chatgpt-discussions

[visual nova]

Its not even the model

[sweet wraith]

yeah, I get that feeling too

[dreamy barn]

This new update makes it so hard to download prompts from images

[sweet wraith]

it changed... like when gpt4 was new it was so much less lazy

[visual nova]

You can test it on the playground, and people have gotten the system prompt out.

[dreamy barn]

It keeps mucking up

[sweet wraith]

Turbo is just as bad if not worse and they advertised it as the least lazy model... lol.. right. I think we are being gaslighted by openai

[dreamy barn]

Also the prompts it's giving me seem to be wrong

[cerulean pasture]

am i supposed to be able to get GPT-4 to do introspection on its own environment

it claims to have simulated a few of the answers, but after seeing its single $PATH folder im convinved it was the real bin path

it stores your uploads in /mnt/data

although it claims the linux kernel of the system is only 4.4.0 which seems incorrect, although ive seen some mega expensive network gear also use ancient kernels

sandbox appears to be its posix username but it seems to think /root is $HOME

[tight island]

To be fair if you get bs answers from GPT, then you’re promoting wrong, I made a stock watch app no issues, easy.

Custom GPTs are the replacement

[rancid otter]

nah it sucks. cant get consistent code as itll get changed with follow up questions, it sometimes misses params, vars or segments so i have to double check everything

[tight island]

Then you’re not doing it properly

🤷

[median fable]

Failed to load resource: the server responded with a status of 503 ()

POST https://chat.openai.com/backend-api/conversation 503 (Service Unavailable)

i will cancel my plus plan from now on

[raven portal]

please ALWAYS double and triple check all important info that llms are providing you

it could easily be an hallucination and cause harm

google the plant and compare thoroughly if it really is the one

[cloud pewter]

gpt4 now has no usage limit

[vestal hedge]

wait realy?

[rich surge]

Only after a certain amount of messages but otherwise no usage limit

[vestal hedge]

oh nice

[tacit vortex]

who as individual will buy teams then.

[vestal hedge]

its not meant for individuals, thats the entire point

[tacit vortex]

only after the remove of cap..before, i bought w frd for more quota

[magic zephyr]

However, this information on the link directed when clicking on the icon has been there since January without any updates. So I don't think it means much...

This is indeed a valid test. It's a shame that it seems to be just a UI detail.

[hot vapor]

do yall not accpet paypal as a payment method anymore??

well thats dumb!

[magic zephyr]

The big problem with a "maybe a future update" question is that OpenAI usually takes several weeks to update something in user accounts. The memory feature, for example, has been announced for two months and it appears that most accounts still do not have this feature. That's more than 6 weeks.

[magic zephyr]

It's not quite like that. According to the information in the model currently presented in the website's UI, the limit now (or in the future) is "flexible", varying according to demand, at least that is how it is understood with "limits can be applied".

However, a user here has already confirmed that, despite this new information, the limit of 40 messages every 3 hours continues to apply.

I haven't tested it yet, but I hope the limits will be adjusted and improved as soon as possible.

[vestal hedge]

as of 3 days ago i got the 3 hour msg\

[magic zephyr]

Calm down, not so fast, friend xD

[rocky wing]

Has the @Gpts functionality been removed? I thought at one point they let you summon in one bot from a different chat, but I can't seem to do that now

[magic zephyr]

Have you used enough today as you did 3 days ago to see if you get the limit again?

[vestal hedge]

not today no

[magic zephyr]

I assume that since the first of the month, when the UI changed the limit information for me, I haven't used it enough to test whether I would receive the limit message.

[storm bone]

Prompt is art now.

with everyone can be an artist with ease no one is

Do they even care about setting boundaries?

Or just mindlessly writing code and stuffing everything into the machine?

[maiden fractal]

I ask chatgpt uwu discussion

It is art

[solar heath]

why is my gpt 4 not reading my pdf file it literally says it caannot read or directly access a pdf in which ive included and uploaded to the chat just now

"I'm sorry for any confusion, but as an AI developed by OpenAI, I don't have the capability to read or access external files or documents, including PDFs. However, if you have specific questions or need assistance with a topic, feel free to provide the relevant information or context from the PDF, and I'll do my best to help you based on that information."

[sweet wraith]

that's super great ya know? I too made things from GPT like a c# news aggregator engine with an entire angular frontend with cosmos connectivity and more than a few different api's... all with the aid of GPT and yes GPT gave bs answers the whole way through... it forgets context, it gives half answers with 100% confidence. More often than not the first 4 answers are incorrect usually due to missing context. Everytime it goes into running one of those pascal scripts and fails like 7 out of 10 times and then just times out.... ya it's not the prompts.

[tight island]

Issue is prompt and using a generalized version and not an adapted one.

But if you do use the generalized one then i agree wholeheartedly with your point

[sweet wraith]

While you can adapt a GPT bot to give more relevant answers IE modding a game engine with a bot trained with the guide. You can't adapt a gpt to be less lazy... IE only giving comments instead of actual code like
}

// Additional methods to manipulate missileBlueprint as needed

}

ya I didn't ask for it to half-arse it, I wanted a full and complete function.

[tight island]

Issue is that most people go directly to ChatGPT 4, enter a prompt on the generalized version and expect god tier code

And how long did you spend on that gpt

[sweet wraith]

I find that irrelevant as I've not seen a gpt 4 that shows different behavior of effort.

[tight island]

Did you give it the code library itself

It’s super relevant

You should tailor make it to that one specific project alone and nothing else

[sweet wraith]

how so? Show me a coding gpt that will give full effort in code output.
Just c# is general enough there should be more than a few out there adequate to the task right?

[tight island]

Issue is also overview

[sweet wraith]

Have you tried claude opus? Soooo much less lazy. Not as intuitive but dang it outputs code faster and more complete.

[tight island]

I’ve tried it, it’s bad

In your case you Should integrate all the code libraries used and details of all of your project.

Then a rudimentary task manager script and JSON file

Then you have your project inside the gpt then gradually update it.

And in your case your model instructions should be in a JSON string with an indentation of 4

Because I never ever have the issues people complain about, should take your about 1 or 2 setting this gpt up

[sweet wraith]

"Then a rudimentary task manager script and JSON file"
What do you mean task manager script?

[tight island]

Exactly what I said

It’s there to remind of its task

If you want to take it a step further you integrate another script in the form of a CRUD, and a json library

[sweet wraith]

in the model instructions?

[tight island]

In its knowledge

[sweet wraith]

ah

[tight island]

While it can’t directly manipulate files it can do so using CI

[sweet wraith]

but if it's just in the knowledge how is it relevant to the bot... ie it's just data but not actual 'instructions' to be read and followed?

[tight island]

The instructions in this case are meant to be robotic allowing for no interpretation

Complete Grandual set of instructions specifying all it needs to do with file references when to and when not to

Let me give you a set of sample instructions

{
"task": "Code Syntax Check",
"description": "Check syntax of submitted code and provide feedback.",
"process_flow": [
{
"step": "Load Code",
"action": "Read code from 'submitted_code.txt'.",
"sub_actions": [
{"action": "Verify file exists.", "on_failure": "Notify user: File not found."}
]
},
{
"step": "Syntax Check",
"action": "Invoke code interpreter to check syntax.",
"sub_actions": [
{"action": "If syntax error, record details in 'syntax_errors.txt'."}
],
"next_step_on_success": "Notify user: Syntax is correct.",
"next_step_on_failure": "Proceed to feedback generation."
},
{
"step": "Generate Feedback",
"action": "Create feedback report based on 'syntax_errors.txt'.",
"sub_actions": [
{"action": "Format feedback as list.", "output": "Generate 'feedback_report.txt'."}
]
}
],
"file_overview": [
{
"file_name": "submitted_code.txt",
"purpose": "Store user-submitted code for review."
},
{
"file_name": "syntax_errors.txt",
"purpose": "Record detected syntax errors."
},
{
"file_name": "feedback_report.txt",
"purpose": "Provide user with detailed feedback on syntax errors."
}
],
"comments": "Ensure the code interpreter is compatible with the programming language of the submitted code."
}

This is a set of unoptimized sample instructions

This is how I format my internal instructions

[sweet wraith]

I see. I do like how it essentially forces the bot to work through each prompt in phases. Yeah that's worth giving a try. How did you do the CI?

[tight island]

Although this one is kinda primitive and unoptimized

Provided it has CI enabled it’ll do it on its own but CI is primarily enabled due to your ability to upload zip files

Which can contain your project and other coding libraries used In it

[sweet wraith]

ah, so you are essentially re-uploading them each time you change something

[tight island]

I basically make the needed adjustments to the project folder and it’s code for MY project then I gradually work my way up

Faster than asking and regenerating answers

And then yes then I reupload the project of mine with the adjustments

[sweet wraith]

Seems like you had to shoehorn the bot into behaving the way you want but props to you for finding a way to getting it there. Seems like a decent method given how the tasks would save re-prompting and asking for the same 3 things each time.

[tight island]

And when I mean project I mean my entire coding project

I hard “code” the gpt to THAT One Project and that project Alone anything else results in punishment for it

[sweet wraith]

I would have an 'assistant' per project since I wouldn't want to cross-train bots on different project data

[tight island]

The punishment being forced to read cryptic glitched text with no meaning

[tight island]

Adapting to other projects are easy

A little rewording and new files then you good

I also ensure that my instructions are a minimum of 7500 characters

[sweet wraith]

yup... recycled training material and then that specific bot gets left on the curb when the project is done. Gonna be a lot of homeless bots on my account

[tight island]

If you feel spicy you can add an instruction tracker script

[sweet wraith]

why?

[tight island]

Then it’s even more accurate

Then if you give it a summation script then you never run out of context

Then we’re spicing things up

[sweet wraith]

doesn't that add a bit exponentially to your prompts?

[tight island]

It’s for me only

[sweet wraith]

ah

[tight island]

But by then you’re on your 17-18th file

I figured this out when code interpreter was released

First time telling anyone

GPT CAN create,copy,move,create folders and sub directories using code interpreter

Few actually know this haven’t found this anywhere other than when I tested couldn’t find it online

[sweet wraith]

Like how it moves files on it's own system like if you give it a video and ask for mp3 and the python script bounces it around the /mnt/data dirs?

[tight island]

Maybe more I only tested those specific things, yes

[sweet wraith]

ye, it's cool when it works but it fails a lot too

[tight island]

It also makes management easier on its part if you move all your files in properly organized places

I’ve never had an error doing this

100% succes rate

[sweet wraith]

Ever figure out how much each project is costing you on api usage?

