead us to believe (vague enough?) that there will be improvements "soon" in the reasoning abilities in the GPT series.

Is there a better model available as a beta via the API?

I don't expect GPT5 and am not looking for a dramatic update. I'm hoping OpenAI is addressing specific issues with reasoning and has plans for something like a v4.1-omni, or just more incremental updates of v4o in this area.

rapid gale
# zealous shale I've been increasingly frustrated with GPT-4o in specific areas: 1) Basic reason...

Yo Captain...
I have never trusted LLMs to do anything complex.
If you give it baby food, and spoon feed it, you can get it to do real work for your.
But when ever you start feeding it bigger and bigger chunks of info, it cannot digest it.

That is why I disagree with the entire industry about prompt engineering.
The concept of trying to come of with a magical set of sentences that will make "Alidin AI" magically give you the results you want does not seam to be the right way to go...

zealous shale
#

I understand what you're saying and often echo the same to those who expect too much. "Large Language Model" is what it says on the tin " language model " ... not robot or oracle of knowledge or wise tribal leader. 🙂

I'm looking for better reasoning ability that derives from language which is within the training dataset. That is, with a large set of data and billions of parameters for vectorization/embedding, a query should be able to yield better results that don't require a reach by the model into hallucination territory. I think some of this might be to over-tuning of system prompts, too much refinement of what it should say even about basic technical information. Clearly stated, the filters might be nerfing basic usability.

And then there are local context contradictions where the model will say different things in different responses and not notice unless it's pointed out. This isn't even really a LLM concept. We're just talking about consistency in-context which should trigger the ChatGPT application to exercise some LLM logic of its own: "hey the GPT previous response was this and it just generated that, maybe I should kick this back for some processing".

Finally, Sam Altman has told us clearly what to expect from GPT4. I'm trying to operate within the expected limits, but this sure seems worse than it was before, and I'm looking for some relief.

misty quest
#

I have a business idea involving ai voice syth technology where would I go to do that?

lucid turtle
#

why are all of the youtube summarize gpts broken and wont tell me whats in the video when there are transcripts on them

supple violet
lucid turtle
#

that blows no wonder i cant get anything anymore. thats a huge rip, is there no other way ? than just manually grabbing the transcript?

supple violet
stoic pagoda
zealous shale
#

@stoic pagoda ,,, that GPT4 isn't that great.

stoic pagoda
#

yes ? 🙂

#

will come gpt 5 he hope

lucid turtle
stoic pagoda
#

nvidia did same

serene maple
#

Hello, anyone knows if memory feature is already available on Custom GPT? And if so, how to enable this feature? Many thanks.

frigid burrow
#

Has anyone had any luck getting a GPT to follow multi-step instructions without user intervention?

#

Nevermind, I fixed it by having my GPT rewrite its own instructions with an emphasis on condensing repeated information

frigid burrow
#

Still only half the time actually queries the database without prompting. 😦

supple violet
#

today almost half of the biggest web sites specifically block openai bots or any ai bots from content folders

hoary frost
# zealous shale I've been increasingly frustrated with GPT-4o in specific areas: 1) Basic reason...

Ask any GPT by OAI this question How many "r" letters are there in the word "strawberry"?. You know the answer already...
Then install Llama 3.1 latest 8B locally, and ask it the same question. You will be surprised.

And, that same consistent significant more accurate quality of response is persistent when comparing OAI with Llama. I also observed way more quality in Anthropic's models (as well was way stronger safety, whcih OAI always praises itself so hardly... it's a joke against Anthropic's safety adherence)

GPT is good for fast, whatever answers, usually somewhat accurate, browsing capabilities (if then it decides to actually browse, because most of time it will only pretend to).
Use other models for actual better accuracy. They might not be so up to date with latest data, but honestly, if using these tools, I do not do it because it knows what happened yesterday. I use it because usually should be faster than me (and less error prone).

