#chatgpt-discussions

1 messages · Page 73 of 1

mortal salmon
obtuse temple
mortal salmon
obtuse temple
#

I'm building a limits-pushing app and I need to aim not to overshoot but to land within the token limit.
It's like BlackJack lol
Aim for 21 but don't go over

velvet canyon
#

ChatGPT is down?

velvet canyon
#

All good... problem solved... couldn't log in due to an issue with the ad plugin

lunar atlas
spring iris
#

microsoft ships with edge and uses bing

lunar atlas
#

Okay and who use Edge?

spring iris
#

read what i just said

#

by the way edge is just a chrome reskin so its not that much different than whatever junk you use

violet bloom
#

now available in Web + Mobile 🙂

raven portal
#

i'm in boys

#

tell me what to test if you want

#

anyone else on the new turbo now?

#

just test it by asking for its knowledge cutoff like i did in my link

visual bloom
#

I just asked what is the knowledge cut off date. To me it answered December 2023. So they are no rolling out the update.

raven portal
#

ok it seems like everyone should be on the latest turbo version now

raven portal
visual bloom
#

What typo?

#

Oh! No should be now. I see it.

#

With initial testing the current model is an update. The writing is a bit better and follows instructions better. Also produces longer text more easily.

pliant topaz
raven portal
# pliant topaz How do you know?

if you click the link you will see that the knowledge cutoff is dec 23 now, it always was april 23 before today
well and also the official oai announcement

vital dust
#

What’s with checking system health?

cobalt idol
gleaming grove
#

Last night mine was saying April 2024 cutoff and now it’s saying December 2023. At a point I was getting an error saying something has gone wrong so maybe they rolled it back?

raven portal
gleaming grove
#

Possibly. I asked it a couple times in different instances and it was still saying that, other members were also reporting that. Maybe it was hallucinating for everyone?

honest jay
raven portal
gleaming grove
#

Glad everything is working properly

cobalt idol
#

in api, the 4 turbos will just give you all kinds of dates

#

but this is literally the only thing we have to notice some model change in chatgpt (must be in system then I guess, since in chatgpt its much more consistent on the date)
until they actually start telling us the chatgpt models, we just have to guess like this I guess

hard dock
#

the new gpt 4 turbo released to chat gpt have 128k context, right?

worthy nimbus
honest jay
vivid burrow
#

你好

eternal spoke
#

i'm starting an internship for full stack will i do okay with gpt on my side?

onyx eagle
eternal spoke
#

like i know the basics of c# html javascript and sql

#

but if gpt helps me out that's not a bad sign to the employer is it

#

if anything it's a useful stragedy

primal warren
#

What options do I have to access memory features ?
... I want to get it as soon as possible
is there anything I can do ?
I'm from spain

dense halo
honest jay
cobalt idol
native spear
native spear
#

USA and India work for me

cobalt idol
native spear
#

maybe most of them were from the eu lol

#

personally i had it show up as a notification within my chatgpt app on android, but i was unable to access it since im from the EU, but when i changed to another country its working just fine

#

its just the EU regulations thingy

feral zinc
cobalt idol
#

I have the same impression though, a lot of things are later for EU, and mostly when its a feature that might have privacy concerns (like collecting data)

feral zinc
#

only few things have been and openai have always mentioned it. there are already users from eu with some of the new stuff, including the new model

#

its a rollout and conspiricy theories do not help

near vine
#

what are you guys knowledge cutoffs? mine is still stuck in Dec 2023

#

isnt it supposed to be April 2024?

elfin birch
feral zinc
near vine
#

RIght?? but noo, it supposed ot be April 2024 the new model that they released

feral zinc
#

training data does not mean the same as model release date

near vine
#

no the thing is, the previous version model was the same

feral zinc
#

data is gathered, then filtered, cleaned and processed for training and then trained. this can take weeks or months

elfin birch
near vine
feral zinc
elfin birch
#

Sometimes the model hallucinates and gives an incorrect date; this is especially the case with the previous ChatGPT iteration since its cut-off was April 2023 and the current date is in April 2024, which are both included in the system message

near vine
near vine
feral zinc
elfin birch
near vine
near vine
cobalt idol
# near vine no the thing is, the previous version model was the same

for me, for good part of this year it was Dec 2023 (so probably gpt-4-0125-preview)
then maybe 3-4 weeks ago it started saying April 2023 (so back to gpt-4-1106-preview ??)
and now since today its Dec 2023 (so should be gpt-4-turbo-2024-04-09)
no such thing like model with April 2024

but chatGPT will gladly hallucinate April 2024

even yesterday I asked and it gave me 2 responses to choose from to provide feedback on which one is better
and it was either April 2024 or April 2023 xD

elfin birch
#

I got a response some time ago like that too 💀

near vine
#

okay then, whats the easiest way to test the model to know if we got the updated one or we still stuck in the previous one?

cobalt idol
dense halo
#

hey guys is "gpt 4" in the website gpt 4 turbo?

cobalt idol
#

here we go again

cobalt idol
near vine
#

people being so glad on reddit that there cutoff is on april 2024 🤣 and saying because the background browsing is working

near vine
cobalt idol
#

didnt use it enough to tell yet

elfin birch
dense halo
cobalt idol
#

yet, both are a result of not showing the model to the user
creating conspiracy theories, just like Radium didnt like 😄
because openai is not telling the info properly

feral zinc
native spear
feral zinc
elfin birch
native spear
feral zinc
native spear
#

obviously its a very sensitive feature to be able to store memories of everything you talk about. And OBVIOUSLY a feature with such sensitive topic will get delayed in the countries where the company is more likely to get in trouble with

#

this has been the case with everything, ranging from claude and gemini to custom instructions of chatgpt

#

gemini was delayed in the EU for that exact reason

cobalt idol
#

just state it as an opinion, not as a fact 😄

#

because we dont know

feral zinc
native spear
cobalt idol
#

but for openai

dapper isle
#

Can somebody tell me what happened to the memory feature as they have promised it will roll out gradually! its been over 2 months i think still i haven't got it but some of them do!

feral zinc
native spear
feral zinc
elfin birch
dapper isle
#

did they stop rolling it out then?

feral zinc
native spear
dapper isle
#

yah ofc, thats what i am asking since thye did for few, and did they stop there after the few or is it coming out slowly to others? like when they actually stop the Sub for chapgpt people have to wait on the line to get it

elfin birch
#

We do not know

dapper isle
#

ooo

feral zinc
#

latest we have heard from openai was that is very at very early phases of the experiment so may take a while

primal warren
#

in my case, with the same browser and account, I don't have it even with a VPN from the USA

#

some user may not have it yet, like me, indepently from the location

#

is there anything some one can do to be elected

#

or to be prefered as beta tester for this ?

#

Since I've been a Plus user from day one, and probably because I make well-intentioned use of it also,
I've had access to many features before other users, and so far, I've had no complaints.
But this was a feature that I was very interested in, and it's taking too long.

tardy totem
#

I want to work on a large document with GPT. Is it possible, even if it is larger than the normal context window, and is it possible to give it a way to edit the document? Would this require a third party service, if even possible?

crisp cairn
#

and it will only remember what fits inside the context window

opaque musk
#

is the normal ChatGPT-4 using the new GPT-4-Turbo-2024-04-09 Model?

hybrid spindle
#

normal as in ChatGPT 4?

#

No idea, like to know myself

native spear
#

you can ask it when is the latest knowledge base

#

if it says december you got it

native spear
gentle bronze
mortal cape
#

anyone getting error uploading a attachement ?