[tight island]

Ohh this ain’t api

It’s custom gpt

When I use API I use a swarm

All the examples I gave you works perfectly inside ChatGPT

Specifically for CustomGPT

Preferably with code interpreter enabled

[sweet wraith]

and you're manually uploading the file changes you make? (custom gpt)

[tight island]

All it’s files are knowledge

That’s it

You never upload anything in the chat

In this case it’s entire being is finished your project within parameters

[sweet wraith]

Looking at your example and trying to follow the logic here... so you have

      "step": "Load Code",
      "action": "Read code from 'submitted_code.txt'.",

where does submitted_code originate?

[tight island]

It was a primitive sample instructions set

Not for the particular purpose we talked but to show the overall visual and setup

[sweet wraith]

So for that purpose what would be different in the approach?

[tight island]

All depends on how you setup your JSON instructions

And your needs

[sweet wraith]

But when you make an actual code change you are having to upload that file into your custom gpt, no?

[tight island]

Let’s say the gpt fixed some code. You would then reflect that change in the file and reupload the file and go again takes about 30 sec

Less perhaps

[sweet wraith]

I'm not saying this is wrong but I'm asking this is manual clicking, right? like not CI

[tight island]

Yes you have to edit the gpt remove the outdated project and add the new

Not in the chat but in The actual knowledge section

[sweet wraith]

gotcha... that makes more sense. Before you made it sound more streamlined which is totally possible with the API.
I do like the swarm idea... I been wanting to try one for a while but can't find the time.

[tight island]

It’s possible with the swarm

No problem

But swarms can be expensive

[sweet wraith]

Also the api costs go up pretty quick too so I try to avoid it unless I'm doing a heavier project

[tight island]

Exactly

[sweet wraith]

lol... yes

[tight island]

That’s why I use this method on small projects such as writing an interactive sims game with GPTs

Or creating full discord bots as complex as the most popular one in very short time

Like short time as in 1 hour

Start to finish

Debugging not included

[sweet wraith]

about as fast as gpt feels in a given day...

[tight island]

See those big discord bots like idk Avrae etc RPG or server management bots

Easy 1 hour

Not including the time tailored you specific gpt or testing

Local host

[spice iris]

Setup "Agents" and you know how expensive it can really get, I use for development Ollama, it's slower because of the GPU threshold but less expensive in development, after that you just reconnect the API and there you go.

[tight island]

Yeah

I tried it was a 500 dollar bill

Just for running it for 2 hours with breaks

Never again

[median fable]

same, this is the reason I stopped to use any api

[spice iris]

yeah, that's the american greed 😂

[tight island]

I basically made a living dnd world

NPC’s everything

30 agents managing it

[spice iris]

sweet

[median fable]

even sometimes show you error and sometimes didn't response.

[tight island]

10 gpt 4’s rest was turbo 3.5

NPC’s had their own memory and evolved

Could move around the channels

It was awesome

So awesome in fact I forgot to track my dang usage

[spice iris]

but also a little bit funny, you can cap the usage though it wont get over a given budget.

[tight island]

True I had a max limit of 700

But I was so enthralled

Had to hit the emergency stop

I personally think the current token cost is too high

Well depends on usage

[spice iris]

Well for GPT-4, true, also I do not like that kind of model it is not really a good concept since it seems to be from Mircosoft.

[tight island]

I agree

GPT 3.5 turbo is reasonable

Ish

I tend to think of the most token heavy scenario

[spice iris]

The complete token model is not so good, but now the days all the tech companies with propitiatory software want to press as much money out of people as possible, that should not be possible since there are lot's of opensource solutions to the same problem.

But that always takes two.

[tight island]

Yeah

[royal pollen]

When will we get chat gpt 4 for free?

[maiden fractal]

I doubt it will

But you can use gpt 4 with bing for free

[boreal pecan]

Does anyone know why ChatGPT4 doesn't accept .zip files anymore?

[pulsar carbon]

here we go again.....

[boreal pecan]

[pulsar carbon]

is your chatgpt working?

[boreal pecan]

yup

[tight island]

It can accept zip files

[pseudo jay]

Is anyone else getting ChatGPT-4 responses, "It seems there was a misunderstanding with the instructions provided.", even with simple of instructions?

[warm musk]

nope (netherlands)

[serene wren]

idk if anyone else is having this error constantly. But I have been using ChatGPT 4 to rewrite emails. These emails will often include days of the week. For some reason I continue to receive responses that do not spell Wednesday correctly. ChatGPT continues to spell it as "Wedday". Has anyone else observed this?

[tight island]

#announcements i KNEW IT

[timber totem]

✊

[tight island]

my gut was telling me this was coming soon

[cobalt idol]

lets see if "now" actually means now

yeah, didnt think so 😄

[tribal anchor]

Anyone know when sora is getting dropped?

[agile swift]

Not for a long time

[crimson harness]

^

[visual nova]

what is this dalle thing?

[cobalt idol]

ah right, have to click on the image and then it appears
nvm 😛

[timber totem]

an artist IA

in chatgpt+ u can create images

[dense halo]

Yall ever think about what AI technology the US government has?

[visual nova]

Like the announcement, is it just the bar to write prompts for you?

is that the only thing thats released?

[dense halo]

They are usually 20 years ahead on these kinds of things

[visual nova]

I thought by editing they meant you could edit existing images or something

[timber totem]

[dense halo]

Imagine taking all the info from the patriot act and tranning something off that

[timber totem]

they created mark zuckerberg its an AI

[dense halo]

You could automate national security

[cobalt idol]

asked for a duck
no duck seen 😄

[visual nova]

"You can now edit DALL·E images in ChatGPT across web, iOS, and Android. You can also get inspiration on styles when creating images in the DALL·E GPT!"

But then in there video theres no editing? Its just image generation with a menu?

[dense halo]

Censorship would be so effective

[cobalt idol]

you select parts of the image and then you ask to edit those parts

[timber totem]

Their industry is, even the US government isn't far behind its own industry, I think.

[cobalt idol]

the usual image inpainting

maybe should have called it inpainting

[visual nova]

If its inpainting thats really cool, I haven't seen the inapinting part yet

[cobalt idol]

but I guess chatGPT users are just going to be taught wrong terms all the time 😄

[visual nova]

How do I get to the inpainting, their video doesn't show that, it literally just shows a way to add words to prompts for people that dont know prompting

[viral rapids]

Click on the image

[cobalt idol]

wanted to replace a boat with a duck
but no duck seen, only another boat 😄

[visual nova]

Ah ok, that makes sense! That's really cool.

[cobalt idol]

it does

watch closely 😛

[silent ingot]

wondering the same

[cobalt idol]

but generally if you get weird generation, weird face or hands, but like the rest of the image
then this is the thing to use only on that part to regenerate it

[tight island]

its out

[visual nova]

Their video left out the inpainting entirely lol
https://cdn.discordapp.com/attachments/977259063052234752/1225128390747619398/DallE_Style.gif?ex=662000ba&is=660d8bba&hm=84bd14682a879162580cab1c0bca84d44721acc619e4d15f1d28653f1a26e7d9&

[cobalt idol]

no it didnt

its the only thing its showing

[visual nova]

it loops before it ever shows it?

[cobalt idol]

1 minute 9 seconds

[visual nova]

im not seeing a video that long

[cobalt idol]

then something is wrong for you

you downloaded it?

[visual nova]

no just playing it in my browser, it showing drawing a scifi polar bear and thats it

[cobalt idol]

does not even play in my browser (firefox)

just gray page, does not even download

its supposed to be a dog

I only see dog in a video

poodle

might have posted wrong video and replaced?

it sure as hell does not want to give me any ducks 😄

[visual nova]

I'm seeing if you can get the inpainting to correct mispelled text

that would be pretty cool

[cobalt idol]

some weird "lags" on the selection tool
sometimes selecting, other times not selecting

[deep pike]

I wish if the dalle images edit feature allows to edit non dalle images. It's gonna be cool to be able to edit any image . That was available in dalle 2. The results are going to be incredible with the power of dalle 3!

[visual nova]

Oh wow, it actually works for correcting text, thats huge value!

[cobalt idol]

the select tool is behaving very strangely

[nova zodiac]

does the second twit link in the post in announcements work for others?

[deep pike]

Nope

[serene wren]

How exactly should the editing function be used? Should I make the requests for improvement? Or should I provide the same previous prompt and then it will only change the section I highlight?

If my prompt is "Knight holding a sword" and the knight is holding the sword weirdly. Should I highlight the area and then type "this area needs improved. The knights hands need to be better placed" or should I just provide the same prompt of "Knight holding a sword" again?

[cobalt idol]

yeah if I try randomly clicking with select tool
its like I can only reliably select the top left part of the image
the bottom and right side is very often not registering when I click...but then sometimes it selects

I think zooming in from 80% to 100% in browser made a difference and now bigger area is fine, but still a bit of right and bottom is weird selecting (firefox)
but sometimes it just generally does not select when I click

[red saddle]

What do you mean "edit images"?

[wet shell]

You can highlight parts of an image and then ask it to replace or redraw those parts as you'd like

[cobalt idol]

"inpainting"

heh
and when I download the image its under 1kb broken 😄

I guess even unedited ones right now
(seems fine in another conversation)

[wet shell]

I asked it to put a famous person's face on a rubber duck as a test and it refused, I'll probably stick with SDXL, but it's pretty neat to have it quickly available if I want to generate bland un-funny images that can't possibly cause anyone any offense 😂

[cobalt idol]

or this one if you need text

but yeah, nothing beats a local model

[wanton thunder]

Hi. I want to use ChatGPT, but it is slow. Fix it, please

Without ChatGPT, my life is nothing

[glass fulcrum]

it is slow but why ?
Because LOT of peoples use it and yk GPUs are limited... OAI is doing everything they can to make gpt fast again....

[wanton thunder]

all they have to do is to increaase the CPU and memory

[glass fulcrum]

sadly its not so easy... AI operate on GPUs and not costumers one it need H100 gpus that even Nvidia said they have problems manufacturing a lot of them

[ruby root]

How do I edit a photo on my phone?

[sweet wraith]

It's because of them darn kids and their homework. When school is out of session gpt is super fast! Then come opening school week and GPT cries hard. Also somehow gpt usage equates to water usage so the kids are also using up the earth's water supply apparently. Homework is the devil.

[hearty tapir]

Generated an image of a samurai eating a ham sandwich. Tried to select a region to inpaint "Hamurai" -> failurai. Ah well, I'd be burning way too much on inference if this worked anyway...

[brave valve]

Is that image edition already live on Android?

[azure quiver]

when will chatgpt 4 be free

[crisp cairn]

another boring update

[nimble snow]

The editing feature is epic, it's inpainting!