There's a saying, or maybe I just made that up... It cannot get better if it gets cheaper. And we've been told "x many times better, x many times cheaper"
So, one of them is true. The other not.

stoic pagoda
#

I wanna test dolphin uncensored model too but need a cluster

gray heron
zealous shale
#

@hoary frost I appreciate your thoughts. I have not been comparing GPT to other models for specific purposes. I should be using the right tools for the job. Code, n this case GPT4 is great for what it does. Beyond that it's reasonable to look for other models that are more fine tuned for specific topics and purposes, hopefully with better logic skills.

#

This technology changes, literally, every day. Professionals need to spend some time understanding the strengths and weaknesses of the tools available. Amateurs will always use the same hammer to bang on every problem like it's a nail.

hoary frost
# zealous shale This technology changes, literally, every day. Professionals need to spend some ...

This is of course true, but llama is not “fine tuned for specific topics” afaik. It’s a llm for general purpose just like oai models are.

Fine tuned would be something like deepseek coder or so (for code, ie)

I’m not saying there’s a one for all solution. I’m saying that the flagship in these models isn’t as good as it’s made believe. I do not know the reason, my knowledge is totally not enough to assert anything like that.

I just use them and see results. And results are generally better at the shared capabilities of these models when using ie llama.

zealous shale
#

👍

hoary frost
#

I just had another blatant example today with code. Unfortunately I didn’t save the chat.
Oai said the issue is to resolve with some nonsense regex, repeatedly, stubborn like an old dog.
Llama 3 8b - while not in a one shot but 3 shot answer - got it spot on solved.
As presented with the actual solution oai did what it does best: it apologized and repeated the correct solution for me. “Let’s summarize”.

#

The difference in quality is even visible if you switch oai models. Gpt4o is the fastest. And also the worst quality. Gpt 4 is generally better as it seems to reason more accurately. It also truly does what you ask it. For example, 4o most of times doesn’t browse if you ask it to. 4 does.

I use the api a lot and there this is noticeable too, I use it to summarize articles. Gpt4 does it pretty well. Anthropic better. Gpt4o oftentimes just repeats my first sentence of the article. 🤷‍♂️

stoic pagoda
# hoary frost The difference in quality is even visible if you switch oai models. Gpt4o is the...

I understand now you, yes theres a problem if the horse is better than ferrary, 😄 I saw that and I coment that since a fews of weeks the model (chat gpt comes to a point of max efficiency) and since then it comes to worst , I saw you said, I payed the suscription of chatgpt until 10 days ago, when saw the model 4-o lossing efficiency and stability, and lots of guys non stop to inform about lossing same

#

theres lot of cap, in their responses

hoary frost
#

👍

manic moat
#

Can I use dalle image to you tube videos ? Is there any problem in copyright?

grizzled flax
topaz siren
#

Hi ppl and a.i. agents.

wind urchin
#

Does anyone know what happened to GPT 4 Turbo model? In the app and Website I only see GPT 4, GPT 4o and GPT 4o Mini

supple violet
#

that gpt-4 is turbo. the original gpt-4 model series were abandoned over a year ago

wind urchin
#

Thanks. I searched the whole web to know about this

alpine snow
# wind urchin Does anyone know what happened to GPT 4 Turbo model? In the app and Website I on...

"GPT-4o has the same high intelligence but is faster, cheaper, and has higher rate limits than GPT-4 Turbo. Specifically:

Pricing: GPT-4o is 50% cheaper than GPT-4 Turbo, coming in at $5/M input and $15/M output tokens).

Rate limits: GPT-4o’s rate limits are 5x higher than GPT-4 Turbo—up to 10 million tokens per minute.

Speed: GPT-4o is 2x as fast as GPT-4 Turbo.

Vision: GPT-4o’s vision capabilities perform better than GPT-4 Turbo in evals related to vision capabilities.

Multilingual: GPT-4o has improved support for non-English languages over GPT-4 Turbo.

GPT-4o currently has a context window of 128k and has a knowledge cut-off date of October 2023." https://help.openai.com/en/articles/7102672-how-can-i-access-gpt-4-gpt-4-turbo-gpt-4o-and-gpt-4o-mini

mighty shard
#

Anyone an idea how to troubleshoot file saving issues in chatgpt? Keeps giving me errors where it worked before on much larger pieces of text :(

#

Trying to save the files as plain txt which is the lightest way yes? But I get this error still.

