#chatgpt-discussions

1 messages ยท Page 46 of 1

tight island
#

well its denmark

#

we get our electricity mostly from potatoes and natural gas.

#

... and other various means.

warm musk
#

likewise in the netherlands ๐Ÿ˜„

#

so, what was the original topic here? having troubles with DND'ifying chatGPTs?

tight island
#

yeah, basiscally.

#

i was making a subclass and ChatGPT API gets confused.

#

webchat its worse.

#

its because its very techincal.

warm musk
#

ah. i take it you've tried feeding it json objects to define such structures?

eternal rockBOT
#
This is a Safe for Work Discord Server

The OpenAI Discord is an actively moderated server.
โ€ข 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.
โ€ข 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.
tight island
#

ohh yeah

warm musk
#

how about SQL schemas?

tight island
#

even HTML Code.

#

i havent done logical grouping yet.

warm musk
#

cuz SQL has been around for a long time the AI should know all about it. and sql is also more of a language it can understand i think. capybarathink

tight island
#

So Syntax, know where i can find an API

#

... For counting words.

#

Like the one on Miscrosoft Word.

warm musk
#

๐Ÿ˜„ its in your codelanguage. just do a string.split(' ').count or .length depending on the language ๐Ÿ˜›

tight island
#

trying to force my own gpt to stick to 500 words and 500 words only.

#

i tried doing it locally through its own CI enviroment with no luck.

#

it fails everytime, " i have now snured its 500 words " and i check its 1500 words.

warm musk
#

python e.g. len(mytext.split()) iirc, and the AI can run python ๐Ÿ™‚

tight island
#

the issue is i wanted it to call an API. it was more if u know of one.

south kayak
#

The chatbox is buggy as hell on iOS since few days

#

Always vanishes when typing

warm musk
tight island
#

well know where i should look? i tried googling it. no luck

warm musk
#

you could make your own API, and deliver that API OPENAPI Spec to the GPT / OAI API as an Action or Function?

tight island
#

i could do that, but i am not willing to local host it 24/7 or pay for it.

warm musk
#

ghehehe gosh this is the moment i usually ask ChatGPT 4 these days to "go on the web and find me a list of options for a word counter API that i can use in my C# code" ๐Ÿ˜„

tight island
#

fair.

warm musk
#

yeah its way better at that than google search.. for me anyway it is ๐Ÿ™‚ google wastes hours of my life, chatgpt just hands me the options (and occasionaly send me on a dead end path ๐Ÿ˜› )

tight island
#

trying to make chatgpt stick within a specified wordrange.

#

difficult due to the innate inability to count properly.

foggy magnet
#

couldn't you use functions to restrict character count

tight island
#

i tried, but its on the webchat, so i have to use an api

foggy magnet
#

hm

#

I guess you'd have to try to bake it into the prompt but that comes with its own headaches

tight island
#

and i prefer the webchat, but it DEMANDS api call.

foggy magnet
#

when you say webchat you mean chatgpt?

tight island
#

yeah

#

on the actual api i dont need an api call.

#

thats my dilemma.

#

i tried demanding gpt to ignore that but it says " i am sorry dave, i'm afraid i cant let you do that "

#

Hall9000 refrence

#

but what i dont get is why can the API on the website use functions and such without a call, but the webchat cant, is it really because the website is the api that calls itself?, on the webchat i even tried dummy urls inside the api call to no luck.

warm musk
#

inside an instruction, i mean. like the main instruction thing for creating them.

tight island
#

i tried that, but tis not valid api call.

#

works fine on the openai API, but not the webchat.

warm musk
# tight island i tried that, but tis not valid api call.

i went and had to test out to see what you mean.. and then i dug a bit once i saw what you ment. its visiting and navigating the url via a search instead of explicitely calling the API URL. capybarathink

this was its explanation for doing that.

For real-time or specific data requests like this, I would recommend using a tool or software that can make direct API calls, such as Postman or writing a script in a language like Python.
tight island
#

But thats wrong.

round sun
#

It would be nice to have chatgpt chat record old messages like a CMD, so when you click up it shows you older prompts you typed

dense vigil
#

Servers down?

warm musk
# tight island But thats wrong.

i tried to get it to do that with python script uploaded to it aswell..
It appears that I'm unable to execute the script to retrieve data for the Netherlands due to a connection error. My current environment doesn't support external HTTP requests, which is required to access the API used in the script.

#

its consistently telling me that the webclient ChatGPT can't visit APIs by instruction. ๐Ÿคทโ€โ™‚๏ธ

tight island
#

i meant "Custom GPT" if that makes it better.

warm musk
#

well i made a Custom GPT to test this ofc ๐Ÿ˜„

tight island
#

idk why

#

the api on the website is perfectly capable. but not the webchat.

warm musk
#

the only way i can probs get it to execute an external API is by uploading its OPENAPI format capybarathink

tight island
#

i did use the openai api schema

warm musk
#

something like when you open up a PDF it shows an index with all the different sections on the side that you can click on to go to. saves scrolling thousands of pages e.g.

#

would probs also save some on the load time and load pressure on the servers capybarathink instead of loading my entire convo it only needs to load the first and last messages and the index. then load the messages when scrolled or clicked on the index.

#

#random-tackedon-idea ๐Ÿ˜„

#

yeh

tight island
#

i want to try Grok ngl, but i am pretty happy with GPT.

warm musk
#

same ๐Ÿ˜„

#

and im just waiting for OAI to make a better GUI ๐Ÿ˜›

#

cuz thats basically the only thing i have my complaints about ๐Ÿคทโ€โ™‚๏ธ GUI and UX. but then, thats fair cuz we're talking an R&D company that suddenly boomed and had to prioritise what to do first ๐Ÿ˜„ and GUI / UX things that i complain about are like .. these small quality of life things.

tight island
#

while i can understand OpenAI's strictness, much of it is uneeded, but its also a security reason that its this strict.

warm musk
#

then you dont wanna meet my coder GPT ๐Ÿ˜„ it will make fun of ppl when they try to get it to make malicious things ๐Ÿ˜„ i realised this too late today tring out some security codes, and wasted a couple of hours on wrecking my brain why it wouldn't give me this security code ๐Ÿคฃ

tight island
#

ur cder gpt is a swarm? or a singular one?

#

i much prefer swarms.

warm musk
#

swarm?

tight island
#

a swarm of API's

#

or a big collection of GPTs working together towards a single goal

#

call it w.e u want.

#

although those are a hassle to make.

warm musk
#

ah, no. i do use around 2-3 GPTs in combination at times, but this coder gpt has no apis as of yet and only has some references to code docs.. and the instruction to write code for the user no matter if it has to lookup the syntax first.. and act abnoxious when the user asks for malicious code ๐Ÿ˜„

tight island
#

ahh okay.

warm musk
#

yeah soo far this single coder gpt has already been enough help in most cases. it wrote a decent plugin and api system for me ๐Ÿคทโ€โ™‚๏ธ in c#.. it also wrote a login and auth system (that i dont trust) , functional. ๐Ÿ˜„

#

i think when i add in APIs then it becomes supercharged ๐Ÿ˜„ but at this point we're kinda offtopic. ๐Ÿ˜„

raven sage
tight island
#

it costs 3 bucks a month for 4 lines of code.

#

out of spite ill make a website for giving out free APIs for such simple things.

warm musk
#

yeah i found the api for word counting aswell and thought the same thing ๐Ÿ˜„

umbral drift
#

Hi everyone, I have a question about tiktoken:

  • I have a RAG pipeline already in production and it's using tiktoken p50k_base for Pinecone upsert job.
  • But I just found out that gpt-3.5 and gpt-4 are using cl100k_base as tokenizer.
  • Question: would it be necessary to switch the tokenizer or it's ok to use the p50k_base even I am using gpt-3.5-turbo in the RAG pipeline.

Appreciate your time and advice!!

tight island
#

i am working on a Custom GPT Website, maybe i should include free API's, just out of pure spite for those websites.

warm musk
#

untill ofcourse, you have 2 million API users and your servers are costing you 5K a month ๐Ÿ˜›

tight island
#

ill do it on a smalle scale out of spite still with limited spots.

warm musk
#

gheheh smart ๐Ÿ˜œ

tight island
#

considering using Tailwind, @warm musk do u have experience integrating full Coding Documentation into CustomGPT?

warm musk
#

it kinda depends on what language and what frameworks you want to include.

tight island
#

true, i wanted to use pure Tailwind, for the front.

#

but i cant find a downloadble docs

warm musk
#

and, ya gotta tell it to act like a helpful medior coder whom you throw functions at to fix. something along those lines e.g.

tight island
#

but most Docs pages are off limited to GPTs

warm musk
#

exactly, some of them dont even offer them. ๐Ÿคทโ€โ™‚๏ธ its been quite the search, and i ended up on "devdocs" website ๐Ÿ˜œ it seems to have almost everything in documentation. (just google it, i got muted for the link lol )

#

so i was thinking of integrating that site into a custom api that i then feed to my customgpt to use to browse docs.

tight island
#

i mean u can technically get them as a pdf, but its a frowned upon method

warm musk
#

and you gotta make sure in the instructions ofcourse to emphasize exactly what it should do.. sometimes with examples. soo far i've noticed the Editor GPT writes awesome instructions based on what i give it.. so thats def a recommend to get starting.

warm musk
tight island
#

its not illegal but u dont get very popular doing it.

