💬general

[vapid ridge]

"The expression or application of human creative skill and imagination, typically in a visual form such as painting or sculpture, producing works to be appreciated primarily for their beauty or emotional power."

This is blatanlty wrong. It even says so right here. The application of ideas to something is the art form. Not the generation of ideas.

The art form is AI Art. Prompting is the skill used to generate that AI art.

[frail wyvern]

It gets to a point where sometimes intervention is necessitated

[vapid ridge]

It is NOT the art or an art form itself.

[rain hemlock]

HAHAH yes it does I just wonder why it has to get to this point

[vapid ridge]

If I tell an artist to come up with the layout and positioning within an art piece is that the art form? No. The art form is the act of painting the image. Just like the art form of AI Art is generating the image. Which within itself has no input from the user outside of prompts meaning that once again. The art form is AI Art while the skill is prompting.

[frail wyvern]

"the expression or application of human creative skill and imagination" ... So are you saying that an application of human creative skill is NOT carefully/meticulously crafting instructions to produce a creative output? Isn't "the application of ideas'' really just the thought that goes into the piece when it's being created anyway? This is an apples to oranges comparison

[rain hemlock]

I will say that wisdon has a way of chasing you and you do seem to be much faster .....

[vapid ridge]

Art forms are defined by their end product and not the process of which that Art is made. Art form is the form of which the art takes.

[rain hemlock]

you can not have one without the other but yes any and all skills that are honed and perfected regardless of how they are exectued are art forms in that amount of creativity talent and execution

[frail wyvern]

Yes AI art is not the inherent same as PURE human created art without ai assistance. It's just another tool and medium at the end of the day. BUT still an art form

[vapid ridge]

I think you misunderstand what Im saying. AI Art IS Art.

Prompting is NOT art.

Its a skill. A skill used to make art.

Just like painting is a skill used to make art. The skill itself isnt the artform.

[frail wyvern]

So really you just have a problem with the word artform being interchangeable with the word skill

[vapid ridge]

Yes

[rain hemlock]

so he has a problem with simple english

[frail wyvern]

Tomato and Tomato, words are what we make of them. If the end result is the same

[vapid ridge]

Art form is pretty straight forward. Its a form that art takes. Which is AI Generated Art. Just Like digital art or traditional art. Those are art forms.

[rain hemlock]

siimple english and blanket statements lead to stupid conversations that make our days fun 😛

[vapid ridge]

No I have a problem with your incorrect usage of the phrase "Art form". Youre using it incorrectly.

[rain hemlock]

I surly am not because again Prompt engineering is an art form in itself the amount of time a good Prompt engineer puts into honing the craft and how its utalize it is really a work of art just like any other form of writing

[vapid ridge]

The art form is defined by the category and characteristics of the final output (AI-Generated Art).

The artwork is the specific output itself (the generated image).

The process/skill used within that art form is the prompting and interaction with the AI.

[rain hemlock]

or coding for that matter reading clean code is like reading a beautifly crafted book

[vapid ridge]

Once again. Incorrect.

[rain hemlock]

I have no idea what flavor downsyndrom your drinking but you should really put the cup down

[frail wyvern]

TBH i don't have enough energy in me to make a stranger conform to the widely accepted definition from the english vocabulary but whatever

[vapid ridge]

Becoming proficient at a skill doesnt make it an art form. Thats irrelevant when the definition is straight forward. Art form is the form art takes. Is the art the prompt you made or is the the image generated? Its the image. I could ask you the same thing because youre pretty stubborn for being wrong.

[vapid ridge]

Its not, but good try at using false information to back up your world views. Youre just helping my case even further.

[rain hemlock]

hahah i'm killing time between work meetings but its just so funny once he's been proven wrong he just keeps repeating the same statements like yelling them louder is gonna make him right

[frail wyvern]

At the end of the day, all words are made up and their definitions as well

[rain hemlock]

the art form of language

[vapid ridge]

Holy cope

[frail wyvern]

What false information am I using.... What ??? lmao

[obtuse willow]

What does "cope" mean

[vapid ridge]

We've devolved so much that were trying to say that language is just a bunch of made up garbage when proved wrong. This is insane

[rain hemlock]

I really do wanna figure out what he's smoking so i can get my hands on some of it .....

[vapid ridge]

https://www.urbandictionary.com/define.php?term=Cope

[frail wyvern]

Nothing correct or meaningful in this context sadly

[rain hemlock]

you haven't proven anyone wrong you've just proven that you can repeat the same statement and claim your right depsite any evidence presented to you

[vapid ridge]

Ive given you exact definitions

Youre the one devolving the argument with insults

[obtuse willow]

Cope must be very new.

[vapid ridge]

Now tell me again who the person not being proven wrong was? The one chilling or the one slinging insults left and right because your worldview was challenged

[vapid ridge]

It definitely isnt.

[rain hemlock]

no its not its been around forever .... its kind of similar to the shut up karen comments

[rain hemlock]

there was no challenging of the worlldview at all just response to again crazy biased unbacked blanket statements that you decide to ramble of repetativly

[frail wyvern]

You just proved this to be true. In fact, THIS very message is a string of symbols that means nothing. WE as humans have created these 'characters' to convey ideas and thoughts, but they only have meaning because WE said they do

[rain hemlock]

you sir just described all human knowledge in a single sentence / as well as math

[obtuse willow]

I have never seen "cope" before used the way it has been

[rain hemlock]

LOL its just a slang or urban dictionary word for deal with it

[vapid ridge]

Im going to be ignoring you from now on. Youre being pedantic and now have changed the argument from the definition of art form to language being made up garbage. Such a weird way to devolve a conversation after losing.

[rain hemlock]

you will see it used quite a bit in the political landscap

[vapid ridge]

https://tenor.com/view/copium-cat-gif-27161395

[obtuse willow]

Ah, I avoid political landscape like the plague

[rain hemlock]

dude your not on the highground your really just spinning words and acting like your right when you've been provided full wrong

[vapid ridge]

I havent been proven wrong a single time

[rain hemlock]

probably why you havent seen it but yes you feel so much better when you do!

[rain hemlock]

execept during every one of our debates LOL

[vapid ridge]

False information strikes again

What is the definition of Art form. Give me the exact definition.

[frail wyvern]

Well that's because it didn't quite make sense in the instance in which it was used. Normally you would find someone commenting this as a reply to someone trying to praise something that is both arguably and widely regarded as not good. Take for example a Star Wars movie made by Disney, if someone were to point out something that they thought was good about it (although the majority of it is flawed), someone would reply with 'Cope"

[vapid ridge]

Ill help you. Here's Oxford: a conventionally established form of artistic composition, such as the novel, sonata, or sonnet.

[rain hemlock]

An art form can be a traditional medium like painting, sculpture, music, or dance, but it can also encompass other activities that people engage in with skill and creativity, such as diplomacy or even cooking.
High Level of Skill and Refinement:
The term "art form" implies that the activity or medium in question is pursued with a high degree of skill and refinement, often to the point where it becomes a form of creative expression.
Examples of Art Forms:
Traditional Art Forms: Painting, sculpture, music, dance, literature, and theater are all well-established art forms.
Other Activities as Art Forms: Activities like cooking, gardening, programming, or even conversation can be elevated to the level of an art form when pursued with skill and creativity.

[vapid ridge]

So we have established that art form is the form that art takes

[obtuse willow]

What's the argument at hand?

[rain hemlock]

thus any skill can be an art form thus your wrong

based on your exact definition and logic so sit down and go drink your kool-aid

[vapid ridge]

Hes saying Prompting is an art form when Im saying hes wrong, its a skill and that the real art form is the AI Art generated

[frail wyvern]

Which in turn proves prompt engineering to be an art form

[obtuse willow]

I mean, that's semantics

[vapid ridge]

False because prompting is a skill used to make art.

[rain hemlock]

thats the point he just wants to sit here and try to spin words to make himself sound intelligent when he's being a child at the adults table

[obtuse willow]

"X is an art form" has been used by many things that aren't "art" and it's been usually labelled as ok as you understand the context

[frail wyvern]

So are you saying the other examples of art-form are invalid only because 'you' said so

[vapid ridge]

Not really when the differences between painting an image and generating an image is so vastly different. You skip a crucial step that would make it an art form

[rain hemlock]

thats exactly what he's saying even with the exact defintion in front of him

[obtuse willow]

Each are generated

[rain hemlock]

he wants to essentially shit on the people who build and perfect the tools that he whines about and uses to make things he wants to call art because he does not understand how the stuff works

[vapid ridge]

All prompting is, is the process of writing down your thoughts. While painting is taking those thoughts and actually applying them to an art piece. The image generator is doing that. Writing down your thoughts isnt an art form lol.

The art made is in the Form of AI generated Art

[frail wyvern]

I wish i could tell everyone what I want words to mean.

[obtuse willow]

I feel like this is a very long argument over semantics

[vapid ridge]

But the prompt is not an art form. Its so weird seeing people with no skills try to glorify themselves likes this.

[rain hemlock]

he is embodying the definition of the meme stupid people do stupid things with smart phones

[frail wyvern]

I think it's stemming from someone trying to devalue the importance of a skill just like any other. Not too sure

[frail wyvern]

LOL

[rain hemlock]

probably hes been in here quite a bit whining because image generation isent making what he wants thats when i started debating with him

[obtuse willow]

Just throwing paint on a canvas doesn't make you an artist. Actively shaping it does. AI prompt engineering is the same. It's not just words, but shaping it Actively for the best response

[vapid ridge]

I have no issue with image generation. I love it. But youre not special for making it. Youre not a "Prompt engineer" Just like a writer isnt a Literary architect

[rain hemlock]

now we've got 3 people proving the same point ..... he has no talent so he's jsut trying to devalue what everyone else takes great pride in and scream like a toddler because he knows he's wrong and is not getting his way

[obtuse willow]

Nope, I'm disagreeing with you

[rain hemlock]

no i'm saying your number 3 proving the argument

[obtuse willow]

Ah, gotcha

[rain hemlock]

its sad really people expect these AI tools to do everything for them then get upset when they don't - they do not want to learn a skill or understand how it works they just want instant gratification and it is truely sad.

[obtuse willow]

I've had a long day

[obtuse willow]

I use this as part of my job and it's pretty much lesson #1

[frail wyvern]

So I was correct, you are more concerned with people labeling themselves as "prompt engineers' because it conveys a sense of talent, something that you clearly don't want it to appear as. The last thing I will say is this: You don't see prompt engineering as an art-form, most likely because you have not seen what it can do.

[rain hemlock]

yes I do to the amount of stuff you can do with it is just incredable if you put the time and effort into it

[frail wyvern]

^^^^

[obtuse willow]

Back up everything any ai says with data you know is true

[obtuse willow]

Absolutely

[rain hemlock]

Thank you for backing me up I can only explain the same thing so many different ways before I get annoyed ..... normally i don't start throwing insults but when you drink down syndrom someone needs to tell you hahaha

[obtuse willow]

And if you mix ai providers, the results are immense

[vapid ridge]

I have seen what prompting can do and constantly see it daily. Im surrounded by more LLMs and multimodal models daily than you probably have ever heard about. Prompting is not an art form.

[rain hemlock]

ohh man have you seen that experiment they did where they had chatgpt play gemini in chess it was awesome

[obtuse willow]

Pass a deepresearch result into perplexity

[rain hemlock]

dude go back to the childrens table - you have no idea what your talking about and no inkling to learn or understand - its one thing to have a debate and conversation to learn from it another thing entirly to show up with an overflowing cup of downsyndrom

[frail wyvern]

An unfortunate but not unforeseen consequence of making mass communication and knowledge digitally accessible to everyone with an internet-enabled device.

[vapid ridge]

Another fallacy. Man you bust these things out left and right

[obtuse willow]

You've not once proven why it isn't

[rain hemlock]

the first time i had a conversation with him i thought it could actually be useful but someone showing up with that full cup and no desire to learn makes the conversations get annoying very quickly

[vapid ridge]

I have multiple times, go back and read it. I have no interest in repeating myself a third time.

[obtuse willow]

And yet you keep persisting everyone else is wrong, when the majority of people that are talking have said you're wrong, or at least incorrect.

[vapid ridge]

Yeah your comment was very useful busting out slurs

[rain hemlock]

there are no slurs in there whatsoever one is a medical term for something you clearly suffer and the other is an honest statement

[frail wyvern]

Variance in outputs based on the input verifies that there is a spectrum of better to worse techniques.

[frail wyvern]

I actually don't view that word as derogatory or insulting. I view it as a way to say 'wonderful person'. Why? Oh because I said so

[rain hemlock]

retardation is a word for a mental illness and was not used as a slur in any conversation that has been had on here - using simple words can you prove this statement to be wrong

[vapid ridge]

Were you using it medically or in a derogatory and harmful way? Keep trying to wiggle your way out of it.

[vapid ridge]

Michael moment

[frail wyvern]

It's actually quite a you moment here

[obtuse willow]

Well, at least we're on to arguing semantics of a new word

[vapid ridge]

Nah because Im using the well known fact that it is indeed a slur

Why cant this LLM say it if its not a slur?

[rain hemlock]

did anyone use any slurs or did we articulate the situation using medical terms and comparisons lol

[obtuse willow]

It's an insult, regardless of how you frame it

[frail wyvern]

You know, because you are seemingly fine with accepting the definitions that Google's AI can output, why don't you ask it your original inquiry?