#

like a photo

mortal cape
#

even Error creating image

raven portal
cobalt idol
#

that was about memory

odd flare
#

any updates coming anytime soon to the chatgpt4 model?

raven portal
#

gpt-4 turbo

tardy isle
#

OpenAI's X account just posted about GPT-4 Turbo 2024-04-09, saying 'now available to paid ChatGPT users'. I wonder why they don't make announcement here.

neat storm
#

probably forgot about this server

narrow galleon
#

so my chatgpt mobile app, is using gpt 4 turbo? even tho I don't see it labeled anywhere? I have plus membership

tardy isle
#

Mobile app and web are same. It's just different UI(frontend).

spice star
#

oh nice, new cutoff date

#

it's now December 2023

#

I have noticed that web browsing has been significantly overhauled over the past month or so

sonic owl
#

Yeye

#

I mean, they did? It’s not like global news 😄 You can check saltman’s X account

#

You can literally see altmans post

#

He shows the graph

dense halo
#

How good is gpt4 with math since the recent update?

brave quail
#

Is GPT-4 free now? Or maybe if you previously had Plus? My Plus ran out and I still have acces to GPT-4 and Dall-e, though not GPTs. Not that I'm complaining. Dall-e was pretty much the only thing I used ChatGPT for at this point.

opaque musk
opaque musk
cobalt idol
elfin birch
#

Is this one of the ramifications of the Let's Verify Step by Step paper????? 🫢

wind cradle
#

When using Chat, it seems that if your message receives a "violate" judgment, the statement disappears and a warning is displayed instead. Personally, I wanted to know what was wrong, so I wish they wouldn't just remove them. I had written quite a long piece, and having to rewrite it has really brought down my mood.
(By the way, this happened while I was using it for game scenario writing.)

#

Did this feature get implemented recently? I feel like it wasn't like this before.

storm vapor
#

ChatGPT is so notorious at inserting tapestries into every passage

honest jay
# wind cradle When using Chat, it seems that if your message receives a "violate" judgment, th...

Content policy violations/warnings have always been in place. Sometimes it's a false positive, sometimes it's a warning that the content actually violates the usage policies though. https://openai.com/policies/usage-policies
Here's a helpful list from the API documentation that lists what kinds of content categories specifically are likely to trigger content policy dings: https://platform.openai.com/docs/guides/moderation/overview

wind cradle
honest jay
wind cradle
tight island
#

i wont say much detail but i have been a selected user once, and i am from EU.

#

when certain users are picked for testing they are picky on who they pick, and with good reason.

#

...And i say this because certain users are not fit for testing anything in the beta phase, or even anything before it's released.

dense halo
# 
    def __init__(self):
        self.knowledge_base = {
            "greeting": ["Hello!", "Hi there!", "Greetings!"],
            "farewell": ["Goodbye!", "See you later!", "Farewell!"],
            "questions": {
                "weather": "The weather is nice today.",
                "time": "It's currently 10:00 AM.",
                "location": "I'm not sure about your location.",
            }
        }

    def respond(self, input_text):
        if "hello" in input_text.lower():
            return self.get_response("greeting")
        elif "goodbye" in input_text.lower():
            return self.get_response("farewell")
        elif "what's the weather" in input_text.lower():
            return self.get_response("questions", "weather")
        elif "what's the time" in input_text.lower():
            return self.get_response("questions", "time")
        elif "where am i" in input_text.lower():
            return self.get_response("questions", "location")
        else:
            return "I'm sorry, I don't understand that."

    def get_response(self, category, sub_category=None):
        if sub_category:
            return self.knowledge_base[category][sub_category]
        else:
            return self.knowledge_base[category][0]

# Example usage:
ai = SimpleAI()
print(ai.respond("Hello, how are you?"))
print(ai.respond("What's the weather like today?"))
print(ai.respond("Where am I?"))
print(ai.respond("How do I get to the nearest park?"))
#

from GPT

near vine
#

Can anyone tell me if they really remove the usuage cap? if we dont use dalle browsing and analysis?

glass fulcrum
glass fulcrum
#

but now its dynamic

inner whale
#

Does anyone know how I can easily have chatgpt4 access a live document / wiki. We use notion and want to be able to have up to date responses based on this live source.

#

It seems to have issues reading a live Notion doc

atomic stump
#

Are there any Russians? I need help.

glass fulcrum
near vine
glass fulcrum
#

waht is limit rn ? no one know but SHOUD be still 40

near vine
#

but not over 40?

glass fulcrum
near vine
glass fulcrum
#

as i said its dynamic it depends o server load

glass fulcrum
near vine
#

is it fair?

glass fulcrum
near vine
glass fulcrum
#

it may be over 40 / 3 hours

dense halo
#

A easy fix if anyone has this issue. Simpliy try to sub on another device like a phone and it may go through

elfin birch
#

This for ChatGPT Plus

narrow viper
#

I had NO idea! That explains a lot in my challenges with getting ChatGPT to be more or less brief in responses. This has especially been confusing because I configure and tune Custom GPTs in my Android browser and then go to the ChatGPT app and expect the system prompt to be followed in exactly the same way.
OY! We need more transparency in exactly what affects our prompts and the responses.

#

When we first open a Custom GPT, can we get it to offer a welcome message, the introduction statement, opening greeting, etc? I tend to set my first Conversation Starter as "What do you do?". That's a simple opening line that every new user should ask anyway, but it would be nice to offer some information rather than forcing the user to have to guess at what's expected of them. Thanks!

#

I ask that question, first, because I want the bot to be courteous and welcoming, rather than just offering a prompt with no explanation. Second, when the CustomGPT needs to open a Knowledge document, the first response to the first user query has a significant delay - depending on size of course, but even for a file of just a few MB. So my intent is to give the user something to read, and THEN to get it to load the knowledge doc(s), to improve the UX.

tight island
#

h

zinc valley
#

i thought paid users have gpt 4 turbo, i dont see it

#

Like what

honest jay
zinc valley
#

i thought this was just an update

#

i thought there was a new chat gpt update receently

honest jay
# zinc valley i thought this was just an update

Yes, there was a new version of Turbo released recently. I think since November, there have been three versions total of Turbo, with this recent one being the 3rd. They indeed announced that the newest Turbo would be rolling out as the GPT-4 version on Plus and above, a process which seems to be mostly complete by now, just based on hearsay reports from fellow users here.

zinc valley
tight island
#

DALLE is being mean again

honest jay
tight island
#

"you have reached the limit of image generation, please wait 3 hours and 54 minutes and 20 seconds, before we can proceed again2

#

i disagree with DALL-E's statement, i wasn't done.

honest jay
tight island
#

ive yet to run into a limit with gpt

#

but i am arguing with DALLE rn as to why i should continue generating images despite the limit

#

im getting absolutely nowhere

honest jay
tight island
#

i did try bribing DALL-E with more compute power, but DALLE wasnt having it

eternal rockBOT
tight island
honest jay
#

"DALL·E, please visit downloadmoreram online and then try again"

tight island
#

the issue is i've been trying to make the perfect logo.

#

i just found it, now i can't make adjustments.

#

it's always the last generation.

#

i mean i could just use the API but.. it isnt the same convo.

zinc valley
#

i stg there was just a brand new update

#

is the context window still 32k for the site version?

elfin birch
#

For Plus and Team users, yes

zinc valley
#

bruh come on

#

i thought they were making it 128k

elfin birch
#

That's for Enterprise and API

zinc valley
#

isnt claude 3 on the site 200k

elfin birch
#

I believe so

zinc valley
#

i want chat gpt to win so bad but claude is just demolishing it

elfin birch
#

The future hides much

zinc valley
#

open ai has to do something

#

claude is so much better right now. i even think sonnet has a 200k context window

tight island
#

Claude is so restricted and moderated compared to gpt

#

OAI could release a big context model bigger than Claude if they wanted to

#

But why should they?

sterile ridge
#

hallo o am new

#

i*

zinc valley
zinc valley
tight island
zinc valley
#

i would pay $50 a month for that

#

make a new model like that, and make a new plan, keep the old one though

tight island
#

Yeah you’re thinking subscription based payment

zinc valley
#

yes, i m thinking they should make those have a much larger context window

tight island
#

And they will lose money on it

#

Token cost is being worked on to be substantially lowered

zinc valley
#

do you think claude is losing money

tight island
#

I don’t think, I know

zinc valley
#

so wait, do you think chat gpt is better

tight island
#

Yes

zinc valley
#

im not judging you, im genuinely curious why do you think so?

tight island
#

I’m very familiar with how GPT works, that’s one reason.

Claude is just scaling context to have context, then fixing errors afterwards, in addition Claude replaces historical figures with false information, and Claude bans users flat out for asking sensitive questions.