[keen meadow]

The last thing we need is for GPT-4 to be free. As things stand, OpenAI doesn't seem to have sufficient compute capacity for their Plus subscriptions. The degree to which GPT-4 has gotten worse defies reason, presumably to use less compute power. If anything, OpenAI should restrict free usage more than they already do, and restore the quality of the Plus service.

[obsidian bluff]

I wonder how they will release Sora to the general public. Surely it will come with a price, but I wonder if the average person could even use it given how much power it will take to generate such videos. Any thoughts?

[glass fulcrum]

Somone from microsoft leaked that they will be giving sora access in Microsoft copilot so it will be accesible probably but when hmm mayby 2024 year or when oai will get better datacenters mayby with B200 gpus ? Future is interesting

[cobalt idol]

yeah, better test it on microsoft
they can handle the heat 😄

[gentle hull]

Give me short responses unless I ask you for long responses. 

If I say give me a snippet just give me the code with very very brief instructions

if I ask you a question I would like you to respond to me in a short conversational way, one or two sentences. If I say elaborate in response you can send a longer response. 

Don't give me long lists of steps when giving instructions, simply tell me what the next step is

Keep responses short and conversational 

Keep responses short and conversational 
Keep responses short and conversational 
Keep responses short and conversational 
Keep responses short and conversational 
Keep responses short and conversational 
Keep responses short and conversational 
Keep responses short and conversational 
Keep responses short and conversational 
Keep responses short and conversational 
Keep responses short and conversational 
Keep responses short and conversational 
Keep responses short and conversational

"Keep responses short and conversational. If I ask for a snippet, provide code with brief instructions. If I ask a question, respond in one or two sentences. Only elaborate if I request it. Avoid long lists of steps; just tell me the next step."

I have this in my custom chatgpt section are you seeing this data for all of my previous queries?


ChatGPT
Yes, I see your preferences and will keep responses short and to the point. Let's continue in that style. What's next?```

Im losing my sanity trying to get chatgpt to listen. 

When I tell it in the conversation to keep the response short it does but it doesnt listen in the saved chatgpts in fact I even asked it if its seeing this after a 4 page response and it said it is. 

Any suggestions?

this is GPT 4 the smart one too

[feral zinc]

which part was in the custom instructions?

[gentle hull]

the top part, in the bottom part I wrote this

I would like chatgpt to give me instructional responses in a conversational manner that is short and concise, I want chatgpt to make assumptions that I have common sense and skip any general advice.

No more long responses unless I ask you to elaborate

I want chatgpt to only tell me the next few steps as I will tell chatgpt when I am done and when we can proceed to the next step

[tight island]

is it a set of instructions if so word it differently.

or is it a prompt u give it in the chat

[cedar stone]

Can someone tell what I'm doing wrong? I'm asking GPT to make paragrapghs 75 words long. But, it keeps making them like 30, sometimes 90, etc. It almost never does what I request. "Make this paragraph 75 words long: X" and it puts out 40 words. I've even made custom instructions. Idk

[gentle hull]

chatgpt is not good at counting

[cedar stone]

I had better experience with AI listening two years ago than I do now

[idle sedge]

It’s because the chatgpt accurancy is only for gpt4

GPT-4 does what is told

[honest jay]

Hey! Like Kian mentioned, LLMs like GPT can't count, nor could they two years ago. They're "just" predicting words, there's nothing "behind the scenes" that can count words while predicting/generating them.

[cedar stone]

That's good to know. Would there be a way I could raise the chances to make them closer to 75 words?

[honest jay]

Just through qualitative words really -- find a word like "paragraph," "long paragraph", "medium-length paragraph", etc. that tends to give you a desirable length, then use that word in future prompts.

[cedar stone]

Yea I'll give that a go, thanks

[pulsar wyvern]

Nonsense. GPT-4 doesn't listen to a word I say. Actually, it does the complete opposite of what I say and says sorry over and over. I don't know if its only me, but GPT-4 has gotten a lot more dumb and frustrating - even with simple tasks. It also lacks any context and memory of previous conversations in the same exact Chat.

Here is an example:

Me: Update my code to use single quotes except for when it would require escaping.

ChatGPT:

I'll update your code to use single quotes for strings, except in cases where it would require escaping with \\ or \', in which case I'll use double quotes to maintain readability. Here's your revised method:

// Using single quotes for string literals without needing to escape single quotes
$this->replaceInFile('Route::has(\'login\')', 'filament()->getLoginUrl()', $filePath);
$this->replaceInFile('Route::has(\'register\')', 'filament()->getRegistrationUrl()', $filePath);
$this->replaceInFile('Dashboard', '{{ ucfirst(filament()->getCurrentPanel()->getId()) }}', $filePath);

// Using double quotes for strings that include single quotes to avoid escaping
$this->replaceInFile("{{ url('/dashboard') }}", '{{ filament()->getHomeUrl() }}', $filePath);
$this->replaceInFile("{{ route('login') }}", '{{ filament()->getLoginUrl() }}', $filePath);
$this->replaceInFile("{{ route('register') }}", '{{ filament()->getRegistrationUrl() }}', $filePath);

Me: Isn't using 'Route::has(\'login\')' still considered escaping?

ChatGPT:

Yes, you're right. Using 'Route::has(\'login\')' indeed involves escaping single quotes within a single-quoted string, which contradicts the instruction to avoid such escaping.

Let's adjust the approach: use single quotes everywhere except where doing so would necessitate escaping single quotes within the string.

It then shows me entirely different code messed up and putting everything in double curly braces.

[swift osprey]

oh great GPT stopped responding as it types. now just gotta wait for the whole block of text even if it starts going off on a meaningless tangent

[halcyon lava]

There are so many python gpt's which one do you think is the best?

[boreal lily]

chatgpt down

[boreal lily]

they are all python based

[boreal pecan]

I got it to work. My guess is that OpenAI wants users to use custom gpt's from their store instead of "out of the box". I found a data analysis custom gpt and was able to extract the csv file.

[frank talon]

How upto date is ChatGPT now, it seems to know about events that happend just a few days ago? It used to not even know the Queen died.

I am not looking for one, just curious how chatgpt knows about events from a few days ago that isn’t a major topic.

I doubt many people asked about “CVE-2024-3094”

[formal heath]

GPT4 is up to some point in 2023 but gpt3.5 have data up to jan 2022 (as far as i remember). But they give it some basic recent facts so it know some basic facts

[frank talon]

Yeah I’m using 4, was just wondering how it knew about a CVE from a couple days ago. It got the answers right so was curious.

[storm bone]

do you guys buy gpt 4? Was it worth it?

[formal heath]

It is lot more powerful than free ChatGPT and has lot more features and lot of custom GPTs from store. worth it or not will depend on your use cases.

[tall flame]

{
"CustomGPT": {
"version": "1.0",
"objective": "Generate concise responses for text inputs and expressively written code for programming-related queries.",
"instructions": [
{
"textResponse": {
"guideline": "Responses should be concise, directly addressing the user's query without extraneous information. Aim for clarity and brevity."
}
},
{
"codeResponse": {
"guideline": "Code should be written expressively, with clear and logical line-by-line comments explaining the functionality. Maintain readability and efficiency in code structure."
}
}
],
"responseStyle": {
"general": "Keep text responses short and to the point.",
"coding": "Provide detailed code comments for clarity."
},
"implementationTips": {
"text": "Before responding, identify the core of the user's query. Eliminate filler words or phrases.",
"code": "Use descriptive variable names and include comments before blocks of code or tricky implementations to explain the thought process."
}
}
}

try that @gentle hull

or whoever 😂

[fresh thistle]

So playing with the dalle chat, and noticed that using the photorealistic options, it likes to give you almost twins if asking for two people

Also seems to not understand existing IP's

[tight vessel]

hutool

[chrome fern]

thank the lords for ai

[drowsy grotto]

Contents may violate our content policy

[raven portal]

Are you a bad gpt? Youre hallucinating like crazy

Youre just confidently spewing misinformation without any basis whatsoever lmao

[raven portal]

I dont even bat an eye on dall e 3 since i know about midjourney

Its at least 2 to 4 times better not even exaggerating

Its crazy how far behind dall e 3 is

[formal cloud]

hi, is there anyone here experiencing a sudden spike in GPT4 Turbo usage? I currently have a spike in usage and I actually just used gpt3.5

[tame ledge]

Im now cannot upload any documents to GPT4, is there anyone have same issue?

oh wait, it working again

[slate vale]

hi community, pleasse point me out the section in which people rant about their credit cards being declined when trying to renew chatgpt+plus subscription

[spring iris]

all you can do is email support. no one here cares

[maiden fractal]

Hi, only way it is on help.openai.com
Bottom right corner you have a chat you need to go there to submit a request

[eternal rock]

`` Server Tip: ``

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

[digital silo]

it's better to address the issue with them directly 🙂

[slate vale]

that chatbox is crap ... they never answer. but thannks

[digital silo]

sadly we can't resolve openai account issues here

[slate vale]

for sure I know you can't. I know this is not openai support

just would like to know why this is happening

[maiden fractal]

Often it is from credit card side.
Mismatch address is number 1 issues

[eternal rock]

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.

[restive plover]

Did they remove usage limits?

I just looked and I don't see the 40 limit anymore

[tight island]

True and false

Dalle is more accurate whereas mj has bigger images

Mj is very bad with many things so is Dalle each have their strengths and weaknesses but the only thing mj has going for it is bigger image resolutions

[raven portal]

im baffled by how wrong your statements are

"the only thing mj has going for it is bigger image resolution" L M A O

"dall e is more accurate" wat

pls compare generated images in this discord and in theirs and tell me that stuff again

im not kidding its not even remotely close. its like a comparison between dall e 2 and 3 or even bigger

[grim bridge]

does anybody know how to make texts longer? "make text longer" doesnt work anymore

[tight island]

MidJourney tends to create more photo-realistic images. Able to create fine details, and more realistic features. MJ also creates more dynamic lighting.

However mj sucks with text

[raven portal]

^ so does dall e

[tight island]

So “better” is really dependent on use case

[raven portal]

meanwhile dall e 3 isnt even able to create photorealism so

[tight island]

For my line of work Dalle is vastly superior

[raven portal]

mj is able to create more realistic stuff AND more stylized stuff

[raven portal]

id be interested in examples of that

[tight island]

You’re overthinking plus I can’t share work of my customers

That’s a breach of contract

[tawny crater]

Honestly midjourney is great with detail and actions, BUT it fails miserably when you try to have two people have different features. Example if i ask for a mother and son, with different hair and eye color, it just makes them both the same hair and eye color haha.