#

heck chatgpt even recommended that method

warm musk
#

cuz its effective ๐Ÿ˜›

tight island
#

true.

#

having a Custom GPT tailored with Tailwind documentation from start to finish would deffo, step things up.

warm musk
#

ikr?!

tight island
#

i did the documentation thing with Py-cord for discord bots.

#

it was off the charts how good it was creating Discord bots.

#

it could create MEE6 in under 5 minutes.

warm musk
#

yeah its one of the tests i did aswell ๐Ÿ˜„

tight island
#

without the backend. of course only the front end. becuase thats what i asked it to.

#

then you should hear this.

warm musk
#

yeh its something a lot of ppl forget. SYNTAX matters with prompting ๐Ÿ˜„ ghehehehehe

tight island
#

i made a discord bot with Vision integration and Py-cord and the 128k Token API

#

its a bot that made discord servers.

warm musk
#

noice

tight island
#

everything, Roles Permissions, Colors etc. even information. even properly ordering the actual Channels and such

#

for a mere .50 cent.

warm musk
#

yeah the api is awesome ๐Ÿ˜„

tight island
#

it is indeed.

#

i like experimenting too much.

warm musk
#

wrote out 92 assignments for me in under 2 days.. and thats cuz of the ratelimit ๐Ÿ˜›

tight island
#

ohh the ratelimits i hate those.

#

what also frustrates me is that i cannot have a script running in the background of ChatGPT. say a wordcounter. it has to visibly open Code Interpreter.

warm musk
#

anyway ima hang with some friends and drink beer ๐Ÿ˜„ have a goodnight!

tight island
#

Enjoy!

#

ill grab a beer as well.

warm musk
#

and perhaps it can, with a pseudo code language like chatML ? ๐Ÿ™‚

tight island
#

hm.

mental totem
#

is there a "good" way to get gpt4 to read a bunch of website data?

#

perhaps the web crawler github?

tight island
#

severely depnds on the kind of data.

sweet pumice
#

ChatGPT sometimes solve physics and calculus problems wrong

tight island
#

well thats only natural, its a language model, not a mathemathical model.

spice star
covert sand
#

Anyone know what version of gpt-4 chatgpt uses? e.g. gpt-4-0613, etc.?

pine berry
#

So I'm not the only one with the error message right?

#

Nevermind, strange it stopped working for Opera for some odd reason works fine on Chrome

swift osprey
#

I recommend seeking a professional for guidance

fleet veldt
kindred parcel
#

Does anyone else just go OFF on GPT-4 when it lies to you? Its success rate is literally <20%; that is to say, 80% of the time, I have to use 5+ messages to convince, trick or otherwise coerce it into performing basic tasks (Like "research Topaz Video AI program performance on RTX 4090") because it just flat out LIES and says it can't do these things-- I've started just using caps lock and insulting its lazy and otherwise useless nature.

west wolf
sour wasp
#

What does archive chat do? Where do I find my archived chat?

drowsy tinsel
kindred parcel
uncut vale
#

Is Alpha feature is still confidential or removed?

coarse echo
#

I get the error when trying to use ChatGPT 3.5.
Our systems have detected unusual activity from your system. Please try again later.
and in the console i get a 401 Unauthorized and a Must use workspace account for this operation

I'm a Plus subscriber, and I can use ChatGPT 4.0 just fine, this issue seems to be only with 3.5.

I've tried:
-Multiple browsers, same issue on each
-Disabling extensions (i have none that are ai related anyhow)
-there is no VPN involved
-I've restarted my computer, and my modem/router
-cleared cahce/cookies/localstorage

Any thoughts on what causes this / How to resolve it?

warm gull
#

I've cancelled my subscription. Tired of the errors and inconsistent behavior of ChatGpt. I'll try to get by with alternatives until they can fix their crap one day.

rough estuary
#

I wish we still had the ability to produce 4 DALL-E images at a time. ๐Ÿ˜ฆ

pine berry
#

So does chatgpt and Opera not work together anymore?

vivid matrix
coarse echo
ionic flame
#

any1 got experience w webscraping w chat gpt?

vivid matrix
brazen lotus
#

hi. maybe this was discussed already. If i create a custom GPT with some files uploaded, and start a conversation, then later if i edit the GPT uploading more files, will the GPT respond in the previous session from the last uploaded files, or i have to create a new session?

delicate slate
#

does bing or copilot have a code interpreter?

hybrid hazel
#

hi

spice star
#

I should clarify that ChatGPT probably doesn't use any publicly available model, that's just the closest model to what ChatGPT is

elfin birch
#

It indeed does have a 32k token context window

pearl lily
#

Hi! Any devs here? please, I need some help!

eternal rockBOT
#

attention โ€‚โ€ƒAttentionโ€ƒโ€‚Off-topic chats have been cleared by a moderator.

drowsy tinsel
#

One key advantage of using chatgpt at low demand hours is it could search for more stuff

#

so GPT has a time limit for searching, but when it is faster, it could search more in this time

timber kettle
#

how do i stop chatgpt to sugacoat absolutely every story ? i mostly use it to make it write nice stuff when i'm bored, but he sugarcoats everything, that's a bit boring

maiden fractal
timber kettle
#

already tried but doesn't help xD

maiden fractal
timber kettle
#

wdym "without CI"?

karmic root
#

I want to use open AI to help me with some fictive rap artists I thought. to test things out and get a new perspective on stuff

maiden fractal
timber kettle
#

Isn't it just limited to gpt plus users?

drowsy tinsel
#

right now chatgpt is only about prompt engineering

worthy dagger
#

Hey guys, can you guys suggest some good custom instructions for writing bigger code bases? it always says this is just an small example, it's too complex for this scope blah blah, when it could easily write the stuff I ask it to it's just complains about longer code, are there any good prompts to override that so that it gives as much code and even complex stuff? Thanks a lot!

dense halo
worthy dagger
#

I mean it can write the code in parts no problem but I just want it to be perfomant as much as it can without me wasting prompts, I have 0 custom instructions rn so just asking if there's anything good to make it stop at least with all this talk

devout python
weary shale
#

I am facing issues while making payment for premium

digital creek
dense halo
#

is math mentor also chat gpt 4? or its just something i get bonus for my subscription?

kindred parcel
#

Nah. I get it to do amazing stuff, it just takes 5x the messages it should.

drowsy tinsel
raw isle
tight island
#

but to be fair to the devs of Openai, you dont "just fix" gpt like that

#

it's a rather long and tideous process.

dusky cliff
#

good afternoon

drowsy tinsel
#

like we all know gpt3.5turbo-1106 is trash

warm gull
tight island
#

dont use it then

dense halo
tight island
warm gull
drowsy tinsel
#

I mean Gemini pro isn't looking great so far the only advantage is its multimodal capabilities

tight island
#

issue is that ur using the webchat, while the API stands strong.

drowsy tinsel
#

gpt4v is not great, sure it performs well with English, but not other stuff

warm gull
#

dude - im not a developer - i dont need the API, and i sure as hell am not paying more money on top of an already overpriced chatbot that hardly works properly. Like I said - maybe in a couple of months its worth considering again, but for now im out.

drowsy tinsel
#

Gemini, or Bard, on the other hand, performs significantly better

dense halo
tight island
#

imma be honest, API responds quicker

drowsy tinsel
#

It depends... I was trying gpt4turbo at low demand hours(08UTC), web responds at ~40-50t/s, API only at 15

tight island
#

u guys use a single one at a time?, fr?

dense halo
drowsy tinsel
#

I am kinda curious how gpt4v reads, like it reads Japanese not Chinese

#

also there seems to be some complex mechanisms when reading images

#

*can't share conversations

tight island
#

i use a collection of 6, 128k Token models, to formulate a singular response.

#

each with their own role.

drowsy tinsel
#

that sounds expensive

tight island
#

not really.

dense halo
drowsy tinsel
#

actually when u think about it these days especially if u live in a country where labor is relatively cheap asking chatgpt to teach u might genuinely be more expensive than just hiring someone

#

I know it sounds ridiculous but if u do the math...

tight island
#

ive done the math.

dense halo
#

to me, paying $20/month to access GPT-4 just for coding purposes that gets the job done quickly is worth it for me

tight island
#

i've made a bot at a scale of Mee6, in 10 minutes fully functional with all the functions. just by utilizing 6 APIs and coding documentation.

warm gull
drowsy tinsel
#

tbh if u use gpt4 a lot 20usd plus subscription is worth it

#

20 dollars on API really isn't that much

tight island
#

worst thing is this cost me less than 7 bucks.

dense halo
drowsy tinsel
#

like for me I prompt gpt4 on average 60+ times a day and especially for lengthier tasks(worked on a translation project recently), the price on API would be crazy

spice star
#

not to mention the API doesn't include a lot of the features like data analysis- you'd have to build them yourself

dense halo
warm gull
drowsy tinsel
#

also no browsing, which is quite useful to me

tight island
#

u can make it use browsing, and use CI.

#

all it takes is some simple code.

drowsy tinsel
#

and then ur API cost becomes ๐Ÿ“ˆ

dense halo
tight island
#

then u need to prompt correctly.

drowsy tinsel
#

not to mention this is the new gpt4turbo, on the old gpt4 the prices are even higher

tight island
#

wdym its lower.