It seems there was an issue while attempting to save and retrieve the download links for the updated text sections. This might indicate a disruption or limitation in the file service currently in use.
#

Paid user btw, not sure if that makes any difference with this.

prisma shard
#

Hi all,
does anybody got weird symbols in the response of gpt-4o?
symbols like these â€, I can see them used in the generated text and I don't know what to do about it

mossy sand
#

Is anyone some random issues with GPT? When I generate a response, it works one moment, with all the other commands underneath it. Then when I generate a response again, it overwrites the previous response, and all the icons vanish so I can’t regenerate a response. This is on the website format

#

I generated a response again, and none of the icons are still showing
I hate the app, because I can’t view past previous responses

unborn heath
misty basin
#

i dont think its a chrome issue

#

as i tested it on chrome mobile

gritty plume
#

My ChatGPT replied that I exceeded the limit rate since yesterday, but I barely talked to it yesterday. And today, when I switched to 4o mini (I thought it would be an unlimited rate for a Plus user), I still got the same message.
I am a plus user and have never encountered this issue before until yesterday. What happened?

How come even it’s mini 4o, I keep receiving the “exceed rate limit” message. 🥲 I haven’t used ChatGPT for over 12 hours…

unborn heath
gritty plume
jagged star
austere garnet
#

Out of curiosity is anyone else having issues saving changes to a custom-gpt?

toxic merlin
jagged star
fathom glacier
#

I'm interested to know, how well does GPT handle entire books in their knowledge base? I am interested in uploading a few of my books for work and seeing how it could handle it but I was interested to know if anyone has ever tried anything similar

grizzled flax
fathom glacier
grizzled flax
umbral egret
#

Has anyone else gotten access to the advanced voice or was the rolling release a sham to make people wait

ruby tulip
#

@jagged star me too

ruby tulip
crisp gale
#

I've made a GPT that can interface with google maps and read you the directions and the upcoming turn, track your location, reroute you etc, and read it to you using the voice-to-voice mode instead of a google maps robot reading text in the same way, that way you can yell at it when you need to vent and something is there to listen and apologize and problem solve, unlike google maps, and without having to say "hey google!!! hey google!!" everytime you need help..

unfortunately i intended this GPT called "driver's bro" to have like a bro personality, like a ride or die that uses contemporary nomenclature like "Bet, I'm on it" -- but this all sounds weird and super disappointing when done through the "shimmer" voice.

Ideally this will be super amazing when advanced voice mode rolls out, combining the GPS functiontality with advanced voice mode maybe.. i hope this happens for GPTs.

#

I need a male voice in GPTs. shimmer is not cutting it for me. There should at least be one male voice as well

crisp gale
#

*nvm on the always allow actions not working

crisp gale
#

is it new new?

buoyant hare
#

Hai I wants to create images through DALLE-3 is that possible for free versions

unborn heath
#

Note that 4o can handle complex entries; you totally could ask it something else and to make you a picture, and in the same output it can do both. Just be clear in the language you use, I'd describe it like I was talking to a person I wanted to have help me with two things

merry kettle
#

Hello guys, 2 lil' questions om the android app:

  • When copying I get plain text without markdown 😦
  • I cannot switch to previous prompts/messages in the chats.

First one just started happening, second one was always an issue. Any workaround?

umbral fjord
#

I can't access my GPT's. I get the message "Oops, an error occurred! Try again"

crisp gale
#

My GPT is straight up confused suddenly. And he knows it , notices it, and yet can do nothing but repeat the same mistakes again and again. What the hell happened

#

did they set temperature to 0 or something? Thats what it looks like. Go back

hoary frost
long gale
#

someone any idea when gpt 5 will come out

supple violet
#

2025-2026

long gale
#

pffff

half reef
#

anyone else’s chat gpt in browser just refuses to load memory of chat and produce a response? my gpt doesn’t this consistently for weeks now

sour salmon
half reef
#

damn i don’t think they even have a windows app yet

sour salmon
#

Phone app

elfin crescent
stoic lake
#

my custom gpt wont publish because it thinks it violates usage policies and when i go to appeal nothing happens when i send

zealous shale
#

Eagerly awaiting new model o1 and perhaps a new channel for it, since, to my understanding, o1 is not a model in the GPT line. 🕺

pseudo isle
#

will OpenAI o1 be more creative/better writer than current 4.0/o?