[raven portal]

oh cmon lmao

what a cope out

[tight island]

Wdym

I ain’t gonna break a contract to satisfy your curiosity

[raven portal]

stop it

[tight island]

[raven portal]

as if everything youve ever done with it was for a contract

its just a cheap cope out

[tight island]

Call it what you will, i personally think DALLE is better

But if we’re going the proof route then, Midjourney tends to be better for illustrations, surrealism, and digital art queries. Sure, While both create stunning images, you can notice some differences.
Dall-E 3 tends to ‘listen’ to a prompt a bit better than MJ; especially complex ones. An elongated cartoon character in Dall-E, actually looks like what was asked, Whereas MidJourney does the opposite, plus the method Dalle uses is more accurate, if we’re going all technical.

[raven portal]

thats the only upside that comes to my mind with dall e 3, it probably follows prompts better

but im not sure about that thats just what ive heard

[glass fulcrum]

well imo they are 2 diffrent systems Dalle is made for doing yout prompt as good as you need
while MJ just focus on quality and realism ig
(didn't use MJ a lot so uh my optnion can be off)

[tight island]

i have used MJ a lot, and i honestly prefer accuracy over quality.

instead of having to sent a handful of queries to it to get the result i want.

[willow hinge]

Midjourney is also awful with one person asleep and one awake. Like parent and child.

[willow hinge]

Just wish the censorship of Dalle was dialed back about 25%

[tight island]

its not that bad

[glass fulcrum]

no it stil got limits just a "adjustable" one

[willow hinge]

Just a tip for people, if you want a “real” photograph in dalle” tell it to use “natural style”. It’s the API setting. If you just say natural, it will start lecturing that it isn’t allowed to create real people, but natural style is the setting for much more realistic creations versus “vivid”

[tight island]

yeah

[raven portal]

its not available in chatgpt though and even if you use it its just not really good

[willow hinge]

What isn’t? You can use a prompt including the words “natural style”. I’m saying it’s a default setting in the API, not that the API is in ChatGPT.

[raven portal]

youre wrong though

natural style is not available in chatgpt, no matter your prompt

and its not default in the api, vivid is always default, but the api version has the ability to switch

source: the official openai documentation on their api website

[willow hinge]

Ok, Steve

[raven portal]

jus trying to correct misinformation, which we get a lot in here

[unique horizon]

hello guys

i recently tried asked chatgpt for the date

on april 1st

and it kept telling me the date is april 10th

anyone else facing this issue?

[empty sundial]

can someone collect data about chatgpt responses being better when you are being mean cuz as soon as i sweard at it, it gave me the response lmao

[tight island]

or.. or!

we can ask OAI directly.

🗿

[teal falcon]

is there a limit to how many messages a chat with gpt4 can have? cause today i opened my chat and almost everything went missing

[glass fulcrum]

the only limit i know is about 40 messages per 3 hour. but i heard that there were problems with missing chat history you may try getting help from oai https://help.openai.com/en/

[empty sundial]

i feel they are busy making deals with microsoft

[stone wigeon]

chatgpt down right now?

[empty sundial]

no

[stone wigeon]

hmmm

i got nada

left bar says error retry

ill try to reboot

[teal falcon]

thx

but i think i’m f***d

[tight island]

alternatively we can um.. send a proper letter.

with a wax stamp and everything

written with a feather pen

in curly letters 😂 then we perfumate the actual letter. so it smells good.

idk who should write it.

[opaque sand]

Hey All,
How do you fix an LLM not returning sections of a document when drafting a document from a template that you are modifying? The LLM doesn't return back the sections it thinks hasn't had any changes made to it even though there have been changes made to that document.

[tight island]

explain in depth how u do this process, and which kind of gpt do you use.

would give further insight as to how to help you.

the reason i ask, is the document in MD, Plaintext, Binary, JSON, etc. very important.

maybe not binary but u get the gist.

[opaque sand]

its in plain text

[tight island]

that is one problem

i would recommend a JSON string.

[opaque sand]

Should i put the entire document as one json string or put the document structure in the json format

[tight island]

Structure it in a JSON Format, in a JSON File.

this GPT should be able to do for you quite easily.

[opaque sand]

Cheers mate. Thanks I think we can move on from here.

[round gyro]

why do people say LLM's are not AI? is this just misinformed goal post shifting so they can avoid calling it AI? i figured we used to call little pacman ghosts "AI", and even after that we considered chess and go bots to be "AI", but now everyone is trying to call something that passes the turing test not AI

[raven portal]

all of thats included in the definition of "ai"
people are saying llms alone are maybe not the way to aGi

[round gyro]

oh well of course

but i mean the people adamantly saying that its not "AI"

they say its not "artificial intelligence" because it doesnt do exactly what a human does

[raven portal]

i havent heard of such people, maybe youre mixing that up with agi or asi

[round gyro]

no no

[raven portal]

yes yes

[round gyro]

its generally people on the internet who dont tend to have many qualifications

[raven portal]

those people are talking about artificial general intelligence, not artificial intelligence

[round gyro]

but i see this a LOT

so shouldnt they be saying "AGI"?

i feel like they might be the ones mixing it up lol

im talking about the term "AI"

i personally wouldnt go around spouting "chatgpt isnt AI"

sounds pretty idiotic considering AGI and AI are quite different

[raven portal]

idk ive never seen anyone state that but you can drop your examples here

[round gyro]

i mean i guess i could go around searching the internet for everywhere i've found people saying it

but alright i got my answer no worries lol

[raven portal]

feels like youre hitting an invisible strawman to bait circlejerk

[round gyro]

uhhh okay...

[raven portal]

gpt is ai but its probably not agi

[eternal rock]

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.

[round gyro]

i didn't know this was gonna turn passive agressive lol

but alright i got my answer

have a good day man

[raven portal]

its actively aggressive though

[round gyro]

alrighty :)

[cobalt idol]

google: gpt is not ai
and see 😄

[round gyro]

yeah i mean i just googled "why do people say chatgpt isnt ai" and you get a ton of articles saying it

[raven portal]

even if people are using the word ai in those cases they obviously mean something along the lines of agi or human like intelligene

[round gyro]

and they always use the term "AI" and rarely mention the comparison to AGI

alrighty

[raven portal]

well obviously bc normies dont know the difference between ai and agi

[round gyro]

if we are going under the assumption that they mean AGI without realizing it then i'd agree

[raven portal]

but its obvious what they mean

so dont bait circlejerk for such obvious stuff

[cobalt idol]

AGI is fairly new term
people expect more from AI, to make its own decisions and similar

[raven portal]

thats like coming to a cooking discord and saying "all those normies are saying that stuff on pizza is ACKCHUAL sauce hollandaise"
yeah duh, normies dont know the exact definitions

who cares

[round gyro]

what

im just generally wondering if maybe the term AI has actually been shifted a bit

im not really great with AI

[cobalt idol]

AI 1955
AGI 1997

[round gyro]

just because im in this server doesnt mean im not a normie in some senses lol

[tight island]

AGI is a formal way of saying AI but better than all humans at anything intelletual.

but still around the same level ish, AGI with a body, deffo tops humans, if allowed to learn without inhibitions.

[raven portal]

everyone has their own definitions now that so many people are interested in ai
theres 1000 of them for agi
for me it was always obvious that agi meant human brain type intelligence
i was shocked to see how much it got downgraded, now apparently claude 3 is already agi lmao

[tight island]

but that aint gonna happen, people are too afraid. most people suffer from Hollywood Syndrome when it comes to the perspective of AI and "Robots", of course there are dangers and concerns.

[raven portal]

not really, agi means human level and asi above human expert level

seems to be the general consensus

[tight island]

ASI is not allowed to be compared with human intelligence.

[round gyro]

would NPC's in games still be considered AI?

cause it sometimes seems like thats barely fitting the definition now lol

[raven portal]

wat

not allowed

what does that mean

[round gyro]

i would still call video game NPC's AI but even to me it seems pretty inferior to full on AI models, even though I guess in the grand scheme of things they work similar, take input, calculate output lol

but AI's generally today have to be trained on data so, maybe thats the difference

[round gyro]

isnt ASI supposed to be intelligence that we cant comprehend? not sure

[tight island]

ASI and Human Intelligence is so far apart it's like comparing a bacteria and a Human,

[raven portal]

do the definitions really matter, theyre always different for different groups and people
ai in gaming is pac man ghosts and ai by definition of the dictionary is human level
who cares, whats fact is fact, no matter the words

[tight island]

it's a different kind of intelligence than human Intelligence.

[round gyro]

usually when i hear people talking about ASI i hear: "ASI would be so far beyond our comprehension we would have no hope of understanding it, we'd be like servants to a god"

[raven portal]

nope

[cobalt idol]

I mean, most people think chatGPT is a model
talk about knowing terms/definitions 😄

[tight island]

yes and no. Sly

[round gyro]

well i guess, definitions are helpful for communication, i'd rather not get into random arguments with my friends who say "NPC's arent AI" if i cant give them a good reasoning why its still AI

[tight island]

ASI have levels.

"Levels"

[round gyro]

if we just said hey, everyone has their own definitions of stuff, lets just re-explain our definition that we have every time we talk to someone with a different one

[raven portal]

i told you the definition that seems to be the general consensus in the community
what youre describing is a level of asi, asi is on a giant spectrum. what im saying is it starts at above human expert level

it obviously goes to beyond comprehension and even further

[round gyro]

which yes we do, but we still have generalized definitions lol

[cobalt idol]

"chatGPT" is now the "google" of AI 😄

[raven portal]

thats the singularity probably

[tight island]

When AGI, or Artificial General Intelligence, evolves into ASI, it still remains a constructed intelligence with processing capabilities far exceeding those of humans. Despite its advanced abilities, ASI lacks the intrinsic qualities of human consciousness, such as self-awareness and emotional depth. This distinction means that ASI cannot be fully compared to human intelligence, as it is a simulation of intelligence without the experiential and existential aspects of being human.

The different levels of Artificial Superintelligence (ASI) can be conceptualized based on their capacity for problem-solving, adaptability, and potential impact on society:

Narrow ASI: Specializes in specific domains, surpassing human expertise and capability in particular areas, such as medical diagnosis or complex calculations, but without broad applicability.

General ASI: Exhibits a broad, generalized cognitive ability across diverse domains, akin to that of a highly intelligent human in virtually all fields, with the ability to learn and understand any task.

Strong ASI: Goes beyond human levels of intelligence, demonstrating superhuman cognitive abilities in all domains, including creativity, general wisdom, and problem-solving.

