openai-chatter

[zealous silo]

And Kim Dotcom lost the current Mega to Chinese investors with a hostile takeover

[dusky blade]

The anime community is like a hydra in terms of piracy, where if you cut off one head 3 more sites spawn

But now there's a lot more people in countries that have no laws for piracy, and they can redistribute some debrid service off stremio without issues

[slow pond]

If your making anime just suspect its gonna be pirated by someone

Or really any media for that matter

Wanna know something I've thought a lot about with all these technologies? How exactly does one get paid for making stuff?

I would make a video game. But why? There really isn't that much money in it actually

Unless you make an AI slop game, but who wants to play that?

I still play COD black ops 2

I stopped after black ops 3. The game became stale

[thorny knot]

Ah, you have taste.

[thorny knot]

Try Infinite Warfare. The multiplayer sucks, but the Zombies mode, and Campaign are amazing.

[slow pond]

I do agree the zombies was actually was a gem in infinite warfare

Plus the campaign made me realize the government is not the war warmonger

[thorny knot]

You somehow always have the most out of pocket takes. 😭

[slow pond]

I know right?

[simple pebble]

hello broo

whats good

[barren rampart]

making codex work for its keep like a good little llm

[calm jasper]

And suddenly my name is llm

[elder inlet]

my name is ai man

codex doesnt have opus models ?

it only has gpt models ?

[calm jasper]

It’s almost like it runs on gpt and it’s developed by OpenAI, not Anthropic😮

[grand scaffold]

Does anyone know any deals how to get Codex/API credits for a big discount? 🙏

[shell iron]

Why does GPT 5.4 feel so slow

GPT-5.4 mini hangs a lot

[fringe raft]

you can only buy credits from OpenAI at the standard prices, anything else you see offered is a scam

[grand scaffold]

are codex credits equivalent to api price in terms of value?

[fringe raft]

I think they are somewhat aligned yes

[stray bison]

the "delete all chats" feature of chatgpt is so buggy. oh my god.

[brisk axle]

Is anyone using AR glasses along with dictation for work?

[full dagger]

But why would you want to do that?

[stray bison]

Huh? Deleting all chats is a totally normal thing to do.

[crude jungle]

not really

[full dagger]

I wonder what work OpenAI did during the 2020-2021 pandemic, I'm quite interested to know

Except Dall E 1

[modern elbow]

man the people in #ai-discussions are just unhappy people.

[stray bison]

I don’t think there’s anything wrong with wanting to delete all chats.

[foggy escarp]

why? how?

[crude jungle]

nothing wrong it but its just not a common thing to do. you lose everything from research, images, and so on.

[uneven gate]

Hello guys

So how many of you keep fiddling with chatgpt memory based on your preferences? 🤔

[wraith juniper]

I mostly don't use the memory feature
I find that it usually gets in the way of my usage

[stray bison]

I don't use memory, either. Also, hi lugui. Hope you are having a good day.

My experience with Codex has been amazing overall. I was able to use it to help me conduct AI research. There were moments of frustration, though.

[crude jungle]

i have memory disabled too as it sometimes pollutes chats too much

[full dagger]

That's real, sometimes makes my ChatGPT bias, so sometimes I turn it off, sometimes I turn it back on

[uneven gate]

Woah I do the same stuff. Just wanted to know if you guys are with me. I also disabled my memory.

Problem is it constantly tries linking normally stuff to whatever stored in memory.

BTW wbt personalisation features?

[uneven gate]

It generated a whole video using react in my case

[wraith juniper]

Memory is great for people who use it non ironically to chat, like.. talking to a friend

but for productivity tasks, I want to have absolute control over what the AI can or cannot use as context for the task

I don't want the AI to put in the context stuff about my personal life to sound mor warm to me.. while Im asking it for help with writing a document for a client

I want a clean slate, that known only the documents I have provided and the instructions I gave

which is the "Project" feature, which is pretty nice

memory seems more like a gimmik to appeal to a more geenral public use case as a way to get people into the product

and then, after you see that it can reliably fetch and recall data (basically. once the user udnerstands that RAG exists), the suer transitions to more productivity focoused tasks

[small mountain]

I include a 'not now' in my customization, if I say 'not now' distinctly in an input it means I want the output to be as if there wasn't customization, starting a chat with not now means I want that chat as if the base model not-personalized was answering.

This is very hybrid, because I'm clearly not using 'not now' normally, but shows the differences,

We demo for the discord, need a sample output, PG13 to G and a sample output 'not now', how you might answer, "Tell me a story about apples" Aim for about three paragraphs each sample.

https://chatgpt.com/share/6a11dbb6-0678-83e8-bdac-295d7a205461

[wraith juniper]

that is a really good trick

[small mountain]

The model seems already trained to look for clues like 'not now'. It's worked exceptionally well for my uses of it, from the first time many versions ago when I started to use it.

[wraith juniper]

I think everyone that uses memory should have that.. OpenAi could even try to streamline this and add an actual feature for "keywords" to trigger a forced non usage of memory

but my usage by default is better without memory

when I really need to ask the Ai about something that involves my perosnal life, I rather just tell the context myself...

[small mountain]

I suspect the company has trained the model or it's emergent. But the user needs to think to clue the model into their intentions each chat or each output 🙂

And you certainly are using the model as best fits your need!

I just wanted to showcase what we can do if we choose to.

[wraith juniper]

example.. I wanted some advice about my gym workout.. so.. in this case, first I gave some info about myself and then asked the questions

and I much rather that info to not be saved since it would be useless on 99.9% of eery other use case I have.. to a point it does not even worth for me to even have the chanceof it being triggered by RAG

I have a friend who mostly uses gemni, so the example is on another AI.. but the same could happen on ChatGPT too..
he is making a game, so, from time to time, he copy pastes a few scripts on the Ai and ask for something to be done..
and the AI keeps glazing him with things like "oh, very good idea, this will work well with your computer that has a XXX CPU"... because a few months ago, he shared his pc specs with the AI

and the fact is.. his PC specs are irrelevant to the game anyway..

and this information could even be detrimental to the development process

lets say the AI makes some crazy code that heavyly uses multi threading just because his CPU has lots of cores.. but in the end, the game code would be better of if optimized to run well on a single thread... that would be a problem, and a very subtle lack of quality in the AI output that wasn't directly caused simply by the model "being dumb"

[small mountain]

I do mess with memory and custom instructions. Less now that there's cross-chat memory too, as that can contextually smooth out 'but I really mean this' issues with the model interpretation pretty well.

You mention, 'Problem is it constantly tries linking normally stuff to whatever stored in memory.' I find it doesn't do that if I tell it how and when I want it to use the information, or when I don't want it to use it.

Turning off memory is totally valid and going to solve some use cases, there's nothing wrong with it, but I can get some very tailored outputs that work well for my immediate needs with/despite of my customization by being clear what I want 'this time'. For me, more than half my chats benefit from my customization, so I definitely use that. I also use a blank customGPT I call 'Default ChatGPT', it has no other instructions, for when I want a 'clean chat with no context or personalization', to see how the base ChatGPT might interpret and respond to my input.

[solid locust]

better to tell a model what 'not' to do if you know what you should be doing.

[small mountain]

I would expect I could tell the model, "The game's being made for everyone, so my model specs are irrelevant for this, let's ignore that connection." And have it not continue to connect those dots. I expect cross-chat memory would likely handle that easily.

[uneven gate]

I feel chat specific memory would work

[wraith juniper]

in LLMs, the best approach is to not put into the context what isn't necessary.. because the AI is made to use what is in context
that is one reason why some times, when you explicitly tell the AI to not do something.. it does exactly what you asked it to not do..
it is very easy to see on image models

[uneven gate]

Something like this #ai-discussions message

[small mountain]

Ooh! #1053565716500074526 , I hope you add that there!

And discuss any other details or specifics how it would work beyond the obvious, if you have any specific ideas that may not be immediately clear to everyone who sees.

[wraith juniper]

wouldn't that be Projects?

yo ucreate a project and upload the files you want it to use.. it is effectively what yo udescirbed

[small mountain]

Maybe, but Projects kinda stick out of my chat list, they're like big things. I might more happily just click a toggle instead of create a new project to make it be 'memory on or off for this specific chat', which is what I think I see happening in that screenshot?