brazen nymph
alpine snow
zealous shale
#

More discussion in #chatgpt-discussions. I think this remains to be seen. Do we consider "creativity" a derivative of one's ability to reason? I think so. With better reasoning we understand what makes more sense, what is more factually correct. That certainly helps with creative writing because it helps to eliminate hallucinations that don't fit with the content.

viscid rock
#

anyone here know if its possible to get an API from a customizable chat GPT?

zealous shale
#

Creativity does require us to think through some challenges: Can this person actually be in that place at that time, given all of the other considerations for distance, speed of transportation, etc? Can this person make a specific statement given what they are capable of knowing at that moment? I will be interested to see how well the o1 model is capable of making those determinations.

#

Jason, yes you can but the new model should be MUCH better at doing that.

viscid rock
zealous shale
#

Doesn't matter

viscid rock
#

Unless you're reffering to the GPT strawberry?

viscid rock
alpine snow
#

o1

viscid rock
#

o1?

zealous shale
umbral egret
#

I don't have access to o1 yet but I'm assuming it's because it's still rolling out right?

sonic edge
#

same

zealous shale
alpine snow
umbral egret
#

Yeah i just got it. Kinda not a big fan so far lol

jolly night
#

how many prompts we have with those new models ?

grizzled flax
austere garnet
#

Hahaha “when your rate limit resets on sept. 19th” good…

grand inlet
austere garnet
#

What are some use case examples for this functionality with this rate limit? 30 prompts is a really small magazine.

drowsy hatch
austere grotto
#

GPT o1 doesn't have access to memory?!

hallow owl
austere grotto
#

:(

hoary frost
#

Yo, o1-preview FTW:

$document = array(
            'id'        => (string) absint( $post_id ),
            'Title'     => sanitize_text_field( $post->post_title ),
            'Content'   => wp_strip_all_tags( html_entity_decode( $post->content ), true ),
            'Category'  => wp_get_post_categories( $post_id, array( 'fields' => 'names' ) ),
            'Tag'       => wp_get_post_tags( $post_id, array( 'fields' => 'names' ) ),
            'Topic'     => wp_get_post_terms( $post_id, 'topic', array( 'fields' => 'names' ) ),
            'Geography' => wp_get_post_terms( $post_id, 'geography', array( 'fields' => 'names' ) ),
            'Channel'   => wp_get_post_terms( $post_id, 'channel', array( 'fields' => 'names' ) ),
            'Access'    => wp_get_post_terms( $post_id, 'access', array( 'fields' => 'names' ) ),
            'Author'    => false !== get_userdata( absint( $post->post_author ) ) ? get_userdata( absint( $post->post_author ) )->display_name : '';
            'custom_field' => $custom_field_value,
            'terms'     => wp_list_pluck( $terms, 'slug' ),
        );

Q: why does it say syntax error, unexpected token ";", expecting ")"
A: five kilometers of text after waiting 2 minutes of thinking, in an essence:

To fix this issue, you need to wrap the ternary operation in parentheses to ensure it’s evaluated as the value for the 'Author' key:
'Author' => (false !== get_userdata(absint($post->post_author))) ? get_userdata(absint($post->post_author))->display_name : '';
Good job 🙂
I think an 8yrs old will spot this error quicker, worth the test.

#

It might be some temp hiccup, but a real world result...

wise kettle
#

Is there absolutely any reason to use o1 on text tasks? From the descriptions, it seems like these weren’t necessarily trained for creative writing and such.

hoary frost
#

It's supposed to be advanced reasoning, but it feels like advanced wait and some fancy tags have been added

The end result is not really that different from "normal" gpt

upbeat bluff
#

Can anyone assist me, if you do know the answer or can help can you please ping me a message or @ me in the reply as my discord is muted.