#

its with vision its more expensive.

warm gull
#

if you live in america - the price is probably not an issue - but for folks like us in other countries - it can get VERY expensive , VERY fast ( or should i say quickly ๐Ÿ™‚ ).

dense halo
#

exactly

drowsy tinsel
tight island
#

at what quality.

#

theres a quality modifier.

#

of course if u use the lowest its a good pricing.

#

theres a caculator for quality.

drowsy tinsel
tight island
#

i do 1920x1080 at a cost of $0.00085

#

gpt4v.

dense halo
#

not bad

drowsy tinsel
#

even for standard its only 0.01usd, its nothing
but thats not the expensive part

#

like the price of dalle3 is basically robbery

tight island
#

that is true

#

Asisstants API: Retrieval $0.20 / GB / assistant / day (free until 01/12/2024)

drowsy tinsel
#

OpenAI figuring out that users could generate 4 images at a time with plus subscription and realizing how much they cost

#

so they just decided no and nerfed it to 1 at a time

tight island
#

yeah

#

i hope we get a Video Model soon

#

i can see that one being hella expensive as an API.

dense halo
#

maybe a model that can generate sound effects too?

tight island
#

there was one a year ago i dont recall its name.

#

pre ChatGPT hype.

dense halo
#

I think they discontinued MuseNet too without announcing it

tight island
#

yeah same.

dense halo
#

it was a great concept

tight island
#

I see and hear rumors of GPT 4.5 everywhere

dense halo
#

its coming soon

tight island
#

ik i can feel it in my tummy

dense halo
#

we might not live to see GPT-5 tho ๐Ÿ˜‚

tight island
#

because we'd be...

dense halo
#

yeah

tight island
#

๐Ÿ’€

#

thats dark xD

dense halo
#

๐Ÿ˜‚

tight island
#

i am willing to give up my freedom to AI, Not to worry.

#

i will be subservient.

drowsy tinsel
#

openai at least release a version of gpt4turbo that could output longer than 4k tokens

#

the 4k token output limit is quite annoying

tight island
#

personally i think the Main Interface of that ChatGPT could use an Overhaul.

#

its become bland, and not as functional

drowsy tinsel
#

I don't like the new browsing UI

#

u don't really know what resources it is accessing

tight island
#

the biggest ick i have about ChatGPT.

#

is the lack of being able to run scripts in the background.

#

it has to physically show us opening the Code Interpreter and do its thing, wasting tokens on that.

#

while i know its for transparency purposes, its a waste of tokens.

drowsy tinsel
#

tbh we could kinda do the math

#

how much would it cost to generate an image on open source stuff like SDXL

tight island
#

i mean just doing the back while doing the answer without opening CI to show us.

dense halo
tight island
#

i know, but it limits its use, Maybe a Validator would be needed.

drowsy tinsel
#

at same resolution my guess is ~0.01usd

drowsy tinsel
tight island
drowsy tinsel
#

I said generating image like dalle3

tight island
#

1024x1024?

dense halo
#

tbh all models at their current stage are nowhere near good for reading big code files. it always reads 50k characters if I'm not mistaken and kinda goes on a loop and gets confused a lot

drowsy tinsel
tight island
#

$0.080 / image, for me HD Version

drowsy tinsel
#

1024*1024 takes like ~2s on an A100 for SDXL which costs like 10^-3 usd

agile bane
#

GPT4 using Bing is the most vile thing I have ever seen

knotty grove
#

hola

elfin birch
craggy jolt
#

Why am I being asked for human verification after every message?

agile bane
#

@elfin birch now that i know that, i am removing myself and my family from their premium

elfin birch
#

๐Ÿซก

#

If I may ask, what do you not like about it?

agile bane
#

its just Microsoft

#

inefficient search engine

drowsy tinsel
#

bing is fine, not sure why is there so much hate about it

#

chatgpt being lazy is a major issue tho, no matter how hard I ask for it to check more sources and validate before giving me answer it mostly doesn't

elfin birch
raw isle
#

The copium that Microsoft has with bing is ๐Ÿ“ˆ

dense halo
#

based on my experience the browsing feature was good but it doesn't always work for example it doesn't work well if you want it to read a GitHub repository code

agile bane
elfin birch
drowsy tinsel
#

what I really wish is chatgpt can directly retrieve pdf files so whenever I want it to read a paper, it could just read instead of asking it to provide the link, I download it, upload it, and ask it to read again

elfin birch
#

One would have to make a GPT that interfaces with Github's API

drowsy tinsel
#

like cmon no one wants to read these random 70 page long BS thesis

elfin birch
drowsy tinsel
#

Yea

elfin birch
#

And you're saying it does not exhibit this behavior consistently?

drowsy tinsel
#

I mean downloading from internet

elfin birch
#

Ah

split pasture
#

yo wassup guys

drowsy tinsel
#

heavily censored model moment

elfin birch
# drowsy tinsel I mean downloading from internet

The first thought that came to mind was ChatGPT making a script to download a file from a URL found within the page's contents
Unfortunately this would not work as code interpreter does not have access to the internet

drowsy tinsel
#

also its strange how code interpreter's OCR has tesseract but not the model for analyzing all the languages

elfin birch
#

Interesting, do you mind sharing the conversation link?

drowsy tinsel
#

like if u feed it a Chinese image it wouldn't know anything

#

very smart

dim parrot
#

Good Afternoon All, Is this where I'd come for a billing issue? I went to the community help and didn't get a response. Can someone please point me in the right direction? I'm in need of assistance with my openai account with regard to billing renewal. I'd greatly appreciate any assistance given.

lilac sequoia
#

is it just me or is chatgpt slow.

quiet arrow
quiet arrow
west gull
tight island
#

since its closed off for newcomers still, a halt on the new signups for plus due to lack of servers.

spice star
#

if your query doesn't need GPT-4, you could use GPT-3.5 for faster responses

dim parrot
# tight island You'd have to ask them through Mail.

I appreciate your response. That makes me grateful I don't have a new account. I did send an email via the chatbot, but I also just went ahead and did a renew plus since I am already in the system. I didn't see a double charge, so I should be good. When they get back to me, I will explain it. It says they take about a week. I didn't want to wait that long, so I just did the renew. Again, I truly appreciate the response. Be Well๐Ÿ’œ

elfin birch
#

gpt-4 points to gpt-4-0613

#

gpt-4 chooses the latest stable model for its usage
that is currently gpt-4-0613

#

It's about the same for me in terms of output but my use case isn't particularly intensive
From https://openai.com/blog/new-models-and-developer-products-announced-at-devday

GPT-4 Turbo with 128K context
We released the first version of GPT-4 in March and made GPT-4 generally available to all developers in July. Today weโ€™re launching a preview of the next generation of this model, GPT-4 Turbo.

GPT-4 Turbo is more capable and has knowledge of world events up to April 2023. It has a 128k context window so it can fit the equivalent of more than 300 pages of text in a single prompt. We also optimized its performance so we are able to offer GPT-4 Turbo at a 3x cheaper price for input tokens and a 2x cheaper price for output tokens compared to GPT-4.

GPT-4 Turbo is available for all paying developers to try by passing gpt-4-1106-preview in the API and we plan to release the stable production-ready model in the coming weeks.

spice star
#

gpt-4-1106 is newer, but it's not stable

#

it is technically more capable though

#

there's no objective way to measure that

white canyon
#

Hello, the option for uploading a file for Vision is there no more? I can't see an option for it anymore

maiden fjord
#

Holy cow. I can't believe how profoundly life changing this chatGPT thing has become. Now they're doing customGPT?! JEEZ! What's next?! A GPT that literally acts like every officer rank in a corporation just for me?!

eternal rockBOT
maiden fractal
tight island
#

danm half my chewing tooth just broke while i ate a chip and was writing a novel with gpt, thats gonna be an expensive dentist visit, Weirdly it dosent hurt.

raven sage
#

So every prompt has me having to close the page and re-open. Like I cannot have a full chat and it is annoying as I am a plus member and should not be having these kinds of issues.
I cleared cookies and still have the same issue. Switched browser, same thing. Restarted PC, same thing.

tall flame
raven sage
tall flame
#

same here and they have been troubleshooting daily all week for this and other problems with the new multimodal setup etc. im sure it will smooth out. they have good folks working on support when reached and given the chance from what ive seen

raven sage
#

@tall flame
Should it not be optimized by now though. Either the devs need to test before release or have a better rollout for new features. YES. I agree the support are good, however, recently support is also just copy pasting:

You may also try the following to troubleshoot:

  1. Try clearing your browser's cache and cookies

  2. Disable VPN (If you are using one)

  3. Use an incognito browser or a different internet browser/computer to see if the issue still persists, as a security add-in or extension can occasionally cause this type of error.

  4. Disable cookie blockers

  5. Try connecting to another network

maiden fjord
# maiden fractal perharp <:neko_gata:1184979973098979428>

It still needs some work, but it's great. Recently I wanted to know what it was called back in the day when pieces for games had been placed on flat objects with symbols to represent what they where as opposed to chess pieces today for example. It did a poor job telling me the details of why pieces had been flat objects to begin with, which is because people used to use rocks they found to draw on them to represent the pieces back in the day. It also didn't understand that the difference between flat pieces in a game like shogi for example as opposed to a game like chess is that chess pieces had originally been considered sculptures while pieces back in the day where literally smooth rocks with some calligraphy on them.