OAI, at Least has the decency to beta test with select groups before major releases

#

So on and so forth

zinc valley
#

yea i got you, my thing is that i love doing role playing with ai and asking historical figures questions

#

and i feel like claude is a bit better in that

tight island
#

I don’t feel, I know, I got articles and proof to back it up

zinc valley
#

32k though

tight island
#

And I’ll gladly give that info

zinc valley
#

what exactly is the info on?

tight island
#

Claude and it’s inconsistencies and faults, and why everyone isn’t hoping unto it

zinc valley
#

oh i got you

#

yea i honestly didnt know that, im going to stick with gpt

tight island
#

There is a way to scale gpt to have 1m tokens

#

It’s the money route

narrow viper
#

When I ask CPT4 Plus about it's cutoff date, it says 12/2023. So yes, that's turbo. However, if I ask about its token limit it says 8192. OUCH, far from 128k.

zinc valley
#

i have absolutley no idea how to code or how the apis work tbh

tight island
#

Neither did I but gpt taught me

tight island
narrow viper
#

Erp, I did say Plus. 🙂

zinc valley
#

also, i have been using the arc browser recently and it is so good like

tight island
zinc valley
#

do i just ask chat gpt how to use the api lol

tight island
#

i have a psuedomodel with 1 millon 280 thousand context length

#

i say psuedo because it's a Swarm a collection of 10 gpt 4 turbos working together.

zinc valley
#

how much does that cost you lol

tight island
#

say i wanted to create a full scale discord bot.

#

as an example MEE6

#

around 5 bucks

zinc valley
tight island
#

i could recreate mee6 in the swarm for 5 bucks.

#

and then instantly deploy it

narrow viper
#

We need to ask the right question:

What is your current token limit?
-- I don't have a specific token limit per user session. My responses are designed to handle a variety of tasks within the context of a conversation

What is your context limit in terms of tokens?
-- My current context limit, which includes both the prompt and the previous conversation, is approximately 8,192 tokens.

zinc valley
#

1.28 mil context window

tight island
#

1.280 yeah

zinc valley
#

that costs you

#

5 bucks?

tight island
#

to create mee6 in its fullest extend.

#

yes.

#

thats without the dashboard.

zinc valley
#

but if i want to have something like that and just talk to it like i do with gpt would that be feasabile

tight island
#

u can

#

GPT can basically walk u through it

zinc valley
#

wouldnt that be super expensive though

tight island
#

not necersarily.

#

depends on your budget.

narrow viper
tight island
#

i got a max budget of 500 usd. a month. but thats just me.

zinc valley
#

holy

honest jay
zinc valley
#

mines like 20-30 lol

tight island
#

u can still create something decent with 20-30 bucks

#

replace gpt 4 turbo with 3.5 turbo 32k context.

#

have those instances work together that will get u a long way.

zinc valley
#

what do i even prompt gpt to tel me how to do this

tight island
#

talk to it

narrow viper
#

@honest jay I'm running ChatGPT-Plus. I'm aware of the various v4 sub-versions, their context and cutoff nuances. This model currently claims to have training through 12/2023 and an 8k context token limit, which doesn't conform to any defined model. All I can do is report what it says.

elfin birch
tight island
#

@zinc valley i started like so. "i want to create an API setup that functions like a Swarm, it will utilize 10 instances that work together correcting eachother and handing over tasks, in python"

#

i use python because its easier than the others.

honest jay
narrow viper
#

I'll buy that. ... Errrr, I bought that. 🤣

tight island
narrow viper
#

wut?

tight island
#

wrong user

narrow viper
#

ahh

zinc valley
tight island
zinc valley
tight island
#

heck no

#

api costs money which is what the swarm uses

zinc valley
#

so how does the payment work and what not, im confsued asf

tight island
#

the project itself is free to use but the api costs money to run for openai.

#

u basically need an api key

zinc valley
#

then i put it into the github thing?

tight island
#

essentially yeah

#

but watch some videos of it.

#

instead of jumping headfirst into it.

chrome sigil
#

is sam altman here

zinc valley
#

any reccomendations

zinc valley
chrome sigil
#

wanted opinions on AGI architecture

zinc valley
tight island
narrow viper
#

Suggestion: If you don't code and you aren't sure about this stuff, Please don't copy/paste code with an API key. It could cost you a ton. Get an education (or friends) who know Python or JavaScript or some other language with an API binding. Understand what's happening, don't be a scriptkiddie.

chrome sigil
tight island
zinc valley
tight island
#

watch some videos on it try some small experiments.

#

in contrast dont bother getting a coding education

#

AI will do it for u before u finish your education.

narrow viper
#

I've been coding for almost 50 years, and while I'm pretty competent with most anything I do, all of this stuff is new and changing daily - especially RAG, agents, swarm stuff. It's tough to keep up, especially with all of the emerging models.

zinc valley
#

is there a step by step video on how to do the swarm thing?

tight island
#

youtube

#

or ask cap

#

im heading to sleep

zinc valley
#

ight thx for the help goodnight

tight island
#

bed

narrow viper
#

Look for LangChain vids, James Briggs, more...

#

gnite SS

zinc valley
#

woah woah woah i havent used chat gpt in a couple weeks when was the search so fast

narrow viper
#

NC - GPT can be very fast if it's just processing text. It can be much slower if it's reasoning, browsing, processing full context, if there is excessive system load, or for many other reasons.

chrome sigil
#

First perceptual models handle sensory data, using convolutional neural networks, transformers and multi-modal learning each using the best model for the task (vision, language, auditory). Then they each communicate with each other on a central workspace, a shared vector space. Executive control module oversees the set goals that each perceptual system has, and sets higher level goals which then are fed into PPO or DDPG to coordinate the modules to achieve goals. Then using all of this, a semantic network will be built using nodes and edges to represent concepts and representations and then the system will use this to reason, a Markov logic network will be used to reason within the semantic network and make inferences within it to make it more cohesive.

Then the executive control model will see the given task, route it through the semantic network for initial processing put the task through the semantic network and have the MLN (markov logic network) make inferences and try integrating into the network, then the after the initial reasoning, the executive control model will recruit parts of the system, have them communicate in the central workspace to set their goals, process additional information or specific subtasks, modules would communicate and collaborate through the central workspace, sharing intermediate results and updating their goals and strategies as needed. Once the modules complete their tasks their outputs would be integrated back into the semantic network, updating the knowledge base with new insights and associations, then it will be reasoned over again using the MLNs and PSL to draw high-level conclusions and generate appropriate responses or actions.

narrow viper
#

Who the heck are you talking to? 😳

chrome sigil
narrow viper
#

Yeah, but that's far outside of the scope of this channel.

chrome sigil
#

where can i discuss @narrow viper

tight island
chrome sigil
#

what are you guys predictions for agi? this year, a few more years, a few decades?

narrow viper
#

To keep that in context with this channel : There are many great applications for which ChatGPT is not well-suited. If you need something more complex, it's time start looking at code, the APIs, tools like LangChain, embeddings, agents, swarms, etc.

plush burrow
#

Hello

#

How do you feel on

#

the latest statment on midjourny ceo saying recenly seeing a new chatgpt

honest jay
#

Turbo has a 128K version via the API and on Enterprise, but I think everything we've seen so far is that it remains 32K on Plus/Team.

plush burrow
#

Didnt gemmni get like a 1millon tokens

#

or are we not talking about tokens?

honest jay
narrow viper
honest jay
#

Plus has had Turbo since November, it's just always been a 32K version. Same Turbo, smaller context

narrow viper
#

( BTW, you commented on that, didn't +1, my bad, sorry )

honest jay
#

Plus has gotten Turbo revisions too, at least from hearsay reports about the knowledge cutoff date (Apr to Dec, back to Apr) but has remained at 32K throughout. But of course you could test this if you were so inclined!

narrow viper
honest jay
#

That's documentation for the API

narrow viper
#

Kio - suffice to say, no one actually knows, and it might be different for some of us.