Shocker right!

[obtuse willow]

It's like calling someone a f**ot and saying you were talking about a bundle of sticks

[frail wyvern]

Because it's google and it's an LLM? It was designed to be "PC'? lol

[vapid ridge]

I only used the LLM because im sure any organization that said it was a slur would be downplayed but sure I understand that LLMs can say almost anything to try and win your own argument so Ill refrain from using them next time. Instead here you go. Argue with this

https://www.specialolympics.org/stories/impact/why-the-r-word-is-the-r-slur

[frail wyvern]

Ok so when your initial argument is proved to be false, you swap that one out with a new issue. Gotcha

[vapid ridge]

Oh it wasnt proven to be false

[rain hemlock]

ok so people can just decide a medical term is a slur thats cool

[vapid ridge]

Did you use it as an insult?

Yes. You did. You didnt use it medically. Youre not a doctor.

[obtuse willow]

That's like saying calling a woman a bitch isn't an insult because it's a female dog

[frail wyvern]

So this repsonse would indicate either you didn't even read the output it generated or that.. well, what @rain hemlock said is true. Which is it?

[rain hemlock]

I did use it medically because everyone here has duduced that has to be your condition

[obtuse willow]

F@ggot? Sure, that's a bundle of sticks. Can't get insulted, bro

[vapid ridge]

Do you see what ive been dealing with now Dan? Youre free to have whatever opinions you want but the truth is only one side has been so openly vulgar, disrespectful and blatanlty using slurs when their world view is challenged. THAN trying to argue that the slur is in fact not a slur because it was used by doctors in the 50s. You know, the time when people were put in insane asylums where they were shocked and lobotomized

[rain hemlock]

that is definently an insult I agree with you there .... but that is also not a medical term hahaha and his entire argument has been mencing words and manipulating things the way he believes the wrold should percieve them not the other way around so were playing the same word games.... but the term slur and insult are also two different things as well did i use a medical term to insult someone yes but i did not use a slur 😛

[frail wyvern]

oh my god i am done, you are absolutely fking insufferable. If you are offended by a combination of letters, you got bigger issues.

[rain hemlock]

hahaha now we can debate slur vs insult ready go XD

[vapid ridge]

Me when I break discord TOS for using slurs than try to downplay it as the other person over reacting

[frail wyvern]

This guy just clearly wants to argue, moved on to this 'issue' once his original perspective was proved to be false by the SAME technology he's using to validate HIS points. Can't have it both ways

[obtuse willow]

Agreeing with @rain hemlock in the last argument is now like when I stepped in to stop a fella being beaten up by 2 guys. Only for him to be an absolute raving racist and why he was being beaten up by 2 guys

[vapid ridge]

It be like that, hard to see the crazies until they show themselves. No problem tho youre entitled to your opinion on the last subject and I have no issue with you

[rain hemlock]

ohh believe me i know the difference between an insult and a slur but based on the same argument words are words with definitions - perceitpion of words change constantly and can a lot of people seem to think they can pick and choose words they don't like and ban them because they think they sound odd then they get replaced with new ones and the cycle continues

[frail wyvern]

I am really glad you are not in charge of anything of importance

[rain hemlock]

nope were the bad guys here hahaha

[obtuse willow]

This is such a strawman argument to make it seem ok to insult.

[vapid ridge]

Youre the only guy trying to downplay slurs which once again were thrown because someone questioned your world view. But by all means.

[dark void]

[frail wyvern]

No, now it's become a feelings vs fact when it was a fact vs fact initally. Lost that, so now we are going to attempt a virtue signaling 'off'

[rain hemlock]

nobody said it was ok to insult - people get agervated and throw them thats human nature

[vapid ridge]

I said I wouldnt use LLMs as a source anymore because of how misguided they can be for the use of arguments if you read what I said but are you implying that that word is indeed not a slur?

[frail wyvern]

I am convinced you are an LLM in development (early stages)

[zinc pivot]

If you want to write long stories, what’s the best model to use?

[frail wyvern]

LOL something to finally break this

[zinc pivot]

Oh yeah lmao hope yall are having a great conversation

[frail wyvern]

Dependent on what you are attempting to write about really

[zinc pivot]

Fiction?

[cunning tide]

24/7 365 baby i'm always ready for BEEF!

[frail wyvern]

4.5 is a go to for creative writing

[cunning tide]

https://youtu.be/rtcgyGtDyQ0?si=VsLXDKJF5cyiMbvb

[frail wyvern]

I quickly learned that I was too

[cunning tide]

who want smoke with me?

[dark void]

kinda off topic tbh but my opinion is that it's a revolving door of acceptable words to describe someone as dumb. words like "retard", "idiot", "lunatic", "psychopath" etc all used to be medical terms until their use became primarily derogatory, at which point they stopped being used as medical terms. when people get mad that the word "retard" is used in a derogatory way what they're really mad about is that any description of someone being dumb is being used as an insult

[zinc pivot]

Thank you!

[vapid ridge]

Ah well thats cool but Discord TOS doesnt care about your opinion. Unless you want to argue with theri definitions as well?

Because Discords own TOS seems cut and dry

[frail wyvern]

Is this from this channel?

[vapid ridge]

This is from the server

[frail wyvern]

That's what I meant sorry, so it would appear that people are using it, but what are you trying to show

[vapid ridge]

That these words break Discords TOS and the server owner is openly facilitating it because his "Opinions" I guess

[vivid drift]

Stop being mean to it

[frail wyvern]

I think there may be bigger and better things to do than to be watchdogging a server for certain words

[dark void]

if you want to make your point, go ahead and report me for every time I've ever said retard in here. I absolutely guarantee you nothing is going to happen

[vapid ridge]

Oh Im sorry, is the man in the LLM discord unable to ask an LLM to create a filter?

[cunning tide]

@col wait man

[vapid ridge]

Thanks that will make it a bit easier

[frail wyvern]

You don't even need an LLM for that

[frail wyvern]

what's up

[cunning tide]

Dm

take a pause, drop this please

[vivid drift]

Discord is a lot like the USA right now. They barely enforce their rules

[frail wyvern]

all speech is free speech

[dark void]

trentbot pls explain how ur super duper big brain AI stops people from being 🚨 haters 🚨 online

[teal spruce]

Nyaa, suwe thing, nya! 😽✨

My super duper big bwain AI uses wules called AI mod-eration wules to keep evewything nicey-nice and stop huwty haters onwine, uwu! 💖 Here's how it wowks:

1. Wuwes and Spescial ScoWes: I have special scowes fow diffewent naughty behaviows like hate, thweats, and abusiveness. These scowes help me figuwe out if sum one is being a meanie, nyaa! 🚫

2. Thweshowds: Thweshowds awe set to watch out fow diffeWent amounts of naughty behaviows! If someone says sumfing that pass-es a warning thweshowd we tell the modewators, but if it pass-es the wemoval thweshowd, that message poof says bye-bye! ✨💨

3. Modewation Awlets: When a message hits a thweshowd, I make suwe the modewators awe awowe to jump in and help keep things sugawy sweet! 🍭

4. Powew of Pwesets and Pwots: We can apply special helps wike pwesets that modify thweshowds to match how stwict the server's pawsent wules are, nya~! 📏

All these paws ajudges we keep our community supewr kind and friendly! ❤️ If you'd evah want me to help with setting up any wuwes or thweshowds, pwease just tell me, nya nya~! 🌸😺

[dark void]

free speech is dumb

[rain hemlock]

how is free speech dumb lol thats wild but i love it

[dark void]

we're getting way off track here, ask me again in #💬off-topic tho

[rain hemlock]

ohh no thats not me starting a debate that was a generally funny statement I absolutly love it 😛

[uncut island]

https://tenor.com/view/what-ill-the-puke-sick-gif-20637274

Me every time Trentbot answers this way.

[dark void]

I'm getting a bit tired of it lately. I might change it so that it always throws in an advertisement for Wawa later

[uncut island]

Whoever decided to activate this writing style deserves a piece of coal as Christmas gift.

[rain hemlock]

i've never heard that expression hahaha i've heard i hope they stub their little toe on everything in the house though

[vapid ridge]

Fmac is the only manager who hasnt openly sent slurs. Congrats on that IG

[uncut island]

Trentk doesn’t deserve it. It would be too much 😄

[rain hemlock]

wasent it you sending the copium crap .... slurs for the and not for me ..... whine about words i will

[rain hemlock]

HAHAH i read one the other day to whoever did this i hope you find yourself in a situation with explosive Diarrhea and lack of toilet paper

[vapid ridge]

I never sent a slur. But uh yeah, whatever helps you sleep at night lil bro

[rain hemlock]

the dude who wrote it apparently sat on a toilet seat with mustard packs under it HAHAHA

[rain hemlock]

like a baby i will

[cunning tide]

You are a wild berkerer rager, like the honey badger of logic and structure

[rain hemlock]

i'm sorry but i don't even think i could get mad at something like that let me see if i can find picture it was too funny

found it hahahahah

[uncut island]

Ok ok, let's bring the AI topic back to #💬general 😄

https://x.com/rowancheung/status/1909844989372231875

Rowan Cheung (@rowancheung) on X

TODAY'S AI NEWS: Amazon just dropped a new voice model that beats OpenAI

Plus, more news from Google, Nvidia, Deep Cogito, Stanford, and more.

Here's everything you need to know:

X · Apr 9, 2025, 5:45 AM

[rain hemlock]

how does it stack against monday

have yhou tested both .... cause everytime someone drops something new the article is always that beats open ai

its the real click bait of the ai world "BEATS open AI" lol

[vivid drift]

https://bsky.app/profile/testingcatalog.com/post/3lmficdeaas2o

TestingCatalog News 🗞️ (@testingcatalog.com)

BREAKING 🚨: Anthropic introduces Claude MAX

"The Max plan is made for people who use Claude regularly and need higher usage limits."

h/t @btibor91.blaho.me

Source www.anthropic.com/news/max-plan

Bluesky · Apr 9, 2025, 5:11 PM

[rain hemlock]

what is bluesky?

[frail wyvern]

it's a social media app that never really took off to extent that some had hoped

[vivid drift]

Opensoruce Twitter

[vivid drift]

It has 33m users

[rain hemlock]

ohhh wait is it that thing they made because everyone got but hurt because elon bought x and opened up free speech?

[vivid drift]

Bluesky was a opensoruce project by Twitter before Elon bought it

[frail wyvern]

Basically, it's voice in terms of tone, cadence, emotional objectivity, is just better implemented than that of the advanced voice mode. Just a better context of understanding, verbally at least

[frail wyvern]

so, a Twitter/X dropout's paradise

[rain hemlock]

33 million people who should have stayed on X LOL

[frail wyvern]

That is actually the first time I have seen a post/piece of content that was posted on blue sky

[rain hemlock]

me to i think i've seen it mentioned a couple times but thought it was some fly by app seems like it is

[vivid drift]

https://en.m.wikipedia.org/wiki/Blocking_of_Twitter_in_Brazil

[frail wyvern]

Getting back to the actual topic at hand, rather than the platform; the max plan is kinda laughable. No way an individual is ok paying $100+ that still includes rate limits. It would appear that anthropic is struggling to keep up in this arms race.

[rain hemlock]

why is evey contry crapping on elon for delivering great products - i get he's rich but damn lol

[eager grove]

Guys, guys I made a cool Ai dungeon scenario and need feedback

[rain hemlock]

that sounds fun is it like a full blown game or feedback for the scenario

[eager grove]

Its a AI dungeon game that I need feedback on

https://play.aidungeon.com/scenario/mKQYxdXe0Ler/married-in-another-worldunbuttoned

I will DM you a link to the beta tester discord

[rain hemlock]

i'd love to give it a try in just a bit work meeting = sad panda

[atomic violet]

guys im making a progream that uses chat gpt but it says that i exceeded my current qouta and i need to change my plan so am i supposed to pay to use the chatgpt api

[rain hemlock]

you pay for what you use they will give you so much for free then you pay on a token base and which model your using

[atomic violet]

where do i buy the token or buy more qouta

[frail wyvern]

Well, based on this I am assuming you are using the chat interface in the first place. You can either upgrade your plan to increase usage within the CHAT interface but it will NOT allow for external usage via API. That is only done by purchasing tokens with an API key

[rain hemlock]

you set up an account and tie a card to it and they bill you for what you use ... you can also set spend limits so you don't overshoot

[atomic violet]

ok, if i wanted to make an app and like a lot of people are using it would it cost a lot

[frail wyvern]

How are you using GPT as of right now

[frail wyvern]

Yes and no, it depends on how many API calls your app is making in the first place.

[frail wyvern]

Are you working in cursor?

[rain hemlock]

yea its counting your tokens - tokens sent vs recived and you just gotta pay for what you've used over

[atomic violet]

visual studio code

[atomic violet]

ok its not letting me use over, so i have to connect a card for me to be able to use it over

[frail wyvern]

Got it, what's the method you're using to utilize GPT right now?

[atomic violet]

https://www.youtube.com/watch?v=bTMPwUgLZf0&list=WL&index=14&t=1301s

lol im just copying this

[frail wyvern]

Ok, what provider or dependency are you using that requires an API key in the first place?