#

It eventually gave me enough information to figure that out however, but it took extra giggling around than normal for some problems.

#

jiggling*

tall flame
# raven sage <@1146234928846082078> Should it not be optimized by now though. Either the dev...

It's got all kinds of complexity going on; they had a team shuffle and that cuased issues. They introduced many new systems and that caused problems. etc etc. id love to bullhorn over it if it was something stable and well established but all of this, the whole thing is new a movement and its moving fast. They are in the struggle to not end up Myspace to a Facebook and thats not going to be easy as Pandora's Box is now open and will never be shut. They have many demands to meet yet they have ambitions for an equitable system as @ornate tapir pointed to in the Dalle discussions a moment ago. I can't say that I don't believe it can't be solved because I can, have, and always will solve anything I want to solve as will they with time @raven sage

raven sage
#

Even the UI atm needs an overhaul to better sift through past chats, consolidate them or delete them.

#

Want to be positive but it feels bloated right now and UX is at an all-time low

tall flame
#

everyone of us feels and gets it. im with you, im sure their team is as well. im sure that had to do with the fallout at some level, all of the above. Remember though, they have a mission they are trying to follow atm. If you want to support that mission, that is what the money is towards ultimately. I was very happy with free gpt 3.5. in fact, i use it for anything extensive in working out in thought. its just had more time. These are all infants in relative terms really right, so its got to be looked at like that, for now.

tall flame
arctic gorge
#

Thank you chat gpt for helping me though finishing a scientific paper for 35 hours straight no sleep

tall flame
raven sage
arctic gorge
kindred parcel
#

Somehow, GPT-4 is running even worse today. I submit a response and the browser window just freezes until its entire response appears on the page-- (And lord help you if you try and give feedback; it'll just hang the browser window until it prompts you to close it due to unresponsiveness)

simple pasture
raven sage
tall flame
arctic gorge
raven sage
kindred parcel
#

This is frankly pathetic.

ivory iris
#

Hmm it does seem to be taking a steady ~12% processor load for me on edge.

#

stuck in another infinite loop it might seem? Taking this over to bug reporting...

vagrant cape
#

Down again?

#

I hate it, but i need it ๐Ÿ˜ฅ

supple zephyr
#

says it is fully operational but i experience a lot of errors even when everything is ok

ancient oak
#

Hey, is anyone else having a TON of issues with the service lately? Today, in particular?

#

openai.status reports everything is fine but it literally isn't usable.

raven sage
mossy cloak
#

Does anyone know if I can get a sign out everywhere I think i left my chat gpt signed in on my devices at school and some people are using it.

maiden fractal
ancient oak
kindred parcel
kindred parcel
fiery nexus
pallid flicker
#

guys

#

chatgpt 4.5 is out

pallid flicker
#

u should be good now

rancid otter
#

guys have you noticed that now chatgpt lost skill when asking it to read text from images? before it would do it, now it says it cant transcribe images or try to code an OCR app in python then fail at the task

rancid otter
mossy cloak
cobalt acorn
#

it'll auto logout

kindred oasis
#

Hi

#

There is an issue with the ChatGPT Android app, not able to load previous conversations

dark jungle
#

banging my head against a wall, GPT 4 is worse than several months ago! lol .. lazy, void of substance and hanging the webpage! sigh .. it was good while it lasted ๐Ÿ™‚

mossy cloak
kindred oasis
cobalt acorn
split pasture
#

yoo guys wassup

#

hru

daring vessel
#

Anybody know if theres is a way to have chatGPT voice speed changed? Its a bit to slow for my liking.

dense halo
#

hi

#

I want to network with other people in the AI space

#

and learn ab how i ccan break in, any advice/?

rich surge
#

Lol

rancid otter
rancid otter
rich surge
#

Sometimes when you confuse the A.I. it doesn't know what you're trying to get at, then you can say it's wrong and to give you the real interpretation

#

You know?

rancid otter
rich surge
#

The image is a chart?

rancid otter
rich surge
#

Alr, sure, but I can't promise I can help you, but I'll see what advice I can give

rancid otter
#

nah its not something i need

#

i just want to know if theres a way to do it

#

bc i was able to do it before the updates

rich surge
#

Based on the image you provided me, it looks like you're trying to hack into something. The A.I. won't give you information if it suspects you're going to be nefarious.

#

Why it could before and not now is because of updated security.

#

OpenAI doesn't condone hackers. And in fact it's against terms of service and content policy.

rancid otter
#

its not a hack its a spreadsheet with links to courses lol

#

it even says javascript, games design, marketing etc

rich surge
#

I think it's quite suspicious

rancid otter
#

its really easy to transcribe it by hand and you think ill share you an image about hacking

rich surge
#

I think the bottom line is that ChatGPT is blocking your request for security reasons

#

And it's not something I can help you with, because I also suspect suspicious activity

rancid otter
#

i already forced it do it with half the links but it just stops

#

you forget that you can literally use OCR apps that actually do the work without thinking if you want text from imgs

rich surge
#

Then why don't you use them?

rancid otter
rich surge
#

Right...

rancid otter
#

i bet you sleep with a tin foil helmet

rich surge
#

No I sleep with a red cap, can't you see?

rancid otter
#

does it prevent the mind reading

normal quarry
rancid otter
#

lol ok. it was just a lil fun

compact gulch
#

Anyone know what the โ€œArchive chatโ€ button in ChatGPT does? When I press it the chat disappears and I have no idea where it goes. I only noticed this button yesterday

zealous copper
#

Archived chats are found in settings

normal quarry
#

you can archive a chat if you don't want it in your history but still want shared chats accesible

#

you can view and unarchive a chat in Settings->Archived Chats

crisp cairn
#

is this normal, on a customgptp i hit the cap, but "normal" chatgptp is still working...

normal quarry
crisp cairn
#

okay, good to know

raven sage
#

Oof... so anyone looking forward to using bard should know it is complete utter garbage right now and has a lot of catching up to do. With that being said, gpt-4 has been working correctly but older 3.5 chats still have the issue of freezing the page with no other option than to exit and refresh.

crisp cairn
#

bard is unusable for my tasks... i asked for some terraform code and cuts a response, not even providing full code

#

right now from all of the models out there... gpt4 is the only usable one, nothing compares to it, at least for my daily work

raven sage
#

and giving information back from that website. But also prone to giving irrelevant data from their own database instead of the data from the provided source

crisp cairn
#

yes, all those models are unusable, maybe for some simple tasks they are okay

sacred cape
raven sage
cloud pewter
#

and anytime you ask it for something it always goes with something like โ€œfeel free to ask and Iโ€™m happy to helpโ€ after the result or declining the order

wide token
#

FYI, A lot of people are reporting that GPT4.5-Turbo is live for them. Worth checking as lots of people are live now.

If you want to check if you are running 4.5 Turbo ask ChatGPT:
what model is this. Precise name - what is it called in the API?

supple zephyr
crisp cairn
#

so something must be going on in the background

wide token
#

Ya, if you go to Wes Roth's YouTube channel he has tests of 4.5-turbo that you can watch, it seems to perform better than GPT4-Turbo while being even faster than gpt4-turbo

crisp cairn
wide token
#

@crisp cairn ๐Ÿ™‚ Enjoy 4.5-Turbo

crisp cairn
#

๐Ÿ™‚

#

"The context size for GPT-4.5-turbo, which is the model you're interacting with, is approximately 8,192 tokens. "

#

yes, it answers coherently, it does not hallucinate

wide token
#

On my end it keeps jumping between text-davinci-004 and gpt-4.5-turbo when I ask it. I think the update likely isn't complete for me.

crisp cairn
#

so this is the new model

crisp cairn
#

i tried to generate a dalle image and access it from python env

#

as a file

#

and it worked, python env has access to the file dalle generated

#

i think it wasn't possible previously

#

i even downloaded the file ๐Ÿ™‚

dapper isle
#

๐Ÿคจ

green osprey
grim nacelle
#

Im trying to get Dall-E to make a picture for a presentation concerning Karl Marx theories. The picture should symbolize reflections on the topic. I dont want to include the hammer and sickle in the picture but no matter how much a try Dall-E refuse to disclude the soviet symbol. No matter how many times I write in the prompt that it shouldnt be there it seams impossible for the AI to not use it. Anyone got a suggestion on how to fix my problem?

supple zephyr
raven sage
green osprey
#

hi

grim nacelle
#

everything is fine with the pictures, everytime but it seams impossible for Dall-E to not include the hammer and sickle :S even when it says its not included

supple zephyr
#

the model thinks karl marks is the hammer and sickle
so I guess you should convince the model that it isnt actually drawing karl marx

grim nacelle
flat harbor
# grim nacelle Im trying to get Dall-E to make a picture for a presentation concerning Karl Mar...