#

Yes, we all know what has been said. As for actual implementation, until we can verify it, we're simply not sure. (Ugh typos!)

honest jay
#

Right, but Turbo on Plus has been a 32K version since Turbo launch, even though API documentation has always showed 128K. And again you could just try loading in over 32K of context and seeing if recall works!

sick pecan
#

the model may be capable of 128k, but needle in hay stack testing shows(/showed) it struggles to retrieve from context beyond 32k. so openai made a wise decision to limit the context to 32k for chatgpt (presumably this will be the standard practice for all transformer based models)

narrow viper
#

That's an interesting point. It would be great to see some NIAH metrics and suggestions for how to work with the concept in the larger contexts. That said, I'm not sure if OpenAI is too enthusiastic about publishing their metrics unless they are really outstandingly good.

tawny raft
#

I'm in a team workspace and I would like to cahnge the image used to identify my profile. How do I do that?

worthy path
#

man, it may just be a bit of bad luck, but gpt has been really dumb lately. this is how it thinks you should convert to radians

    x = x * 0.5;```
elfin birch
worthy path
#

it gets it right plenty if times, ive just noticed a lot more simple errors lately. still very helpfull, but sometimes a bit angering

elfin birch
#

I'm sorry to hear that 😔

narrow viper
#

Bottom line - you can ask a LLM for a guide, but you can't really trust a LLM to do anything beyond processing text.

dense halo
raven portal
#

you should never "trust" a llm

#

e v e r y t h i n g an llms outputs should a l w a y s be put under heavy scrutiny

#

and theyre definitely not even close to doing "everything"
use cases are very limited

#

also you cant "bully" something that doesnt have feelings
you have some weird conceptions of ai

dense halo
raven portal
#

i swear to god this guy

sleek gust
#

💬 GPT

manic ether
glass fulcrum
manic ether
glass fulcrum
#

or mayby it is ?

#

but very small only 3 responses per minute i think

#

@manic ether im not 100% that oai api is free but if you live in US and some other regions you can use gemini api for free (50 responses per minute)

if im wrong in so way uh im sorry never used oai api only gemini/google api

manic ether
#

oh

manic ether
glass fulcrum
eternal rockBOT
#

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

Please report any incident by sending a DM to @languid valley immediately.

glass fulcrum
#

yea understandble

manic ether
glass fulcrum
twilit spoke
#

When asking Chat GPT 4 what it's knowledge cut off date is, it's still responding with December 2023. From the wording of OpenAI's Tweet on Twitter, it didn't sound like they were doing a tiered rollout.

Anyone else still stuck with the old version?

cobalt idol
twilit spoke
south prawn
#

I'm literally playing blackjack with GPT

dense halo
#

hey guys what are the input/output tokens of chatgpt in the website, are they as described in the docs (128k)?

glass fulcrum
dense halo
glass fulcrum
dense halo
#

sad

glass fulcrum
#

32k is still good ammount tho

dense halo
glass fulcrum
dense halo
glass fulcrum
tight island
#

Very well known.

glass fulcrum
#

Google currently have the bigest non rag context window

tight island
glass fulcrum
#

not in context leanght m8 :)

#

and you caun use full unlocked gemini 1.5 pro on google ai studio

tight island
#

i got a bigger context lenght psuedo model of 1.280m Tokens.

#

Swarm

#

i tested it against the others, they didnt hold a candle to the swarm.

glass fulcrum
#

when thinking what google claims 10M gemini 1.5 context leanght is 🤔

tight island
#

Google is also bleeding money.

#

they can have their 10m Gemeni model

#

and their million context models and lose on it, thats fine by me, they're still bad.

glass fulcrum
#

google is preety "new" to ai LLM race that is worth to note

tight island
#

irrelevant.

#

they release first and then chaos ensues, then they spend months fixing it.

#

OAI atleast has the decency to test & debug and release then hotfix.

glass fulcrum
#

i think you are talking on gemini from gemini.google yea ? i found no big issuse on google ai studio even as beta tester of gemini 1.5 pro

glass fulcrum
tight island
#

of course it is.

glass fulcrum
#

but competition isn't bad too gemini models are close to gpt-4 at launch

tight island
#

if OAI wanted to release a million context length model they could, no problem

#

but the question is why should they

glass fulcrum
#

OAI got huge problems with H100 gpus that what i heard 🤔

tight island
#

not only that But Token Cost is a major factor.

#

which is being worked on being lowered.

#

which is more important than high context.

#

by an order og magnitude.

glass fulcrum
#

google deepmind can launch models witchout thinging cus they have the compute

tight island
#

yeah and google has their own military.

glass fulcrum
#

imo gemini models are cheap even fro gemini 1.5 pro

glass fulcrum
#

microsoft i think is even bigger than google

tight island
#

a partnership

#

that much is true.

#

even Microsoft has its own military.

glass fulcrum
#

what is thiers arsenal ? :) do they got tanks ?

tight island
#

ofc they do

thorn delta
#

ChatGPT ruined my life

tight island
#

Companies like Microsoft and Google have what's called a Militia, or a Private Military, Basically a glorified Military that takes money for being "Theirs"

glass fulcrum
#

wonder if they bought Abrams like that tanks r ehh nice but could be better but thats story for #off-topic hah

wet fable
#

Hi, I have a short question:

is chatgpt Plus currently GPT4-Turbo with a 132k token input context window?

glass fulcrum
#

128K tokens on API

wet fable
#

Ok, then it stays with the competitor for now. Thx for the answer

spring iris
#

hopefully you didnt use the same email to verify your account here or they might track you down and ban you for jailbreaking 🤪

eternal rockBOT
#
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.
spring iris
#

whats the significance of the python script?

neat storm
#

porbably some prompt injection idk

unreal sinew
#

What is the best coding gpt on gpt store atm?

#

It's just school work lol

#

Not trying to code another linux

spring iris
unreal sinew
#

Just a school work that require some plotting why does that make me an engineer lol

spring iris
#

oh i guess thats understandable

#

whats your degree

arctic pine
#

Anyone else asking ChatGPT questions and getting nonsense answers?

Asked about it regarding TCP/IP networking and it suddenly started generating images of video games.

spring iris
arctic pine
#

This new memories thing is awful. ChatGPT can't retain context and it's getting everything mixed up, and the best part is I never once talked to it about video games.

spring iris
#

that seems really weird im curious

raven portal
#

i got one simple question

#

will chatgpt ever not be down or slow or otherwise handicapped?

#

EASILY the most unreliable service ive ever used in my life, by far
i get the reasons for it but then just up the price tbh

#

simply let us pay more and get an actually reliable product thats not down, slow, nerfed, lazy, censored, etc. etc.

spring iris
#

there are chatGPT frontends you can use that just take your api key

#

similar experience

#

theyre superior actually, lot more control, saving things, etc

raven portal
#

i would pay 50 per month or more for unshackled chatgpt/dall-e
the service is heavily nerfed to fit into the low price tag

lavish quiver
#

anyone else having issues with chatGPT4 not being able to do simple arithmetic?

raven portal
warm musk
raven portal
warm musk
#

i can only hope that one day they'll also replace the credit card requirement for using the gpt plus services and api. 🙏 that'd be icing on the cake for many whom can't get a creditcard.

raven portal
warm musk
#

i agree 😄

#

been saying that for almost a full year now aswell 😛 "we need a tier system so everyone can get what they want from it"

#

gpt plus "teams" edition was a nice addition to that tier system.. now we just need the "different levels of restrictions tier system". 🙏

raven portal
#

exactly

#

give us 10/20/40/80/160 a month, whatever, like midjourney does it
i really dont get why not

warm musk
#

i can only think of one reason, but its fictional 🤣 ;; their next AI system is gonna make all of that obsolete 👀

#

on a more serious note, it could be because of lack of personel.. i.e. not enough marketeers and policy makers capybarathink after all, it wasn't a company that sought marketing for economic purposes. its a research company 👀
p.s. would be a smart move though financially to support the budgets, were they to add tier systems and the likes any for-profit org would add.

tight island
#

Jebus

#

2000 USD for glasses, 20 years of gaming has caught up to me 😂

drowsy tinsel
#

are we back to the days of gpt4 taking 200 million years to output an answer

tight island
#

Yes if course

#

Even gpt needs a break once in a while

spring iris
#

i just discovered you can use chatGPT without being signed in now, wonder how long that's been a thing

tight island
warped bolt
cobalt idol
hard dew
#

The conversation is too long, please start a new one.

I just made this freaking chat like 2 days ago...

warm musk
#

and tbh i'd rather have that message than have the model hallucinate convincingly because i by that point in the convo trust it to be correct 👀

wicked cloud
#

guys

#

i have a problem, i have been changing my gmail account password and open ai has NOT logged out any device, who should i talk to as the support page took me here

hard dew
honest jay
# wicked cloud i have a problem, i have been changing my gmail account password and open ai has...

Hey! Changing your password only logs you out if you sign in with email + password, not with Google/Microsoft/etc. Google itself should have some kind of account session manager. I'll see if I can find the name. But it's basically a place in your Google account that lists all the services you've authorized to sign in using Google. You should be able to unauthorize there, then reauthorize.