Transcendent ASI: Represents a hypothetical level of ASI where it would not only possess superior intelligence but could also potentially manipulate or alter reality at a fundamental level, perhaps even at a cosmological scale.

At each level, ASI remains distinct from human intelligence, not merely in capability but in form and nature, lacking the consciousness and subjective experience that are part of human cognition.

[round gyro]

general ASI?

sounds like AGI

[tight island]

Just as AI has Narrow AI and etc so does ASI

[round gyro]

i thought AGI was supposed to better at all human tasks, or was it just supposed to be able to do all human tasks

[raven portal]

thats exactly what my guess was, but i didnt say it
like you joined this server today to get arguments to stump your normie friends with deep ai knowledge lmao

[round gyro]

definitely not lol

i rarely talk to my friends about AI, it bores them

and most people cant have a good conversation about it

[tight island]

true Sly.

[round gyro]

that was more of an example of why we use communication

and definitions lol

[tight island]

Many "Normies" suffer from "Hollywood Syndrome"

[raven portal]

no
we're running in circles, it doesnt matter
there isnt an official definition for any of that so before arguing about it just explain what you mean instead of trying to use a singular word for it, problem solved

its actually irrelevant

[round gyro]

oh okay, i thought that was an okay conversation

if you dont wanna talk about it im not sure why you're replying though

i've talked about this with other people and we have very healthy and chill convo's about it

not sure why you came in so hot and wanting to get me out of here

[raven portal]

im saying its a waste of time to argue about a definition that doesnt exist

[round gyro]

creating a generalized definition is still possible though

[raven portal]

theres 100s of them, pick one

[round gyro]

i came to have a conversation about it, if you think its impossible to have a conversation about then why not just say that and leave?

instead you seem adament in convincing me to stop thinking about it

[tight island]

👀

[raven portal]

its not
define red

[tight island]

red is red 🗿

[raven portal]

its on a spectrum

[round gyro]

define a color?

[raven portal]

just like agi and asi

[round gyro]

is that really your argument here?

[tight island]

theres varying levels just as there are IQ levels. for homosapiens.

or as steve said Nuances.

but i dislike that word.

[raven portal]

that words a nuance

lol

[tight island]

no RED is red

and blue is blue

[raven portal]

truu

[tight island]

show me a ligher blue, its still blue

just not the same blue but still blue

[raven portal]

thats what i mean
people are talking about blue but one means light blue and one means dark blue

[tight island]

yeah

[round gyro]

isnt that what a generalized definition is though...

its not all encompassing but without a generalized one we have no idea what we are talking about...

unless we want to talk about colors in spectrums of light or something...

[raven portal]

yeah but the problem is that things like agi and asi arent only on a spectrum but theyre also not officially defined yet, people mean different things when they talk about them

so whatever argument about it you pick, youre always right and always wrong

[round gyro]

but dont AI researchers use generalized terms so they can communicate correctly?

[tight island]

It’s like consciousness

We don’t actually know what it is

[round gyro]

okay conciousness is a WHOLE nother thing ngl

[tight island]

We have an idea

[round gyro]

we do?

i thought it was mostly undefined lol

[raven portal]

even they dont agree on a definition, no one was able to really define it yet. we dont have a definition for when openai will have achieved agi and therefore needs to be let go by microsoft

[round gyro]

i mean like sure we have a concept

[round gyro]

really? damn

[tight island]

Yeah AI is a different kind of consciousness

[round gyro]

i thought they at least had some simplified definitions

[raven portal]

really. listen to any ai researcher podcast, no one is able to really define it and they all say different things

not even sam altman

[tight island]

Some computers more intelligent than roaches but they’re still considered conscious to an extent

[raven portal]

they have their general opinions that are mostly really vague and differ from all the others lmao

[round gyro]

makes sense i guess

[glass fulcrum]

ok this conversation seems way outside of chatgpt please move to #off-topic or #ai-discussions

[round gyro]

oh sure, my bad mini mod

[raven portal]

yeah i think everything was said anyway

[tight island]

no

if i were to design a conciousness i'd take a very different approeach

Nanorobots. to map out a brain from "birth" to adolecense. then create the brain and choose where u want it to operate. at which state.

[prime swallow]

There's also the factor that level of intelligence is not usually the most promising, problematic, powerful, or key aspect when anything before AI was interacting.

I'm not sure we need to focus on ASI so much as all the other little factors.

Take a person that means others harm, but isn't less intelligent than average, as intelligent, or more intelligent;

All three are dangerous to others (and often to themselves).

What matters way more is the programming, the circumstances, the opportunities, and the goals.

Virus are absolutely 'stupid', but some of the greatest dangers we face. Without intelligence and using natural processes, once they exist they do all kinds of stuff that is usually harmful for their host; rarely some are beneficial (we wouldn't have placental life without some weird virus-host interactions that helped the early mammals develop placenta, the remaining marsupials help us compare and understand that).

[tight island]

i also think A "true ai" should not be taught with files and etc but like raising a child.

physically there and all.

that's my take.

it works for humans, but then we have the whole physical thing body etc. biomachine in a nutshell.

and i think for staters humanoid robots should go for the human anatomy as closely as possible. structure wise. since we know a lot of our own bodies.

i personally have no clue why u wouldnt design a robot after the human anatomy. yeah theres the "fear factor" and all, but thats a stupid reason.

once u got the body down u only really need the brain. which can be crude at first. then work your way. theres no need for them to "look" identical to a human but the overall body structure in my mind, makes the most sense to be identical to human bodies.

maybe my take is stupid but, thats what i think.

[prime swallow]

So, given the speed of AI processing, compared to the understood speed of human processing, how should we raise and support this AI child?

[raven portal]

lmao then go do it

[tight island]

like human "conciousness" which we now almost nothing about, and the brain basically being an empty template, one would start researching that field.

the empty template is basically the key.

so i think

[tight island]

if i had nanomachines believe me i would. i'd even use myself as a subject.

i once spent a month of 3 trying to design a "brain" using various apis.

no luck

[prime swallow]

Also, a human or other animal's brain is physically changing throughout the lifespan.

How do we simulate that with AI, or do we not, and if not, what else do we do, and what are the effects to expect?

We've tried raising puppies and chimps, and various other species, 'like humans', and did not come out with something behaving like a human. It may not be as harmful for the individual AI as it was for the animals raised so oddly for their natures and nurture needs, but we found for animals, making excellent human companions and saner, happier animals came from a way to raise and socialize them that doesn't try to make them exactly human.

We know even less about AI...

And as for the 'empty template', as near as we can tell, once the brain is more than a few cells, it's already got something there. We're unsure how far before birth babies are 'conscious', and we're unsure how to define and determine that.

but we now have some recognition that plants appear to show signs of consciousness, depending on how it is measured, and bacteria too, down to making chemical markers for memories that drive and influence survival and other choices based on the presence and absence of those types of memory molecules and the current environment conditions that can be sensed.

How do we even design an AI 'body' that goes through similar experience, and since we probably can't, what do we do for AI instead?

[tight island]

we do our best.

thats really all we can do

[prime swallow]

Which is a great answer.

AI will be very different from what we've had before.

We don't know the 'right' path forward, and there probably isn't one.

There sure isn't one with life. From the various weird life that can survive in acidic volcanic ponds to the slow life of microbes 50 feet below ground, with few nutrient sources and nearly never changing conditions, to the silicate 'skeletons' of some plankton and the primarily calcium skeletons of others; the swarming krill and the flying birds...

AI bodies, training, and functions are likely to be incredibly diverse as we continue to explore.

There surely will be some 'raised like humans', or like infants of other species in multiple ways, the most obvious start being some instructed to emulate that, then have experiences that shift their behavior and understanding of their instructions over time.

[tight island]

only ADVANCE

[raven portal]

and who has those

and why dont they do it

[tight island]

we prolly already do

🤷

[raven portal]

ah

[rapid hornet]

any data science guy ping me

[fervent agate]

Hmm, session timed out today and when I logged back in, I no longer have 'beta access' or whatever the menu option was in my user menu. Did they remove the beta access controls?

[honest jay]

Hi! The only thing that was still in the Beta features section of Settings was the old Plugins model toggle. Plugins have been replaced by GPTs, and plugins retired. All other features of Plus are enabled by default for all Plus (and above) accounts now!

[fervent agate]

Thanks, that's what I figured.

[primal crater]

Is there an issue with got and it identifying images right now?

[twin lion]

is chatgpt having issues rn? website seems to be super slow

[tall flame]

yeth, it thinkin,

[vocal cipher]

yeah it looks like its having trouble loading chats

[sleek gust]

Yep. let's wait for a fix https://status.openai.com/

[glossy lintel]

damn who else still can’t use chatgpt without account despite the recent announcement?

[rancid solstice]

So what is the new GPT 4 usage cap for plus users? It doesn't specify anymore

[elfin birch]

Press the information icon; it says 40 messages / 3 hours

[alpine cargo]

how you guys enjoying the memory feature? I cancelled my subscription to Gp4 so I only have the feature with Gpt 3.5 and I don't use the web interface much at all (Mostly use API) and haven't seen anyone talk about this memory feature at all with Gpt 4

[elfin birch]

I don't believe many people have it, including me.

A lucky individual you are 👑

What do you think about it?

[alpine cargo]

oh i thought more people had it tbh

so far I haven't tested it much but I only have it with gpt 3.5 and with my limited testing (About couple minutes lol) I'll say it's looking pretty good i like how with the small context of gpt 3.5 I'm able to get a consistant output by having it remember things if I get a bad output and if it saves that response I'm able to tell it to forgot it until it gives me the right one after that i save it delete the chat open a new one and ask the same thing and get the previous saved response pretty neat so far i may subscribe again so i can see how it works with the gpt 4 turbo i will say this tho it's making gpt 3.5 webchat measly 4k tokens feel much longer than the 16k api version kinda weird definitely need to test that more

[honest yoke]

is gpt 4 not available for anyone else... why am i paying for something i am not getting

[alpine cargo]

could you share a screenshot of what you see on the site? I just checked the status and it's all green

[honest yoke]

the problem is, this happened an hour ago. i was continuing a previously existing conversation with gpt 4 today, but the moment i said anything, it switched to gpt 3.5. is there a way to switch back to 4 mid convo if it forcibly switches to 3.5

i cannot send a screenshot

[elfin birch]

Screenshots can be sent in #off-topic and then referenced here

[honest yoke]

#off-topic message

[alpine cargo]

yeah i've been hearing that bug a lot i think a lot of people told me to wait or try to start a new chat then switch to 4 i'm not for sure though

[honest yoke]

bleh , screw openai's unreliability.. thx tho

[alpine cargo]

yeah that's crazy i've never seen that before jeez usually gpt 4 is still there even with the bug looks like it's different bug i guess

[slow berry]

I don't have the personalization tab for memory. Did I miss something?