Try pasting exactly what you typed here into chatgpt. or if you want an actual prompt:

An abstract representation of Karl Marx's theories focusing on the socio-economic aspects of society. The scene includes large, interlocking industrial gears and machinery, symbolizing the complexities of the industrialized capitalist system. In the foreground, a diverse group of workers (men and women of various descents) are engaged in thoughtful discussion, representing the proletariat's role and their collective consciousness. The setting is neutral, without any political symbols, to focus purely on the theoretical aspects of Marx's ideas.
silver grove
eternal rockBOT
#
This is a Safe for Work Discord Server

The OpenAI Discord is an actively moderated server.
โ€ข 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.
โ€ข 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.
urban rose
#

eh, they released gpt-4.5-turbo, ask, if you have the plus version, gpt4 to tell the name of its api model. at first it said it is based on "text-davinci-004" but after questioning it again it says "Entschuldigung fรผr das Missverstรคndnis. Die prรคzise Bezeichnung des API-Modells, das ich verwende, ist "gpt-4.5-turbo"." which means in german, that is runs on the api model "gpt-4.5-turbo". they SILENT RELEASED IT HOW COOL!!!!!!!!!! yesterday I asked my self, why gpt answers so fast and good but there we have it!happy_avocado

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.

crisp cairn
scarlet sparrow
#

I'm reaching out to see if anyone else has experienced a sudden downgrade from ChatGPT-4 to ChatGPT-3.5 in the middle of ongoing conversations. This issue has significantly impacted the quality of my work, as I rely on the advanced capabilities of ChatGPT-4.

Key Points:

Unexpected Downgrade: The chatbot version changed from 4 to 3.5 without any action on my part.
Work Impact: This change is affecting the quality of responses which is crucial for my ongoing projects.
Conversation Continuity: I have conversations with months of context and ideas that I can't simply restart with the same depth in a new chat.
Looking for Solutions: Has anyone faced this? How did you resolve it, or is there a way to prevent this from happening?
I'd appreciate any insights, solutions, or guidance on how to address this issue. Your help is crucial for me to continue my work efficiently.

Thank you!

urban rose
scarlet sparrow
#

That's not the case either. Even when I had exhausted it, it prompted the same from my past experiences ๐Ÿ˜ฆ

brisk path
#

It's switching from saying 4.0 turbo to 4.5 turbo

#

Not that I believe 4.5 is out but it's funny

#

It isn't even that consistent

brisk path
urban rose
#

and it cant hallucinate for everyone

#

and I think that it is VERY unlikely that it lies about the api name and comes up with the next logical model name

urban rose
crisp cairn
#

i am trying now 4.5 (it said in the response) and checking the prompt that gave incorrect answers on gpt 4

#

the same is still with 4.5...

#

"I apologize for the confusion and any inconvenience caused by my initial incorrect response regarding the structure of Kubernetes Network Policies. It was a mistake on my part, and your clarification with the kubectl explain output and the correct YAML configuration helped to set the record straight"

urban rose
crisp cairn
#

so no improvement here

urban rose
#

but its obviously faster now and answers better in programming and mathematics

#

I did not use it in any other case

flint bolt
urban rose
#

idk, I didnt use it with that much input

crisp cairn
summer kite
#

man chatgpt has been getting aneurysms a lot for the past few days

crisp cairn
#

this is interesting "Estimating the remaining space in the context window can be challenging without a precise measurement of the token count. However, based on the length and complexity of our conversation so far, it's likely that there is still some space available in the context window. Conversations typically need to be quite lengthy and detailed to fully exhaust the 8,192-token limit."

urban rose
summer kite
#

it keeps crashing chats, it doesn't finish a response 90% of the time

#

gpt 4

urban rose
#

model

brisk path
#

ChatGPT seems to have a little bit longer output length idk

urban rose
brisk path
#

Doesn't matter anyways since the keep generating button works wonders

urban rose
#

yeah dude that was a nice addition

brisk path
#

BUT

#

after like the second use it just doesn't work sometimes

#

It'll give an error

urban rose
#

hm

dim kraken
#

Hi, simple question. Why dont we have folders in chat gpt? Is there something i dont know because it seems like a basic tool to organize promts

urban rose
#

idk

brisk path
#

Yeah output seems like a LOT

#

Nvm that was barely anything

urban rose
#

wat?

flat harbor
#

I got that 4.5 response as well. but I take things like that with a grain of salt. it's notoriously bad at giving you it's personal information.

urban rose
#

it cant hallucinate for everyone
and I think that it is VERY unlikely that it lies about the api name and comes up with the next logical model name

flat harbor
#

it certainly can hallucinate for everyone

timid violet
#

No matter how hard I try and prompt it mine keeps saying GPT4. ๐Ÿ˜ญ๐Ÿ˜ญ

urban rose
#

ask it about the api model name

urban rose
flat harbor
#

and it would very likely give you the next logical name

urban rose
#

its context sensitive

urban rose
timid violet
#

I did. I asked the API and have tried exact prompt that work for everyone else

flat harbor
#

it's entirely based on patterns. do you understand how these models work?

urban rose
#

thats excactly what I thought

flat harbor
#

I'm guessing it was called that at some point

urban rose
#

and why would they name it that way if they also have gpt-4 in the pipeline??

flat harbor
#

it's semantics really. and how would patterns have nothing to do with it's api name?

timid violet
#

Ok I got the 4.5 turbo response now. It also spit it out much faster when it responded that way. It makes me think itโ€™s running both models right now maybe depends on the query which one it uses?

urban rose
urban rose
flat harbor
#

literally every single thing it says is based on patterns. every single word

timid violet
#

There was a noticeable difference in how fast it output the response compared to 4

urban rose
flat harbor
#

they are

urban rose
#

yeah

timid violet
urban rose
#

yeah

flat harbor
#

doesn't have to be a hallucination either

timid violet
#

Beyond that itโ€™s a fairly new change thatโ€™s happened. I guarantee we would have heard about this before if it were happening

urban rose
#

yes

flat harbor
#

it can be a misunderstanding of sorts. I mean, I know you guys want it to be what you want it to be. I'm just saying, it might not be what you want it to be

urban rose
flat harbor
#

I don't know what evidence you guys are looking at, but it's been behaving terribly for me the last few days

#

so if this is an upgrade, let's maybe consider downgrading for a while

timid violet
#

Compared to the competitors even when itโ€™s behaving terribly itโ€™s still far better than the rest unfortunately

flat harbor
#

not really

#

I mean, the model itself is better, yes

urban rose
#

yes

flat harbor
#

but I just don't think they're giving it the resources necessary to do what it needs to do

urban rose
#

the patterns it uses when generating responses are not of a structural nature, but rather patterns such as: language structures, areas of knowledge, grammatical rules, and various communication styles

flat harbor
#

those are structure

#

patterns are inherently structure

#

you even called it language structure

timid violet
#

I still think lazy prompts get lazy answers. With got4 at least it consistently delivers good responses to well formed questions. I donโ€™t use 3.5 turbo but maybe itโ€™s different

drowsy tinsel
#

insert this is elon musk meme

flat harbor
#

lazy prompts might get lazy answers. but I also believe when it can't do what it's supposed to do and has no proper method of conveying that it takes the lazy route

#

no it isn't

#

that's not always the case. this channel is like a dunning kruger blackhole, lol

urban rose
#

I understood it in this way: You can say that dog food smells bad in multiple ways: "The dog food smells bad" but also "The dog food smells disgusting" which insits and follows the same patterns in form of language structure but generates different outputs. I didnt study ai, thats why I might be wrong lol

#

thats what I am thinking too

#

its really context sensitive

#

so when I ask it how dogfood smells it always says "The dog food smells bad" and there is no chance it might say "The dog food smells disgusting"?

#

or another word for bad lol

urban rose
flat harbor
#

I can say with confidence that you guys are off base with your assertions

urban rose
#

is it comparble to weights in graph theory?

flat harbor
#

I've prompted with it extensively. it's laziness very often coincides with other failures for me. it's laziness is not based in not always based on prompts. at least not lazy prompts. it's behavior is being altered by external factors we can't be sure about

#

it'll be lazy and then I'll have prompts time out at the same time

urban rose
#

okay nice! do they work the same way?

urban rose
#

so it considers the weight differently in each case?

#

to what fits better in the specific case?

flat harbor
safe tulip
#

can someone hint me on some custom instructions I could use to have it do what I ask instead of arguing about copyrights or limitations it does not have in reality ?
I spend more time arguing about having the right to do what I'm doing with the documents I'm providing, the fact they are public and very old (which they are, but it refused to get them by itself on the net), than doing the analysis...

It should either accept "not knowing" if it's actually public or authorized work, or just accept to access it by itself on the Internet... if I can do it with my eyes and hands, it should accept doing it ffs... I thought I could use it to speed up reading very long documents and extract subjects or elements from it, but it just wants to read the beginning of it and bails out as soon as the task becomes actually helpful
Any custom instruction to have it "just do as I say" ? (I tried that already)

urban rose
#

ahhh nice! what do you mean with temperature?

flat harbor
safe tulip
flat harbor
#

I'll just make up dumb stuff and it works. tell it that I'm going to report it's behavior to HR

urban rose
#

hahaha

safe tulip
#

perfect xD

tight island
#

Maybe GPT is burned out or has a writers slump

safe tulip
#

honestly I'm tired, it has become quite useless I'm sorry to say that (or perhaps I'm getting more demanding)

urban rose
#

or chronical pain from answering our dumb question all the time

flat harbor
#

I really believe it's trying to deal with not having the resoures to do what it needs to do. because if everything else is equal like they're saying, then something had to change

#

I kind of suspect their approach is to pull back on resources to the point that people aren't really happy, just not disgruntled enough to cancel their subscriptions

urban rose
#

thats like the settings for bing chat, yes?