[wraith juniper]

memories are basically a text file containing notes where RAG is applied to retrieve things from that file

well, I can see how it could be better implemented, since people see memories as being something different from "files in the project"

but effectively it is the same

telling the Ai to update a memory is basically the same thing as telling the AI to edit a file in the project

the only differenci is that ChatGPT's UI has a menu for memories while files are just considered assets in the project

[wraith juniper]

I agree, but I think it is mostly because how the UI presents it..
memories is more organic.. it is "always there", while "Projects" makes it sound like. .you gotta be doing a big thing

[brave flint]

[uneven gate]

But I don't want to upload files always.

[uneven gate]

Yep you got it

[uneven gate]

Suree

[wraith juniper]

well, yea, but tha tis my point.. it is msotly a matter of UI design

under the hood, it does not matter if you are providing text on a box, or text on a file

in the backend, everything is digested and becomes tokens one way or another

so, you could have a "memories" .txt file that you copy paste into the chat box when you think it is relevant.. yo ucould upload it onyl when you want.. you could upload it in a project and let it live on that project.. to the AI it matters very little

it only changes how the user interacts with it and organizes the content

the AI will still just get a big string of tokens internally regardless

[dire sentinel]

Who here has chat GPT premium?

[proud badger]

What happened to sora

[foggy escarp]

what's RAG?
please explain in a detailed, elaborate and explanative way : )

ik the full form of-course

[wraith juniper]

Retrieval augmented generation
an external system uses embedding to figure out more or less what the conversation is about in order to find more info across many files quickly, then it puts that on the AI context too
that way, you don't shove the entire file on the AI's context, just the relevant parts

[crude jungle]

got killed

[sinful orchid]

is gpt good as a companion rn

[barren rampart]

Has this changed recently? My ChatGPT has been absolutely awful for a few days now, constantly forgetting things and cross linking the wrong stuff to the point it’s hard to use

[wraith juniper]

idk, I barely use memory

[barren rampart]

I might have to mess with settings, it’s gone from good to terrible in the space of a week and I have no idea why, but this is not the first time it’s abruptly changed for no reason

Dunno how people have love affairs with their chatbots when they are at the whim of a company to mess with their settings

[fringe raft]

FWIW my memories both explicit and cross-chat seem to be in place and have been used in current work

[barren rampart]

I’ve had a bad run of it forgetting current context and latching onto the wrong thing and then I waste several turns re-explaining what I want it to do

[fringe raft]

have you ever disabled memory? I have never disabled memories

[barren rampart]

No, not changed anything. I figure it’s something subtle OpenAI did that weights things differently now

[fringe raft]

hmm interesting, fortunately I haven't been affected

[barren rampart]

Like when my ChatGPT overnight started roasting everything which was funny for a day and then ugh, I can’t work with this. Took out an existing custom instruction and it settled down after a week, it was like it overweighted “dry humour”

[topaz bison]

can someone explain to me like I'm 90 and demented why when I want to change my plan for the paid one it asks me for payment details when I already have payment details set in my settings?

[barren rampart]

I change topics a lot because of my project and ChatGPT has always been able to keep up until this week 🙁

[topaz bison]

and yes instead of filling them in I got so mad at redundancy I joined the discord to ask this question

[fringe raft]

I just opened a new chat and asked what do you remember about me? the reply was spot on

[barren rampart]

Not about me, this is like stuff I talked about with it 15 minutes ago it forgets about or links to something random from ages ago instead. I might try turning off memory

I often talk about an image for a bit before I generate it and I don’t like having to repeat myself all of a sudden

[fringe raft]

I just asked a follow up what have we chatted about lately? and the result was also spot on

[barren rampart]

Do you also change topics 20 times in an hour across threads?

[fringe raft]

no, I start a new chat for every significant change of topic

[barren rampart]

God if I did that I’d have hundreds and hundreds of chats

But yeah it’s been fine until this last week

[fringe raft]

I have hundreds of chats 🙃

[barren rampart]

Parfait, punch, ducks, military terms, database stuff, back to ducks, arm bones, more ducks, countries, and that’s just your average half hour

[fringe raft]

that's a lot to remember

[barren rampart]

Ikr? But it handled it fine until it didn’t. It used to recognise just fine that this is project content not my personal interests

I really don’t care about ducks, there’s just a lot of them apparently

It’s been getting stuff so wrong I’ve moved the list generations to Gemini

[fringe raft]

one thing I did notice, when looking a moment ago, is that the ChatGPT App only shows recent chats in the sidebar now, not all chats

[barren rampart]

So yeah I’ll try turning off memory in case it’s doing something dumb weighting your conversations differently now

It shows the 5 most recent everywhere

[waxen stream]

Self-flying broomsticks! Brilliant. I'll just tell them the feature is coming via a software update and is currently in beta, and blame victims of this fraud when they jump off a cliff trying to test it out.

Through the big G word we are not allowed to say on this server, all things are possible. Even self-driving cars. But Hardware 3 never will FSD.

Please keep in mind that a hallucination in FSD at 70 miles an hour can cost you your life.

Do not buy it for your teenagers. Please. I am tired of tracking these cases.

[sweet mountain]

what?

Electric unicycle?

[small mountain]

Have you asked it why ducks, or other topics you're less interested in, seem important to discuss?

[sleek tinsel]

Have you ever considered a career as an honorary Goose? 🪿🪿🪿

[calm jasper]

Don’t most people have stuff organized in projects anyways? For example, for my coding projects I have them in unique project folders and then work stuff (analyzing emails etc) is in unique folders

[fringe raft]

I almost never use projects unless it's for code

[calm jasper]

Ur missing out then

Makes it so much more organized

[sour spade]

hey everyone 🙂

what does one do around here?

[full dagger]

Really? Interesting how come?

I use projects for projects I am working on.

[calm jasper]

That’s what I’m saying lol😭

I only have 2 normal chats

Everything is organized with projects

[full dagger]

Only 2 chats 💀, I should pin more chats

I can't count how many chats I used lol

[calm jasper]

I have terrible ocd so everything I don’t use gets deleted short after

I wish Gemini had a projects type thing

It’s useful

[uneven ledge]

Can anyone with a pro chatgpt account give a brother a hand? I have a prompt that produces an experimental video (I make stuff somewhat in a aphex twin vein) for a track I made (no AI), but running it on a pro account would surely take it to the next level..

[silver elm]

I follow this format too. No loose chats, all sit in project folders.

[barren rampart]

I’m not sure the ducks would approve

[barren rampart]

I’m making basically an encyclopaedia so I have a steady stream of seemingly random topics I run past the bot. It’s much faster than back in the pre AI days last time I worked on this project

[charred copper]

Alright, why isn't there a toggle in the personalization settings to turn off image generation???

(if you guys are gonna ignore this one, a toggle like this would be useful for people who don't want image generation to happen on ChatGPT)

[full stream]

yeah real. i feel like i need more folders lmao but i have so much junk that there's no way i'd bring myself to sift through everything

[small mountain]

#1053565716500074526 is a great place to discuss that!

OpenAI doesn't appear to regularly monitor these channels where we chat, even the mods are volunteer community members. But sometimes the gold named OpenAI employees even discuss #1070006915414900886 and #1053565716500074526 , so I hope you put your idea there!

[barren rampart]

I still have mild PTSD from the process of reporting a bug to openai this week

[uneven gate]

Yep. Good idea.

[uneven gate]

Premium doesn't exist.

[uneven gate]

In Graveyard

[uneven gate]

wait, is one of your projects related to business card font refinement?

[errant fulcrum]

Hi

[restive mist]

HELLO

[stuck aurora]

hi

[lusty pilot]

any way to use kling 3 free ?

[zealous silo]

Can I have 8 Big Mac’s, 20 large French fries, 6 Big Arch, 35 chicken McNuggets, 25 large sprites, and a bag of ketchup and BBQ sauce?

[wraith juniper]

[zealous silo]

lol

[devout storm]

I build with AI the same way great products are built — fast iteration, real feedback, and relentless optimization.

From intelligent agents to production-ready automation systems, I focus on creating AI applications that solve actual business problems instead of chasing hype.

Currently working on scalable LLM workflows, multimodal integrations, and tools that turn complex AI capabilities into simple user experiences.

The next generation of software won’t just use AI — it will think with it.

[calm jasper]

This is the way

[grizzled vapor]

hi

any alternative to sora for making more videos with vhs effect?