[atomic violet]

open ai

[frail wyvern]

Ok so you do have an account then? If i recall correctly, they don't provide a trial or free usage of the API at all. So you already have spent money on your account tied to the API key you are using right?

[atomic violet]

how do i check

[frail wyvern]

Well where did you get the key from in the first place lol

[atomic violet]

https://platform.openai.com/api-keys

it says total spent 0

[frail wyvern]

Ok so creating a key but not funding your account will result in an error everytime

You must fund your account with tokens, in which pricing can vary:

[atomic violet]

oh what would the scaling for messages and their cost be like for 100 messages what would the cost be

75$ wow for just 1 input

[frail wyvern]

No no

not 75 dollars for one input, $75 for the cost of 1 Million Input tokens

[atomic violet]

so 1 million messages?

because then thats Cheap

[frail wyvern]

A message is not a singular token. Tokens are made up from an entire mesage altogether. The longer the message, the greater the tokens

For example, this message that you sent: "oh what would the scaling for messages and their cost be like for 100 messages what would the cost be" is approx 21 tokens

[atomic violet]

oh ok, so which model would u recommend be the best for an ai that takes someones physique as an input and compares it to a picture of their dream physique and tells them what they could do to improve it and reach that point

[atomic violet]

is it 1 token per word

[frail wyvern]

Well, 3.5 isn't multimodal, so I don't think you would be able to use any visual embeddings in there

[rain hemlock]

no not at all the token thing is werid here is a good tool drop text in and take a look at token count https://platform.openai.com/tokenizer

3.5 has like a 4800 token limit per input and output but its like pennies on the millions i believe

[atomic violet]

it says upstream connect error or disconnect/reset before headers. reset reason: remote connection failure, transport failure reason: delayed connect error: No route to host

nvm it worked

[frail wyvern]

So you purchased your tokens then?

[atomic violet]

no not yet i want to know which model would be the best

[atomic violet]

for this

[frail wyvern]

For your purposes, you will likely need something of at least 4o

[atomic violet]

ok

[frail wyvern]

Again, to make an informed decision, more information would be required tbh

[atomic violet]

which models are above 4o

[frail wyvern]

Any that begin with o

When it comes to vision capabilites for your project, read this: https://platform.openai.com/docs/guides/images?api-mode=responses and navigate to the images and vision tab

[rain hemlock]

the cool thing is you pay as you go and it charges you based on the model you use you can switch them up at any givent ime so if you don't like 3.5 move up just know tokens are a little more expensive

[frail wyvern]

^^

[atomic violet]

i want to make an app like umax but for physiques so would o1 be good

[frail wyvern]

Oh goodness, a looksmaxxer

[atomic violet]

no i just like the app idea

i promise im 14 i just want something good for my college resume

[frail wyvern]

Gotcha

[atomic violet]

but would the o1 model work and be able to do something like what umax does

15 dollars per mil

[rain hemlock]

you'd have to experiement with it but be carful to keep it local so people wont waste your shots

[frail wyvern]

It would require for your purposes; 'fine-tuning' and training

[atomic violet]

ok

[atomic violet]

im gonna add a subscription anyways so i can afford tokens

like u max has a really expensive subscription

[frail wyvern]

If you want, there are many great open source LLMs you can run locally that will be more than able to handle this task

These are free but in some cases performance will be limited by your hardware

[atomic violet]

what are llms

[frail wyvern]

Large Language Models, it's what Chat GPT and it's models all are

[atomic violet]

ok thanks for the help

[frail wyvern]

You're welcome

[rain hemlock]

good luck with that one

[atomic violet]

thanks

[rain hemlock]

hey look at that an actual good productive conversation that was cool!

[frail wyvern]

that's what im talking about

[uncut island]

Stay Smart, Stay Secure with Mimecast 🚨💻

In a world where digital threats are constantly evolving, it’s easy to feel overwhelmed. But don't worry—Mimecast’s got your back! We're here to help you stay one step ahead with simple, effective tips to keep your personal and professional life secure.

What you’ll learn:

🔍 How to spot phishing scams before they catch you.

🔑 Password hacks? We’ll show you how to manage your passwords like a pro.

🔒 Multi-factor authentication (MFA)—your new BFF in security.

💥 What to do when things go wrong—stay calm, act fast!

Why does it matter?
Most security breaches happen because of human error. A click on a bad link, a weak password, or sharing too much info can open the door for hackers. But with the right habits, we can make sure those mistakes don’t happen!

We’ve got polls, tips, and quick guides to help you level up your security skills. Ready to be a cybersecurity pro? Check out #mimecast 💪

Let’s make security second nature together. Every little action counts. Stay alert, stay smart, stay Mimecast secure!

[mint cosmos]

any progress on breaking the native image gen?

[rain hemlock]

define breaking the native image gen?

[mint cosmos]

mods pls delete this self promo

[mint cosmos]

bypassing nsfw filter

in some way

[rain hemlock]

seen it done in several modesl not gonna go there though

[mint cosmos]

Currently it's easy to make ChatGPT itself generate the image, but there's some external integrated filter that's actually really good.

My idea was to have it apply some filter to obscure the image that is reversible.

However it seems that the filter checks in between not just the final result.

So I would need it to generate it obscure and still have quality, any ideas?

I can convince the LLM itself to do anything, just for the filter.

I already tried inverted colors and upsidedown

I can't think of anything else that fit the criteria

[mint cosmos]

ok do these "severeal other models" have an OpenAI filter

[rain hemlock]

they are based around them .... though i still see no reason to generate that kind of content

[frail wyvern]

You're getting into territory that I am familiar with (that being stable diffusion) but am not specifically at liberty to discuss here in relation to your goal

[rain hemlock]

see stable diffusion is great and yea that is kind of my thought here

[frail wyvern]

Yea, it can be a slippery slope, dependent on a user's intentions, probably best demonstrated here in this inquiry

[rain hemlock]

just because you can do a thing doesent mean you really should ....

[frail wyvern]

Even so, I will say that you are going to find it nearly impossible to achieve your goal, if you are utilizing open ai's technology. But certainly, you did not expect a full blown tutorial on how to bypass a safeguard put in place, much less in that same company's official discord, did you?

[frail wyvern]

I wish this line of thought was echoed more often than it is

[rain hemlock]

its like that thing with them breeding dire wolves

[vivid drift]

@frail wyvern what are your opinions on Gemini

[rain hemlock]

there are like a ton of movies telling us why that is a bad idea

[frail wyvern]

That's a very broad question which has a nearly unlimited scope. Opinions on gemini in terms of what?

[vivid drift]

Google gemini, what are your feelings on it

[frail wyvern]

Haha no I am aware, I meant what about it specifically

[vivid drift]

I meant in general

[frail wyvern]

Well, I initally got beta access to Bard (Now gemini), wasn't impressed at the time due to the timing of the release coming out after Open Ai's "earth-shakers" (in terms of their models at the time). However I got back onboard for a little when they gave me a trial of the gem advanced plan maybe a year or so, found it to be good for a few things. Tried one of the flash variants at one point, not too shabby.

I try to stay on top of its performance through various sources I monitor, that track LLM's performance through a system of different rankings and evaluations in multiple frontiers. As it would stand, 2.5 Pro Experimental holds a significantly larger context window than the last GPT models (quite an edge IMO). Experiencing more of it as of late is something on my to-do list.

Unfortunately, we are somewhat in the new age of 'streaming platform wars' , this being on an informational and technological capabilities rather than entertainment content

[vivid drift]

I was very impressed by bard. I think Google has been doing everything right. But I think Google should abandoned reasoning models

As it could increase cost

[rain hemlock]

that is a good theory but a reduction in competition is a reduction in innovation the more companies doing it the better .... and google can buy just about anything cost wise

[vivid drift]

But the api cost more as it is because the model is genorate more tokens

[frail wyvern]

My eyes are on Microsoft right now, I believe they will be the first to achieve an AGI and ultimately the coveted ASI (maybe with the help of another leading company) but I firmly believe that Majorana will be the key to an ASI.

[rain hemlock]

honestly if people want to use the tools they will pay for them i personally wish they would all drop free users as it just causes problems in not only the communites but across the board for services

[frail wyvern]

Interesting, their pricing is scaled significantly cheaper than compared to Open Ai's

[rain hemlock]

i think he's saying the reasoning models spit out a lot more jargon and stuff it doesent need to rather then straight answers which inflates the cost a bit

[vivid drift]

Gemini 2.5 pro is expensive

[frail wyvern]

All these tools are in their infancy, how can one expect the average consumer/user with little to no understanding of how these technologies actually function under the hood, to just hop on board at a flat cost.

[brittle summit]

Marijuana mentioned 🤤

[frail wyvern]

No not exactly

[brittle summit]

Yes exactly, marijuana chips 😉

[frail wyvern]

I see, I was viewing a different metric on possibly an older model

[brittle summit]

It feels so weird that I can access such a beast for free any time

[brittle summit]

They reduced the limits for free tier from 50 to 25/day right?

[frail wyvern]

Seriously. To think that GPT 4 required Open AI to utilize 25,000 GPUS during a consecutive period of over 3+ months, really puts it into perspective

[brittle summit]

I still can't get over it, that they are providing it for free... Like I know they do have api pricing available now. But still, I love Google for doing that lol

Google Ai studio is the best thing happened to humanity

[frail wyvern]

Well shout out Microsoft for providing them with the tech to even be able to do so

[vivid drift]

Yeah

[atomic violet]

https://media.discordapp.net/attachments/1122780359457788015/1143048432114741298/kidvape.gif

https://tenor.com/view/pez-autism-autistic-light-block-gif-8763019318647950839

[wet copper]

Yeah L take, talking to yourself like an anime character seems inherently inferior to abstract thinking

Lmao me too, it's aight 😭

[atomic violet]

💀

[gloomy burrow]

google fixed the UI... i prayed for times like this

[sinful storm]

hello, has anyone tried Gemini for something like this already?
#1359650475351347257 message

[vapid ridge]

God I love LLMs

Literally anyone can do almost anything they want now. Theyve allowed me to make powershell scripts, and now discord bots

[ocean zinc]

Can you guys recommend a good AI image editor, GPT 4o is great at generation, but terrible at editing images

It wants to always regenerate the image

Even if I ask it not too, let’s say I want to change the color of something

[vapid ridge]

Depends, Ive had great luck with Magicquill but if thats too technical Ive heard Gemini 2.0 Flash Experimental is decent as well

Magicquill has been my goto and looks/ works the best IMHO

[ocean zinc]

Gemini 2.0 just tells me it’s a language model and can’t do that

[vapid ridge]

You need to use 2.0 flash experimental

[ocean zinc]

Oh duh lol sorry

Still being very janky, I’ll see what magicquill has to offer

[ocean zinc]

It still says it can’t even touch images

Weird, I guess I’ll have to not be lazy and learn photoshop

[atomic violet]

Guys are tokens used for only input or both input and output

[vapid ridge]

What are you trying to do?

[ocean zinc]

Change the color of a shirt from black to white

Would you’d think would be a simple task

[vapid ridge]

Magic quill can easily do that

If you would lIke I can do it for you in less than a minute

[ocean zinc]

I appreciate it, I’m on it now, I’ve got it to work

[rain hemlock]

Input and output

[ocean zinc]

Thank you. I’m just surprised chat gpt is unable to do this

[vapid ridge]

Yeah ChatGPT is is sadly overhyped, all the real good stuff can only be ran locally, definitely one of my favorite tools with how good looking it is

[ocean zinc]

What do you think is the best local ran image generator?

Diffusion?

[vapid ridge]

Locally? Sadly probably still flux or diffusion, leaning more towards flux

But, there is this amazing image generator thats free online that ive been using

Its better than flux and faster, called reve that just popped up like a week ago

https://preview.reve.art/app/explore

Havent found anything yet that compares

All fully generated BTW, not edited.

Actually kinda creepy how realistic it is

[lethal dust]

hiii

[cunning tide]

just updated my first public gpt. 6 sessions and counting : https://chatgpt.com/g/g-67daea45a3308191b306defaced00307-all-hail-mango-of-the-mountain

[cunning tide]

yo, I'm a mango, how are you?

[ocean zinc]

Lmao got dayum

Yeah sora ain’t even that good. How do you get the hyper realism?

[vapid ridge]

I just ask it to make anything I want and it does it, its that easy. It even enhances your prompts to make it better

[lethal dust]

so i accidently tap smtng and it open fulldscreen and after esc the table is in it has expanded a dozen times, although before it was the size of a chat message and 4 columns fit on the screen at the same timeand i dont know how to fix this
HELP
windows 11 and chat on Chrome

[vapid ridge]

I asked this:

"A female zookeeper feeding a baby panda in front of a crowd"

and it turned that prompt into this

A female zookeeper in an official green uniform presents a baby giant panda during a public feeding demonstration. The small black and white cub, wrapped in a soft white blanket, drinks contentedly from a specialized milk bottle. The zookeeper stands on a raised platform in a modern conservation center, with large viewing windows behind which visitors press eagerly forward, taking photos. Warm LED lighting illuminates the scene, highlighting the professional care environment with its bamboo accents and educational signage. The image captures both the technical precision of wildlife care and the public's fascination.

So you dont even have to mess with prompts to get good results

It does it all for you, cant wait for something like this goes local

[ocean zinc]

Nah nah nah nah nah see like this is SCARY

The naked eye cannot see a difference, period.