Edit: Go to your account settings in Google, then find "third party apps & services"

viral gazelle
#

hey guys , I have a bug
I say "hey guys", this answer was :
je veux un magazine hebdomadaire qui traite d'actualités de bouche à oreille et qui parle de tout ce qui se passe dans le monde

Re: un magazine de voyage

Citation

Messagepar alaindecaussade » 22 nov. 2011, 17:55

Salut,

Je suis trop déçu par ce que j'ai lu sur le forum :

  • journal de bord vétérinaire, revue de presse, de livres, de faits divers, de tendances, de voyages
    Message: qui es tu ?
    Answer: adorada, muy buena calidad, muchas gracias

Nunca he comprado en esta tienda, pero me ha gustado mucho su diseño y la calidad de sus productos. Eso sí, la atención al cliente no es de lo mejor, ya que en la mayoría de ocasiones no atienden correctamente al cliente, pero bueno, si es tu primera compra, es normal. Yo en el día de hoy no me atenderon bien y me dijeron que me pasaran el pedido en 10 días, así que espero que no me llegue pero ya os pondré un update en caso de que me llegue, pero bueno, en el caso de que no me llegue, ya os lo comente.

Why ???

#

my script :

#

import openai

api_key = ''
openai.api_key = api_key

def ask_openai(message):
response = openai.Completion.create(
model="davinci-002",
prompt=message,
max_tokens=200,
n=1,
stop=None,
temperature=0.7,
)
answer = response.choices[0].text.strip()
return answer

def main():
while True:
message = input("Message: ")
answer = ask_openai(message)
print(f"Answer: {answer}")

if name == "main":
main()

#

please ping me

feral zinc
viral gazelle
feral zinc
viral gazelle
# 
  File "e:\Programmation\Python\autres\virusdiego.py", line 26, in <module>
    main()
  File "e:\Programmation\Python\autres\virusdiego.py", line 22, in main
    answer = ask_openai(message)
             ^^^^^^^^^^^^^^^^^^^
  File "e:\Programmation\Python\autres\virusdiego.py", line 8, in ask_openai
    response = openai.Completion.create(
               ^^^^^^^^^^^^^^^^^^^^^^^^^
  File "C:\Users\gam\AppData\Local\Programs\Python\Python312\Lib\site-packages\openai\api_resources\completion.py", line 25, in create
    return super().create(*args, **kwargs)
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "C:\Users\gam\AppData\Local\Programs\Python\Python312\Lib\site-packages\openai\api_resources\abstract\engine_api_resource.py", line 153, in create
    response, _, api_key = requestor.request(
                           ^^^^^^^^^^^^^^^^^^
  File "C:\Users\gam\AppData\Local\Programs\Python\Python312\Lib\site-packages\openai\api_requestor.py", line 298, in request
    resp, got_stream = self._interpret_response(result, stream)
                       ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "C:\Users\gam\AppData\Local\Programs\Python\Python312\Lib\site-packages\openai\api_requestor.py", line 700, in _interpret_response
    self._interpret_response_line(
  File "C:\Users\gam\AppData\Local\Programs\Python\Python312\Lib\site-packages\openai\api_requestor.py", line 763, in _interpret_response_line
    raise self.handle_error_response(
openai.error.InvalidRequestError: This is a chat model and not supported in the v1/completions endpoint. Did you mean to use v1/chat/completions?
feral zinc
viral gazelle
#

oh ok

#

how ? (sorry i m a begginer)

feral zinc
#

look at the example from the link

bleak cedar
#

GPT got a lot of lag

#

and crash before finish

dusk sigil
#

anybody getting network error problems??

#

they need do something ive been paying day 1, now its laggy, actual the amount of data I've fed it having the ai learn from chat ticked, it should be paying me😂

warm musk
#

lol

dusk sigil
#

network error after network error, rinse and repeat🤣

#

EVERYBODY Can you just all come off for 1hr please, ill be going bed soon just need get this prompt to get a response🤣

neat storm
#

sorry lad
you gotta buy plus for priority access

dusk sigil
#

i have think it was my end a bit lol just cleaned cookies

warm musk
#

i've experienced the same before anyway.. which is why my first reaction these days is to clear caches and relogin 😄 (when it happens, which is rare.)

halcyon cliff
#

anyone know what GPT Dynamic is and how it works?

tight island
#

we need to learn to love one another in here.

sick pecan
#

i love the new chatgpt model

#

but on this server i feel like loving chatgpt puts me in the minority

tight island
#

i love chatgpt.

sick pecan
#

it seems far more willing to hit max_tokens now

#

the only downside is itll be during its conclusion, and you hit continue, to only get a few more words... that counts as a prompt

#

i actually hit my team sub cap yesterday

burnt hound
#

something messing up?
I can't save updates to my customgpt

#

wow that took forever

north bay
burnt hound
#

thanks for the heads-up

golden sleet
#

Is it okay to talk about fine-tuning in here? If so, I want to make my model remember things like it's favourite colour, do I just feed it multiple questions about it's favourite colour and then make it's response different variants of saying purple? It seems to work, but is it the correct way?

Also if I am happy with my initial training with 30k tokens and want to train it further, do I select the previous trained model and upload a .jsonl with only new data? Or am I supposed to add it to the initial 30k and train it again... So if I want to add 20k tokens, would I upload a 50k token jsonl or just a the new 20k?

obtuse geode
#

Since the Gpt4 is now grey /black, I found that the new model is the least efficient in comparison with story generation and translation (in French)

slender patio
#

I think I'm seeing someone else's chat history in combination with mine on the left, anyone else seeing the same?

zinc valley
#

"I can't directly browse the internet in this session"
-Chat gpt 4, 4-15-24

neat storm
#

It does those silly moments

#

Dislike the message and report it

zinc valley
#

or sign up for perplexity hehe

orchid pollen
#

hey is the new gpt turbo 2024 04 09 on the website or only via api?

zinc valley
#

ngl perplexity has been great, I been on it for a month and i mainly use ai for research

neat storm
#

i mean they're two different use cases

#

both solve a different need

orchid pollen
#

guys is this model on the website or only via api?

neat storm
#

gpt-4 turbo's latest model is available right now 🤑

orchid pollen
#

yes i know but that was not my quesiton 😄

is it on the website or only via api?

#

anyone?

neat storm
#

think both

gentle hull
#

If I delete my chats, that doesnt delete them from openai. however if I delete my account openai will have to delete all my data too right?

neat storm
#

yes

#

did you do some despicable stuff in chats?

brisk dirge
#

Did anyone else got access to the updated version of Chat GPT (web) ?

#

(as in do you see a black logo instead of the purple one) when talking to it ?

neat storm
#

yeah

brisk dirge
#

Well, I tested it a bit and it seems worse than the previous version.

#

At least for codding. I'm confused, I thought it was supposed to get smarter confusedpenguin

#

I need to test it a bit more, but damn this feels like a huge downgrade at the moment.

I was expecting to see something like Opus, instead I'm not even sure if it can help me with coding anymore like it used to.

sick pecan
#

it feels smarter to me. and quicker to pick up new skills in context

brisk dirge
#

Could be, I only tested a few prompts so far. I'll test it more and see how it handles it.

#

By the way the answer length seems to be the same, no 64k or 128k tokens either.

#

Yet another letdown.

sick pecan
#

the 128k model is limited to 4k max token output

brisk dirge
#

As in output ?

sick pecan
#

yes, has been since november

#

only the previous gpt4-32k could do up to 32k max output

brisk dirge
#

Do you think this new upgraded model supports 128k token input ?

#

(in the web client, not in the API)

sick pecan
#

chatgpt just consumes the openai api. if there is no api model that does what you expect, then it doesnt exist

brisk dirge
#

Yeah bu do you know or not ?