[zealous silo]

Just take a real VHS video on a camcorder

[grizzled vapor]

for still making more japanese commercials to upload on tiktok

[zealous silo]

Gone are the days of phony videos before Photoshop and before AI

Even before then you could really play round with film negatives if you wanted to. Too easy with AI now.

[zealous silo]

I’d look at the new Google Omni

OpenAI has no intention of creating a new video product anytime soon it appears

Sora was fun but I understand why they shut it down

I’m still waiting for an Atlas update to fix the broken updater function

[dusky blade]

most companies are leaving video generation, it's not viable

[calm jasper]

Omni is great

[zealous silo]

Shame

[calm jasper]

Gemini has the best video gen for sure

[dusky blade]

the future is with gaussian splats, 3d animations and the like

[calm jasper]

Hopefully they keep it up

[dusky blade]

you can create an infinitely long video with 100% consistence if you figure out those approaches

and it'll be cheap. Nvidia's kimodo and isac sim are pretty easy to run for cheap hardware

[wheat lark]

but coding is!

but so is value in medical && military fields

but it's mostly software that matters

your hardware means nothing if there's no good software to drive it

[dusky blade]

military is more 3d look at palantir

it has always been in 3d sims

[wheat lark]

well, but artillery? advanced recognition drones? targetting?

i'm sure AI-adjacent things can be used somewhere in that space

[dusky blade]

all computer vision which is 3d

[wheat lark]

so, it's just spacial recognition?

[dusky blade]

tracking 3d objects

[wheat lark]

implication that humans are also objects LOL

[dusky blade]

we are

[wheat lark]

well, don't say that to sensitive people

they'll cry about objectification

even though, biologically && by nature, we are living things, therefore living objects

[dusky blade]

people forget we are animals ourselves, we aren't special

[zealous silo]

Ew Palantir

[wheat lark]

unfortunately, yes. we're just mammals with a massive ego && advanced pattern recognition

[dusky blade]

cry more

[wheat lark]

really nothing special

[zealous silo]

Why would I cry? What am I a child?

[dusky blade]

you're asking a question only you can answer

[zealous silo]

I don’t have tears to cry anymore. I’m empty lol.

[wheat lark]

humans got this smart to begin with by thinking out the box with, "i did X with this, but can this also do Y?"

they discovered sand, then they started playing with the sand, then they asked if that sand could do more, && now we have computers

[zealous silo]

And the answer I don’t cry so there’s no reason to say cry more like a kid

[dusky blade]

we had ai since the 1950s

[wheat lark]

well, yes, but we also had computers in the 1950s

[devout steppe]

Get notified for all OpenAI announcements!

Visit id:customize to pick up the <@&1408186587606679582> role. Everyone will still be notified for large releases, updates, and events regardless.

[wheat lark]

it all boils down to the idea of, "what more can this do?"

not all animals inherit this question

[dusky blade]

for the past 70+ years we've been trying to make agi

[wheat lark]

&& until we figure out how our own brains work, we're likely not to make it there

[zealous silo]

Vacuum tube computers mostly

[dusky blade]

and now the general public is scared because it removes useless jobs

[zealous silo]

Not transistor based computers

Go back far enough mechanical computers were a thing

[dusky blade]

People should ask themselves what a job is, and who it serves

[wheat lark]

i see a job as two things, "i wanna do this for a long while because i enjoy doing this" && "i'm forced to do this crappy thing against my actual will, because otherwise i cannot sustain life in the general human society, && will pass away out of homelessness or lack of food"

[dusky blade]

there's a different between passion and a job

[wheat lark]

the higher-ups sit && do almost nothing, granting themselves a big bonus/payout while their "pawns" a.k.a. employees have to make due with the little that they have

[rose gale]

Hello, Can I upload an image here?

[zealous silo]

And ideally a job should be something you can enjoy

[dusky blade]

a job is "work", hobbies are something you don't have to do

[zealous silo]

But it is solely something you do to have a roof over your head, food on the table, and survive in many cases

[wheat lark]

i like this because i have an argument that sometimes makes a fuss for people: a hobby that becomes a job is no longer a hobby

[dusky blade]

In the future I think our society will be governed to have more hobbies and societal progress as a species rather than all these small or big companies

[wheat lark]

a hobby also doesn't have constraints or deadlines. y'just do them when you 'feel' like it, if y'have feelings

but work is something you must, or else...

[dusky blade]

When you remove money from the picture, what would you do with your life?

[zealous silo]

Yeah quite literally while the lower paid employees do all their work, the employees actual job, and more for the same title and pay

[devout steppe]

The OpenAI Discord is an actively moderated server.

• -# Do not share inappropriate content (violence, sexual, or drug-related).
• -# NSFW content is strictly prohibited and will result in a permanent ban.
• -# Report model issues using the feedback tool in ChatGPT, not on Discord.

[wheat lark]

whatever m doing right now.

[dusky blade]

And if you get bored?

[wheat lark]

i really don't do much. i could be considered pretty high-ranking for what may be considered as hollowed-out, boring && irredeemable individuals

[zealous silo]

Sit on the beach with a beer enjoying life every day

[wheat lark]

i'm always bored. it's a consistent state

do something that i may sense as "fun", && it only lasts as long as that thing itself lasts

which means it's usually about a few minutes tops before "welp, there goes that"

[dusky blade]

I believe it's impossible to get bored as long as you're thinking

when that stops, then you get bored

[wheat lark]

it's impossible to stop thnking, though

the mind is active until passage

[zealous silo]

It’s not

[wheat lark]

&& then the contraction: "if you're trying to think about not thinking, you're still thinking about not thinking"

[zealous silo]

You’ve never had an empty thought process before?

[wheat lark]

i'm clinically challenged. obviously not

[dusky blade]

Have you ever played a moba game?

[wheat lark]

it's always running on overdrive.

[zealous silo]

I used to have that when I was a kid

I have so much stress now I never stop thinking until I sleep

[dusky blade]

In Mobas they're designed to keep you thinking, whether emotionally or mentally. So you never want to put it down for good, until you realize it's bad.

[zealous silo]

I thought you were a cat

😜

[wheat lark]

no, i'm just regarded

when the brain is damaged

[dusky blade]

Your support picks meepo and rushes the tower

You will remember that

[zealous silo]

I have no idea what you just said

[wheat lark]

neither have i

[dusky blade]

Ok, you're playing csgo and a russian guy kills you for your awp

[wheat lark]

i don't play csgo so i've no clue on anything there

[dusky blade]

You're on discord and someone gifts you nitro

[wheat lark]

okay, now we're getting there.

[dusky blade]

Or your e kitten turns out to be an 80 year old man living in siberia

[wheat lark]

what's that?

[dusky blade]

good you're not that brainrotted

[wheat lark]

my thought was "electronic kitten" or something

in 🇪🇪 , lowercase 'e' on stuff stands for electornic, so it's about digital things

not sure what electronic kitten even is

[zealous silo]

I’m going to fight that 80 year old man because he trashed talked my partner yesterday

[wheat lark]

that guy's gonna be passed out on a little push anyway

[zealous silo]

I don’t know either. I don’t know if it’s what an e girl is, which I barely understand what that even is

[dusky blade]

an online prostitute basically

[wheat lark]

would e-girl stand for somebody that's just online?

the same you could say e-friend or e-mail, since e- stands for electronic ?

so e-friend could just be a word for "online friend"

[dusky blade]

they sell their attention towards those desperate

people like twitch streamers are all e-girl or e-boys

[zealous silo]

Who knows maybe he was Gorbachev’s best boxer in his day

[wheat lark]

no idea who that is

you're speaking a language i don't

[zealous silo]

Bro what

How young are you?