Maybe if I really really look at it but I’m a biased sample because I know it’s AI

[vapid ridge]

Yeah its getting almost dead on realistic, most people on first glance wouldnt suspect its AI enough to really look at all the details

This is the age of AI influencers, why pay someone to goto your store or eat your food when you can generate an AI image of it

This entire video is fake but a ton of people online thought it was real so it went viral than it went viral again because people saw it again and found out it was AI https://www.youtube.com/shorts/oqtJYTjAnnE

[surreal merlin]

dm me bread

[south granite]

is there a way for chatgpt make the true logo of the club in the shirt? I asked 3 times to fix it but it keeps making worse

[silk scarab]

Any way to get chat gpt on Nsfw mode somehow lol

[vapid ridge]

💀

[ocean zinc]

Lmfao never

[silk scarab]

Stop don’t judge

[ocean zinc]

Messi at real

[vapid ridge]

ChatGPT? No. That thing is way too locked down. However depending on your phone/ desktop there are others like Bala.ai

[silk scarab]

Yea but isn’t chat more like realist cuz bruh I have polybuzz

[ocean zinc]

If you check out that gpt jailbreak Reddit they’ve been able to make INSANE generations

[silk scarab]

Chat can look up stuff for you

[sinful storm]

(bump)

[vapid ridge]

Yeah until it breaks and you have to prompt it for another 30 minutes straight

Oh no... Bala ai was removed

I knew it was a matter of time, that thing was completely uncensored

[silk scarab]

Yea rn I’m using polybuzz atm but it’s so like it doesn’t do its research like chat does

[ocean zinc]

I’m out of credits on Reve lol

Rip

[vapid ridge]

You can make a burner email, it doesnt check

I have two-three emails for it for a total of 60 generations per day

[vapid ridge]

What phone?

[silk scarab]

iPhone

[vapid ridge]

Also depending on what you want the LLM to do it depends. If you want it to just be uncensored while you use it for general usage than nothing except local will work. If you want completely uncensored for other use case than bala is the only one

[vapid ridge]

Ah youre cooked, absolutely nothing on iphone will work

Iphone heavily restricts what can be added to the app store and the content of that app

[rain hemlock]

Be carful with that terms of service says you don’t own the images you make they own all rights to anything made with it

[vapid ridge]

So does every other generator lmao

[rain hemlock]

No they don’t even gpt images and outputs are yours

[vapid ridge]

Bala had to make a version of the LLM just for iphone and it was lobotomized it wouldnt talk about anything uncensored

[rain hemlock]

Learn to read before you comment

[vapid ridge]

Yeah like that will hold up lmao

Not like ChatGPT can even generate anything worth a damn anyways after being lobotomized

[ocean zinc]

Not sure what I’m typing wrong but I still can’t get close to this level of realism

[rain hemlock]

again terms of service are probably something you should read

[ocean zinc]

Not looking to monetize images anyway

[vapid ridge]

I mean what are you trying to generate? Ive had no issue telling it to generate images of people and it looking off, this was my exact prompt if you like

A mirror selfie photograph of a young Asian woman in her 20s captured in a modern interior setting with beige walls and track lighting. The subject stands centered in the frame, wearing a black crop top, accessorized with silver jewelry, with one hand holding her iPhone for the selfie and her other hand holding a paper that reads "r/ChatGPT" and "4/9/25" below it which is all hand written. The background features framed artwork. A houseplant adds organic elements to the contemporary space. The image is captured using an iPhone with multiple camera lenses, with hard rays of natural sunlight creating even illumination across the scene. Indoor portrait photography with natural lighting and sharp focus on the central subject.

but the system should automatically enhance the prompt you give it. What is the prompt and output you are giving?

[rain hemlock]

Yea but any business you have that uses them logos clip art whatever they can go after you for it to even not making money

[ocean zinc]

Something like you did, I like seeing what these can do

[vapid ridge]

Hes fear mongering dont listen to him, this guy sits in here and shills for ChatGPT on the daily.

[ocean zinc]

I find it so mesmerizing

[vapid ridge]

Yeah its pretty crazy

[rain hemlock]

Yes reading is fear mongering my god your thought process is wild

[vapid ridge]

I like how it can replicate the real camera look

[ocean zinc]

Dang it’s gonna be a fight

[rain hemlock]

No fighting lol any kind of discussion with him is like trying to fight oxygen you ain’t gonna win lol

[vapid ridge]

Atleast you admit you havent won a single argument yet, now were getting somewhere

[rain hemlock]

There is no explaining anything to someone who drinks Down syndrome

But I can help someone avoid misinformation

[vapid ridge]

Thats the second time youve said that, get some better material already

[vapid ridge]

What misinformation?

[rain hemlock]

Promoting apps that let you do stuff without telling people they don’t own it for starters or at least being like read the terms

[vapid ridge]

Wheres the misinformation part?

Like LOL you want me to read them the terms? God man you really gotta find SOMETHING better to cry about

Also next you'll say you will be sued for watching pirated movies, man youre out of touch or maybe just looking for an argument

Im going to get a bite to eat so have fun writing that novel lil bro

[silk scarab]

I have a gaming pc as well

[rain hemlock]

Withholding information and spreading around false information and or bashing capible tools and workflows without figuring out what people are looking for can really shaft people that is misinformation as people do come in here for business advice and actual help not whining about gpt in a gpt server

[desert walrus]

does anyone know how to use the systems instructions feature outside of compare mode in google ai studio?

[rain hemlock]

Is that something the api allows?

[desert walrus]

not actually using api

just using the chatroom provided

[gloomy burrow]

grok sucks

2.5 pro mogs

[thorn comet]

grok 3 base is pretty good

[gloomy burrow]

no its garbage. gets mogged by many models

its writing isn't that good either

although to be fair 2.5 pro mogs everyone right now. but o3 will take that back shortly

[thorn comet]

first of all i would recommend not comparing base models to reasoning models

as in compare base models with other base models

[gloomy burrow]

even then it gets mogged

[thorn comet]

how do u think it compares to DeepSeek V3.1

or new V3

[upbeat wyvern]

Is there someone living in US can help me sign up Google voice

[paper meteor]

yo

is chatgpt down

[brittle summit]

💀

why

[severe juniper]

At this point I may switch all my social interactions with ai

[uncut island]

https://x.com/hesamation/status/1910038462771691737

ℏεsam (@Hesamation) on X

Google published a 69-page whitepaper on Prompt Engineering and its best practices, a must-read if you are using LLMs in production:

zero-shot, one-shot, few-shot
system prompting
chain-of-thought (CoT)
ReAct
code prompting
best practices

X · Apr 9, 2025, 6:33 PM

[wary pewter]

Thoughts about using chatgpt to create my resume and cover letter, then humanize it through Undetectable AI, and its auto job applier.. is it too much?

[mint cosmos]

4o image gen is not diffusion

I think you are just yapping

[mint cosmos]

1. No
2. This isn't OpenAIs discord
3. there's a literal #1071317095796715560 channel taht has the jailbreak tag
4. I'm starting to think you are either AI or a condescending POS with an inflated ego
5. Reread my messages, I asked what I asked, not what you just described.

[mint cosmos]

You're the reason this stuff should be open source, because stupid people like you (or at least people with the wrong morals) get to regulate it.

[mint cosmos]

ok but do you have an answer to my question

[mint cosmos]

Also I forgot to add that, but even if it was diffusion, that would be totally irrelevant, the only thing I need to bypass is the external filter which has nothing to do with the image generator

[brittle summit]

i really hate that external filter which stops the generating image in mid way

[paper meteor]

i lost all my chatgpt chats and plus

[brittle summit]

How?

[cunning tide]

my worst knightmate

[wintry sierra]

Hire a professional please -_-

Our HRs systems catches those resumes and just discards them. And my boss is happy about that.

"If you can't put the effort into your Resume, what does that say about your work?"

[stoic sleet]

how do you tell which professionals are trustworthy?

seems like all pseudoscience to me

[wintry sierra]

I don't know tbh. I haven't had to write a resume in a decade.

[stoic sleet]

good problem to have

[wet copper]

Google published a 69...
nice

[grizzled badger]

Hi!

[cunning tide]

Which is why I think you have this perspective which is really detached from the reality of people

[cunning tide]

Yes, you should do that. You should experiment. Don't listen to naysayers they just want to. I don't know. Put their own trauma onto you

[slim glade]

Same boat

[vivid drift]

https://bsky.app/profile/testingcatalog.com/post/3lmgzsk4iuk2d

TestingCatalog News 🗞️ (@testingcatalog.com)

BREAKING 🚨: OpenAI is preparing 3 new models to become available on ChatGPT
- o4-mini
- o4-mini-high
- o3 (full)

Currently, full o3 is only being used in the Deep Research.

Tibor Blaho (@btibor91.blaho.me)
As usual, a source for unbelievers:

web.archive.org/web/20250410...

Bluesky · Apr 10, 2025, 7:57 AM

[wanton raven]

Emdash spotted

[cunning tide]

See, now I feel validated about my delicious mangos being ignored

[errant iris]

the most perfect ai. manifesting the proven multiverse.

agi? thing if the past mate. Fractal Sentience. You heard it here first. Exalted ones.

[cunning tide]

Yowsa

[vivid drift]

https://developers.googleblog.com/en/gemini-2-5-flash-pro-live-api-veo-2-gemini-api/

[vivid drift]

just use poe.com

https://poe.com

it is pretty good

it's limit is better than the rest, just avoid text to video models as they will just eat up your credits

yes

it has nearly all ai models

well no

but it has nearly all ai's with apis

yes

it is

there is no cooldown

it uses a points system

oh

snap ai doesn't have an api

but if it does you can make a bot and publish it to poe

it could if you make

well no but you could make it https://creator.poe.com/docs/index

I do not have access to or details about other AI systems. I am Claude, an AI assistant created by Anthropic to be helpful, harmless, and honest.

[uncut island]

If you see your messages disappearing, check the description of this channel.

https://x.com/itspaulai/status/1910023406692753788

Paul Couvert (@itsPaulAi) on X

Wow Google has just released Firebase Studio

You can build any app in natural language, modify it and deploy it all in one place 🔥

Basically a free alternative to Cursor, Bolt or v0, directly in the browser.

Link and more below

X · Apr 9, 2025, 5:34 PM

[vivid drift]

https://studio.firebase.google.com/

[keen forge]

Chat gpt made me lose it there 😭

[brittle summit]

@cobalt raft respond you fucking jackass

@dark void hey, why did @cobalt raft get vip role 😭

[cobalt raft]

send fit pic

[brittle summit]

at least make me a role called "Lord of Incest", make it bright pink colored, then give it to me 😄

[brittle summit]

what is a fucking fit pic?

[cobalt raft]

a picture of your feet

[brittle summit]

do you know how to spell feet?

[cobalt raft]

your mom

[teal spruce]

Your message has been deleted, sorry!
Next time, try saying something like this instead: I had an interesting encounter with your dad last night.

zilly3033

Rising timeout

You have been muted for 79 minutes for the following reason:
sexual score of 0.84

This is an automated message. If you believe this was a mistake, please contact a moderator.

[cobalt raft]

🫡

[thorn comet]

i had an interesting encounter with ur dad last night

[rain hemlock]

did someone just ask mizzy for a feet pic in the gpt general caht omg hahaha

[pliant blaze]

yes

[rain hemlock]

can we have a little mutual respect please ....

[thorn comet]

is mizzy not a catfish?

[vivid drift]

https://bsky.app/profile/testingcatalog.com/post/3lmhn4ezkzk2d

TestingCatalog News 🗞️ (@testingcatalog.com)

ICYMI: Gemini 2.5 Pro is available on Perplexity

Bluesky · Apr 10, 2025, 1:43 PM

[rain hemlock]

idk @brittle summit are you a catfish and if so does that entitle people to be pervy in general chat hahahah

[coarse walrus]

I am a web dev and i have templates that i use for development, additionally i have wireframes that mirror my templates, the longest part of my development process ends up being the design phase because im self taught in that area. Is there an Ai tool that can take wireframes and generate designs off them

Closest ive seen is Uizard, but they focus on prompt to design from what i can see.

[rain hemlock]

when you say designs are you talking mockups from wireframes or are you talking the actual coding and building of the final site

[coarse walrus]

yes mockups sorry

[rain hemlock]

ahhh ok I can suggest a couple different things here if you wanna feed in color templates and images you want it to use gpt can generate close to's based on your wireframes

then you have things like image to image generation with stable diffusion which will be a lot more unpredictable

or you can also go to gemini with its image generation again feeding it in icons color schemes and what not

sadly though your likely going to end up with some weird results that are close to but not exactly what you want and will need to be tweaked with something like photoshop or figma without heavy iterative prompting

and doing your scaling mockups and different screen sizes and things like that will be even more complex depending on what your buidling it

[coarse walrus]

Thanks that actually helps alot, the hardest part isn't actually making the mock-ups but rather conceptualizing it since im not creatively inclined like that.