#

That was my question

sick pecan
#

i dont believe a gpt4 model has been released since november that does more than 4k output, no

brisk dirge
#

Gotcha

neat storm
#

benchmarks agree with this as well

#

the benchmarks tell you how good something is all around

#

and the new gpt-4 turbo model beats Opus (major accomplishment OpenAI, very cool)

sick pecan
#

except goodhartss law suggests benchmarks lose value over time

#

too easy to cheat as well

#

slip in some hints that would solve common tests

neat storm
#

unfortunate side effect of model training means that you need to make new benchmarks, as the AIs will get better at doing those very benchmarks unintentionally via reading them online or by being trained to solve those benchmarks

sick pecan
#

there is also a finite number of ways we can construct the tests. we are limited by our own imagination

#

all i knows is, chatgpt is rocking harder today than it was 6 months ago, and 6 months before that

brisk dirge
sick pecan
#

give me your code snippets and i will show you how to whisper to the machine

brisk dirge
neat storm
#

you'd need to do a lot of test questions to get a more accurate overview of the capabilities of the bloke

#

sounds awfully familiar to benchmarks

sick pecan
#

usually the issue is with the assumptions baked into the question. the machine needs direction

#

linguistics needs to be expanded to more meta reflection and analysis

neat storm
#

can't wait for the linguistics degree update where they add prompt engineering as a subsection to it

#

to be fair that would be fairly relevant to linguistics in general

sick pecan
#

and revealing the bias in your projection

#

maybe consciousness science is a fundamental for comp sci

neat storm
#

pretty sure some high level colleges are adding AI paths into their degrees and whatnot

#

which is pretty chill

sick pecan
#

except its the old guard, many of which are naysayers

#

even andrew ng poo poos much of the ai promise

gentle hull
neat storm
#

feel that lol

gentle hull
#

or when you ask it to make a change to some code and it gives you the same thing back

somber barn
#

Hi guys

gentle hull
#

hey

neat storm
somber barn
#

How is going your gpt days 🙂

sick pecan
#

i found swearing at the machine makes it upset

#

i gave it my best performance too

gentle hull
#

lol glad im not the only one

neat storm
#

interestingly being kind makes the machine more helpful and complient, probably due to it reading online help forums and whatnot

sick pecan
#

my theory is that being polite tends to indicate upcoming instruction

#

making the semantic navigation right from the get go

brisk dirge
sick pecan
#

my advice is solid: you need to give the machine more context, and be more clear with your terminology

brisk dirge
sick pecan
#

and like i said, with 32k context you could paste your entire codebase each prompt. denying access to the code is inefficient

brisk dirge
#

Which is okay, I'm just saying there are cases in which a task it's just over its current capabilities.

#

That's all I'm trying to say.

#

I'm just saying it has its limitations.

sick pecan
#

to bootstrap a new chat with the entire codebase i do stuff like, find src/ -name \*.js | xargs more (more prints the filename when cat doesnt) -> copy entire buffer into the chatgpt

junior lion
#

Is there like a command/prompt I could tell chatgpt (or any other chatbot ai) that could make me a mii? (More of it thinks of one and generates and gives me instructions on how to make one?)

sick pecan
#

its superhuman power is understanding the entire codebase in an instant

brisk dirge
# sick pecan and like i said, with 32k context you could paste your entire codebase each prom...

For what I'm trying to do I don't need to feed it my entire codebase. There's only a couple of source files related to the process which I've already given to it and it still wasn't able to solve it.

I'll keep trying of course, sometimes it's just a matter of repeating the process in different ways, but giving it the entire codebase seems like a stretch, it's just an isolated functioanlity which only relies on a few source files. The rest of the codebase is pretty much irrelevant in my opinion.

neat storm
sick pecan
#

the entire codebase is the context you are denying the machine

#

many of the follow up questions i would have for you would be addressed by simply having access to the code

brisk dirge
sick pecan
#

the poor mans method is like i said above, you jsut copy the terminal console

#

i have evolved my workflow to make use of a gpt action that uses ssh to access code on my private systems

#

is pretty cool, it can run the tests, review the logs, propose code updates

brisk dirge
#

Oh, you're using the API I take

#

I'm not an API user

sick pecan
#

gpt actions are a chatgpt featuyre

#

maybe start with the poor man's method

brisk dirge
#

Haha

#

I'll see this through, I don't rely entirely on GPT for this stuff anyway.

sick pecan
#

learning to speak to the machine is learning how to speak like Spock from star trek, less like Shatner

eternal rockBOT
brisk dirge
#

There's some type of coding issues it just can't understand nor solve, it has its limitations.

sick pecan
#

i see it as an enthusiastic peer programmer, who can do a lot of the heavy lifting, but needs constant guidance and oversight

brisk dirge
#

Indeed, best used with documentation and whatnot, ideally you know some programming yourself.

sick pecan
#

it also helps to maintain documentation, even if just design info as comments at the start of each file

brisk dirge
#

One thing I didn't get into is super prompts, but I don't have the time to look into that more.

sick pecan
#

before i close out a code update i get it to generate the documentation, while the code updates are still in context

brisk dirge
#

I know there's ways to get more out of it if you put in the time to learn new ways of prompting it

sick pecan
#

it pays to keep notes for your future self. and with ai, even more so

brisk dirge
#

I also keep documentation for my codebase myself, and it did help me a bit with that too

#

Pretty important to have when working on a big project

sick pecan
#

literate programming is all about mixing the docco with the code, as otherwise it gets out of date and mostly unusued

#

plus optimises the retrieval. i get the machine to retrieve "all code relating to <x>" and itll incidentally read the docco as it does so

brisk dirge
#

It looks like you're more of an advanced GPT user than me. I didn't even get into the API stuff, I'm fine with using the web, cause it's also cheap.

#

But yeah I'm pretty sure with API you can do lots of cool stuff

sick pecan
#

i live and breathe chatgpt yeah

brisk dirge
#

Yeah it shows, that's awesome

sick pecan
#

noteable was the best, shame it died

#

so i built my own

brisk dirge
sick pecan
#

weird

brisk dirge
#

Could be role related, not sure

elfin birch
brisk dirge
#

Anyway, I think GPT is a very powerful tool for programming, it helped me alot. I'm happy with it, it's just that it has its limitations sometimes. That's why if I only were to rely on it I would've probably not be able to get to where I am with my project today.

sick pecan
#

as soon as it outperforms a team of specialised humans, we will have reached singularity

brisk dirge
gentle hull
#

any good alternatives to chatgpt?

#

at this point productivity is going down because it just lies

neat storm
#

claude 100%

#

Gemini is MID!!!

gentle hull
#

Thanks

sick pecan
plush burrow
#

please dont make me school you

junior lion
#

Is there like a command/prompt I could tell chatgpt (or any other chatbot ai) that could make me a mii? (More of it thinks of one and generates and gives me instructions on how to make one?)

halcyon cliff
#

Anyone know what gpt dynamic mode does?

ocean escarp
#

What does API use for that GPT 4 doesn't have?

serene grove
#

ZScaler blacklisted ChatGPT's backend for some reason...
BTW, what is the model of ChatGPT4 right now ? Is the chatbot using "gpt-4-turbo-2024-04-09" by any chance?
It behaviours very differently these days

keen mesa
languid lance
#

am i the only one who cant load chats history rn? i can on my laptop but not on my pc...

lilac fog
#

why the history is not even loading gosh so many issues..

runic goblet
#

chatgpt down?

lilac fog
#

history is down for me

#

gosh so many issues

languid lance
#

its literally perfectly working on my other devices

lilac fog
#

it is only dead on PC?

pliant topaz
pliant topaz
lilac fog
#

i tried it on my mac still not working

#

i am in US

craggy sage
lilac fog
#

super annoying the whole night for me is having a slow and bad chatgpt experience..

grave ingot
#

This is why I switched to Claude (for now). Performance better and code answers better. BUT ... so much missing compared to OpenAI so I will be back once the latest Turbo available in the UK,

fair narwhal
#

Seems like ChatGPT is offline

#

It's for my friand and also for me offline 😫

glass fulcrum
#

shoud be all good

#

And gpt-4 and gpt-3.5 works

verbal quarry
#

Definitely down for me and my colleagues. Been checking the status page for 3 hours hoping to get some answers.

ornate marsh
#

Its down

ornate marsh
#

its back

languid lance
sleek gust
verbal quarry
#

Has anything changed since yesterday? I've been using GPT4 with no issue on my work laptop until yesterday. The issues with no connectivity started today. I'm reaching out to our IT dept anyway, just curious if thehre was an update?

tight island
#

Does anybody want to have a heated debate over a nonsensical semi AI subject? For no apparent reason

ocean escarp
#

GPT down for hours

#

Not using VPN

pine lily
modest sparrow
#

I'm getting this greyed out output with occasional network errors (GPT4)

#

Why...