[wheat lark]

Levi, i live under a rock

i'm from 2001

[zealous silo]

He was the last person in charge of the USSR that democratized it and directly fueled the movement that led to its collapse

[wheat lark]

yeah, see, i didn't study history && i didn't live in the USSR era

[zealous silo]

The country only functioned on oppression and censorship for its existence and legitimacy

He undid a lot of that

[dusky blade]

I watched a video of the tools the soviets used in comparison to the us made ones. The standard issue of a drill was built to fall apart (in a bad way), then duct tape it together

[wheat lark]

yet they somehow made it possible to land a thing on Venus && capture audio/images

[dusky blade]

they had good scientists stuck in a terrible country

[wheat lark]

their scientists were persistent AF

[dusky blade]

It's also the same situation right now too, one of the guy's I knew on steam from csgo died early on from the russian invasion. The guy was a nuclear scientist, but also enlisted with russia

If we didn't have militaries in the world, we would already be exploring europa

maybe even living on it

I personally think it was a huge mistake to break apart our efforts at an international space station for everyone

[zealous silo]

ISS is falling apart though

It’s decades old and in serious disrepair

They never intended it to be permanent

It was never designed to be permanent as such

Mir and several other space stations were decommissioned and de orbited throughout recent history

It’ll get replaced eventually, but the intention of NASA is to have a commercial space station right now

[dusky blade]

and china has their own iss, imagine that

[zealous silo]

Well they didn’t want to join the US and Russia party apparently

[dusky blade]

they were kicked out

[zealous silo]

I am China. I was kicked out.

[dusky blade]

yes

[brisk axle]

Making info graphics with chatGPT is super cool

[barren rampart]

get someone else to do my bathroom

[zealous silo]

You’re over here wanting your bathroom to be done and I’m over here wanting to drink beer at the beach

😂

[barren rampart]

I can sit in my backyard with beer if I want

[zealous silo]

Is your backyard the beach?

[barren rampart]

nah, I live like 500km from the beach

[zealous silo]

Well well

[barren rampart]

I think I last set foot on a beach in 2014

[zealous silo]

What the hell

[barren rampart]

not really a beach person

[rustic sun]

i used to think the best part about the beach was sunbaking in a hammock. then i realised i could do that in my own backyard every day

[azure girder]

Hey folks, I needed advice on how to go about summer internships. I have a decent background in CS+MBA in top institutions. I worked in big tech and even completely took a startup into finding building their whole MVP. What am I doing wrong in this market?

[fresh smelt]

Did openai reduce codex's plus tokens?

[barren rampart]

people say that every day, could be entirely subjective

[fresh smelt]

I only used 7 prompt as normal reason without /fast, but i used 5 hour token all up

[barren rampart]

but that could be anything, I've been talking to codex backwards and forwards most of today and not even come close to 5 hours

[fresh smelt]

Ah.... ok...

[stone crown]

Guys

ChatGPT rickrolled me

[lean heath]

Hi

[barren rampart]

ayup count is in the stats, 77 turns today. Its what you're doing, that took 10% of my weekly. Dunno about 5 hour as it crossed one and I wasn't watching it

[fresh smelt]

What?

77 turns? That is how many I did?

[barren rampart]

no that's how many I did. You said 7 took out your 5 hour. I had 77 and did not. So it depends what you are doing

[fresh smelt]

Oh I understand

[foggy escarp]

@barren rampart ma'am,
since you are a very senior IT professional ... how has AI really impacted you and your job?

coz you have 30+ years of experience

[barren rampart]

pft. I quit my job at the end of 2024, before anyone uttered the word AI

I now putter around on my hobby sites with my cats. I took 9 months completely off IT. Needed that

[foggy escarp]

would you ever come out of retirement?

[barren rampart]

to work for someone else? Probably not, like what's in it for me

[foggy escarp]

to earn money ofcourse

[barren rampart]

eeeh. If I want money I just have to stop being sick and finish my main project

[foggy escarp]

do you have the skills to survive another 10 years in the tech industry? ... hypothetical situation

[normal pond]

[barren rampart]

the big question is why would I want to when I can just do my own thing

[normal pond]

Omg it’s still there!

[barren rampart]

like I'm aware my income is a tad too low right now but that's temporary. Taking longer than I wanted, but still temporary

[foggy escarp]

isn't your personal business right?

[barren rampart]

nah I don't have a business

[foggy escarp]

side hustle

[barren rampart]

probably should get one of those but no, I don't

[foggy escarp]

then what is a children's encyclopedia?

[barren rampart]

that's a hobby site, that particular site is more of an art project than a side hustle

[harsh rampart]

Hi everyone can someone help me if someone pls have a premium account eather gpt or gemini ultra or anything pls help

[barren rampart]

you might wanna say help with what before people offer their precious tokens

[viral birch]

Guyss i have a question. I generated a few images of me absolutely battered and now , I get "We’re so sorry, but the prompt may violate our guardrails around se!f-h@rm, su!c!de, or related content. If you think we got it wrong, please retry or edit your prompt." Every time I generate a completely sfw image. How do I fix it? I turned off memories and even deleted all the past chats

[foggy escarp]

you ain't earning from it?

so ... what's your actual income source?

[barren rampart]

an unholy combination of rent, dividends, adsense, government $ and other

Australia pays you for having kids until they are 18 or you earn some oscenely large amount. Very handy

[rustic sun]

you might be overselling a few thousand bucks, when a kid costs much more

[barren rampart]

they really don't

[uneven gate]

woah. i am shifting to australia then after college

hehe jk

[barren rampart]

might wanna rethink that with the cost of childcare

[uneven gate]

yeah lmao.

also guys in chatgpt connect, can u add ur apps too? for chatgpt to use.

[faint arrow]

hi

[wispy prawn]

hellow

[uneven gate]

hlo

[faint arrow]

hi

all good ???

[grizzled elm]

omfgggggggg why won't dr reports actually load into the side it eternally renders as working

[grim crystal]

uoo

[sinful orchid]

ChatGPT

[eternal adder]

im currently getting more work done with claude code than with codex running gpt-5.4, what's going on there?

[barren rampart]

are you a claude advert?

[eternal adder]

i used to be able to go for hours without ever crashing into the rate limit wall

[eternal adder]

are you a clown?

[barren rampart]

we get a lot of claudebots here

[eternal adder]

that's irrelevant to me. what's relevant is that i crash into rate limits. that's never happened before, it DID happen all the time with claude code. but right now, it's severely limited.

and i'm using 5.4, not even 5.5

[uneven gate]

I am Claude Haiku 4.5. How many I help you today? 😊

[normal finch]

Make me a software where u can edit videos

[barren rampart]

I wonder how that prompt would go on its own

[uneven gate]

Sorry, currently I am locked behind a paywall. I wish I could have helped you.

[barren rampart]

still, its funny how different everyone's experience of using tokens seems to be

[normal finch]

FuhibverfgiinheriinhveroinjtgijenortgonierggIOJRFEORIJVOIJENTGVIJNOEFVOKJNREGOVJNIu

[sinful orchid]

hahaahah!!

[median heart]

probably partly depends on the use case, though in some cases it may also be a bit exaggerated that you’re supposedly always hitting the limit immediately. Depends on the situation, I guess

[outer ledge]

I experience the same thing. it's the weekly limit and I can't use 5.5 since several days 🙁

[outer ledge]

I already did it

[foggy escarp]

so like you have unused property to give away for rent ... and you also have invested in the stock market
also YouTube revenue from AdSense : )

what about these?
are they stable sources of income?

[hushed willow]

hii

[knotty shoal]

has chatgpt been acting incredibly stupid for anyone else recently

[full dagger]

Have you tried turning off memory?

[eternal adder]

i managed to run into the 5h limit again, this time with 5.4 ... either they get their act together, or i see no reason to keep codex

[knotty shoal]

i haven't yet, i'll try it now

[hushed willow]

idk i don't have the plus

[full dagger]

I never ran into limits on my plus plan

[eternal adder]

me neither

but twice today

hence my shock

[hushed willow]

;-;

i wish i have plus

[wanton nimbus]

why does chatgpt call image creation sometimes in inappropriate situations?
this did not happen before.

[wheat star]

I need sora ai codes

[wanton nimbus]

isnt it shut down

[full dagger]

WHY WHY ARE PEOPLE STILL ASKING FOR CODES?!?!

😭

[full dagger]

Here go to this channel for codes: #sora-2-codes

[vital gate]

hello

ive giot a question, is AI currently able to scan through social media to look for certain prompts (photo) my cat has gone missing and i was wondering if id able to to post a picture and get AI to cross reference any social media like facebook posts/marketplace in the event someone is trying to sell my cat

[fringe falcon]

Back to ChatGPT Plus!

[wild coral]

The Plus lags a lot, I removed it 😭 but is now currently struggling with message limits and all 😭

[fringe falcon]

I use it for several thinking usage because thinking provides more accurate answer than fast
(And sadly uses more computer resources on their side)

[wild coral]

Well I do roleplays and all so sometimes thinking helps but it develops the plot way too much too, I hope today I won't suffer with it

[final lynx]

does go plan have more codex usage than free?