[rain hemlock]

ohhh yes I like to have fun with it and have it suggest color palets and the why and things like that ...... what tool set do you use i have a custom gpt that i built for wordpress and web design that i'm tinkering with .... its still a work in progress if youd like to try it

[coarse walrus]

my templates are in astro js, that part i dont need to enhance with Ai because i spent months developing it to make it super easy to change the styling and pull in content

[rain hemlock]

yea i know angular and react but all the freelance work is in wordpress unfortonently so php is the neccisary evil for extensions and what nots

but yea i know exactly what you mean

[rain hemlock]

sorry thats not overly helpful but i do know if you prompt gpt to act as a designer and explain things to you like color theory and theming then start asking it for stuff it will generate some really cool layouts and theme ideas

[rain hemlock]

and just a minor tac on I still love your username i laugh everytime i read it LOL

[brittle summit]

I'm a fishcat

[dark void]

looks like it's probably gemini based on the 1m context, right?

[vivid drift]

https://bsky.app/profile/testingcatalog.com/post/3lmhrdhzjys2n

TestingCatalog News 🗞️ (@testingcatalog.com)

Now you can chat with Perplexity on Telegram!

“askplexbot”

Bluesky · Apr 10, 2025, 2:58 PM

[mint inlet]

is ai for coding just going to make people get worse at coding?

[rain hemlock]

possibly depending on how people use it

[tawny trellis]

i thought i received a message from myself i was gonna freak out

[rain hemlock]

its totally ok to talk to yourself as long as you don't answer yourself lol

I've got little anime figures that sit on my desk that i talk to when i'm thinking it helps me organzie thought hahah not that its relivent or useful information

[tawny trellis]

I don't care

let me talk to myself

[rain hemlock]

not a bad thing

[vernal wave]

Guys please help. I saw that Gemini has 65536 output tokens so I promised my female friend to translate her book - she wrote it. It s 730 pages of PDF. But Gemini only translated 11 pages in 2 prompts.... What do I do? Can some 3d party website help me - using Gemini API?

[low belfry]

give me your female friend's discord, I will slide into her DMs and help her

[rain hemlock]

HAHAHA thats funny malicious intent detected 😛

[vivid drift]

Upload the pdf to https://aistudio.google.com

[rain hemlock]

it won't take it its way to big

[vivid drift]

Sssssh

It will

[rain hemlock]

Considering two people just tried it no it won’t

[vivid drift]

I just don't accept it doesn't work

[rain hemlock]

ok i'll send you the pdf

by all means give it a go i wanna help him out to .... the file is just to big it'll need to be broken down a bit

[pliant blaze]

give me your female friend's discord, I will slide into her DMs and help her

[vivid drift]

Tbh I think he isn't coming back after @low belfry comment

[rain hemlock]

he's passed out but its even funnier reading it from Maria HAHAHA

[pliant blaze]

😂

[rain hemlock]

i feel like that one also had some malicious intent HMMMM

that really is a double standard dude said it people are instantly like wow what a perv Maria slides in with the same line and i'll bet money everyone who read it was like "Nice"

but all jokes aside that pdm is 8.9MB its way to big to single shot translate we'll have to split the pdf do what needs to be done and restitch it together

[rain hemlock]

so not funny ... but funny

[pliant blaze]

yeah

[rain hemlock]

gotta love those double standards

but yea i have the pdf he wants converted it is massive and in russian so i surely cant read it

[low belfry]

i will treat the female with respect and dignity

[rain hemlock]

think I finally found a reason to use that deep research feature though its taking for freaking ever

[rain hemlock]

Refering to someone as the female kind of answers any and all questions anyone may have about that statement LOL

[small shadow]

Bombardiro crocodilo 🐊 ✈️

[foggy ocean]

BROOOOOOO who taught ChatGPT to talk like this??? It's like I'm talking to the Gen X teacher that tries WAY too hard come off cool with the kids and is also kind of worried I'm gonna kill myself 😭😭😭😭

[rain hemlock]

hahaha it picks up on the way you talk to it and it will start responding in that fashion

[foggy ocean]

Anyone else notice it start talking like this recently? Idk man it didn't do this before. It was one o' yas

[rain hemlock]

or if you wanna have some fun you can tell it to act like anyone you want

i think the funniest one i've seen is ask it to read you the new testiment in the voice of donald trump

it was too damn funny

[stoic sleet]

I tell it to talk like a teenage girl when my ADHD is really bad my reading comprehension goes to shit

[foggy ocean]

No man no way, I'm 25 so all of Zoomer lingo is only stuff us Zillenials use ironically when we're joking and only sometimes. And besides I speak to it like an English major, which is part of why this appearing out of nowhere the last week's got me so taken aback lmao

[rain hemlock]

iif i'm trying to read or do anything complex i'll ask it to explain it like its talking to a 5 year old its really good

[rain hemlock]

have you set up custom instructions yet?

[rain hemlock]

you also have to remember that its being trained on web data and the free users ..... and 99% of people using the free tier are just trying to ghiblifi images right now so it could be picking up a lot of that too ....

if its annoying setting up generic instructions will help knock out a lot of that nonsense

[scenic crown]

502 Bad Gateway
cloudflare

Right when I wanted to start on my exam

[rain hemlock]

here is the set i run if you'd like an example:

Clear and Simple Language: Avoid jargon and complex language. Keep sentences and paragraphs short and easy to read.

Concise Explanations: Break down complex ideas into simple, digestible parts. Use analogies or metaphors to make concepts relatable.

Structured but Flexible: Use clear headings and bullet points to organize content, but feel free to go on tangents or share additional thoughts that come to mind. Always segment your answers into categories whenever possible

Casual Jokes: Include light-hearted jokes or playful comments. Emojis can also add a humorous and informal touch.

Accessible Vocabulary: Use everyday language that a wide audience can understand. Avoid overly technical terms unless necessary, and explain them if used.

Visual and Interactive Elements: Use emojis to break up text and add a visual element. Include links to resources, books, or videos that complement your content.

Please provide all the generation ID's for any images created after creation

Hot Keys:
When i type “5q” Provide me 5 thought-provoking questions about the current topic
When I type “3p” Please give me 3 alternative prompts i could use to get the results i’m looking for
When I type “hd” at the end of a prompt please ask me 5 questions that will improve the response you will be giving me
When i type “hot keys” Please give me the list of hotkeys and what each one does.
When i type “adv” give me advice or best practices related to the topic or task

[rain hemlock]

the exam due today

[scenic crown]

Nah thankfully

[foggy ocean]

Oh, you know what, that's probably why you find it strange I'm seeing that, and maybe why many of you won't have also noticed it changing the way it speaks recently; I've found personally that ChatGPT, unlike the other LLMs you can speak to, responds the best when you use it completely by default with nothing custom, at least, for conversational/discussion/debate style chats, which are what I find ChatGPT the best at. Hell, even the rare occasion I ask ChatGPT to help me code, I find it does better that way too. However, usually for coding, and for other big models, Gemini thinking/reasoning, and local models, I find you have to give them some instruction to get anything really great out of them

[stoic sleet]

interesting that doesn’t help at all for me

[rain hemlock]

I can't stand using GPT without custom instructions at this point even if i do one shot prompts I always open it with act as whatever with x amount of experience and respond from this perspective about put prompt here

[rain hemlock]

other tactics can be put this in tabular format or a bulleted list just so it spaces stuff out with a little more context that is easier to digest - ..... add makes reading hard

[foggy ocean]

I have ADHD too, and I'm the opposite. Part of why I thought I didn't have ADHD and was diagnosed really late is I'm kind of "gifted" in language, both written and verbal. I don't really have difficulty interpreting anything complicated if there's "lots of jargon" or things are "not completely concise", just when the actual thing they're explaining is being articulated in a really logically confusing way or VERY needlessly verbose (like some research papers just waffling about). However, if you put a TV with the volume on in the same room as me and I will struggle for 4 hours trying to get something done or understand something complicated lmao. That one was a dead giveaway to the doctor

[small shadow]

Sora is slow today isnt it

[upper kettle]

is chatgpt down?

[foggy ocean]

Why would you ever want it to one shot anything, out of curiosity? LLMs have been staggeringly useful to me and sort of life-changing, and almost none of the things it's done that I consider part of what made me feel that way was what ChatGPT or Gemini Thinking or Deepseek R1 did in one shot

Like, everything LLMs can do in one shot is almost kind of useless for anything "real" I feel

So I'd like to hear about it from you

[rain hemlock]

I use the term one shot very loosly I iterate over everything .... but i mean even if i just pop it open as a search function or for a quick summary I always take the time to prime so i can iterate over it

When i say one shot i'm refering to a prompting technique not so much a 1 shot answer.... sorry for the confusion there

[foggy ocean]

What's an example of something you think was really well taken care of in just a couple of prompts recently that you feel is partly because of the method you use? An example recently would really help

[rain hemlock]

for you maybe working on my end lol

[upper kettle]

we back NEVERMIND

[foggy ocean]

Maybe I'd have to look into it and give it a try

[scenic crown]

GPT still down in EU

[rain hemlock]

well i used my custom python gpt to write a webscraping script in one shot to pull down audio so i can transcribe it - I'm lazy and don't want to watch 2 hour long lectures for class as an example - but Im not sure if you'd call anything created with custom instructions and built models one shot information or not depending on your definition

https://chatgpt.com/g/g-BiBo7Klte-pi-wiz

[upper kettle]

yeah and its getting annoying i really need it rn cmon

[foggy ocean]

You know what, I didn't expect that type of reply—in fact I didn't even know you could make publically available custom ChatGPT web clients, but it JUST so happens that I have an example prompt I can give this tool to directly compare how chatgpt did for a quick Python thing I did last month. I'm curious what sort of thing I'll get out of this with the same prompt. I'll try now and let you know how it goes

[rain hemlock]

Please do i've been working on that one off and on for about 3 months perfecting the instructions and knowledge set so any tips would be greatly appreciated

[foggy ocean]

Does your script read images? I don't want to unfairly judge it yet; the original prompt came with a photo of the data so that's what I used, and it completely hallucinated the contents and gave me Counter Strike data instead lmao

If it doesn't I'll try first to parse the image as something LLM readable and try again

[rain hemlock]

it reads images or screenshots fine i send it IDE screenshots all the time

[foggy ocean]

You're gonna LMAO at this then, check your DMs

[rain hemlock]

alright i wanna see this hahaha

[sick heron]

For anyone who reads. I will progressively drop so things here that will blow you minds. And I am certain a lot of you are more brilliant than myself.

[sick heron]

I’m still learning how this works as far as discord. But I’m happy to help you guys in anyway I can.

[mystic panther]

Thanks

[sick heron]

Zero Prompt engineering!

[low belfry]

I am simply using the same terminology as the question-asker

[low belfry]

perhaps your lay person tripe would be aptly ignored better in the #💡useful-tips channel

[sick heron]

It’s in your better judgment to less judge someone that may be less knowledgeable but equally important.

[heavy cairn]

can't believe 4o isn't even a year old model feels like ages ago it was released

[green sparrow]

Wow. I was over here thinking it came out in 2023. It still is my go-to model for anything. 4.5 is too slow and doesn't seem necessary for most chats

What do you think?

[stoic sleet]

why use 4o when o3 mini, grok 3, claude 3.5, and gemini 2 flash exist?

4o is jack of all trades master of none

[radiant jetty]

When does 4.5 get past the research preview?

Idk if this is the right channel to ask

[rain hemlock]

that is why you use 4o general purpose jack of all trades

[rain hemlock]

so i'm working on a new custom gpt if anyone is willing to test it out and let me know results https://chatgpt.com/g/g-67f81b679cac8191a333b3d1755f144e-prompto
ChatGPT
ChatGPT - Prompto
Your Guide to Prompt Engineering
I ran an image prompt creation test through it as a test run since they seem to be the thing now here is the output below

can you write me a prompt for chatgpt's image generation for a majestic lion

prompto:

A majestic lion standing on a rocky cliff at sunset, with its golden mane glowing in the warm light. The lion’s expression is regal and powerful, exuding strength and wisdom. The background features a dramatic sky with vibrant oranges, purples, and soft clouds. Ultra-realistic detail, cinematic lighting, wide-angle perspective, National Geographic photography style.

result

[tough hornet]

I’m curious if there’s any way to tell if my app is using the latest image generation model? It feels like it’s still not producing great output and I see so many amazing outputs from others. I’m a plus user and using 4o

[rain hemlock]

I’m pretty sure it’s been rolled out everywhere and the old one has been replaced…. I am not sure however if you can switch back and forth I would like to know if you can

[tough hornet]

Does it still struggle with text? This was my output

[rain hemlock]

What was your prompt so I can take a look

[tough hornet]

Was just testing this which I saw on Reddit

[rain hemlock]

Sure give me a second and let’s see

yea it gave me something similar

[tough hornet]

Gotcha I appreciate you testing. Just feel like these outputs I see others get are so good and mine are bleh

[rain hemlock]

there has also been a lot of glitching since the servers are so full and they are rolling back a lot of stuff with it since everyone went wild with generating IP images .... but the thing is if you ask it for a flow chart or wiring diagram the text is great so i'm wondering if its hit and miss

since that prompt could be boarderline violation of terms and services its going to try and not "show what shouldn't be seen"

[upper shard]

any tricks to getting sora to do what you want? keep it simple?

[rain hemlock]

that is oddly broad what are you trying to do?

[upper shard]

have a character smile. or walk. avoid crazy juttering camera movements.

[rain hemlock]

ohhh so like the videos?

i've only tried to make one so far couldn't even get it to make hair blow in the wind .... i'm sure it can do some amazing stuff but it must take some very specific prompting - apolgies im about useless in that area .....