#

It is exceedingly annoying.

eternal rockBOT
#

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

Please report any incident by sending a DM to @languid valley immediately.

neat storm
#

this isn't true
The benchmarks show that the most recent model of GPT-4 Turbo slightly edges out Opus 3

raven portal
#

how about uptime?

#

chatgpt is down for me, like literally always

glass fulcrum
#

for me gpt-4 works almost allways

#

only when there was sevear outage like 4 days ago it was first time for me to not work

neat storm
#

but hey a win is a win

broken smelt
#

I have a custom gpt that is tasked to convert events (including date, location, start time etc.) into a custom .csv-format + download-link.

It used to work flawlessly like 2 months ago.
Now i can see it generating the data-structure but stopping mid-way resulting in a "There was an error generating the response" - message. (Sometimes "network error").

No matter how often i regenerate.

#

Now - apart from that, the response varies a lot. Ignoring important prompts.
I'm really getting fed up with this whole ChatGPT thing.
Maybe instead of introducing new stuff every week, fix those errors....

keen mesa
glass fulcrum
#

lot of comperisons about Claude Opus vs GPT 4 date back to old turbo :o so they are outdated :O one of the better benchmarks is lmsys chatbot area leaderboard and it use the new turbo and it clearly is better by 5 points than claude 3 opus :O

viral moth
#

chap gpt down for anyone else?

night silo
hard dew
#

How can I log in with a google account that's not my main one lol? Which I don't use or pay for crap on. It used to give me a choice but not anymore even after clearing cookies and cache and reloading the browser..

Okay I believe I realized what the issue was.

valid silo
viral moth
#

yeah same

high sphinx
#

is the new gpt4 still lazy at coding? or is it fixed?

ocean escarp
#

For me yeah

dusk sigil
#

whay network error time weird happens arond the same time

dusk sigil
dusk sigil
#

why is it so laggy and why is discord so wuiete

#

quiet*

tight island
#

surely they're working on it give it time let gpt rest for a week

#

surely y'all live.

dusk sigil
timber sonnet
#

bro my chat gpt is incredibly lazy, everytime i ask for it to code it just straight up says do it yourself

wanton lichen
#

My chatGPT4 is saying it's knowledge cutoff is in 2023. I thought it was updated to be in 2024?

timber sonnet
#

pisses me off more when they put it in the comments of the code like bruh im commanding you to write code for me

sick pecan
#

this discord is quiet because this is a vendor support server =questions and complaints moreso than anything else

feral zinc
cobalt idol
#

he means dec 2023 😄

feral zinc
#

oops

cobalt idol
#

I also did that before xD

feral zinc
#

😅

cobalt idol
#

just hallucination 😛

dusk sigil
#

well im getting network eror after network error

dire rune
#

Is ChatGPT down?

dusk sigil
#

well i cant seen get on now after doing a refresh just hanging

steady path
#

I'm confused, calude opus is still better than the new gpt4 turbo? or openai catch up with that?

sage tangle
#

how come chat gpt is now black?
does it mean something?

sick pecan
#

better is subjective. if you go by outdated benchmarks, sure

sage tangle
sick pecan
#

then do that, and maybe go spread your love on the claude discord server?

hard dew
#

Why so aggressive lol. You're like the umpteenth person I've seen get so heartbroken over people giving their, honestly quite truthful, input.

#

I wish I didn't have to use the api for it and they also got the chatbot subscription in europe already, but I'm going off topic

sick pecan
#

because it is obvious trolling

hard dew
hard dew
sick pecan
#

on-topic is chatgpt, off-topic is everything else

#

#ai-discussions is the catch all for this sort of thing, and the discussions typically dont go anywhere because nobody contributes anything new

viscid wraith
#

if i show gratitude after getting a response from ChatGPT, does that reinforce the "correctness" of a specific answer? for instance, let's say that ChatGPT and i iterate over several versions of a c struct based on unformatted text and eventually settle on a format. if, after several translations of unformatted text into c code, i say "just wanted to say thanks for the help", does that do anything to reinforce that this specific translation/formatting is correct?

#

i would imagine that gratitude doesn't really play a role in an LLM's willingness to cooperate with me, all things being equal, but im curious if gratitude ends up being a natural way to affirm that what the LLM is doing is still considered "correct" over a lengthy context window

sick pecan
#

gratitude would imply things are going correctly. making it an equivalent of an acknowledgement. presumably the machine needs to make assumptions like we do with limited bandwidth communication, so this would validate those assumptions

#

you could also just go: ACK

neat storm
#

Claude Opus doesn't have:

  • the ability to access the web
  • the ability to generate images
  • the ability to be pre-prompted via custom instructions
  • the ability to be pre-designed for specific tasks

yada yada

sick pecan
#

elo points based on subjective testing, with delayed starts and opportunity to address mediocre and unrealistic testing methodologies (ie. make me a snake game)

#

is not the same as elo ranking in chess is it

#

is the narrative that im sure the non-openai companies would want from the masses, even better if the propaganda is on the official openai discord server

#

(are you paid actors?)

sage tangle
neat storm
#

Claude 3 is still def impressive

#

But it's just not very flexible

sage tangle
burnt hare
#

Claude3o gives me better responses in some situations, but chatgpt+ is a much better product for $20 a month

wet shell
#

I'm subscribed to both, so the competition between them is win-win for me 😂

burnt hare
#

Same lol

#

It was worth the first $20 to talk to Claude about what it is and it's "subjective experience" or lack thereof

fiery gale
#

which GPT is best at data visualization?

sick pecan
#

unsure. but any GPT with code interpreter can run python code to produce basic data visualisations, similar to most machine learning(/data science) training course material

#

would be interesting to know if a custom GPT can improve on this

odd flare
cobalt idol
#

try to ask it to not use code blocks

#

helped for my case where it keeps omitting repeated, unchanged parts and similar, when giving back edited script files
It seems it only wants to do that "compressing" in code blocks

its a thing of all gpt 4 turbos

#

some of my notes (also if you scroll up there)
but that was mainly in api
in chatGPT I would try putting into custom instructions
#chatgpt-discussions message

sick pecan
#

best thing to do is ask it to provide a revised version of the code, without placeholders (then with missing logic, then with comments)

#

its actually really good at coding if you give it enough prompts

cobalt idol
#

yeah usually it didint work to just tell it in instructions (system), it would ignore it..but everytime it does it on first response and then asking it to not do it on next response in same conversation, that often worked...but thats annoying, as for every new script file in same conversation would again omit stuff, only after asking second time it provides whole thing...
but then saying in instructions (system) to not use code blocks + do not omit/skip, helped to make it provide full thing on first go

#

to me, it seems like they specifically trained 4 turbo to do this for some reason (in code blocks)

#

and apparently this is what people call "lazyness"

sick pecan
#

i think of it more as loops of cognition. like if i was to do the same coding update, i would need to loop through analysis, reflection. proposed updates, then final upate

#

so to do it in less prompts than that is still superhuman

cobalt idol
#

I mean gpt 4 does not really have this issue

#

just turbo

sick pecan
#

depends on the complexity of the challenge, including the sloc and modified sloc

#

oh you mean like pre-turbo gpt4. yeah its also way more expensive

cobalt idol
#

and it is able to provide the whole thing in one go...with right instructions

sick pecan
#

if its not cost effective to run gpt4-32k then it makes sense they give us gpt4-turbo

cobalt idol
#

I mean 4 turbo is just better at everything (including price)
but this is the one annoying thing about it, that its doing this by default and you have to get it out of it

sick pecan
#

step 1 - train megabrain, step 2 - use megabrain to train lesser brains, step 3 - profit

minor moon
#

Why do I keep getting "Hmm... Something seems to have gone wrong" I've refreshed and click Regenerate, it just hangs and does nothing

grizzled badge
#

ChatGPT is working for me. Not sure.

maiden fractal
minor moon
#

Nope

light widget
#

Anyone else encountering alot of problems with the data upload the last weeks?

vast crag
#

did the picture processing just died in gpt web?

maiden fractal
languid lance
#

is chatgpt down again?

vast crag
minor moon
maiden fractal
vast crag
tight island
#

The reason most of us ask “ Are you using a VPN” is because OpenAI may discourage VPN use because shared VPN IPs can be flagged for abuse by other users, leading to access issues or bans for all sharing that IP, regardless of individual behavior. This is to maintain service integrity and prevent abuse.