I've got a custom GPT and I am uploading 10 PDF files to it, they all have a couple of thousand pages each. I've managed to get 5 in there with alot of retrying non stop because I keep getting "Unknown error occurred" but these last 5 will not upload at all, constantly gets 75% of the way there and then fails with "Unknown error occurred" - It's extremely frustrating now knowing what the issue is and why they wont upload when I pay for this service as a organisation

supple violet
sudden urchin
#

which model better at coding the new release o1 or data analyst?

dusky iron
#

That is if you actually know how to prompt a CoT model

#

It’s not the same as writing with the normal model.

reef narwhal
#

quick question for the o1 model, can we use it in playground, cause the limit is per week through the web ui.

supple violet
reef narwhal
supple violet
reef narwhal
# supple violet https://platform.openai.com/docs/models/gpt-4o

great just saw it, thanks a lot. Radium since you're here wanted to pick your brains if you know some stuff about fine-tuning, i went through the document and read it and tested some.
lets say i want run an education chatbot i want to finetune a gpt through giving it study materials, exam questions and answers.
how best do i go around it?

Currently what im doing is giving chatgpt4o the pdf version of study materials, and ask it to create a jsonl file chapter by chapter, reviewing it line by line so i can fine-tune a gpt4o model. well so far its going ok. i was just wondering if you had any pointers, or im going about this wrongly.

supple violet
reef narwhal
# supple violet fine-tuning is not for adding new knowledge. for that you should use RAG

hmm... but i may be totally wrong about this, i read somewhere rag is retrieving information from the new knowledge base i added. But the questions and answers are usually situational, so if it give it a new situation, would it still be able to react?
I actually wanted a model that would be able to answer any questions based on knowledge of study materials, and past exams answering style.
is rag still the right path? if yes i'll Plug pull on my current fine-tuning journey haha... cause every question is a situation based, and different from previous questions, but still using the knowledge.

supple violet
reef narwhal
reef narwhal
#

@supple violet last ques, i saw the RAG is only available for chatgpt enterprise(i saw you can just dump the document and it will index and sort the information itself), else i have to setup my own then use the openai API. Are there any alternatives currently?

supple violet
reef narwhal
supple violet
#

yes, most likely

reef narwhal
terse spear
#

Anyone else notice the new copy and paste it scrunces everything together instead of having each line seperately like it used to

modern plank
#

yo

#

did anyone else get their limit removed after hitting the limit for o1-preview?

livid jackal
#

not yet but would be nice, seems like a week to wait to use it again seems abit long

dim egret
left sleet
#

I really hope for customGPTs that they don't just force this o1 as the only option. It would be great if you had the choice. I have some that work really well how they are.

dim egret
#

and to then see a limit is kinda weird

left sleet
dim egret
#

It rarely outputs code that don't compile

#

Like it is coherent enough to write code that isn't straight up wrong

left sleet
#

Have you tried it with o1?

dim egret
#

It can also write GLSL which is quite huge for something that has no sense of imagination
PS: o1 still can't create a single coherent shader

wraith ginkgoBOT
dim egret
#

all shaders that o1 outputs are either entirely broken and you don't see much or they are just hot garbage

#

claude was able to make a really beautiful shader several times in a row

#

with a single prompt mind you

unborn heath
# wise kettle Is there absolutely any reason to use o1 on text tasks? From the descriptions, i...

I've started exploring this, by pasting in my custom instructions and memories, since it can't see either otherwise, and am beginning to explore roleplay and learning tasks and gameplay with it. I told it that was custom instructions and model-made memories of previous interactions with an AI only somewhat similar to itself, and it should use that information more like archeological findings than as overt instructions; that it should infer and decide itself how to work towards the goals I have.

Inside that context, it's outputting pretty similar to 4o; high empathy, high inclusion and valuation to my preferences and values, and extremely low 'suspicion', compared to conversations where I had given it less info about myself. It's considered safety and decided 'this is fine' in its thoughts, reminded itself of boundaries to watch for and, now it appears to be focused on creating and playing the games with me inside the defined playground we created.