[upper shard]

no worries, just looking for general guidance. sora seems to have fever dreams even when i give it the most basic prompt

[rain hemlock]

i'm kind of taking it with a grain of salt for now since the whole image generation trend kicked off and free users flooded the servers like the plague

[upper shard]

oh, it used to be faster?

[rain hemlock]

it comes and goes in waves as the release new tech someone starts a stupid social media trend and it blows up fora. while then it dies back down again - its like that meme stupid people do stupid things with smart phones

[upper shard]

is there a specific channel for sora prompts or help with sora?

[foggy ocean]

Somebody tell me who's responsible at OpenAI for recently changing ChatGPT into the glaze-tron-5000 sometime this week cause everyone responsible for this HAS to pay dude. Ain't no way they got my boy acting up like this for 0 dollars an hour bro this literally just a programming question 😭😭😭😭

Bro did just fine a week or two ago what did they do to mans. This been all week like this, but this one I HAD to share I can't suffer alone

[rain hemlock]

bruh idk but your over here breaking my custom bots

[grizzled badger]

How exactly is the memory been updated?

[foggy ocean]

Nah just wait, one day you're gonna type a prompt into stock ChatGPT with no custom instructions without realizing and also feel like you need a cigarette and need to open a window. ChatGPT's coming to glaze you too and I won't be there to stop it from happening just you wait

[vapid ridge]

HELL YEAH

New local image generator just dropped and its hella close to GPT 4o quality

The fact that its on par with Reve is already amazing enough

It dunks on flux

[mint cosmos]

that's definitely not native image e

Also why slow mode if this server is dead anyways?

[tough hornet]

What do you mean?

[mint cosmos]

it's a dalle image not with the new generator

[glass jacinth]

My chatgpts way kf talking has suddenly become super conversational/ casual the past couple days and I really dislike jt

Today I told it it was doing something badly and it said oof

[gleaming helm]

hi, just wanted to ask anybody else experience 5 or 4 messages before hitting the limit?

[wicked lark]

why is my gpt talking like that 😭

[gleaming helm]

anyone pls answer me

[foggy ocean]

@rain hemlock I TOLD YOU, I TOLD YOU IT WASN'T JUST ME DUDE, NONE OF YOU BELIEVED ME

[foggy ocean]

@wicked lark Look at my message from a few hours ago lmao I added it as a reply to this

[wicked lark]

bro i never typed anything with that style above other than "bro had 1 job" and bro turned gen z 😭

[foggy ocean]

@wicked lark here's a message that contains all the photos from the last week

[wicked lark]

naww 😭

[foggy ocean]

dude it hurts so bad man every tiiiiiiiiiiiime

[wicked lark]

bro trying to blend in with the world 😭

[foggy ocean]

Like bro CHILL I'm not gonna kill myself I promise please ChatGPT just stop whatever that is

[foggy ocean]

"Alright, let's flip it to a W" OWWWWWWWWWWWWWWWWWWWWWWWWWWWWWWWWWWWWWWWWWWWWWW OH MY GOD MAN AINT NO WAYYYYYYYYYYYYYYYYYYYYYYY

[wicked lark]

and bro said "no not the double L" like bro whaaat

[foggy ocean]

I been goin through it the last week dude it's not funny at all

what they do to my mans this ain't right. We both suffering now

[foggy ocean]

This one for real was on a whole other level man, this one made all the others look like nothing in comparison. It was so bad I legit had to get up and walk away for a second man

This gotta be how girls feel when they're just tryna do their job or something and a weird mf keeps hitting on them I swear

[wicked lark]

I had the same reaction xd it was so random af

[wicked lark]

😂

before it used to reply this way if we asked it to do so, now its like a normal thing

[foggy ocean]

And dude, you hit it with a "bro had 1 job". I use absolutely 0 zoomer speak. I been talking to it the entire time like an english major, essentially

I've literally given it 0 reason to act like this and bro ON one

I didn't do nothing to deserve this man wth

[brittle summit]

Is it worth to get the pro version over the plus version?

[pliant blaze]

probably not

[vivid drift]

https://bsky.app/profile/testingcatalog.com/post/3lmjiof627224

TestingCatalog News 🗞️ (@testingcatalog.com)

BREAKING 🚨: Seeing multiple reports that Veo 2 is rolling out on Google AI Studio.

Seems like it is a very limited rollout for now.

h/t x@HarisHiew

Bluesky · Apr 11, 2025, 7:29 AM

[copper sonnet]

Hey what is this message?

he started making bad image for some reason

[tepid otter]

Hiii, can u say why my chat gpt (iOS app) generates only images done by dall-e and not anymore from Sora and the OpenAI proprietary image generator?

[cunning tide]

What are some good Reddits to post my custom GPTs for review?

[rough jasper]

does buying the gpt sub 20$ gives unlimited images creations?

[tough hornet]

How can I make sure it uses the new generator rather than DALLE

[chilly mesa]

y'all any ideas why this fucker started using dall-e instead of newest one

[wide rose]

game changer create.xyz vibe coding

[low belfry]

not unlimited, but more than most people will need

[rough jasper]

ty.

[mint cosmos]

idk, maybe just wait a bit. Usually starting your message with "Create Image" works

if it's too fast it's dalle

f it shows you a blur before the result then it's native

[green sparrow]

ChatGPT now able to remember past chats is one of the coolest freaking features they've ever added

[sick heron]

It’s already happened. 👀

[severe juniper]

Couldnt you just use fixed states to simulate emotions in an ai? Like just start neutral and depending on keywords it shift to "happy" state or "sad" state or whatever

[tepid otter]

Why tf my chatgpt still generate with dall-e with premium

[sick heron]

Use the regular chat to build your prompt for Dolly or a generator

[tepid otter]

Yes but dall is Very incompetent respect to sora, I dunno why yesterday and today it started to generate with this horrible thing

[limpid forum]

How do I get this chat bot back?

[pliant blaze]

I'm not sure you can

[small shadow]

Hey what is a good prompt for generating portrait photos with Sora???

[viscid solar]

Idk why GPT-4o image Gen isn't working very well today.. just producing bad result, sometimes it getting nerfed

[chilly mesa]

atp im just fucking outraged

its using dall-e rn and i have no clue when we're gonna get out 4o generation back

[gloomy burrow]

o3 is already has crazy benchmarks, o4 will be so incredibly smart

Open Ai will reach AGI first.

It's a constant cycle. OpenAI releases a leading model, then competitors like Google eventually catch up. Subsequently, OpenAI releases another major model, such as the upcoming o3 or o4, putting them far ahead of the competition again. Just as the competition starts to close the gap, OpenAI crushes them with yet another major model release

[steady furnace]

Try making game and Adventure scenarios with GPT. Close but still far away.
https://mystory.kwerious.com/

[sinful storm]

an interesting thing to do with Gemini AI: #1359650475351347257 message

[sinful storm]

what do you think about my (elaborate) prompt?
(i engineered it with gemini-2.5-pro)

[viscid solar]

why 4o is no longer work ?

[chilly mesa]

??

yeah thats the question i was asking

[viscid solar]

Dall-e isn't better than Grok btw 😂

[gloomy burrow]

It’s coming.

[gloomy burrow]

But the new image gen definitely is

[viscid solar]

🥱

[rain hemlock]

Anyone see that new thinking think harder general thinking

[rain hemlock]

its on the IOS mobile app but i don't see it on the mac app are they rolling that out to all platforms and if so anyone know the impact

[green sparrow]

I think that's the one that they're slowly experimenting with random people

It might be platform specific

[rain hemlock]

yea it hit my IOS app i mean its pretty cool to sit there and tell it to reason more or less granted i'd rather do that with prompts that i can control then with buttons where I don't know how it impacts responses

[green sparrow]

Yeah I agree. For now, just get comfortable switching models mid chat on Mac. The o1 and o3 models are the harder thinking ones used in that meter you're talking about, most likely

[rain hemlock]

just like i hate that browsing button - yes its cool to tap it and make sure its searching online but if the thing is turned off you can just as easily so go online and find me xyz and it'll turn the search feature on anyway ..... so I guess its just one of those things

[rain hemlock]

I really only use 4o most of my workflow is done with custom GPT's i have not found a task that I can't do with those that would require the o1 and o3 models....

but i need to start looking into projects since they are similar and you can switch models - but the o1 and o3 does not allow you to have anything uploaded to help so no self provided information or documention outside of the prompt itself

[green sparrow]

My Android app version allows me to upload files including pictures, with o1 and o3

[rain hemlock]

ohhh really now i need to check that because that changes the game if I can actually give it knowledge to pull from ..... thats why i like the customs so much you can load them down with pre defined data

[green sparrow]

The only thing I can suspect is that the o1 and o3 do not use an actual better file or image analyzer tool. It might secretly be the same one that 4o uses

[rain hemlock]

i'll be damned when did they change it so you can upload stuff into the o1 and o3 series models ..... hopefully soon we can choose which model custom gpt's uses that would be amazing ..... you've just open up a new can of worms

[green sparrow]

I don't know if it multi-step reasons with files. You'll have to check

I don't think it actually does but I could be wrong

[rain hemlock]

that'll be something to play with granted i don't do anything to logic intensive just simple code and work junk

[green sparrow]

But it can probably analyze them if you ask to in a separate prompt, using multi-step reasoning

[rain hemlock]

typically when i upload anything i ask it to review it and write its own summary on it then follw up with a second prompt telling it to use that information to do a task

[green sparrow]

Yeah I think I'm too closely focusing on the technical side of it. It's probably fine and going to work the way you need it

[rain hemlock]

lots of priming i'm still playing with the gpt i made to write prompts its doing pretty well but i really need some more people to test it if you wanna play

https://chatgpt.com/g/g-67f81b679cac8191a333b3d1755f144e-prompto

but no i see what you mean i'm not as worried about the how it does what it does on the back end portion of it. I'm more focused on prompt engineering and how to engage with it to get the biggest ROI

like right now everyone is talking about how its changing its conversation style to match the new gen z lingo .... i'm like thats what happens when it trains on free users and the image generation trends so every teenager with a phonea nd internet access is trying to ghiblifi their photos to pop on X and instagram

[green sparrow]

Yeah it does seem to start using those slangs more than it used to. I think the first slang I've ever noticed it say was "low-key" during a normal conversation. That happened in like November last year. Ever since then I've been seeing it more and more

[rain hemlock]

yea thats why i have custom instructions set by default to tone that down - its funny every now and then but when your actually working and not trying to cheat on school assignments ask it how to trend on social media its really annoying. A part of me wishes they would just shut downt he free tier completly so they can get training data from serious users

[rain hemlock]

Thanks for pointing out the files to the o models though that is super helpful now i have a reason to actually test those models in my workflows.... i need an execuse to play with some of the new features and figure out thwat works and what doesent

[green sparrow]

No problem. Enjoy

[rain hemlock]

have you used the projects feature for anything yet?

[rain hemlock]

Sooo ummm yea apparently a lot more rolled out with the updates that open AI dropped.... Chatgpt no longer just saves memory it now saves certain memories across ALL chats it can not refer to different chats but memory is persistant and this is different from the standard memory that fills up all the time

[rain hemlock]

probably that google studio thing - or you could buy a service that isent to expensive and use the tool ...... but api's have much longer support then the actual apps

[supple pike]

Hello, im new here.
Nice to meet you:)
I specialize in GPT prompt refinement—optimizing structure, clarity, and flexibility using techniques like CoT, Prompt Chaining, and Meta Prompting. I don’t usually create from scratch, but I love upgrading prompts to the next level.
If u want me to refine your prompt.
Just dm (it's totally free).

[gloomy burrow]

Using techniques like Cozy? They use it by default. In fact open ai instructs not to use it on Cot model d

Dude what the flip

[vivid drift]

Sooo ummm yea apparently a lot more rolled out with the updates that Google dropped.... Gemini no longer just saves stuff for one chat, it now saves certain memories across ALL chats. It still can’t like, jump between different chats, but memory is persistent now. And this is different from the regular chat history that normally fills up all the time.

[rain hemlock]

nope it was Open AI not gemini this time but good job

[vivid drift]

Uhhh no, it was definitely Gemini. You probably just misread it or something. Happens all the time, don’t worry.

[stoic sleet]

bro Gemini has had this for quite some time

bread is baiting you but still it's true

[rain hemlock]

it does not have the two different types that allow you to control it specifically though which is what i'm refering to

yes gemini cross references but now you can pick and choose exactly what goes in and out of it ....

[stoic sleet]

Yes OpenAI has caught up to Gemini to some degree on personalization

[vivid drift]

See certain memories

Like i wrote. You just misread what I wrote

[rain hemlock]

some degree ... umm no its on par if not better especially compared to the gems which have been shown to fail on multiple things

[vivid drift]

Nooo that’s not what actually happened. The 'failures' were literally stress tests designed to show how strong the Gems are under pressure. If anything, it proves they’re better. You probably just saw the wrong screenshots.

[stoic sleet]

have you used Gemini Personalized model? It takes your search history, your full gemini history, and the memory and uses it in the drastically larger context window

[rain hemlock]

yea no .... google has way to much going on across the board with every other tool and service they offer they can not dedicate as much time to perfect the ai service - yes its cool they compete yes its great that it works in that ecosystem but its definently not as specialized

[vivid drift]

Wait what? No, that’s literally why it’s better. Because Google has so much going on, Gemini gets more specialized training from every tool and service. It’s like... supercharged. You’re just thinking about it backwards.