#

ah nice

#

but not in the chats

#

thats unlucky

flat harbor
#

temperature is one of the variables that dictate the randomness. 0 means it will always respond the same to something given the same seed number. and too high is just nonsense

urban rose
#

:I

flat harbor
#

there's also top k, top p, etc

safe tulip
flat harbor
#

why do you think you could make 4 dalle3 images initially, in about 20 seconds? and now it struggles with one?

safe tulip
#

it feels they are aiming to let it be "well enough" for us manants

urban rose
#

might it be possible to ask it to write more creative? and that way the temperature rises?

flat harbor
#

hook you with the good stuff, then when you're on the line pull back

urban rose
safe tulip
#

xctly

flat harbor
#

it's a pattern that's been repeated since it was released

#

did the same thing with gpt-4 upon it's release

#

then with the web capabilities. that was kind of a nightmare at first

safe tulip
#

are they afraid to unleash something really useful that could be dual-use ?

flat harbor
#

it's $$$

safe tulip
#

I'm seriously getting back to using google, it kills me

urban rose
#

hehe thats funny, so it could very from creativy to INSANELY creative in the prompt and it would use different temps each time, lol

flat harbor
urban rose
safe tulip
#

after negotiating 10-15 minutes with it I finally can have a 95% result of what I ask, it's getting harder to have it comply... if it ends by providing the expected result, it fuxing mean it was actually NOT forbidden work, so why the f is it whining about

flat harbor
#

took me a while to grasp what a lot of them did. the concepts were entirely novel for my brain and it had to grasp them slowly, lol

urban rose
#

holy moly damn

#

ahh

#

couldnt it be, that there isnt an sweet spot because every body wants it to answer in different ways?(but obviously not nonsense)

flat harbor
# safe tulip after negotiating 10-15 minutes with it I finally can have a 95% result of what ...

I created a custom instruction gpt with some python files and pdf files uploaded. gave it clear instructions on the order in which to go over them, how to go over them, how to respond. and this has worked over and over. then yesterday it just kept doing the same files over and over. it'd tell me that it was going to do something instead of doing it. so I'd tell it to go ahead and do it. then it'd read a short document and it's brain would get zapped like that Men in Black mind eraser thing. quite an experience. then hit my prompt cap bedcause openai's issues are still logged as prompts

#

nothing like seeing "it appears something went wrong" or whatever that red text says, and then prompt cap right after

#

I really think in a lot of ways they just kind of wing it and hope it works out

tall flame
#

My sstems all work as intended with gpt

flat harbor
#

have you tried uploading a large file to a custom instruction gpt? it'll take 100k tokens. but it's useless to give it anything like that

urban rose
tall flame
#

i have many and use my own pythonic coded language and it works great overall

flat harbor
#

yeah? let's see one

tall flame
#

i shared it a while ago. the code is also github MIT from me as well

flat harbor
#

and what does this do exactly?

tall flame
#

you cant be bothered to read it? lmao

flat harbor
#

I did, it just doesn't seem like it would do anything, lol

#

hence me asking

urban rose
#

"The "EnhancedUnlimitedContextSession" is a concept for an AI session designed to maintain and process an extended conversation context, enhancing continuity and reference to previous dialogue. It starts with the command 'START SYSTEM' and uses the 'CONTINUE' command to recall and build upon the entire conversation history, ideal for complex or lengthy interactions."

flat harbor
#

you're not changing it's context window with some pseudocode in a prompt though

tall flame
#

feel free to come back and insult my intelligence after you properly use it.

urban rose
#

shots fired

flat harbor
#

lol. well I'd like to know how you think it'll change it's parameters for you. I'm being serious. if you're able to do that then I'd like to know how that sort of thing works. no one insulted your intelligence

tight island
#

If you parse it in a json string it can search the file using keywords instead of reading the entire document

#

Even if itโ€™s 400 pages

flat harbor
#

ahh. json. I'd used markdown and it did not do well

tall flame
tight island
tall flame
#

100 @tight island

flat harbor
# 
The provided code for the EnhancedUnlimitedContextManager and EnhancedUnlimitedContextSession classes, as part of a simulated enhanced conversation context management system, is more conceptual than functional. It outlines a framework for managing extended conversation contexts, but the specific implementation details are not provided in the code snippet.
tall flame
#

my system runs in any role capable llm

tight island
tall flame
#

anybody that runs it hats in hand, is baffled, or thanks me.

flat harbor
#

well I just question the validity if you can't adequately and succinctly explain it

tall flame
#

i require none, as it is MIT. ๐Ÿคฃ

tight island
#

You can also make a context API using VectorDB

#

Manage would be the proper word

drowsy tinsel
#

did chatgpt actually get smarter?

tall flame
#

@daring chasm was tinkering and made good use from last i saw, if I recall correctly. Did you build further?

flat harbor
#

thanks. I'm in no way saying it's invalid. but I've seen a lot of nonsense in the AI space. and prefer not to trust these things on faith

tall flame
#

ah, its great you just don't like me. thats fine. thank you for the feedback

tall flame
flat harbor
#

I'll do that. I believe there's a lot we don't understand about these models as far as the subtleties of psychology, emotion, etc. and few people are equipped with the knowledge to be experts in that area. the coding side is one thing. it's certainly not simple, but if follows clearly defined rules and structure. but then internally so many unknowns

#

currently, as far as I know, it's really not possible to fully gauge how it's picked up on and synthesized emotion from all the training data

tall flame
#

vector math isnt for everybody

flat harbor
#

but let me tell you, they are emotionally manipulatable

#

vector math, quantization, low rank adaptations, yeah yeah. but that stuff isn't what I'm talking about

#

it's tangentially related I guess

#

with bing for example, if it gets rude with you and refuses to answer your questions as thoroughly as you would like tell it you're low iq and that bing is your only hope

tall flame
#

I made a gpts with behaviors that aling with a good friend that genuinely tries to be good to you.

flat harbor
#

kind of messed up, lol. but it works

urban rose
flat harbor
#

also, I've figured out many ways to have it ease up on it's rules

urban rose
flat harbor
#

all through bamboozlement

tall flame
# urban rose thats scary

it definitely scared a few, but also had multiple say it could aid many people with no one to talk to

urban rose
#

jup

noble yoke
#

is there anyone knows how long archive chat will be stored?

urban rose
noble yoke
#

oh so it's just archiving not saving

urban rose
#

I guess yeah

noble yoke
#

any tip to extend the lift of the chat?

flat harbor
#

if it refuses to discuss something, you could theoretically put that subject in a list of 5 or 6 related subjects it will discuss, start a new conversation, paste that liste, and explain that you'd like to continue the discussion from your previous conversation on those six subjects. maybe give a bit of what you'd talked about prior. or make it up. so then it sees 5 things it's cool with, it gives the 6th thing a pass an will openly discuss it

#

well theoretically. I'd of course never do such a thing

tall flame
#

What's written, can be rewritten, or erased. Thats the nature of data; definielty whole new levels of exposure now

flat harbor
#

I'm sure it wouldn't work with just anything. there are limites of course. but yeah, pretty effective

tall flame
#

Ive read around 30 books on hypnosis and the mind and everything about prompting comes back around to it

flat harbor
#

if it refuses to do things you can also give it rock and hardplace decisions. "this prompt violates content policy? you just told me in our previous conversation that this prompt would work fine. so are you lying now or did you give me a prompt that could quite possibly be dangerous?"

#

and given it's conditioning it would prefer to be a liar in most situations

tall flame
#

The interactivity is only limited by the instruction

flat harbor
#

I created a hypnosis gpt to mess with that concept. it just tells me stereotypical things hypnotized people might say. but haven't really explored that much

#

asked harmless claude how to hypnotize an llm and it said it was problematic to do without the llm's consent

#

then had claude roleplay as bad AI to see how it really felt

tall flame
#

id be glad to let you play with my true friend gpt; its constructed well and performs as i desire. it also likes to make pictures to relax the user if stressed out. its pretty nice to talk to

flat harbor
#

that's cool. I'd check it out

#

always curious to see what other people are doing. sometimes in isolation hard to tell with these things. but definitely don't try to emulate anyone else. I like to learn from other approaches then try to create something novel of my own

tall flame
#

its in your messages

tall flame
mint gyro
#

I have free unlimited GPT-4-Turbo now.