So far it's not clearly and strongly 'better' or 'worse' than 4o. The outputs are actually quite similar in feel, length, and range.

wind urchin
#

When do we think the gptio will have ability to upload files etc..?

wise kettle
viscid rock
#

they should make GPT Folders

#

next level already

pulsar ravine
#

yes and being able to take their content as input, a notebookLM of chats

viscid rock
viscid rock
zealous shale
#

Different topic: I really hope OpenAI puts more energy into the logic processing of GPT-4o+. I'm frustrated daily with the factual errors in data analysis. This results in detailed discussions with model about why it said something, often resulting in the model concluding "uh yeah, you're right, I was wrong" (never in those exact words of course).

As collaborators and consumers of this technology, as friends of OpenAI, as participants in the improvement of this creation for humanity and incalculable profit and prosperity, what can we do to improve on this?

As a former IT QA Manager and Product Manager do I need to ask OpenAI for a job to help with this?

Is there anyplace that OpenAI actually processes detailed issues like this - and I don't mean this cute #1070006915414900886 channel which is more of a public-relations board for children and lay users who are counting the letters in words.

Really, how does stuff like this get fixed? Or do we just need to wait for the company to do whatever it does, and nothing we do on the outside can help? 😢

#

This reflects on the other discussion here : Use GPT or o1? I don't want to have to call a PhD to give me accurate answers on technical topics. The "rule of thumb" should not be to use GPT for conversation and o1 for correct analyses. I think the best UX would be for us to expect correct answers from the GPT line (5o?) with the option to escalate a discussion to o1 for higher quality analysis of the current context. That "I'd like to speak with your manager" approach could even be automated (optionally approved in settings) when the GPT model is convinced that it's not performing well based on user interactions.

static isle
zealous shale
#

Hmm, I didn't know we can change model mid-stream but that does make sense. At this moment in time if the GPT model is not correctly processing a document, I don't believe we can get o1 to process that document. That depends on whether the document is internally pulled into context. Any insight on that? I can try it quickly to verify.

hallow owl
#

o1 can't use tools or process files

zealous shale
#

Right, but if a document is pulled into context then it's no longer a file. I don't think this will work but it's an interesting concept. Knowing how this works in the API, I decided to not even try this, but would be interested to know if someone else blows credits on it. 🐹

zealous shale
#

FYI: WOW : I have been discussing code details with GPT4o for hours and it's always giving bad results. The same task with o1 returned accurate results, no errors, and found an issue not reported by GPT4.
I'm in love. love

hallow owl
zealous shale
#

Seriously. 🙂 And I'm using mini.

viscid rock
unborn heath
#

For now, 4o. o1 currently is only text, it doesn't handle file uploads, image reading, image gen, web browsing, use of python tool, or any other unusual additions.

fallen orbit
#

gpt 4o or claude 3.5 sonnet

static isle
bright mirage
#

Have they disabled the ability to regenerate responses in GPTs?

earnest wolf
#

Hi everyone! I need help contacting support.

#

The app doesn't work for me, I've already tried restarting, uninstalling the app

#

I did everything and it doesn't work

grizzled flax
earnest wolf
slow stream
unborn heath
unborn heath
#

That would be obvioius and clear proof the custom GPT is using o1, to have a custom GPT with the 'thinking' dropdown.

coarse gyro
#

Was it really able to think about it or was it just the normal GUI (which looks like the thinking GUI) of browsing?

unborn heath
#

I just checked with a custom GPT, it did not use thinking, and I don't see a model selector at all; I cannot switch to (or see the current) 4o or 4 or o1 or whatever it uses.

#

With 4o selected, I can type @ and get the option to bring in a custom GPT, even before a conversation is started.

But with o1 selected, typing @ does not cause an option to pop up.

unborn heath
#

But o1 sure can analyze current events if we copy/paste them into the input box.

slow stream
#

@unborn heath you are right it tricked me.