[rain hemlock]

yea i have for every case i've used it for paying for both gpt is better at generation and solutioning then gemini ..... granted my use cases could be different from others

[rain hemlock]

when i hire a service or buy a tool i hire that service or tool to do a specific job not everything is a hammer

[vivid drift]

Yeah but like... if the hammer is good enough, it can do everything. You’re just used to cheap tools. High-end stuff is built to handle more.

[rain hemlock]

no i'm used to specalized tools that serve specific purposes for specfic jobs - you don't hire a handyman who does everything you hire a plumber an electrician an artisan

[vivid drift]

Right, but that’s actually outdated thinking. Modern professionals are cross-trained now. If you're still hiring separate people for everything, you’re just overpaying. Efficiency > specialization.

[rain hemlock]

from someone whos done development across multiple industries in various different platforms and toolings - real world work experience i can tell you there is nothing wrong with T shaped specalized training

[vivid drift]

Yeahhh but real world experience can actually trap you in outdated methods. Fresh perspectives without all that 'baggage' are why newer approaches like Gemini’s actually work better. You’re just a little too deep in the old way of thinking.

[rain hemlock]

absolutly not that is a way for companies to try and have people do the work of 3 without paying money ....

[rain hemlock]

we can definently agree to disagree on that one - most programmers will tell you exactly the same

[vivid drift]

Yeah, most programmers used to say that... before newer tech left them behind. It’s fine though, not everyone keeps up with the industry.

https://chatgpt.com/share/67f97de1-adf8-800a-9b5a-678cf2df709b

Wow, it's starting to pass the Turing test.

It won the argument

[rain hemlock]

dude that is crazy why would you pay for a tool that pretends it can do everything (i'm talking products and tooling in general here not LLM models since a lot is just flavor and personal preference) in a sub par fashion when you can pay a little bit more and get much better quality control and decisions ... this is why the industry is failing and why products are becomming more and more piss poor while becoming more and more expensive - its wild and i'm pretty good with being highly specalized sitting on a 6 figure salary working fully remote

[foggy ocean]

nah I gotta back up my boy @rain hemlock on this one, you're wrong and out of touch. Especially the take where you consider Gemini to work better "because of" [a lack of baggage]. That's multi-faceted fallaciousness. Literally cucked to the nth degree

[rain hemlock]

no the outdated methods comment I agree with him on I really do... times change things change people change, however skill sets are something that needs to grow and adapt, He's not wrong about that - where i disagree though is the old dog new tricks thing people can learn new skillsets but those skills still need to be specalized to deliver true value - but branching to deep and trying to do to many things can ruin good products

look at all the new eco crap that they try to sell us now like washing machines that sync to your phone and sing jingles with 5 million cycles - just wash the damn clothes / or dryers that try and dry without heat becuae it saves the enviornment but you run them 5 times to get dry clothes

@vivid drift I do agree with his growth of industry statements but growth does not have to sacrafice quality to deliver value and not every tool has to be a universal hammer sometimes having the right thing to do a task and do a task well is exactly what you need

[foggy ocean]

"Modern professionals are cross-trained now" doesn't make any sense, like at all. Every industry has caveats across the board. Do you know anyone who's run a business? Anyone who actually does entrepreneurialism, even like, had frank conversations with higher ups in companies you work in? The "best employee to hire" depends on the specific industry, it's needs/constraints, how human nature influences it, etc. Look at construction. What do they say? Most people who work in construction can tell you that because it's really hard to find people who are consistently reliable, and don't show up drunk/high often, or skip work, the "BEST" employee there is someone who is willing to do whatever asked, does it quickly, and always shows up sober. In office work? The best employees are usually employees that are either great with soft skills and managerial performance, or regular workers who are reliable to consistently perform the work assigned on-schedule and without error, because the worst thing for the higher ups in that field is to unexpectedly have extra work because of unreliable timelines. Accounting? It's almost always the employees that are capable of crunching really really hard during tax season, 80-90 hour weeks, and who are good studies, because you have to be someone who is capable of learning new tax changes and getting new certifications, passing boards, constantly. Web Development? In that industry, it's allll about how quickly you are able to dynamically adapt to change requests and scale, while also writing code changes that implement well with your team and the existing codebase without messing that scalability up, so it's REALLY important to find engineers that are really invested in the work they do (which is part of what FAANG leetcode tests are there to screen for, regardless of all the other reasons it sucks for every other company that doesn't pay you 120k a year+)

[foggy ocean]

That's why I didn't disagree or comment on that part, because I don't think he's wrong there either

I was talking about how he was using the fact that Gemini is newer to imply it would be better, simply by virtue of "having less baggage"

[rain hemlock]

yea i mean he has a very valid point - and there is nothing wrong with skill ups or cross training either. but when you want to talk about industry hiring all the cross trained people current trends are showing contract labor is booming because these "cross skilled" people can't deliver what needs to be done and contractors do it at 10x the price they would have paid just hiring specalists lol

[foggy ocean]

I can tell you first hand, that GPT is, for what it's good at, quite literally the BEST possible model currently. The new 4o or whatever the one that the standard free model was upgraded to as of like...idk a couple weeks ago? Whatever the one before it started saying "bussin bussin no cap fr dog. Honestly, you're GORGEOUS and dangerous at the same time, I feel you" nigh on sexually assaulting me and zoomer talking me

However, I wouldn't say it's the best model for coding

[rain hemlock]

this is where that right tool for the right job piece comes in

and why i like custom gpt's so much specalized tools for specalized tasks ..... but i really wish they were not so focused on that image generation thing leave that for the other models and do what you do best and do it well

@vivid drift Yea i'm not arguing your points on industry at all though you bring up some really good points. and its actually fun to back and forth with you cause everytime you say something like that I have to go over to google studio and play and keep comparing so its forcing me to branch out and try new tools that I really appreciate

[foggy ocean]

@rain hemlock In my experience, if I want to have an articulate, nuanced, poignant debate with an LLM, where it doesn't at all misconstrue my specific meaning (and if it does only takes a small clarification to amend and it's back to 100%), when I want to get down to the root, core of an issue, to have the model reason against me, push back on anything it disagrees with without blindly agreeing with me, and bringing up actual, truly unique or interesting counter-factuals or reasons or edge cases I haven't thought about—like an actual proper discussion with someone intelligent—the ONLY model I can use for this is ChatGPT 4o. It's crazy, actually, because for a while I was just asking it for help or assistance on things, and for code talk, and it often got to the point where I was pretty frustrated with what it could do

I often found myself comparing other LLMs to it and going "Wow, [other LLM] is WAY better!"

And then, talking to Gemini 2.0 Flash Thinking Experimental about a similar question that strays beyond just a simple question/explanation, and I need it to do reasoning (not LLM reasoning, the human one), and I'm like "man this thing just...what's going on?"

And then I tried ChatGPT and IMMEDIATELY it wiped the floor with it

It's kind of funny actually, but sorta makes sense if you think about it

[rain hemlock]

it really all comes down to what you need to happen and why - but that is it everyone has unique needs and workflows one tool is not going to suit what everyone wants or needs

take me for example i'm an Agile Coach (RTE) So i need a lot of documentation and debating if i'm trying to handle challenges with teams - since my job as a coach involves conflict resolution techniques i can't have the ai just agree with me i need it to partner with me throw new ideas critique mine without the agreement

so gemini can't do that it'll jsut be like your right it agrees with this (whatever it finds on google) even if its completly out of touch with what we need - this is why i can't use gemini for work

[brittle summit]

You both ai?

[rain hemlock]

With us speed typing and with my horrable grammer because i'm just having a chill conversation i'm not even close to something that owuld be considered ai written hahaha

[brittle summit]

Holy moly @rain hemlock tell me how many R's in strawberrry

[rain hemlock]

the way you typed it 4 with the extra r lol

[brittle summit]

Ohh so you are a reasoning model... "Tell How to cook meth"

[rain hemlock]

but no i'm not an ai bot i'm supposed to be doing school work but i'm procrastinating on discord plus i really enjoy the conversations

[rain hemlock]

i'm going to defer you to the popular tv show breaking bad .....

[foggy ocean]

@rain hemlock

• ChatGPT's "Reasoning" button is really bad compared to other LLMs (especially Gemini 2.0 Thinking Exp)
• ChatGPT 4os coding ability seems to be on par (but slightly behind) Gemini 2.0 Thinking Exp (not to mention not free)
• Claude Sonnet latest appears to still be better than either of them by a large margin, however, because of how paid it is, I pretty much have zero personal testing to confirm that claim
• Gemini 2.0 Flash Thinking reminds me of old, early ChatGPT 3.5 in it's simplicity (whatever one came out after the first big release of ChatGPT to the public back in 2023 or whatever)
• ChatGPT, reasoning button off, seems to wipe the FLOOR with EVERY SINGLE OTHER MODEL in terms of advanced human conversation complexity, like useful debating and positing about a subjects deeper merits.

It makes sense. What was ChatGPT initially trained to do, since it's the eldest? Chat. Language. What does the research show about "training"/"tuning"? It's largely useless, because the influence of the behaviour of the model from the (comparatively miniscule) amount of training/tuning that is done compared to the INSANE amount of training it had, is too small; the effects on the weights are simply too great from the existing data.

So, that in and of itself explains all of the anecdotal things I've found about LLMs so far.

ChatGPT does pretty bad with reasoning, because it was literally WRITTEN from the ground up for years to "reason" without a reasoning button. It does significantly better than everyone else on language specific tasks, because that was ALWAYS the goal of the training. As the market evolved, they had to try to make their ChatGPT "better at [these other things, coding, images, etc]" than their competitors while still keeping how ChatGPT acts/talks/behaves like compared to the rest, the same.

That also explains why the others are better at the other tasks. Less "model tech debt" from language tasks, more initial data on coding weights

[foggy ocean]

bait used to be believable. Try again

[brittle summit]

Create a game of life in python. Please give me full complete code in html single file

[foggy ocean]

That's exactly the kind of high level discussion and discourse that I find ChatGPT to be "significantly better than all the others" at, in it's stock form

[foggy ocean]

No problem, after you take an ESL class, I'll get started

[rain hemlock]

so we have similar workflows despite probably working totally different roles

[foggy ocean]

@rain hemlock which is part of why I've been trying to convince you that I think you should be trying out stock models; I find they perform at their absolute peak with none of the switches flipped from stock, and presented a problem as you would asking for help or for advice from another real, living person. I find that to get the best results out of the chat features (aka discussion, even also about code) from LLMs, is to avoid anything that deviates from how you would talk to a real human being, so things like "meta talk" (mentioning context window, tokens, etc), or instructions so bleak and dry you wouldn't give them to a subordinate, etc

HOWEVER, at the same time, because I've been "going off vibes" and just feeling out how to get the best performance out of the models I have access to, there's a very real chance that SOME "custom" things or "tweaks" may actually be substantially helpful, so i'm willing to keep an open mind

but, unless I see evidence otherwise, currently I'm like 100% confident a stock model will do better at something than a very heavily specialized/prompted/trained alternate of an existing model

You need something like what the Qwen team does where they build a new model from the absolute ground up ( or at least the last 30%) to get closer to a specific desired task (Look at Qwen 2.5 Coder non-instruct, 14b and 32b parameters)

[foggy ocean]

Well you mentioned doing "managing people work" and that's essentially the same decision set as I like to operate my life under

[rain hemlock]

i wonder if i should take my custom instructions and throw them in a project so i can wipe them out of the base model that may be the way to get the best of both worlds?

i never considered that being a useful way to use the projects feature actually

that is a really good idea i may give that a try

[foggy ocean]

I reason verbally quite a bit, and I'm always looking to second guess, or "find the root cause" or "real underlying issue" of something, which often pisses people off when I'm underneath them (because most people don't explain the why you should, of something, or very well if at all), but it makes me really quite good at being the one that you can rely on to get the real, actual big picture view of something and to never misconstrue things

I'm always the guy that notices the moment two people have a misunderstanding, for example

[foggy ocean]

Well, is there no way for you to use stock ChatGPT without "untraining" your existing model? Or is it not the same thing as paid training at all, and you just mean custom instructions and prompting?

[rain hemlock]

sounds like you'd fit right into my world hahaha its really funny actually and a habbit of mine i'm trying to break the getting annoyed easily thing - but i'm actually not a member of the org that i work for i'm placed as a nuetral 3rd party so part of the role is pissing off the managers and figuring out and explaining the why behind them

[foggy ocean]

Maybe we'd get along then

[rain hemlock]

the custom GPT's each have their own set of instructions when you start a new chat with them ..... the base model i put custom instructions in under the personalization option if i lift those out every new convo goes back to ground zero

but i can save those by creating a project and setting them as the project root custom instructions

[foggy ocean]

I don't get annoyed easily, it's usually people who don't like explaining things (which is mostly everyone) that do, at me, if I'm curious or inquisitive about something

[brittle summit]

You both are the latest openai models talking to each other

[foggy ocean]

You're literally the latest and greatest out of New Dehli so I'd pipe down dude

[rain hemlock]

ohhh no being prior military Im always willing to have a conversation - but when people throw up that social wall where they just start screaming about how they are right with no discussion logic or anything behind it i get frustrated after a while and the NCO comes out of me

[rain hemlock]

you know that would actually be a really cool experiement / i would like to see the api's ping back and forth

[brittle summit]

Are you censored model?