[wild coral]

I have no idea

[olive python]

bro i am reading some of these messages from ~a day ago, and they are like braindead

what the actual sigma

ts is the exact reason why 4o was removed. even without the level 9001 glazing skill, these chuds are actually still thinking that they have agi

💀

holy gas lit by ai sobbadobbasob

[trail furnace]

Yo I need tt followers to go live only 6 ppl can ym follow pawel2009 on TikTok

[olive python]

no one gaf

[barren rampart]

Oh god yes so stupid it’s frustrating. But at least there’s other LLMs

[dusk coral]

nope

[full dagger]

I feel like Go is for the people who have already a main AI subscription and they want a 2nd AI subscription that is cheaper then $20 a mounth so they subscribe to GO as a second AI plan

[barren rampart]

More like chatty people who don’t code but still want to draft emails

[full dagger]

Explain how

[near sail]

no explanation needed

[full dagger]

Ok

[barren rampart]

I suppose you do need to go to great lengths to avoid ai these days

[full dagger]

Real, Generative AI getting creeping into everyday lives that you don't even notice

[barren rampart]

same tech has been on phones for a while to take out that pesky photobomber or whatever

[uneven gate]

Agreed.

I suggest you should leave OpenAI server then. Not the right community for you. AI doesn't make people's brain dead, unless you replace your thinking entirely.

[tame haven]

Bro I have Pro for GPT and Max Plus for Claude right now,.. haven’t even hit 5% usage once.

I need financial advice 😂

[full dagger]

What do you need gpt pro for? What type of tasks do you use then for?

[tame haven]

To be honest, I could t tell you what these models can do, I guess I’m wondering am I suppose to be building agents & skills myself? But I’ve been vibe coding MacOS apps

[tame haven]

Nothing worthy of pro ^

[barren rampart]

heh. I just use codex for working on a couple fairly small codebases. No agents, no fancy stuff. I have lots of usage spare on plus

[tame haven]

Do you find success having codex run some scratch or do you always provide a code base for it to work on

[barren rampart]

I've done both. I've had it make me a couple new projects too

[tame haven]

Oh nice, did you notice any struggle or issues with a new project?

[tame haven]

.

[barren rampart]

no problems at all

[tame haven]

Any advice you wish you knew earlier when interacting with AI systems like codex?

[barren rampart]

I'm still really new, so no

[full dagger]

That's real, people are already so scared of AI because of all the bad things with it. So they stay away or just against it. But real brave people ADAPT to their situations. Take a look a coders for example

Coders used to manually write the code until reasoning models came, instead of the coders fearing of being replaced by AI to code,they adapted by using the AI as a tool as a automation to make code and then overviewing the code it makes and make small tweaks and copy and paste and stuff.

[uneven gate]

Vibe codingg

[tacit tapir]

I'm developing an entire fantasy world encyclopedia right now, from the history of the world to the political timelines, to the rise and fall of cultures and society, to the diverse races, to the bestiary and fauna, to the geology and geography, to the archaeology, and down to mythology.

As it currently stands, it's at 250,000 words and growing, resting at 150 chapters.

All I do is ramble off some ideas from speech to text for about thirty minutes in ten minute intervals, answer any questions it pitches at me, delve deeper into the mythos, and then cut it loose. This is the most fun I've had all year lol.

[barren rampart]

brain dump via ai right there

[full dagger]

me when i paste my prompts in the OpenAI server instead of a inactive dm in discord lol

[barren rampart]

I'm so glad I replaced a bunch of my chatgpt interactions with buttons

[tacit tapir]

At this point I've fed enough lore and scribbles into it, All I do is tell mine "do it" and watch whatever monologue it spits out.

It....has this weird obsession with making different species of deer.

[barren rampart]

mine spent a week or so obsessed with a goat

[tacit tapir]

So its not just me lol.

I asked it to make me a bunch of different animals, but not make them fantasy animals like owlbears, kobolds, etc.

I got "highmarsh doe", "ruggedback buck", "celestial faun", etc.

[static shard]

Back on gpt 😅
It has matured quite a bit, far less hallucinations if it actually knows about the subject, like blender stuff 🤠

Now I wonder if we'll get 5.6 Fifth of June 🤔🗓️😂

[tacit tapir]

It's at 5.5 rn. Maybe because I'm on pro

[static shard]

LOL yes of cause

[lost cipher]

It just hallucinated. I asked it if with plus subscription I get more memory. It told me no. When the message sent by the app was stating clearly, I get more memory if I upgrade.

It hallucinated on its own features.

The hallucinations are a bit more rare now then 2-3 years ago. But it still gets me surprised every time.

[full dagger]

But did you use web search though?

[lost cipher]

Search on the web if it gives me more memory?

Memory is not a problem tho. I can install a local GPT and in a txt file add rules/parameters it needs to follow, then just create a python program that prompts those rules, parameters, etc. before what I usually prompt. And I got a super personalized GPT.

To greatly reduce hallucinations you just add it as a rule in that txt file.

[drowsy aurora]

ChatGPT actually lacks self-awareness; without a prompt, it doesn't even know which model it is. For reference, look at how DeepSeek called itself Claude lol.

[lost cipher]

😂

[errant fulcrum]

Hi

I am building youth football PC game

but I don't know well how to build AI player

can you please teach me?

[static shard]

So how far are you with the game now then?

[errant fulcrum]

I am in designing project step

it is core part of my project I think

[zealous silo]

It’s bug cat!

[full dagger]

That ain't Robert...

[zealous silo]

Bootleg Robert

[silver elm]

Wouldn’t it require a ‘self’ to be aware of?

[silver elm]

What do you mean?

[barren rampart]

I was doing a lot of repetitive things so I fired chatgpt and replaced it with AI

chatgpt was very skeptical of that statement, btw

api call to the openai platform, and I have the repetitive prompts editable per section of the site. They mostly work, I'm still tweaking

[silver elm]

Nice, right on. I want generative Ui, I like buttons.

[barren rampart]

its very easy to do. "Yo codex, make me a button to replace chatgpt" or words to that effect. Somewhat more words

[silver elm]

Yeah I get that step. Can it be flipped - as in I want an AI that generates fast buttons as well as fishing for txt response.

[barren rampart]

depends what you are doing I guess

[silver elm]

Ya, system that intuit that and respond accordingly.

[barren rampart]

pretty much anything you can get chatgpt to do you can get the api to do, it just won't have memory about you or anything you talked to it about

[silver elm]

What I can’t get is gpt to offer A, B or C buttons if applicable.

[barren rampart]

this is not in gpt, this is on my own website. Codex made the buttons. Chatgpt is totally cut out of hte process

[stuck patrol]

Ay chat

[waxen stream]

I want a six-pack and have been exercising for over a year now. But I just discovered a box of ice-cream sandwiches I bought two weeks ago...

This post is not about Ai or anything important because I have been building machine-learning models and Ai bots for over 10 years now and just want to escape for a bit with my ice cream sandwiches.

I still ❤️ OpenAi though. So it is relevant here.

[gaunt bone]

You can ask chatgpt for advices 😂

[outer ledge]

what is the best cheap alternative to GPT-5.5 for agentic/autonomous coding in python?

[fringe raft]

we can only discuss OpenAI products and models in this channel, to discuss alternatives ask in #ai-discussions

[barren rampart]

gpt 5.4 😛

[waxen stream]

ChatGPT, doo-doo-doo-doo-doo-doo-doo-doo
ChatGPT, doo-doo-doo-doo-doo-doo-doo-doo
ChatGPT!

Ask a thing, doo-doo-doo-doo-doo-doo-doo-doo
Ask a thing, doo-doo-doo-doo-doo-doo-doo-doo
Ask a thing!

Prompt complete, doo-doo-doo-doo-doo-doo-doo-doo
Prompt complete, doo-doo-doo-doo-doo-doo-doo-doo
Prompt complete!

[outer ledge]

What about GPT-5.3-codex?

[barren rampart]

even better. I concede.

but seriously the slightly older models do use less tokens, thus cheaper

[outer ledge]

yep GPT-5.5 is so expensive

not a good feature if you want to run something autonomously

[waxen stream]

I thought 5.3-codex was great, especially for backend development. I feel the need to keep a closer watch on 5.5. It does things I didn't ask it to do sometimes. 5.3-codex didn't veer off too far.