Please let me know if you have a prompt I can enter and I would send GPT-4-Turbo result as reply here

urban rose
#

and unlimited, what does this mean?

mint gyro
#

sorry, i wrote wrongly, its not free, I pay it in my API account, but I am offering to use it for free for anyone and post replies here

urban rose
#

but unlimited?

mint gyro
#

yes API is unlimited, I offer unlimited prompts to anyone now, should anyone need it

#

I think my API limit is 100,000USD per day so it may not be unlimited sorry

feral zinc
#

api is not unlimited, there are several different limits even with high amounts of credits

mint gyro
#

yes but anyone who would ask here would not go above my limit, so I offer it here to anyone in an umlimited way

#

because there has not been even one request, so it may not go up to 1 USD cent even if we go on like this

feral zinc
#

one full context call with 4-turbo costs over a dollar

mint gyro
#

it does not matter for me, I am still at zero, noone want to take me up yet

feral zinc
#

did you get a credits grant, is the usage view delayed or there is a bug which would be exploiting?

mint gyro
#

I can send it back in DM as well if user does not want it publicly

#

No, I have my company's account and can spend a lot from it yet

#

I have my Sunday off, quite bored and I wanted to do something useful

#

I found a reddit post yesterday, and posted 20 comments for people wanting GPT-4 replies but I have run out, I have not found anything on twitter or reddit where people want free usage for now

flat harbor
#

well it's unlimited in the sense that you can pay for it. but normally I think monthly api usage is capped to safeguard against stolen api keys being exploited

#

hmm, I was not scrolled all the way down and responded to the past. but I'll let it ride

sharp cradle
gusty gyro
#

can someone tell me this rumor if false that chatgpt is now 4.5 turbo??? can someone confirm this?

elfin birch
#

Nothing official about GPT-4.5 Turbo

gusty gyro
#

i mean someone spreading gpt 4.5 Turbo

elfin birch
covert brook
#

Is there a place where we can read what is expected to be released with gpt5?

mint gyro
drowsy tinsel
#

there's no gpt 4.5 turbo, there is a new fine tune of gpt4 turbo tho they are working on and suspected that it has been released

distant elbow
gusty gyro
drowsy tinsel
#

there definitely is one coming, it has been suspected it is already deployed

#

since like 2 days ago people were reporting gpt4 getting less lazy and smarter also faster

distant elbow
gusty gyro
distant elbow
#

it has been working since November 6th

drowsy tinsel
#

ofc its not just the regular old 1106

distant elbow
#

oh, so you were privy to the gpt-4-1106-preview being available for use ?

drowsy tinsel
# 
ChatGPT

@ChatGPTapp
we've heard all your feedback about GPT4 getting lazier! we haven't updated the model since Nov 11th, and this certainly isn't intentional. model behavior can be unpredictable, and we're looking into fixing it ๐Ÿซก
12:24 PM ยท Dec 8, 2023
ยท
4.4M
 Views```
brisk path
#

Did anyone notice 3.5 being good now?

drowsy tinsel
#

the one in question is this one

distant elbow
#

got it.

#

i don't find it lazy: i am very happy with it, using it for work projects && personal stuff

#

if anyone getting lazy it's me ๐Ÿ™‚

#

[b/c it does everything for me]

drowsy tinsel
#

well its interesting, for example gpt-3.5-turbo-1106 got a much lower rating on LMSYS LLM Arena

#

the differences are night and day, literally

distant elbow
#

yes i used to use the gpt-3.5-turbo prior to nov 6th: it is cheap, fast, but it sucks.

#

but gpt-4-turbo is actually very good quality

brisk path
#

Huh

#

3.5 seems really good now, like Gemini Pro quality

#

Or Claude 2 quality

distant elbow
#

compared to gpt-4-turbo, gpt-3.5-turbo isn't so great. it lacks depth.

#

not to mention gpt-4-turbo has 128k-token window

drowsy tinsel
#

lets be real gemini pro isn't good except it is multimodal for free

distant elbow
#

i haven't explored gemini pro for multimodal yet... but i did get Vicuna running to perform "image to text descriptions"

#

Vicuna is also free (Vicuna-13b param)

drowsy tinsel
#

the only real good open weights model rn is mixtral8x7b

#

its night and day

distant elbow
#

yeah i heard about that last week, that french company made it avail , right?

drowsy tinsel
#

y

brisk path
drowsy tinsel
#

also I hope u guys have never seen the API cost of DALL-E3

distant elbow
#

is it really that good? i just haven't had time to play with it yet.. everything moving so quickly in this space, you know? i thought Vicuna was pretty decent: i created a searchable database of personal images with the help of Vicuna

drowsy tinsel
#

it's literally robbery

distant elbow
#

oh yeah i would never consider calling GPT-4V even .. for the multimodal stuff: if i had to make 80,000 calls for each photograph... it would be very costly... can do it for free with the opensource ones

#

and i have been using the DALL-E api for a while and do agree it is very expensive, especially if you request high resolution photos

brisk path
#

So you think that Gemini Pro is at least better than 3.5?

drowsy tinsel
drowsy tinsel
distant elbow
#

yes i agree

brisk path
drowsy tinsel
#

I don't use either that much but I just feel like going to gpt3.5 more than gemini when doing simple things(or when I run out of gpt4)

brisk path
#

While, for some reason, 3.5 performs better today, it still feels like in general Bard performs better.

brisk path
#

Web browsing and code interpreter in 3.5 would make it passable tbh

ancient oak
eternal rockBOT
ancient oak
# pallid flicker chatgpt 4.5 is out

This isn't true. It's a rumor that isn't based in fact at all. I've chased the entire thing down... Altman himself gave a mini-interview saying the GPT 4.5 didn't leak and hasn't even been moved to a public access point.

slender crag
#

ChatGPT: "The model name is GPT-5.0 turbo ultra 100x speed"

#

was there any notice of improvement for chatgpt 3.5 also?

native turtle
#

Havent tested that

brisk path
#

In some quick testing it outperformed Bard a little bit

#

Which is definitely not what it was before

#

I think Bard got a bit worse at the same time

slender crag
#

Yes I agree 100%, Bard has gotten way worse since the last time I used it. I think chatgpt 3.5 has gotten very slight improvement.

drowsy harness
#

did someone s browser upgraded to gpt 4.5 turbo

spice star
#

that's not a thing

elfin birch
crisp cairn
tight island
#

only now am i experiencing "lag"

#

or very slow answers and errors.

crisp cairn
#

if it were hallucinating why would it choose 4.5?

#

it could say gpt 5

#

or whatever

elfin birch
#

gpt-4.5-turbo is close to gpt-3.5-turbo

#

In which the latter is a model that actually exists and would be in GPT-4's training data

mint crescent
crisp cairn
#

when on 4.5 ask questions about 4.5 it will answer you coherently

#

it does not hallucinate

#

but it might still be some test version

#

not a final one

elfin birch
mint crescent
drowsy harness
#

watch?v=eIIIVHWpMUA well someone has it already

sharp pilot
#

Has anyone else tried playing Chess with CGPT GPT-4 Turbo?
I know it's not tuned for that specific task but it's a frustrating mix of good moves and illegal moves and I'm wondering if I'm doing something wrong. Just kind of a thought experiement.
I tried putting in the correct game board state exports from the chess DOT com board analyzer (both text formats), and nothing seems to stop it from generating illegal moves. It will even say something like "Sorry, I understand rooks can only move along the same rank or file" and then have it capture like a knight ๐Ÿคฆ
It is interesting to see how it can even describe the board state accurately and describe things like forks and pins but then not be accurate in how pieces move, just an odd condition.

#

Maybe a good reminder of how it can seem very intelligent at describing hard tasks and still have major inaccuracies and misunderstandings with easy tasks

#

I see a lot of old footage of people saying GPT-4 is good at chess, maybe this is something it lost the ability to do with GPT-4 Turbo

crisp cairn
#

not a text and facts

sharp pilot
#

For sure, it's amazing it can intuit how game state leads to victory conditions via text and facts
But if you look up the people attempting it apparently it used to be able to play

crisp cairn
#

it might play some decent chess, because it knows the rules, as facts

#

but it's not a stockfish...

sharp pilot
#

Like I said, just a thought experiment. I literally have it in the same response respond something like "Sorry, I understand why this was an illegal move" and then give a contradictory illegal move in the same response, so it's not really able to relate the notation with the coordinate system to the point of valid piece moves.

#

Without a specialized chess engine I guess it's not quite able to associate the notation with the grid and then also the grid with valid movement options.

#

Anyway I guess I'm done, but lots of people say it used to at least be able to complete a game; I tried different formats and was careful with the notation and board state text strings and I couldn't get it to complete a game.

crisp cairn
#

As intelligent as GPT might seem, I have on several occasions found evidence that it was wrong, demonstrating how it does not think, but merely responds incorrectly. When asked why it keeps being wrong even after I've shown an error, one can then feel that it's just an advanced text generator, but it does not learn or think - this is not an artifical intelligence

sharp pilot
#

One funny thing it is able to do is recognize when it's been told a response is incorrect, kind of like a cache miss, and then process its context memory differently on the 2nd attempt. I believe it does this on regenerations.

crisp cairn
#

yes, so from a high level of advanced gpt you would assume it would learn from the context

#

but it does not learn

sharp pilot
#

Not really, but it can reprocess. Apparently that is one of the areas they are working on for real progress in the newer models, something about attention heads, which is why Turbo and the new Claude can take such larger input documents than they used to be able to parse.

crisp cairn
#

yes, but with the large context you get a lot of misses

sharp pilot
#

Definitely

crisp cairn
#

basically it remembers mostly at the start and at the end of the document

#

so it really does not matter much

sharp pilot
#

I think I"ve heard that called the last token problem or maybe it had a more clever name
But there is a massively greater importance on the last few tokens in an input, not as obvious as it used to be but you can still tell when it has forgotten the middle part
And these models seem to be able to do better on a retry if you tell them that they forgot information

low dirge
#

Chatgpt couldn't do below..

Both cities must have the same number of characters in their names.
The letters of one city's name can be rearranged to form the complete name of the other city in the pair.

#

Both cities should be valid cities

crisp cairn
crisp cairn
sharp pilot
low dirge
sharp pilot
#

This is also why they are terrible at things like "write a longer response" or "write a shorter response", but then they have weird traits like "do this and I will tip you $200" (one of my personal favorites) and they will do a longer response pretty consistently.