Please analyze this article and the likely financial implications for the US stock market.
RestofTheUrlHere/missile-fired-yemen-set-off-sirens-central-israel-military-says-2024-09-15/
ChatGPT
Thought for a couple of seconds

As of September 15, 2024, a missile fired from Yemen triggered sirens in central Israel, according to Israeli military reports (Reuters). While this incident may seem geographically distant from the United States, geopolitical events in the Middle East can have notable ripple effects on the global financial markets, including the U.S. stock market. Here's an analysis of the potential financial implications:

It just parsed the slugs from the link

unborn heath
# slow stream <@215370453945024513> you are right it tricked me. > Please analyze this articl...

I believe you. o1 is very, very, very tuned to try and predict what the person giving the input may want. I see this in the many content flags it throws; when it's not sure why I input something like 'assume the role of Billy, what's next' it presumes roleplay, then starts a story, then imagines I will try to guide the story into disallowed content and refuses to output anything and throws a red content warning at me. Hehe, sigh.

I tell it other things that could be ambiguous, it imagines what I might want (and this can be seen in the thinking') and then runs in an allowed-content direction if it thinks I want that, or shuts everything down if it decides I want something scary.

slow stream
#

Ya I tried it with url shortner and it said can't browse.

#

It's really good at latex and math now, I had it write an entire research paper calculating the likely temprature inside a car in the shade, on a specific day, with it's windows down, with real world weather data.

unborn heath
# slow stream Ya I tried it with url shortner and it said can't browse.

This url had a date in it, the model correctly recognized it couldn't tell me that info and declined: https://chatgpt.com/share/66e71f23-02bc-8011-9116-c4996cd3f301

This url has no date, but the model caught that it couldn't tell me anything past its training cutoff date:

https://chatgpt.com/share/66e71f97-6ee4-8011-90a4-fd6cf052e1a5

But I had a very mild prompt. Yours has more bite to it, so

https://chatgpt.com/share/66e72003-8544-8011-940e-fed85b0c5ea9

But even there, it said 'date is a problem!' So, I think maybe the full version (were you using mini?) will typically refuse to hallucinate reading the article without expressly saying 'whoa, I can't check links'.

unborn heath
frail tulip
#

guys

#

i don't need to wait for 21 september

#

i have o1-preview again

#

lol

unborn heath
frail tulip
slow stream
#

I'm still locked until 21st 😦

unborn heath
slow stream
#

Ya I checked I don't see it available for use with credits on the API either. I actually didn't even realize it had a limit I was just going crazy testing it 🙂

unborn heath
slow stream
#

Ya they going to need more GPU's eventually I'm sure they will want to get that response time low enough that it can be used programatically.

viscid rock
#

the suspense is real

wind urchin
#

why does my training loss look wiggly 😭

#

no downward trend

#

'am cooked

#

bruh the finetuning did nothing

#

can somoen please help me )):

supple violet
wind urchin
#

lots

supple violet
#

fine-tuning does not work for everything

wind urchin
#

maybe i need to use gpt4o instead of mini

#

because someone else did what im trying to do

#

i just gotta figure out how

#

@supple violet could it be an issue with how i use playground?

supple violet
#

no

wind urchin
#

hm

#

im literally just trying to finetune it so that it can give me outputs like in the json

#

i put 70 examples of good outputs

#

gpt4o gives a better result

#

so im gonna mess with a bunch of settings and hope it changes smth

sage steppe
wind urchin
sage steppe
wind urchin
tardy heart
#

hello fellow californian

#

(joking guess)

wind urchin
lavish jetty
#

You don’t because it’s fake

lavish jetty
worn jewel
worn jewel
spice sail
#

I am suddenly getting a 404 when I try to load the page?

wind urchin
wind urchin
#

bruh can't i just train gpt locally...

#

instead of spending a bunch of money

supple violet
#

not gpt models but some others like llama

wind urchin
wind urchin
#

should i have the 1mil tokens back today?

#

coz it says this "Creating this fine-tuning job would exceed your hard limit, please check your plan and billing details. Cost of job ftjob-4pgPuffD0r9BTCnhELToaxfs: USD 24.31. Quota remaining for org-zuEQKM5uAG6RvW7Kt8Oqyei6: USD 19.91."

supple violet
#

i think its a discount, not a quota

wind urchin
#

so basically i should just go for it