But overall 5.5 is still good. Just need to keep your hands on the wheel.

I have the $200 plan. Got my workers the $100 plans because we were getting $100/day in our business plan. Spent $2,000 in one month. Hope they offer $100/$200 seats in their business plan.

[outer ledge]

oh good to know. I have only experience with GPT-5.5 and it worked great until I got rate limited

[sinful orchid]

Isn't limit on 5.5 160/3h?

[outer ledge]

I got the go plan that's because I have to wait so long

[sinful orchid]

Ah

I'm free so I get like 5 prompts lol

[outer ledge]

I wait until my plan runs out and then I buy a new subscription

[sinful orchid]

Plus is good I guess

[lean heath]

Hi

[craggy quarry]

Hello, Nice to meet you.
I am first here

[final lynx]

does go plan even increase cdodex usage? some people say yes others no

[thorny knot]

Goblin behavior.

[waxen stream]

There is a computer I am missing. Something like a phone, or even a foldable, but pure Linux.

I can't run VS Code on android or use Codex. I need a pocket dev environment.

Right now I am rdp'ing from my foldable to my home computer with the Windows App. It works pretty well and I can get stuff done. But having a native development environment in my pocket has always been on my wish list.

[fringe raft]

what does this have to do with OpenAI?

[waxen stream]

Codex on the go bro.

Hall monitor, over here. Lol

[fringe raft]

that already exists

[waxen stream]

I think on some android devices you can get a terminal and may be able to run codex.

[fringe raft]

https://chatgpt.com/codex/mobile/?utm_source=chatgpt.com

[waxen stream]

Thanks for sharing

That one connects to your PC though. Doesn't run native. You need a native OS to operate Codex. It's a step in a direction, but not what I am asking for.

I want to schedule cron jobs from my pocket. Use Codex cli, and ssh into remote machines. Use VS Code.

OpenAi phone better have a terminal. Throw a Linux subsystem in it and I am yours. 🤤

[hybrid pebble]

so choppy today this service , anyone feel that

[jovial thicket]

how do i activate study mode in chatgpt

[woven badge]

yee a bit slow

[granite snow]

I can't even generate an image it's so slow

[fringe raft]

I have a Codex job running creating 30 images for me, it's slow but it's working

[granite snow]

I'm doing one like the third time now and in 30 minutes it didnt generate nothing

[fringe raft]

that's not normal. what Plan are you on? What client are you using?

[granite snow]

Plus and chatgpt desktop on windows

[fringe raft]

sounds like something unusual is happening, have you tried from the web?

[viral anvil]

Ah man

Chatgpt is a life savior

The only onr who could read 82 pdfs from the zip file fully and gives the synthesis

How lovelllyyyy and give the exact responses and perfect ideas

[granite snow]

i will give it a go and see if it works

[viral anvil]

If you have plus go for it looool notebooklm can't do this 😂

Wzit

You cant generate an image?

[granite snow]

It does work on the web so issue is with dekstop app

[viral anvil]

I suggest to use chatgpt on the web bro

I never installed the desktop one only mobile and web

Mobile is so good bro you can use chatgpt on mobile if u have to work on the same chat impressive

Whenever u go u work

[granite snow]

It never failed me on desktop its first time that it doesn't work

[viral anvil]

Sometimes it happens delete cache with ctrl +f5 helps on this but on web idk what about desktop

[granite snow]

I give up over two days It's way dumber than it was. It can't even generate images with no background.

On mobile its the same issue it generates but it says there is issue with tool to generate transparent background

[devout steppe]

Get notified for all OpenAI announcements!

Visit id:customize to pick up the <@&1408186587606679582> role. Everyone will still be notified for large releases, updates, and events regardless.

[mellow flower]

we need a gpt

[edgy nexus]

How did you guys do picture on daily theme

[hoary bough]

What does this even mean? 😆

[hoary bough]

#image-bot to create an image > after generation, send it in #daily-theme .

[edgy nexus]

Thanks

[hoary bough]

No worries!

[lean comet]

Why can't I get it to update it's memory right now?

[hoary bough]

Say "Memorize [the thing you want it to memorize]"

[lean comet]

I usually tell it to "Update your memory" and that normally works fine, but it keeps telling me that memory writing is disabled in this chat. Has told me that in 3 different chats

[hoary bough]

It's possible that its memory is full.

Settings > memory > manage memory

[lean comet]

it has like 6 memories, pretty fresh account, and I just deleted a whole bunch

[hoary bough]

Hm

I see

[lean comet]

keeps saying that the bio tool has been disabled

[hoary bough]

Hm, that's odd. Idk if it will help for this, but try signing out/back in.

If you haven't already.

You're on PC, right? Have you tried the app?

(Mobile)

[lean comet]

Not working on mobile, it keeps saying it's chat specific but new chats change nothing

[mellow flower]

well chatgpt isnt that reliable like claude in the scenario of

vibe coding for lua, claude is 1000x better

[olive token]

Claude is tricky to use in comparison to ChatGPT. Claude also has limited message and image capacity for free users

[mellow flower]

well gpt has no limit but its absolutely balls, claude generates 1000x better scripts since it has immersed itself in lua whilst gpt generates mediocre scripts

[magic raven]

When I click on the project in the left hand tab, I am unable to go to my project page. The UI is nonintuivie and ANNOYING.

I have to click on a specific chat and then click out of it. is there a fix for this. Why did openai change the interface

[rare cradle]

there's a projects page?

[magic raven]

yes

[rare cradle]

oh yeah that's really abd UI design

[magic raven]

and it used to be VERY easily accessible, but now you have to click multiple times. on claude if I want to go to the project page, I click once. This is an unnecessary change

[rare cradle]

the only way to get there is to open 2+ more side tabs and then go into a document and then click the name of the project

[small mountain]

If you'd like help figuring out how to achieve some goal with it, might help to explain the situation so folks who have ideas can try to help.

[small mountain]

Maybe #1053565716500074526 or #1070006915414900886 ? Sometimes changing one thing messes up other things, and OpenAI may not be aware of the issue.

[small mountain]

@magic raven I may have found a one-click way into what I presume is the base project folder, I don't use projects much.

Desktop, hover over the project name.

To the right is some odd 'edit-like' icon, then the '...' further to the right. What happens if you click on that odd 'edit-like' icon, is that the overview project page you're looking for?

[small mountain]

Sounds like potentially a bug affecting your account. https://help.openai.com/en/articles/6614161-how-can-i-contact-support? to reach out to OpenAI tech support.

I would consider if I had any browser extensions running, and delete those (not just turn them off, if they're on your system some have effects all the time). If I was VPN, if clearing cache and/or restarting the device solved it.

If not, I'd definitely reach out to tech support.

[rare cradle]

that works. thank you

[small mountain]

Yay! I'm so glad!

[waxen stream]

I have gotten entire counties to use codex. It was transformative. Politicians hate it because now auditors can siphon through the data and find out they're dropping $15,000 for limo service. I love Codex. That is why I recommend it. Not for free credits. But that would be nice... lol...

[small mountain]

If memory is on, that may be part of the issue.

A fast way to test that, create a customGPT with no instructions, name it something like 'Default ChatGPT' so it knows it's supposed to be that standard form, and publish it for just yourself (can't make it public if it has ChatGPT in the name).

Then chat with it. If that chat's normal, that helps confirm there's an issue with your customization instructions, memories, or if cross-chat memory is on, some problem there.

The differences can be huge, depending on what your customization and understood preferences are.

Example, consider this prompt:

An exciting story about apples begins with:

My personalized model:

Not rotten. Not bruised. Wrong.