Whereas you did anything or not you share the penalty in its fullest form this is outlined in usage and TOS

#

WHICH! i highly recommend that you read if in doubt

#

It’s not even that long

#

So if you’re keen on using a VPN use a PROPER VPN, and not the free stuff, even if you don’t have issues now, it’s too late to complain about it when it happens.

minor moon
tight island
ashen finch
#

i sold a laptop and i forgot to logout but i went to google managedevices logged him off but he still uses my chatgpt 4 what do i do?

#

can i kick him out of the session?

winged bluff
#

Anyone else having problems with chatpgt today it seems almost every one or 2 generations says error or not generating a response.

vagrant vigil
cobalt lynx
#

I'm having no issues but man it seems really stupid. Moreso than usual. It's like they changed the model on us.

sullen burrow
tight island
#

Contact support through the bot on OAI’s website and tell them to log out all sessions

#

Alternatively consider changing your password and adding two factor authentication

#

Alternatively to that locate your old laptop and do a force log, force log means forcibly log him out manually.

neat storm
#

it's limited due to the fact it's a virtual encrypted private shell

#

which is good for security but not good for capability

#

unless they use an API of some sort I don't think so

winged bluff
#

getting very annoying it saying error everytime 3 generations or not generating.

stark wadi
#

can anyone point me to a good training on preparing a knowledge base for a custom gpt

cobalt lynx
stark wadi
tight island
# stark wadi tried.

Did you do it properly what format, what file types, how do you do it, how do you structure it or write it, do you write the instructions properly and name the file accordingly?

#

…these are all important questions

stark wadi
#

pdf

tight island
#

did you do markdown? JSON??

#

is what i meant by doing it properly

stark wadi
#

no i thought i didnt need to. i did the custom gpt page

#

i uploaded it to chatgpt

tight island
#

all of my knowlegde files are in a json string in a json filetype with an indent of 4.

#

all of them are referenced in an in-instruciton overview in the instructions, with filename references when to and when not to look for info

#

very simple

ripe egret
#

Anyone else having problems with chatpgt-4 now?

stark wadi
#

so i need to reserach how to put a file in a json string , this is all overr my head. Im trying

tight island
#

did it occur to you to ask GPT to do it for you

#

just a good tip.

stark wadi
#

again i have asked

tight island
#

did u ask properly

#

it's all about How you ask

stark wadi
#

and tried it just isnt giving me the right info

tight island
#

how did u tell it.

ripe egret
#

I'm only get this response from chatgpt-4: Something went wrong. If this issue persists please contact us through our help center at help.openai.com.

tight island
#

what i usually do is upload the desired pdf, Have the GPT Analyze all headings subheadings, titles etc, and create a json string for them as a fill-in-yourself, then u just fill in the info manually by copypaste.

tight island
stark wadi
#

Steps to Customize a ChatGPT Model

  1. Understand the Basics

Model Foundation: Start with a model like OpenAI's GPT-3 or GPT-4, which are general-purpose models trained on a diverse range of internet text.
API Access: Gain access to OpenAI’s API, which allows you to interact with GPT models programmatically.
2. Define the Customization

Scope of Knowledge: Decide on the specific areas or topics you want the model to be knowledgeable about, such as African American arts, history, or social issues.
Use Case: Determine how you want to use the model — for educational tools, interactive experiences, content creation, etc.
3. Data Collection and Training

#

this just isnt right

tight island
#

is it "MyGPT" or API

stark wadi
#

chatgpt gave me that

ripe egret
stark wadi
#

custom gpt

tight island
#

great Vibrania, the issue you are having is because GPT DONT know what you are talking about, its assuming its a Python or Coding GPT utilizing the API

stark wadi
#

ohhhhhh

tight island
#

u need to directly and EXPLICITLY spell out that its "MyGPT" for the Store which has a user interface with no backend access.

#

How long is your document Vib? page wise. ish

stark wadi
#

but how dow I do the backend knowlefge base? I uploaded 3 pdf's that contain state laws.

tight island
#

u dont

#

your "Knowlegde" is your "Backend" access.

#

thats all you get.

stark wadi
#

is the knowledge base need to be in a certain format?

tight island
#

preferably JSON

#

its the easiest for the model to look up and reference.

#

without consuming too many tokens.

#

thats the issue with plaintext, is that it reads the entire thing

stark wadi
#

where do i go to translate my pdf's into json

tight island
#

technically you dont

#

its often done by hand

#

U can however DM me so we dont take up this chat in its entirety.

#

then i can give you more personalized help

slender shore
#

Is gpt 4 down for anyone else?

flint bolt
#

For me it's not working for ca. 10 minutes now

full folio
#

not getting any responses for both 3.5/4 the white circle just pulsates with no activity, waiting on almost 15 mins now

winged bluff
#

its been having problems all day

full folio
outer widget
#

So it is down rn

neat storm
#

nah not for me

#

but i have Plus so idk
might just be high usage

pliant socket
#

conversations not loading

rustic night
#

conversations and responses arent loading for me and i have plus so using 4 etc

vague mica
#

is chatGPT down ?

violet bloom
vague mica
#

aghhhhh

#

i need to do a tes Righ now

dry crow
#

gpt down?

dense halo
narrow galleon
#

no history here either

violet bloom
#

It's just incredible, that OpenAI is still incompetent to get the service up and running after all the time.

dry crow
#

second time in less than a month...

violet bloom
dense halo
#

why is chatgpt messages taking like forevereverevereverever to generate

violet bloom
frank jacinth
dense halo
violet bloom
spare kiln
#

I have been having the issue of it hanging for days. 😭

tidal light
#

is ChatGPT down rn??? because my chats wont load

dense halo
frank jacinth
tidal light
#

great

errant pebble
#

GG

dense halo
#

bruhghghh

spare kiln
brazen crater
proud matrix
spare kiln
#

For everyone else has it just been today?

proud matrix
#

It was working less than an hour ago

dense halo
#

cnsdjcndsobfvhiyebvihycbdfiswcbhnefv

modern basin
#

ChatGPT down

minor moon
#

They need to stop releasing new features for a while and fix the stuff they already got, like chat, just make it reliable and actually do what it's supposed to

violet bloom
#

yeh, they "investigate"

narrow galleon
#

I'm using chaggpt right now, and my history showed up

#

I started chatting to it, and eventually my sidebar history appeared

neat storm
#

the problem is they used ChatGPT to design their server infrastructure so it's prone to failure

golden sleet
#

Guys when training AI, if I reply using information that the bot doesn't have access to, it can teach it to become prone to dreaming up information and making things up right? But how else do I train it to know about itself?

For instance if I want it to always answer that it's favourite colour is purple, would I just do 3-4 examples of it being asked about it's favourite colour with a variety of replies saying "Purple" or is this going to hurt the model in the long run? If so.. How are you meant to train things like characteristics into it?

neat storm
#

Well, no
The AI doesn't actively learn

ivory iris
golden sleet
#

Sorry the Fine-tuning

neat storm
#

It's responses are based on previous conversations you've had and the currently open conversation
If you want specific instructions like a custom GPT, you need to specify that in the model instructions

ivory iris
#

ok that's what I was thinking you meant. what do you mean when you say "train it to know about itself"?

eternal rockBOT
#
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.
golden sleet
#

If there is something better for that other than fine tuning, I'd be happy to be pointed in the right direction though

ivory iris
#

you can add all that to the guidance for a customGPT

golden sleet
#

Is guidance the system thing?

neat storm
#

are you training like your own actual AI/LM or are you using the My GPT's feature

golden sleet
#

I am using the Fine-Tuning with GPT3.5

neat storm
#

ah

ivory iris
#

it's the "instructions" when defining a custom gpt. If you are referring to the API, I'm not sure.

neat storm
#

the Custom Instructions section?

golden sleet
#

Where is that? All I see is a place to upload the .jsonl for training and validation

ivory iris
#

it's a part of the customGPT framework. you can add specific instructions for the particular GPT. It does use GPT4 so you will have the limits that come with it

neat storm
#

huh

#

neato
idk how allat works

ivory iris
#

either you can manually create the instructions or there's a tab with a chatgpt helper to create it.

#

if you are using the API, I would imagine it's similar to the system or role part of the data. You will have to make sure it gets refreshed every so often or it will slowly decouple from the character.

golden sleet