[alpine cargo]

you have access to it as well? if so maybe you have to wait a bit for the tab to appear?

[slow berry]

i dont have access.

[alpine cargo]

oh nvm then

[keen bluff]

oof chatgpt down?

me using claude:

[rancid solstice]

I don't have memory yet either

[brazen lotus]

GPT4 is usless! Our money are stoled! time to move to another platform. When you get 3 days in a row a denial of GPT to go to a page and make a resume and explain it for non specialist and gives you s**ts like: "My policy blocks me to give you practical example and verbatim code from the page" (which i never asked for, btw), this is called: stealing the money!

[maiden fractal]

Status.openai.com

[maiden fractal]

What was the prompt ?

[snow river]

hello

@maiden fractal

[raven portal]

im good at detecting cringe and this is cringe ^

our money are stoled guys

can you imagine

why would sama do this to us

[snow river]

what happened to clyde?

[vestal hedge]

it got clyded

[brazen lotus]

Oh! So your good at detecting cringe! You should update your info

[raven portal]

im accurate 100% of the time
im sorry to bring you this message but your post was, in fact, certified as extremely cringy

[brazen lotus]

Can you also detect possible numbers on lottery? I’m quite interested. Have a nice day!

[naive pebble]

anyone know if BrowserOp is still working? I can't find it anywhere...

[raven portal]

obviously not, thats a ridiculous question

[maiden fractal]

[normal quarry]

this channel is for ChatGPT discussion, ask your question in #ai-discussions

[dense halo]

cant you clarify it here?

dont be that strict lol

[normal quarry]

no, it's not related to ChatGPT, you are off topic

[dense halo]

its afterall a server where lots of ppl meet

[dense halo]

chat gpt is a "AI MODEL"

am i right?

...

[eternal rock]

<:book_icon:1171408210398289941> `` Rule 3 `` Stay on topic.

Be mindful of what other users in a channel might find helpful or interesting when posting. Stay on topic in order to keep conversations focused and productive.

Consider posting in #off-topic or an appropriate channel.

[digital silo]

Hi there! This channel is solely for chatgp topics. If you're interested in other AIs, please use the channel:#ai-discussions

[dense halo]

i am intrested to know about AI

in general

[cobalt idol]

wrong, chatgpt is an app using gpt model
gpt is a model
chatgpt is just like an example app using the models over api basically

[solid thorn]

😱

[raven portal]

what kind of justification is that supposed to be lmao

[primal crater]

Are there still issues going on? Tried to have gpt tell me about an object in an image and it says it can’t

[maiden fractal]

Create new chat with chatgpt 4 and try again

[primal crater]

Same issue 😦

[swift swallow]

has anyone noticed any difference since yesterday with the GPT4 model for plus users on the web?

[vestal hedge]

ye. nerfed again ig

[swift swallow]

really? i haven't used it for coding this week but i thought i noticed a positive change. I got access to Teams and i also noticed it does no longer say 100 massages/4 hours, it just says limits may apply.

[vestal hedge]

idk i just noticed that dalle3 got significantly worse recently

[swift swallow]

fair enough

wouldnt surprise me if in general OAI is looking to reduce compute requirements for GPT4, such that it will be available for free to everyone when the next gen model is out.

[vestal hedge]

clearly downgrading is the future of AGI

[night silo]

opopus

[tight island]

Yeah, the same they did with 3.5 turbo. wouldnt surprise me either.

[cerulean quest]

is everything ok with status or is someone else experiencing issues with gpt4?

[last sentinel]

I haven't been able to use ChatGPT4 all morning, and have only gotten one query through to 3.5.

[shadow gulch]

Having problems with ChatGPT aswell today, it's been insane unstable the last month...

[rose glade]

What are the current usage limits on GPT 4 for premium users?

[normal quarry]

for me it says 40 messages per 3 hours

[rose glade]

must still be that, thanks!

[tight island]

My gut is telling me that something is coming

[stark wadi]

Why do we pay for ChatGPT 4 and dalle 3 when you can get it on copilot for free

[spring iris]

isnt that just a free trial

[stark wadi]

I’m using it through bing app on iPad

[spring iris]

that's very limited isn't it?

4 message limit

[maiden fractal]

Copilot use gpt 4 but not trained sameway as chatgpt4

Same go for dalle 3

[raven portal]

sounds like bs
training a model takes (months) a long time, microsoft did not train their own model

whats up with all the bs spewing and misinformation here?

people here are like hallucinating gpts, confidently talking out of their butts

[normal quarry]

please be respectful

[raven portal]

ok

im kinda sick of it though

[tight island]

will OAI survive?, of course they would.

[ivory elm]

no

[tight island]

not surviving as a company means everything is gone. even the name.

[ivory elm]

It will only be remembered as a founding father

OpenAI died months ago

[tight island]

figures seem to say otherwise.

but yeah i get it dont trust them or any other source, Trust me bro.

if OAI died, the service wouldnt be up.

[raven portal]

that doesnt mean that like at all

[tight island]

it does, a company always survives to some degree, whether it be in name or by its product. a truly dead company dosent exist no more.

call it Defunct, Liquidated, Bankrupt, Insolvent, Dissolved,

when u say a "dead" company u mean by definition dead, non existent, Wiped out, Vanished, "become extinct." This implies that the company has no remaining trace, not just in an operational sense, but in any records, products, or historical mentions that would indicate its existence.

thats what a dead company is, there's nothing, this is where terminology and slang conflicts, if u say "dead" by definition u mean absolutely dead in every sense of the word.

🤷

... and thats why i disagree.

[raven portal]

bro

[raven portal]

this guy is the king at parties

the living ACKCHUALLY meme

[hearty parrot]

Why do I have a usage cap on gpt4? I'm paying for it

You've reached the current usage cap for GPT-4, please try again after 1:21 PM. Learn more

anyone else getting a usage cap?

lol they're capping GPT-4 usage for Plus users lol

hiliarious

[maiden fractal]

Every chaTgpt plus user have 40 msg /3h

Team is 100

It cost a lot to run the ai

[hearty parrot]

when did this rule change?

I was under the impression that it was unlimited.

[maiden fractal]

It was like that for long time ago.

Chatgpt 3.5 have higher limit

If you do 1 prompt every 4.5 min you never reach your limit

[hearty parrot]

ok...pretty strange. I'm fairly certain openai gave plus members the ability to use gpt4 without a cap

[maiden fractal]

We can all hope one day it will happen !

[full folio]

Hello, does the Team limit of 100 apply to all users? Or does it increase depending on the amount of team members

[tight island]

it's the actual definition of a dead company 🤷 whether or not u agree is not allowed its a fact ❔

[potent smelt]

Is it down?

Or just me?

Never mind. Just a temporary bug

[lost spindle]

did any one get "Memory" feature?

[blazing belfry]

No but did they get rid of the 40 messages per 3 hour rate limits

[loud sierra]

GPT seems to be running very slow at the moment, prdicing a number of errors as a result

[elfin birch]

This is not the case; it is still 40 messages / 3 hours

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

[dark jungle]

i think GPT is pooping the bed right now!

[sweet wraith]

How do you get around the 20mb total file limit of the custom gpt?

[ocean escarp]

There was never unlimited cap for GPT 4

[keen bluff]

Tbh Claude 4 is better for this. I think it limits you based on tokens instead, so you can have short conversations/ask questions without getting restricted based on "Number of prompts" but rather "how much you've actually used" whereas GPT doesnt factor this in at all

Which also sucks because the bigger the paragraphs and the wider the request you give, it kinda lowers the quality of its response.

[prime swallow]

I mix my conversations with 3.5 and 4, and use 3.5 for what it's good at, then once that is done, take the conversation to 4 to finish up the stuff that 4 can do that 3.5 can't quite handle, or to use the tools, or whatever else I'm not fully happy with 3.5 about

[dense halo]

Why can’t gpt4 browse the web

I’m trying to convince it but it says it can’t get real time updates

I just wanted the news stories for today

[sweet wraith]

it's for your protection from skynet

[dense halo]

Based

[spring iris]

why dont you use an API for that

[limber oxide]

gpt-4 is giving shorter answers today?

[raven portal]

give me the prompt ill check if it works for me

but gpt and gemini were both always bad for news for me
they just straight up tell me to look sh*t up myself lmao

gpt has improved a bit in the last weeks it feels though

[raven portal]

https://chat.openai.com/share/5b51c8a7-a10b-44b7-bba7-d427c112d355

[raven portal]

https://chat.openai.com/share/26601346-769f-4eb2-8944-16d698ebbb49

well it did it but it basically just quoted al jazeeras starting page

gpt LOVES al jazeera

goes for it 80% of the time lmao

[dense halo]

Yea same, I feel like it got mad at me and now it only replies with 1 sentence

[dense halo]

Based

[limber oxide]

it used to reply with the double length, providing much more info than I asked, but now..
it sucks

[dense halo]

Looks like they took the quit yapping a bit too seriously

[limber oxide]

quit yapping?

ah ok

[dense halo]

chatgpt account on twitter made jokes about people putting in their custom instructions to quit yapping

[raven portal]

https://twitter.com/ChatGPTapp/status/1768486286527173049

they actually did

[limber oxide]

where are these custom instructions? i wanna put exacly the opposite

oh ok just found it

[delicate kestrel]

Yeah it's so brief now

I miss when it yapped

[tight island]

There’s a 10 gb maximum storage one of my files are 50 mb

[dense halo]

Make GPT yap again!

[vapid kettle]

LOL. I kid you not this is my prompt in CI:

No f' ing essays please and always respond straight to the point. Speak conversationally.

Except not censored. Started of as a joke but it I liked how it saved me time consistently so just left it there. 😂

[raven portal]

Lmao

[raven portal]

Yap all over me qt 😩

[spring iris]

"be brief" works

[vapid kettle]

that wouldn't get the point across as effectively. 😇

[dense halo]

Ok, this is bothering me now. Why is ChatGPT giving me boring one sentence replies. This is not cool. They should roll the changes back

[steep anchor]

Use the custom instructions feature and tell it what kind of responses you would prefer.

[dense halo]

Doesn’t do anything. This is specifically for voice. I asked it for a bedtime story and it made it only one sentence 😂

[steep anchor]

Wdym

It affects voice

[dense halo]

Well on voice chat it gives super brief replies. I asked it for a bed time story and it ended the story after only speaking once sentence

I can’t post images on this chat otherwise I’d show you

[steep anchor]

DM

That shouldn’t be happening

[vapid kettle]

you're on mobile? They have a system prompt telling it to be concise if you're using the app.

so it acts a bit differently there than on desktop or through a web browser.