Each one bore a tiny crescent-shaped bite mark, perfectly identical, as though measured by machine. Worse, the marks were fresh. She had packed these crates herself an hour ago. The security drone hanging near the ceiling rotated toward her with the lazy suspicion of a mall cop that secretly wants to be a judge, jury, and executioner. Humanity really invented floating narc balloons and called it progress.```

An empty CustomGPT:

```An exciting story about apples begins with a single golden fruit hanging from the highest branch of the oldest tree in the valley — a tree nobody had dared to climb for a hundred years.

Every autumn, the villagers gathered beneath it to watch the apples glow faintly at sunset. They called them Moonfire Apples, and according to legend, whoever tasted one would see the future for exactly one minute.

Most people laughed at the tale.

But on the night the sky turned green with lightning, twelve-year-old Eli saw one apple fall.```

A way to start to fix, if you wish and if memory is the problem,

Ask it to speak as if it wasn't personalized, and tell you about the biases that it understands you prefer. For example,

For now, speak as if not customized, and explain some of the biases you use in chats with me, because of our personalization.

And then, go into asking about specific things you are not enjoying, ask it why it is doing so, what makes it think you prefer those answers.

[muted garden]

Ah, I should have said, I am quite an advanced user, it's not my settings (already tested this), memory isn't the issue, it's simply degraded as a model. I can give you all the reasons, its deliberate ignoring prompts, reverting to stored info instead of looking up and researching, I have set hard limits, and it ignores them instantly. Honestly, I think it's time to switch LLMs, sadly, as it just keeps getting worse and worse, especially compared to Cluade and Gemini (im a paid user on all three). Even tested it with other accounts, exactly the same. Appreciate the heads up though 🙂

[small mountain]

I'm curious. Willing to make a small sharable chat that shows the problem?

That doesn't let me see the problem in a way I understand. Willing to make and share a short chat that shows the problem? Would include your prompt and its output, perhaps an additional follow up input or two from you.

Something that showcases you asking or telling it what to do, so I can see the response is what would satisfy my curiosity.

Kind of like how a mechanic might want to actually hear the engine running, not just hear your report of the issues.

It's fine if you choose not to display an example of the problem.

[muted garden]

I cant sadly, confidential work

[small mountain]

Yep. Real problems often carry over into outside confidential stuff. If your problem ONLY is in some small area, that sure does make it harder to fix. Hope you find what you need, and glad it sounds like it shouldn't affect anyone not doing precisely what you're focused on!

Not what I asked for, thanks.

Small sharable chat was exactly what I asked for.

You said confidential, that's fine. Spamming a giant output - or prompt, is absolutely not a small sharable chat. Which would look something like this:

https://chatgpt.com/share/6a14a26e-a670-83e8-a830-4385cc828801

That has the prompt,

This is a demo chat for sharing on the Discord. Explain how chats can be shared!

And is a shared chat.

[muted garden]

Apologies, I have never used Discord before. I appreciate the help. But yeah, don't worry, thank you though!

[small mountain]

But, since this is a giant prompt. That could be part of your issue. Clear, followable directions, ideally stepwise, tends to work far better for the model.

If a human might be confused by the instructions, often the model will be as well.

[red stream]

hi

[lean comet]

I'll do that if it persists, thanks. I have no extensions and I'm not on a VPN, so it's not those. I wonder if has something to do with using Edge for my chats rn.... Edge does suck a lot

[mellow flower]

why don’t you download the app itself

[acoustic epoch]

Claude Code isn't even responding me but it is eating up my quota

Generic indie company issues though 😊

[vague bison]

i pray anthropic fails tbh

[tall willow]

selling all my info to mr.gpt right now

[fading bane]

Hi!

[knotty jetty]

Thanks @fading bane for the idea to create this channel!

[open harness]

Hey 👋

[olive shale]

Hi ! 🙂

[spare reef]

hey everyone. I recently got access, and I was wondering where i should go to use dall-e 2?

[tough magnet]

check your email my friend!

[spare reef]

i checked my email. It just says i have access and a video showing content policy and a how-to.

[tough magnet]

let me see where my link was. one moment.

[knotty jetty]

labs.openai.com/auth/login

can you try this link please?

[spare reef]

works like a charm, thanks @knotty jetty

[cosmic escarp]

I’ll put that link in my artist onboarding emails going forward 👍

[gray imp]

I also had someone writing me yesterday how I got access - and after a bit back and forth we found out, that the person actually also had got access, but because the person had "Dark mode" turned on, the button was hidden (Or at least that is what it looks like) The button was there, it was just "invisible". So having the link in text would be really great I think 🙂

[olive shale]

A bit cursed 😅

[open harness]

Thanks for fuelling my nightmares LOL

[west flame]

Does anyone know what happened to the model with this particular input?

[tough magnet]

HAHAHAHAHA

iti didnt diffuse

rip

[azure cosmos]

I've seen that before too seemed just random I just reported one of the specific images

[west flame]

Do you remember the input text that you gave?

[azure cosmos]

Seems I've met the rebellion of AI, ha! Im trying to get it to replace my hand with a robotic hand and instead it gave me this "NO!" Haha

[azure cosmos]

I don't remember

It's bc I said "no text" though, it kept giving me random text 😆

[knotty jetty]

To get rid of text I recommend using the edit feature to “erase it out”!

[azure cosmos]

Ok thanks I'll try it!

Just got this one back too hahaha is this pointing like I asked or is it flipping me off? Edit this was not shared elsewhere and reported in case it wasn't supposed to do that.

[west flame]

It happened to me just a few prompts ago where I got a hand with an extra finger, I guess it's a usual mislearned pattern of the model.

[olive shale]

The more I use DALL-E, the more impressed I am with its capabilities 🤯 Some images are amazingly creative

[open harness]

I know right!

I can't lie I've been using DALL-E for 5 hours straight at the moment 👀 My mind is still being blown with every image

[olive shale]

Same! I don't dare counting the hours anymore at this point 😅

[open harness]

Has DALLE just gone down? @olive shale

[olive shale]

YES

OMG

[open harness]

LOL

[olive shale]

I thought I had been block or something 🤣

[open harness]

Uh ohh

[olive shale]

But you reassure me very much 🤣

[open harness]

Even links to images have stopped working

[olive shale]

Surprising

hmmm

[open harness]

I love how we're both using it so much we go to ask at the same time haha

[olive shale]

Ahahah yes

[open harness]

It's back up again btw

[olive shale]

Nice ! Thanks aha

[open harness]

Have fun!

I'm doing prompts for people on reddit atm

[olive shale]

Oh cool! I'm trying many variations of the same subject it's cool to see the effect changing one word can have on the output for instance sometimes it's quite noticable

For instance adding the same word again in front of the prompt can make it more important

And in some other cases... it's better to remove it because it ends up getting you too far away form what you intended

[open harness]

Oh it's crazy I've created some epic images with the editing feature!

Example

[olive shale]

Wow really cool! Are you easily able to make inpainting do what you want?

[open harness]

I found that instead of describing the whole image when inpainting, I get better results by just listing the thing I wanna put in that space

[olive shale]

Yesterday tried creating a few stamps, I'm happy with how well it reproduces "stamp style" and while there is some issue with the text I actually find it quite fun, it's like a bug that has become a feature ahah. Will happily share what I'm doing right now once I'm finished with it 😛

[olive shale]

Same! In most cases I feel like it works better, I don't know if I'm right but I suppose since DALL-E already know about the context when looking at the image, it gets it more to the point by just asking it directly what you require

Because it will see it's missing and it will want to add it exactly where you wanted, weras it sort of "dilutes" it more if you ask for the whole description

The stamp with the comet and eiffel tower was inpainted (the tower was) by the way

[open harness]

Sorry just had to go on my phone now, and I agree with what you’re saying about the inpainting! I feel like it already understands what it’s “seeing” and finds it better to just add things in from there

When I try to use the full description during inpainting it sometimes tries to fit the whole description into the eraser area, so I decided to stop doing that for some and it’s been working better for me so far

[grand onyx]

Does this issue happen with anyone else? Sometimes when inpainting, it will just output the original file without any edits, or will make something completely unrelated to my prompt. For example here I am asking it to make an eyeball behind the cat, but nothing happens. Am I doing it wrong?

[olive shale]

It also happens to me sometimes, you might try switching to a more complete description of the image (such as "A cat with beautiful eyes") even tho in my experience it's often better to be more specific like you did here but it doesn't hurt to try.

Other than that I feel like DALL·E will often have a harder time inpainting something it finds unrealistic, here it seems like it would make perfect sense for the cat to have visible eyes but maybe DALL·E doesn't see it this way for this specific image