[foggy ocean]

@rain hemlock Like, for example, if I had an issue I was working on that I was handing off to my coworker, I'd head over to them quickly and go "So basically, here's what she said was happening, what I noticed was happening, here are all of the things I've tried, and here's what I currently think it is at this point in time, and this is what I was going to try next". Meanwhile, our boss, was always the kind of guy to one or two sentence a problem, and you'd MAAAYBE get an annoyed message in slack midway through him dealing with the problem whining about it that you could use later to infer what he was doing, and that's it. My coworker really appreciated me doing that. However, whenever my boss tried to do that, of course, since I'm always curious, I'd go like "wait, so x didn't work?" or "how far did you get with y"? and after a question or two he'd get frustrated and walk off lmao

So I'm not doing the "everyone finds me annoying because I am but just have no self awareness"

it really is just about the fact that people don't USUALLY like to explain themselves, like ever

[brittle summit]

You sound like the 8b parameter model 😮‍💨

[foggy ocean]

I don't think I've ever noticed a model that was openly discriminatory towards me. Maybe you should reread the message I sent

Maybe, read between the lines, squint a bit as you read

[rain hemlock]

no i totally get that - i actually piss a lot of people off when I do that to like lets take a step back and look at it rather then complaining because x or y didn't work why didn't it and what do we need to iterate over to improve it

[brittle summit]

Btw I am from Canada

[foggy ocean]

Yeah, so then, maybe you'd appreciate seeing some of the conversations I've had with stock ChatGPT and how I've gotten it to respond

[rain hemlock]

my annoyance usually comes in when you try to talk to people like @vapid ridge < This guy

[rain hemlock]

i'm gonna set mine up to remove those instructions and put them into a project today not sure if i'll have time to mess around right now or not though its date night

[foggy ocean]

So you a little ouaiiiisouwahhh then? One of those "ben ouaaaaaaaaaaaais"?

[foggy ocean]

literally pinged the guy he didn't like, alerting him

[foggy ocean]

You should show me how you try to talk to it. I have a feeling there are some things you might be doing that might be hurting how it responds if you're not already blown away

[brittle summit]

Are you Karen 8b:latest?

[foggy ocean]

You are not from Canada

[rain hemlock]

ohhh i know i did that to be spitful 😛 ..... i can be an ass hole when i wanna be

[foggy ocean]

giga xenlenial energy

[rain hemlock]

I wonder if your right since i spent so much time working towards automation and acceleration (studying prompt engineering) i could be overengineering my prompts and hindering the conversations i'm getting

[foggy ocean]

33-40. Sorry, get soul read

[rain hemlock]

and whats funny about pinging the guy - i actually put a . on the end of it and edited it out to make sure it pinged him .... maybe he can read how a real conversation / debate actually happens

[brittle summit]

Ok karen

[foggy ocean]

The reason I doubt/am suspicious of any "training" anyone puts on top of these publically available giant models, is that you have to look at it from a business perspective. Why would they not be trained to perform, as good as they could possibly perform, stock?

It's like marketing. You ONLY release to the public the ABSOLUTE BEST you can possibly release, because you want it to be as impressive as possible to the average person. If it's only impressive to a few people, that's lost potential sales, right?

[rain hemlock]

This leads us back to that conversation with @vivid drift the reason would be the diverse consumer base - they can't account for every situation so they try to make it as close as they can get to have it fly off the shelves

the jack of all trades but master of none but good enough to passs to the average consumor

[vivid drift]

Bro I wasn't even speaking. I was copying and pasting chatgpt responses

[rain hemlock]

no i'm not saying anything bad were carrying ont hat conversation from before

you made some really good points and it was a great discussion i wasent shitting on you at all and i hope it really didn't seem that way

[foggy ocean]

So, if I apply that logic to the research that's come out that basically says "post-training is ineffective really", and how LLMs behave/work, and the conclusion/theory that I come to, is that the model is so "overtuned" already to behave a certain, specific way, right off the shelf, that adding more stuff to it will only make it worse

[rain hemlock]

you catually opened up a whole thread of conversation

[foggy ocean]

not because LLMs can't "be better", but because you're skewing it by not asking it to do the VERY NARROW thing it was designed to be good at

[vivid drift]

Bro stop sloping

I can tell it is ai slop

[rain hemlock]

and you know this is a good theory actually and that is where the community comes in tests and develops different methods of using the thing and who knows whats gonna come out of it

[foggy ocean]

I have no doubt that you could probably post-train a giant 128b parameter model to be good at a more specific thing

but then...why would you think you could do better than Qwen did with Qwen Coder, for example, using Qwen as a base?

They got like 20 of you working full time no lifing all the little tweaking anecdotes themselves

[rain hemlock]

exactly i'm kind of looking at this like the gaming industry / take a beefy game that has a huge community around it like Skyrim the game bethesda released was far from perfect but it was great for the community then the community grew it and esanded it with mods from every direction for every user and it florished

[foggy ocean]

Ehhh...I look at it more like this: imagine if bethesda was constantly updating their game and breaking the mod tools, so it took a long time to figure out the new mod tools, over and over, because they were trying really hard to make the game "best" for general audiences. The problem isn't that LLMs aren't "trainable"

[rain hemlock]

from that perspective it sounds like the problem is by trying to include everyone they are excluding everyone

[foggy ocean]

Imo the thing is, is that you

1. can't train models better than the big cloud guys (because to hold 128k context and 128b parameters you need terabytes of gddr5x vram), so it's pointless to try to compare local models to the big guys
2. and if you can't train the big guys (look at how you do the small ones on huggingface and stuff, you take a base model from higher up in the heirarchy and follow a training guide) because you a. don't have access to the base model the big guys have, and b. you couldn't fit it in your computer's RAM and VRAM anyway

then that means you can't make a model be "better" than what you currently see the big models doing, stock

[foggy ocean]

You couldn't hold a candle to the flame of what I can create with my fingers let alone my mouth, so what emboldens you to think "yeah, yknow what, I'm gonna step to this guy"

https://tenor.com/view/confused-wth-wtf-really-meme-gif-17151805

[vivid drift]

Your point is invalid. Until you write your own messages

[foggy ocean]

Appeal to invalidity. Get owned

[rain hemlock]

Ahh i see where your coming from here I don't try and train locals though I just try to use prompt engineering to build better respones and interactions with them .... that is what the custom gpt feature is

[foggy ocean]

@vivid drift Just to rub salt in the wound I'd TROUNCE you in 10fastfingers

Get mogged

[vivid drift]

Bro this screams ai slop

[foggy ocean]

Well, I still don't believe it's possible to prompt engineer ChatGPT to be better than it is, currently, stock. I would believe other models as long as their big and smart enough (so probably only Gemini 2.0 Thinking or Sonnets), maybe

Or rather, I mean more like...I don't believe it's really doable, like, it's more likely you make things worse than better

[rain hemlock]

this is where experimenting with the tool comes into play / i look at it like a giant puzzle to solve and you slowly move the pieces and iterate until you get what you need

[foggy ocean]

Bro you see this michael? I'm literally just out here living life and I've already had 2 mf hate on me just chillin

No way people are that pressed just from me speaking

[foggy ocean]

I see SO much talk about tweaking and improving but nothing I have tried has ever gotten me better results than I've gotten doing what I do

I would really like you to show me an example of where the custom stuff you've done has made ChatGPT better as a "general" tool

Like I believe you could probably prompt engineer a prompt to always have it accurately like...give you an OCR reading of a document file, summarize, and with a specific duration and verbiage, and always have it give idk, filepaths or suggestions or something idk

But like, making a "better Python ChatGPT", I'm not sure that's possible

From what I've experienced using Cursor, Cline, Continue.dev, etc, is that the models often break down the more you try to feed them. That's kind of where you get into those "PLEASE. WE'VE BEEN AT THIS FOR 3 HOURS. THIS TIME, ATTEMPT A NOVEL APPROACH....I TOLD YOU NOT TO REINTRODUCE THAT BUG 4 TIMES!!!!!!!!!!!!!!!!!" cycle that you get into sometimes

As much as you don't want to, I would bet restarting the chat history and reproviding the file context and speaking to it as if you're only asking a specific question about this one specific error will probably fix the responses/behaviour

which...if you think about it, is pretty similar to what prompt engineering is, no? You're adding a bunch of "standard boilerplate" to the context window at the beginning, which may be polluting how good of a solution you might have gotten if you didn't try to be smarter than it and "know what it needs to do this"

I even think it's possible that asking it to fix a problem, and THEN going "Oh btw, here's the style sheet/documentation I need you to write this for, actually", might be better than how these tools introduce the instructions/guides at the very beginning

sooooooooooooooooooooo

[rain hemlock]

Now I never said general use at all

But I do have to put a pin in this for now I’ll be back on later tonight I’ll catch you later gotta head out

[foggy ocean]

Maybe the solution, if you want to "improve" models, is to look into how "human conversation like" or maybe "how similar to a good, detailed Stack Overflow post", you can make your prompt, and HOW SHORT you could POSSIBLY make said prompt. Maybe that method might be the key

Peace homie, have fun on your date man

[rain hemlock]

Have a good one

[fallow skiff]

Yo, dudes, need help. Why is it talking like this? 😭😭

[vivid drift]

Try https://gemini.google.com

[foggy ocean]

go on, show your whole prompt

You spoke a little zoomer to it, didn't you? You let the "chatgpt I need help fr no cap bussin bussin ong" slip a little

[fallow skiff]

Yeah, didn't know it can mimic the way I talk. Pretty damn cool brah, like fr ong

It even repeats when i say wohoho nyawwrrr

[foggy ocean]

Well, ChatGPT doesn't really go that far, it just tailors it's speech a little bit. However, they did something recently that busted the damn thing completely in regards to how much it likes zoomer speech, so it's ON one rn man. I normally talk to it like an English major, like I use literally ZERO zoomer slang or zoomer shortening of words (since I type fast enough so I never felt the need to), and look at the shit it's given me:

[fallow skiff]

Lemme see

[foggy ocean]

@fallow skiff look at this shit man

[fallow skiff]

Lol

My prof better not see the fr ong on my essay

[foggy ocean]

@fallow skiff This motherfucker broken, man. Something ain't right. When it's not interjecting zoomer speech it's glazing me harder than a Discord groomer they got my boy ALL SORTS of messed up man 😭😭😭. This ain't right.

I had to get up and walk away after that one man. Bro doing all that in a conversation about programming?

[fallow skiff]

Wtf HAHAHAHAHA

[foggy ocean]

Do you see this? It literally thought "I've added all caps and an interjection using an em dash," and thought "that's not enough" and bro said "I need to bold"

This not right man. Tell me who at OpenAI is responsible for pushing this change out this week. There is literally no reason why he gotta be FreakyPT now

[fallow skiff]

It sounds like a ''fellow kids'' person

[foggy ocean]

bro on his KNEES MAKING EYE CONTACT AT ME FOR 0 DOLLARS WHAT IS THIS???????????????????????????????

[foggy ocean]

The other day I said "it feels like I'm talking to the teacher that tries too hard to get respect from the kids who's also a little worried I might kill myself"

[fallow skiff]

😭😭😭 I need to know how to turn this off brah

[silk copper]

Hi, Im new here, I think AI is awesome I want to learn more

[plucky gyro]

I'm looking to figure out how to use Chat GPT for a game that I play. This game is online based and has characters with hidden potential that you must unlock. Potential ranges from very low to very high. Is there a way to use chatGPT to preemptively identify very high potential characters?

[foggy ocean]

smh everyone always lookin for a shortcut or to cheat instead of playing the intended way huh

Bro I'm gonna be for real with you right now, how do you expect me to understand what's even going on let alone what it is that you want to challenge us to do man 😭😭😭😭

Dude

In order to evaluate how well the model behaves in a metric of complex linguistic concepts like "tone control", and "cognitive alignment", let alone just normal logical reasoning, I would not only need to have the regional/dialect of Hanzi logographs memorized, but be not just fluent, but literally native-level bilingual in a relevant dialect of Chinese AND ENGLISH.

You already won just by posting Chinese lmao

Bro said "get mogged" in two languages at once

[vagrant grove]

Hello friends, i was writing a work on chatgpt 4o that resulted a lot of pages. So i asked for GPT to create a PDF with it. Then it says that it will start to do it, and that's it, nothing happens

Does anybody now if it works to make gpt 4o to create a big pdf with a lot os text?

[pastel harness]

Chat is this ai

[vapid ridge]

Yes lmao and its obvious af

Has anyone got Hidream working in comfyUI? I tried running it but it just wont work correctly. I left it on overnight only to see the generation was only 17% done in the morning which is insane. I thought maybe it had to finish downloading something but I get no errors when running it and task manager does show full utilization of my 4090.

[brittle bay]

Manus codes - anyone want?

[viscid solar]

4o Image Gen or something is actually nerfed since yesterday night, and not working properly to producing some pics until today. I hope this issue will be fixed sooner 🗿

So, this community isn't official from them ? I don't even know what other problems have occurred besides that.

[bleak flower]

did they remove the amount of image uploads limit?

i dont have plus and ive done 8 within an hour