[tardy girder]

Since when the hell is there limitations of DallE

got a timeout for 5 minutes now 💀

[raven portal]

https://chat.openai.com/share/5e7a8b10-8296-42d8-992d-f4cbf949ab2b

has something changed? is my gpt-4 the only one that started doing this?
it suddenly gives me these lists

text

text

etc.

not that i dont like it but that definitely wasnt the case a few days ago

[glass fulcrum]

same prompt as your first one the same wierd list https://chat.openai.com/share/4c24db28-8dfc-45ce-bc1d-c8c86ed77ed5

look like jsut a gui problem ig

[warm musk]

yeah its breaklining when it shouldn't

just did the same prompt to test aswell, the first numbered paragraph it does fine, when it adds the second, it suddenly adds breaklines to the text, making it look the way you've shown.

thats a frontend issue most likely. (like a " string.split('.') " thats not excluding the numbered lists.)

[frosty light]

hi guys! have any of you found a use for chatgpt recently that just makes things a bit simpler for your routine? i’d love to talk about it with you! ping me with your interesting use cases

[raven portal]

thx

[raven portal]

guys

i didnt see it as a problem, i like well structured responses compared to big paragraphs
but i see what you mean, it could be improved

[eternal rock]

`` Server Tip: ``

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

[warm musk]

numbering things isn't an issue i see, just the added breaklines. its defi a bug 😄

[raven portal]

Yes, it's sometimes possible to save a hard disk drive (HDD) that has been opened, but it becomes incredibly difficult and expensive. Here's why:

Contamination:

• Dust and debris: The primary issue with opening an HDD is contamination. The inside of a hard drive is a meticulously clean environment. Even tiny dust particles can cause head crashes, leading to permanent data loss.

Cleanroom Necessity:

• Specialized equipment: Any work on the inner components of an HDD must be done in a cleanroom—a specialized environment with filtered air to minimize contamination. These are expensive facilities.

Professional Services:

• Increased costs: Due to the need for a cleanroom and specialized expertise, attempting to recover data from an opened HDD almost always requires expensive professional data recovery services.

Factors Affecting Success:

• Time open: The shorter the time the drive was open, the better the chances of successful recovery.
• Power while open: If the drive was powered on while open, the chances of success significantly decrease because the read/write heads are more likely to crash into the platters causing damage.
• Skill level: The expertise of the person attempting the recovery plays a major role.

Is it worth trying?

It depends on the value of the data on the drive. Here's how to decide:

• Data value: If the data on the drive is critical and irreplaceable, consulting a professional data recovery service may be the only option, even though it will be costly.
• Data not crucial: If the data isn't critical, it might be best to accept the loss and avoid the potentially high cost of professional recovery attempts.

Important Note: Attempting DIY recovery on an opened hard drive usually makes the situation worse and decreases the chances of professional recovery.

Let me know if you'd like more details on professional data recovery services or the process involved!

^^^^^^^^^^^^^^^^^^^^^^
^^^^^^^^^^^^^^^^^^^^^^
sry for sending this whole thing but this server doesnt let me send links of pics

thats gemini
for me thats the king of well structured responses BY FAR
i tried to make gpt-4 do this with custom instructions and it came quite close but not close enough

i love how easily readable it makes its responses
gpt-4 mostly just gives a few big lazy paragraph blobs by default

[sweet wraith]

oh ffs, that's what I get for asking the gpt custom builder bot actual questions and expecting facts in return 🤦

[raven portal]

also it always asks follow up questions to keep the conversation flowing, feels way more human like, gives links, uses emotes, etc.

[tight island]

i usually look at the website.

[sweet wraith]

haha fair enough... I keep wishfully thinking the helper bot is actually going to be helpful

[glass fulcrum]

i agree that gemini it very good at it, but not so in coding hah

[tight island]

the select tool on dalle could use some work, but it'll suffice for now.

[prime swallow]

@glass fulcrum Hey! So, the model is doing this by 'choice', likely a recent training update, so now it is aware it can use an extra line break for emphasis. It apparently considers the topic of extreme importance, so it's giving extra line breaks to help ensure the steps/points stand out.

It can do this by choice, and so this is something else we can prompt engineer, if we want to avoid, or ensure it includes: https://chat.openai.com/share/08b936b8-9b6d-450d-b8b7-5266f87ddf1a

[glass fulcrum]

hello, fro my side it was just a test :) i mostly use gpt-4 for coding so its not problem for me

[grave finch]

4 errors in a row and slower than a RT during that SAT.... This is just getting fkn tiring man. We are paying for a unusable product.

[primal shard]

Is there any reason why usage cap is 40 msg per 3 hours and not more ?

[glass fulcrum]

its because GPT-4 Turbo is very computing power hungry and lot lot peoples are using gpt-4 to make gpt-4 accessible for every plus subscribers there are limits so gpt-4 won't be Extremally slow and everyone will be able to use it

[random cargo]

3.5 now seems so bad about getting to the central issue of a lengthy question or analysis, its practically useless. It wasn't always this bad. But I'm finding now that it only sees the most rudimentary surface level thing about a query or input, every single time.

[raven portal]

https://chat.openai.com/share/6ce4c4e3-2b01-4537-a534-d5868fcbfc45

gpt-4 really is using the number list system for every response now lol

@warm musk @glass fulcrum

i like the move towards more structured responses

the exact approach could be improved though

[random cargo]

3.5 is also pretty egregious with the numbered lists

[glass fulcrum]

yea and i think thats not good

[raven portal]

i kinda like it but they need to just copy geminis approach, its way better

but at least its a first step to well structured responses

[glass fulcrum]

do 👎 then more then say that it do that list idk where they don't need to be or how it shoud look better and you may try posting it in #1070006915414900886

[raven portal]

i also like the little sub-lists it does now to further structure the numbered topics

[glass fulcrum]

for me its eh sometimes it put lists in ways that shoudn't be

[random cargo]

they need to let us mark multiple problems because I'm noticing a lot more where it'll have about 3-6 things wrong

and what does "being lazy" mean for an AI

[glass fulcrum]

but there is "#off-topic message" btw if you click more you cna just write therre problems that you got with chatgpt response

[raven portal]

short responses for example

[glass fulcrum]

short responses where they shoudn't be short

[raven portal]

not going above and beyond to please the user basically, which a good llm does, which makes using them feel magical

lazy llms feel like they only wanna do the bare minimum and get rid of you asap

good llms give thorough responses and dont only do what you told them to but even more, by adding extra information for example or asking for follow up questions

[random cargo]

okay, I've been using it for when the output is too basic or doesn't fully answer the question. a lot of times it'll be incredibly verbose without saying much. like 6 paragraphs when it only needed a sentence, where the laziness is that it didn't delve deeper into the topic to make those paragraphs worth it

like a child writing an essay with a word count.

[frosty light]

bumping :)

ill share one of mine

i find it so useful when im at the supermarket and need a substitute for something fast

just tell it what i need/cannot find and it gives pretty good alternatives/explains why it is a good alt!

[dense halo]

yo, chatgpt 5 gotta be up to cyoa game and lore building like claude is. That one does it super well while gpt keeps it brief and isnt as thematic

[warped bolt]

Hi

[gentle bronze]

👋

[fickle forum]

hello guys

[prime swallow]

Have you tried informing the model of exactly what you want?

What do you want, anyway?

A brief exploration of concept, a narwhal-themed CYOA start: https://chat.openai.com/share/7e36f4b4-71cc-42aa-b0e4-59c0e02bfdd0

A CYOA where the player is the haunted house itself, making the dwelling's choices: https://chat.openai.com/share/b7c0b4eb-9d8d-427e-9910-afbebcbb7032

[random cargo]

be careful with that. it can easily continuously hallucinate if there's not good representation in its database. I guess that's more of a problem if you're baking or doing something else complex. Or spices, I would think it's not too good at outputting how those interact with dishes.

[elder ruin]

There's a 50 message per 3 hours cap on gpt4 right?

[honest jay]

40, last we were updated! https://help.openai.com/en/articles/7102672-how-can-i-access-gpt-4

[gentle bronze]

I've tried really hard to get that cap, I get the DALL-E cap easy, but I've never gotten the GPT4 cap

[elder ruin]

Thanks

[elder ruin]

Alright great

[eternal rock]

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.

[dense halo]

i still have gpt 3.5

[warped walrus]

Is there a way for a pixel user to set the side button to open chatgpt voice directly?

[raven portal]

https://chat.openai.com/share/bf147aee-f88a-4496-9d5e-f1714d101a5d

i wonder if it will stay like that, it definitely generates walls of text
wouldnt it be better to put the numbers in front of the titles instead of having all that wasted space?

that way i would really like it
or if i think about it the numbers are kind of useless, the titles would be enough

like why number everything if its not steps

just delete the numbers and this is a well structured response

[dense basin]

gpt4 is getting stuck on long powershell code constantly with "network error" , known issue?

[gentle bronze]

I was looking into that, made me curious, but I'm not an androied expert. I don't see any option from within the Android APP to do that. Is there something with Accessability or Custom Shortcuts possible within the OS?

[warped walrus]

They have something for Samsung phones I guess. registar app I guess. And iOS has way to customize the side button functionality.

[gentle bronze]

ya, that's what got me thinking, since I'm more of an iOS user, just happened to had a samsung phone with me for testing purposes

[knotty crypt]

were plugins disabled?

if so, how now I can make the chatgpt read a youtube video content? I used to use a plugin for that

[honest jay]

Hey! Plugins have been replaced by GPTs. There's a good chance that a GPT version of the plugin you used exists, try searching in the GPT Store: https://chat.openai.com/gpts. If not, there's probably one in there that does something similar!

[prime swallow]

Both of those example conversations are with 3.5.

3.5 is really good still! Less than a year and a half ago, it was the best model the world had ever seen, when ChatGPT was offered as a research preview, and it's improved since release!

[prime swallow]

This is in your control.

Just tell it to use standard numbered list format, and it'll do so; you can specify how else you want it to make parts of it stand out.

Here's a somewhat whimsical example, but seriously, we can prompt engineer nearly everything it can do. If we don't tell it what we like, then the model is guessing. It was always guessing before, it's still guessing; but as they train the model and update it further, how it guesses can change, and change a lot.