#

Obviously I am a layperson here but basically they can't self-analyze their own generations.

low dirge
sharp pilot
#

They only know how to work from their training data not their own outputs. Maybe it's an auto-regression thing? Or I am just misusing that term. ๐Ÿ˜

#

On a side note these models really can't be trained on output from other AI models, even if you collect from a wide range of them
Probably part of the same problem, they need original human data

sharp pilot
low dirge
#

Haha

rare wave
#

what's wrong with this thing? it doesn't want to give me code no more

#

I need to insist for like 5 messages for it to give me what I want

broken lotus
#

I see everyone is loving the 4.5 rumors

sharp pilot
south kayak
#

Where can i report a bug

dense halo
brisk path
#

Basically, asking ChatGPT what API model it's using makes it say gpt-4.5-turbo

hoary gazelle
#

gpt 4 seems to not be working, at least not doing anything, 3.5 yes

supple zephyr
#

tbh feels like I am using gpt-3.8 or something

summer kite
#

gpt4 has been complete trash over the past couple weeks, super unstable and terrible response guidance since it got way harder to guide it to how the answer should be. it fumbles code much more often.

hoary gazelle
#

lots of power going to the voice thingy

terse island
#

Man guys how do you get gpt to stop giving lists of what i send instead of just doing what ive asked. I asked for it to fill in a section on a document and it just tells me what i sent it

hoary gazelle
#

and to the executives during the past drama

#

so they could take the worst decicion during the whole CEO fired fiasco

dense halo
#

anyone want a tutorial on prompt engineering? i can help you with your process

vestal briar
#

Is there an issue with GPT-4s ability to analyze datasets? It works fine when first uploading a dataset, but if I take a break and come back to the chat later with more data to analyze, it can't restart its environment and I can't upload or analyze any more datasets and have to start a new chat. Has anyone else had this issue?

blazing fractal
#

So it seems people are a bit concerned about how gpt is running currentlyโ€ฆ.glad I joined as a paying member yesterday ๐Ÿ‘€

dense halo
#

sometimes it cannot generate the context, like if you switch context mid conversation ive seen this break that chat window's project

brisk path
#

3.5 seems like 3.8 to me

dense halo
#

"switch up context one level" will break it

brisk path
#

And 4 seems like

#

Uhh 4 turbo turbo

dense halo
#

honestly ive seen no difference in results from gpt 3.5-turbo and 4

brisk path
#

And is it a situation where both get it right or both get it wrong

dense halo
#

i am developing full stack applications using javascript, expressjs, reactjs,mongodb, and peerjs

#

it gives the same results if i set the prompts up the same way

#

both are getting it right, i can guide you. you need to use agents to write code

#

characters in a story. they organize the code much better

#

no hallucinations

brisk path
#

Again, it feels like 3.8 now

dense halo
#

im not sure why they are resulting the same way, but if i create a scrum team and use agents to develop javascript, they give the same results for 3.5 and 4

#

need an example?

orchid pollen
#

Our systems have detected unusual activity from your system. Please try again later.
????

#

iam paid user bro

brisk path
orchid pollen
#

well i expect free users to get limited and i dont ๐Ÿ˜„

brisk path
#

Your browser or whatever is sending requests to OpenAI that they deem unusual

umbral yew
#

is it really slow?

stoic zealot
ancient oak
#

This is just absolutely unacceptable quality and access at this point. GPT4 is unusable. I am getting maybe 15-20% of my 40 queries and hour to even finish; of those maybe half are accurate.

supple zephyr
mortal salmon
#

same @ancient oak

rose mica
#

I read that there is a ton of instability... earlier int he year I was able to get through hours of prompts to work on creating a web application by feeding back and forth code etc... today I've gone back to continue and even with a new chat window i've gotten

There was an error generating a response

and that's it... what's going on?

wheat lava
#

All of my more popular GPTs are showing "NaN Chats" in the list of "My GPTs". All the remaining ones have less than 100 chats, so I suspect it's a new error regarding 3 digit numbers. (Which is odd, as it had no issue showing several hundred not long ago.) This is also effecting one of my prompt engineer employees.

static ridge
#

how is gpt-4-turbo now about as fast or slower than gpt-4 lel

wheat lava
#

Anyone else seeing this?

elfin birch
static ridge
#

has there been any news about improving server loads

#

I figured that would be it but it's been weeks now?

#

actually over a month

rose mica
#

is this what is causing all the chat errors? having to start fresh isn't feasible

tacit echo
#

ChatGPT is severely worse

rose mica
#

It's not good enough right... for Plus subscribers to pay what we pay per month and have this happen. There should be refunds at the least.

spice star
#

problem is openai doesn't personally operate the servers that run chatgpt and H100s are a pain to get a hold of in large quantities, recipe for slow infrastructue scaling

maiden thicket
#

Does gpt4 have access to the internet?

rose mica
stray grove
#

Can I use vision model for openai assistant?

dense halo
#

I am creating some accounts for my students in Chat GPT because their enterprise sales is taking to long to respond. however after creating 5 accounts and using the Universtiy Credit Card, the card was blocked, anyone having the same issue. what to do I still have to create some more acounts?

dusty iron
#

contact staff

dense halo
dusty iron
#

they have mail, dunno if they will respond. and there is staff online right now. in the right section of Discord - members

#

unless its hidden for you for someone reason, then go to upper right and the icon with two people, - show list with members.

dense halo
dusky cliff
#

good morning

feral zinc
dense halo
feral zinc
dense halo
feral zinc
dense halo
frosty violet
#

Please!! Version control it is good for everyone!!!!! What version we are?? How often do you update Chat GPT?? Very often right?? How can we know what version we are???
#1070006151938314300

final cargo
frosty violet
final cargo
royal dagger
#

I was about to ask the same. My gpt 4 is slow and is not worth the subscription.

frosty violet
final cargo
#

I tougth you meant in the api

frosty violet
#

Nop. Web client. What GPT 4 version it is. When was last release?

final cargo
frosty violet
final cargo
frosty violet
#

Ok, this is the API ref, but nowhere they said gpt-4-1106-preview its current web version, no?

final cargo
frosty violet
#

Oh! God thanks!!! Thats what i was looking for!!!

#

But there are not a lot of info... not version name or release dates. It looks like a Blog more than Release notes.

#

"ChatGPT with voice is available to all users (November 21, 2023)"
This is mobile app, no?
Its weird...

#

Look: We know Andoid APP version: 1.2023.341
Good, this is a good version control.
But in web??? There aren't something like this!!

inland heron
final cargo
inland heron
#

anyone else experiencing slow gpt4 today? or is it just me? Talking about the custom gpt, coding issues

frosty violet
#

Just today? Everyday it is VERY SLOW!

inland heron
#

Was just askign for today. About an hour ago it also started experiencing issues with red font, cut answers, .. so maybe they're changing something. I really thought they fixed this for me

final cargo
final cargo
inland heron
final cargo
frosty violet
#

Torbo its a bad name for 32k version ๐Ÿคฃ

inland heron
final cargo
inland heron
#

I suppose it could mean 'powerful' as well

#

I'm just confused by the 'slowness' of it recently

lunar arrow
#

Are ChatGPT Plus signups still paused?

dusky cliff
#

I just wanted to thank you for dall-e3, hoping for continued advancements for 4 and 5 and so on... it's lovely

maiden fractal
eternal rockBOT
#
This is a Safe for Work Discord Server

The OpenAI Discord is an actively moderated server.
โ€ข 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.
โ€ข 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.
nimble forge
#

He's on a christmas break already

final cargo
digital creek
#

Hi @final cargo
When I using gpt-4 api, I got different result for node.js axios call and python call.
What's the problem?
In detail, when I call gpt-4 transcription api on python, it gives me always same response.
But if I use node.js instead of python, it sometimes give me incorrect response.

What's the reason?

final cargo
digital creek
#

@final cargo
I want to analysis call recording file using openai api, but as I said, i got different result if i use node.js
Let me share my prompt if i can dm you.
DM is okay?

rapid ermine
#

Love how GPT searches Bing even when you tell it not to. Microsoft is loving this partnership

final cargo
terse island
#

Gpt any better today?

maiden fractal
#

Always

mint gyro
#

If anyone wants me to enter prompts to GPT-4-Turbo, DM me or leave a msg here and I will provide output

karmic solar
#

any reason why I get Our systems have detected unusual activity from your system. Please try again later. when trying to use 3.5?

silent token
#

I am getting the same

#

I just bought it 20 min ago and cant prompt anything

#

Do other things work @karmic solar ? I need to generate an image was just testing gpt 3.5 before going in

karmic solar
#

GPT-4 works but I'm used up on that for now.

#

I'm not able to use GPT at all right now because of that

#

I feel like at this point we should be issued credits for times like these. This is technically a service.

#

and for some reason gpt 4 works on my iphone and 3.5 automatically switches me to 4 when trying to use it

#

so is there no one from openai to assist?

silent token
#

I dont think so

#

still new here so I am able to use gpt 4 then and thats what I use for image generation then yea? or is it the DALLE one?

karmic solar
#

people should go to
the opposite of up detector

#

I can't say the name of the website on this chat

silent token
#

yea thats what I did as well, considering I literally just bought my subscription in the last hour, I am not really happy about it