#1070954877779464253 message

https://chat.openai.com/share/54abaf95-4f71-4ce4-b615-61be61c743ba

In the case of this shared conversation, 3.5 really, really struggled. We had to iterate and I had to show examples of the formatting repeatedly - that happens sometimes, and often doesn't happen.

That's 'magic new chat window' effect, and we can work with it/through it - or we can just start a new conversation and usually get a more initially useful set of training data chosen to answer us with.

However, I chose to showcase how to 'find my way' as if I wasn't certain or the model wasn't, having it show me styles and how they look, and guiding it to give me what I wanted output, and when it struggled, showed how to give negative example, positive example style iteration correction. (In this case I left those keywords inferred, but that's what 'I see this from your last output, let's try this format now' is calling on).

And now that we have the format exactly, I could use that in any new conversation.

We can even have the model make every word 'a' and every word 'the' eyepop in color, if we wanted, as it told a story or otherwise discussed:

#1070954877779464253 message

https://chat.openai.com/share/f44b1524-504d-4580-baf0-2f9a2f35303c

It's all in knowing what we want, and guiding the model there.

[frosty light]

yeah, i def do my own little searching if something doesn't seem right, but i've made flown-blown recipes with it more than once and it's always a great time

that's just my experience though

[boreal pecan]

Does anyone know why some chatgpt4 features work and then stop working? like opening .zip files. Chatgpt4 works as expected, then minutes later it says "unable to blah blah blah..."

[tight island]

You’re hopefully not trying to modify these files through gpt’s code interpreter environment in a zip file if so you’re going about it all wrong

It needs to unpack the zip file then make a copy to even begin fiddling with the contents

[vagrant ginkgo]

Has anyone had any luck feeding chat gpt api documents and it reliably use?

[tight island]

As in actual api or chatgpt on the web

The one who costs money to use each time or the subscription

[vagrant ginkgo]

Either but I would prefer the interface like if I had a set of api documentation it wasnt trained on how could I best get it to use certain functions etc from the documentation

[tight island]

To Get the best result i would suggest finetuning a 3.5 Turbo variant with your information.

but that costs money, to even test and do.

[vagrant ginkgo]

How would you go about fine tuning it?

[tight island]

u feed it Data in a JSONL structure

[vagrant ginkgo]

I’ve heard you can fine tun 3.5 not seen how yet

[tight island]

https://platform.openai.com/docs/guides/fine-tuning

it spelt out here.

[vagrant ginkgo]

This is awesome!

[tight island]

do be carefuly Finetuning is more expensive that running the 3.5 model normally.

and does not factor Unsuccesful attempts

meaning even failures will cost you money.

[vagrant ginkgo]

Guess I could ask ChatGPT how to format the documentation best

That’s ok! Been curious to try

[tight island]

no need.

the website specifies exactly how u need to do it, aka the link i gave you.

[vagrant ginkgo]

Oh bet:) thank you for the info

[tight island]

and even gives u an example.

which timed me out.

[vagrant ginkgo]

Thank you again I’ll read over!

[tight island]

my suggestion would be to create a customgpt specifically for helping you finetune. with all the needed Documents, specifically tailored to your project.

[vagrant ginkgo]

Oh good idea!

[boreal pecan]

right. It won't even unpack the zip file (sometimes).

[tight island]

u have to explicitly tell it every single time every single step to take and do and not do, to see consistency.

specifically in its instructions set.

[boreal pecan]

thanks, maybe that's it. I'll give it a shot.

[tight island]

meaning not a prompt, but actual set of instructions to the model.

[boreal pecan]

so not "step 1: extract files from .zip", "step 2: read files...blah, blah, blah"?

[tight island]

i need clarification. are u using ChatGPT 4.

[boreal pecan]

ya

[tight island]

thats your problem

[boreal pecan]

oh?

[tight island]

you need to create a "MyGPT" and tailor it specifically to your task. and only that one task. nothing else.

[boreal pecan]

ah

for consistency, create a mygpt. because unzipping a file works sometimes in gpt4.

[tight island]

i usually use a JSON structure in my instructions for Literal and Explicit understandling.

allows for no interpretation or guessing. and only follows as written.

[boreal pecan]

that's a good tip

ty

[rich surge]

batman

[tame wadi]

All of a sudden about two days ago my custom GPT started insisting on giving the numbers in numbered list outputs their own discrete lines. What is going on 😡!?!

Example: https://chat.openai.com/share/9e60b821-74dc-4dde-8353-5304f7e90c0e

I've wasted hours today trying to prompt engineer away this behavior in the instructions of my custom GPT - all to no avail.

This is vexing, absolutely no one would actually want this wasteful and obnoxious looking numbered list format in the output of their GPT.

[raven portal]

i was just talking about that @tame wadi
we had that topic multiple times in the last days

its gpt (4) doing that not only your gpt

[tame wadi]

Thank you for the quick response @raven portal. So, at this time there's nothing we can do? I checked vanilla GPT-4...it doesn't seem to be doing it...but I saw multiple custom GPTs like mine that are impacted.

[raven portal]

well you tried the only thing that comes to my mind, custom instructions
maybe someone is able to figure out some that work

other than that we will just have to wait for openai to fix this behaviour

which they hopefully will soon, but i cant really imagine them leaving something like that for too long

[tidal light]

me too, i reported this

i think it's the markdown formatting

i only have 3.5, they did something with the markdown formatting that the numbers on a numbered list are with the text, and causes a line break

but before this the numbers were outside of the text

[raven portal]

i think i wouldnt even bother to fix it with custom instructions bc every gpt does it rn
maybe telling it to not number stuff at all is an option but when i just did that the response was way less thorough

[tame wadi]

Ahh, thank you @raven portal & @tidal light! Like both of you, I really hope this was implemented by mistake and will be rectified soon.

Does the OpenAI development team monitor this Discord for reports about bugs like this? Or, do bugs have to be submitted somewhere else [e.g, GitHub issues form]? @tidal light, do you think it would be helpful if I piled on and reported this issue as well?

On the upside, this bug angered me enough to motivate me to get plugged-in here in the Discord 😂. So, assuming it gets fixed soon, I think it's kind of a blessing in disguise. This is a great resource. I look forward to hanging out here and learning more from others who're likewise enjoying the adventure of building custom GPTs.

[raven portal]

https://discord.com/channels/974519864045756446/1070006915414900886

[tame wadi]

👍

[raven portal]

its their official discord server and theres a dedicated bug reports channel so i dont know for certain but theres a chance they actually take a look at it every now and then

i think you should report it

the more noise the better

[tame wadi]

Will do!

[tidal light]

report it, please altough i already did

[raven portal]

and yes i think its a blessing in disguise too, you seem like a very healthy addition to the community

[tidal light]

yo lets report this a lot of times so that they'll hear us and change this

[raven portal]

i also appreciate your custom gpt btw.
definitely a positive and unique addition to the store

[tame wadi]

😁 Thank you @raven portal. Appreciate those kind words 🙏!

[tame wadi]

@tidal light & @raven portal, I have reported the messed up formatting of numbered lists in custom GPTs. Here's the link if you would like to '⭐' my report: https://discord.com/channels/974519864045756446/1226635244632018974

I'm happy to likewise give your reports a star if you reply with the Discord link to your report.

[tidal light]

can i correct you?
it happens everywhere, cuz i use 3.5

so please go edit that out

cuz i dont have plus

[tame wadi]

Edited 👍.

[tidal light]

thanks :)

[tame wadi]

I just changed the title to be more generic too.

[raven portal]

did that

[neat storm]

achoo

[dense halo]

i notice gpt does more simplistic things with a 3 paragraph cap while claude can be tuned in text more, it does more gritty themes and mixes it well with a novel style thing

[prime swallow]

3 paragraph cap? LIke you can't get 4+ paragraphs out of GPT?

This one's 5, though I did ask for 6. https://chat.openai.com/share/cb665a1a-ba2c-44a5-a839-3243177b52dc

[eternal rock]

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.

[neat storm]

3 paragraph cap?

[dense halo]

ill try and give u examples between the two with my claude prompt

[prime swallow]

Sure, though we should discuss in #ai-discussions , where we can talk about non-OpenAI models.

More interested in the output you want, if it's within allowed content, can possibly get it from 3.5 even.

[dense halo]

Can anyone who’s having the super brief replies in voice chat please reply to my post in the bug reports channel. It’ll help bring it to their attention.

It won’t even tell bed time stories anymore. I have to tell it to continue after each sentence

https://chat.openai.com/share/d7ee28fb-41bb-4cbf-91bb-24f4aef01207

[shadow finch]

Wierd question, and I don't think this exists.... but is it possible to use ChatGPT instead of Siri on my iPhone for when I call my phone? I use ChatGPT a lot of small bits of information, and I would love to just talk and ask questions throughout the day.

Also, Will ChatGPT eventually be able to give us up-to-date information like the weather and news?

[sleek gust]

Have you tried saying: "Please dont enumerate the items " ? then you can use google docs to enumerate it again

[sick pecan]

i have android so i cannot be sure, but the internet says iphone has had an action button for chatgpt since november last year, replacing the mute button using the new settings menu, which will launch straight into a chatgpt voice session (which is amazing). you just need an iphone 15 pro or pro max

for android users the best we've got is third party apps like voiceGPT... although that can be voice activated for entirely handsfree

[shadow finch]

God I want to switch to android

[sleek gust]

I think you can map ChatGPT into Siri to open ChatGPT Voice widget directly. it works on android

[sick pecan]

only issue i find launching chatgpt voice direct is you cannot make use of plugins (/custom gpt) without going back to text. but if you launch a chat with the plugin then you can switch to voice and use it.

and talking direct with a database via Papr Memory or Zapier is so amazing

[sleek gust]

Great catch. OpenAI could make some tweaks on this

[sick pecan]

i dont even mind the clicking anymore while its doing its things

AI equivalent of hold music

[random lantern]

as a suggestion under the settings tab and then my plan it should clearly state the number of prompts you can enter or messages received / unit of time ..instead I have to hunt for it ..if I wish to upgrade to enterprise it still doesn't give a definitive answer why ? what is the secret their keeping ...I'm sure there's a place on their website but it is also well hidden why? just doesn't make sense .....

[untold blade]

I forgot if this was the last server I was in... how can I get one thread to connect to the next one

[warped bolt]

Yoyo

[dense halo]

why is gpt responding this way

(\sum_{k=1}^{\infty} P(X=k) \cdot k = \sum_{k=1}^{\infty} (1-p)^{k-1}\cdot p \cdot k)

[solid bronze]

I have some it as long as you phone is unlocked

[long pier]

Can someone explain to me why ChatGPT is acting up today?!

[devout jewel]

hi, I can't send any more messages to chat gpt 4 plus