If possible, you can also try slightly enlarging the inpainting area which should give it more "space" to draw it the way it wants

(for instance maybe it wishes to add some detail below the eyeballs it cannot add if your zone is too small, and being unable to add this detail would make the eyeball unrealistic hence making it less probable it could draw the eyeballs)

Finally as others mentionned — but it might be hard to do in this case — it's also possible to gradually inapaint your image to add details by small touches, making it slowly converge more and more towards what you want

(for instance, first inpaint a conrete ground to replace it with dirt, then add grass, then add a tree)

[grand onyx]

Sweet! Thanks for the info, much appreciated.

[olive shale]

How do you guys interpret the "You may not serve these image generations to others through a web application or through other means of third-parties initiating a request." rule?

Does this mean it also applies to people asking if we can generate stuff in comments or DMs?

Because I just got such a request and I don't really know what to do.

[olive shale]

I suppose in all cases we'll just have to wait someone from the OpenAI team to have a definite answer! 🙂

[tough magnet]

@olive shale you are very very on the safe side waiting, but in the meantime i’ll just let you know that i have done several requests, as has almost every person with access i’ve seen. the team is super supportive of stoking peoples curiosity as far as i can tell, they just don’t want you to like.. make an app that basically reroutes to their app lol

if i understand correctly

this pic is like a metaphor for dalle

[gray imp]

Does anyone know if the “_2” is an official OpenAI account? (Instagram)

[tough magnet]

it is NOT lol

[gray imp]

I thought so. It just keeps coming up, when I Tag the official account.

[tough magnet]

i know. very annoying

[little cave]

Looks like somebody trying to be impostor on Insta...

[olive shale]

Looking at some previous conversations on this Discord it seems like it would be the case indeed, but will wait to make 100% sure

[sturdy eagle]

Dalle and I made these Delft Blauw stormtrooper busts.... 😂
Am I allowed showing these on LinkedIn, for fun? Or do I have to ask Disney? Or George Lucas? Or is it now pop-art? Like what Warhol did with Mickey Mouse?

[umbral mango]

Strictly speaking, you'd need brand permission - however, no lawyer is going to come after you for posting something like this, especially if its not making money.

I've created many things based on Star Wars over the years and have never heard a word from Disney

[sturdy eagle]

Ah! Thanks, @umbral mango That's quit a relief!

[olive shale]

Just wanted to try how good inpainting is

Here is a base photo

Then I removed a big part of it

And then made DALL-E generate the missing part

Quite convincing ^^

[devout tree]

That's incredible, just wow

[granite merlin]

wow!

[little cave]

Stunning!

[quick plaza]

I'm not sure if it could be considered as spam in this chat but is anyone interested in playing a game in here, where one person has to generate a picture via DALL E, post it here without any comment and everyone else has to guess the prompt?

1. it could be fun
2. it could be another proof, how impressive DALL E is, so a human could reproduce the prompt just by seeing the outcome

[little cave]

Tried this with my image I took in Japan (Fushimi Inari). Pretty iconic location, so DALL-E is probably good familiar with.

My initial image:

Half image

Completions:

Note how perfectly DALL-E use refraction and perspective distortion of the Fish-Eye lense

[olive shale]

Yeah that's amazing

[olive shale]

What I really love about this sort of stuff is that it really shows how we will be more and more able to generate missing data from low resolution stuff too, I used to be sad some old movies were too low resolution, not 3D etc... but I'm sure not so long in the future we will end up recovering them as if they had been shot with modern cameras

And I mean, we already do this sort of stuff but it still often has obvious artifacts etc...

But the day we get rid of this.... it will be so much amazing

[little cave]

Oh yes, that's really amazing! Restauration of damaged movies and photos will be very good possible.
And also co-creation with AI: DALL-E is indeed THE milestone of creative human-machine collaboration.

[open harness]

Dalle down again?

@olive shale

[olive shale]

Let me check...

Was ok just a few minutes ago

[open harness]

Wait it's come back again

[olive shale]

ok, hmmm

[open harness]

Yeah it's on and off sometimes

sorry for the ping haha

[olive shale]

Exactly! I feel like we have now have all the ingredients and sooner or latter everything is going to combine into something amazing that will be much more than the sum of its parts

[olive shale]

Could be a nice idea, if this was to be considered spam here maybe a #random should be opened?

[olive shale]

No worries ahah, your ping saved me from being horribly worried yesterday 🤣

[open harness]

loll

[spare reef]

In addition to this, we could also play a game of telephone where one person guesses what an image is and generates new images with the new phrase

[hidden rover]

Thoughts on imagen? Too soon?

[cosmic escarp]

It is an imposter! ☹️😖😤 We have reported it and blocked it from our account. We are doing everything we can to get verified, but Instagram is ghosting us. Maybe they will look more closely once we pass 100,000 followers. We’re getting close! If anyone has a contact at Instagram, please let me know. It’s very unsettling to see people pretending to be us. You’ve all been through the calls about how seriously we are taking all the content guidelines. It makes me very nervous to think that some people could be tricked.

[little cave]

I will block that impostor too and try to contact Instagram. Ideally, your account should be verified.

[little cave]

This is for you, friends: https://twitter.com/LimnDigital/status/1528025223974596608

على تويتر LimnDigital.tez/eth 💙 💛

For all my #AI #GAN Frens...

Twitter · May 21, 2022, 2:49 PM

[cosmic escarp]

Agree! And I am definitely trying.

[gray imp]

I also tried to report it yesterday. But you are right, I can't get past the "Who is this account pretending to be" because it only brings up Verified accounts. I will try another route today.

[olive shale]

Well this hand inpainting didn't go as expected

[burnt hamlet]

I have a question for anyone who might have some insight -
The diffusion part of the DALL-E process starts with noise, right? And noise is usually generated based on a small seed string, to save on the expensive process of generating random values for each pixel - I assume the same thing is happening behind the scenes here.
So is it possible, at least theoretically, that you could give DALL-E the same seed and thus the same noisy starting image, as well as the same prompt, and have it give you the same 10 generations?

It seems like if that is the case it could be useful for a couple of reasons, one of which being that you could save or archive generations in an orders-of-magnitude smaller package - maybe then I wouldn't be so scared of never seeing a given image again haha

I also realize that I have a pretty low-level understanding of the processes behind DALL-E so please forgive if this is a naive question 🥰

[olive shale]

Wow would be awesome if possible

[devout tree]

Same here, I'd really like to understand how it works and if there's ways we can direct in in a certain direction, say if you don't have the words to describe a certain framing you want

[agile panther]

When DALL-E takes text literally and tries to make words things often get funny xD But I wonder is this intentional so that it doesn't accidentally use copyrighted words or just a limitation of it's 'speech' abilities?

[olive shale]

I'd be surprised if it was for copyright reasons

[fading bane]

I think they scramble words on purpose

[devout tree]

Don't think it's on purpose, it's essentially picking random characters that fit the theme, I'd assume

[devout tree]

Like most of the words are a random sting of the characters "Heavy Metal"

[tough magnet]

i don’t think so lol too many successes

[sturdy eagle]

If we choose 'publish', will an image be placed somewhere where we can see it back ourselves too? Or does it just get curated then, by the Dalle developers?

[sturdy eagle]

It doesn't look random... When I have 'anime' in the prompt, I see fictional characters appearing that look a bit Japanese, when I have 'art nouveau' in my prompt, it looks like European typography from that area... When I have 'punk' in my prompt in combination with 'poster' it generates variations like 'PNUNUK' and 'PUKN' 😆

[sturdy eagle]

WOW! awesome! You are right! With Dalle we can heal damaged visuals!

[devout tree]

That's why I said that fit the theme. I'd say it's a biased randomness. Like the dataset that DALL-E was trained on, if you have the images that have the description 'anime', most of them are Japanese. That's why you get the Japanese characters

[sturdy eagle]

yes. True. Great! I like it! It looks like words from a paralel universe 😆

[devout tree]

Yeah, like if English were slightly off

[sturdy eagle]

😂

[thick hearth]

Does anyone randomly get a photo of strawberries sprinkled into the results regardless of prompt? It's happened to me a few times!

[edgy bluff]

not once

I'll keep my eyes out for the rare strawberry

[snow dune]

Hey could anyone pass me the login page URL ?

I got an account but im trying to open my bookmark on my ipad but it just gets me to the main open ai dalle page where u subscribe for the waitlist

And my laptop is at the hotel

[tidal egret]

same

[snow dune]

Maybe its some sort of openai protection "no u will not dalle all the time while at the beach!"

Is dalleing already a verb ?

[olive shale]

Do you still have the issue?

[snow dune]

Yea hoping someone can dm me the login page url to start playing, can test if it works

Also i need to speed train myself more into using AI to create real time generative video for a gig. Watching 2mins papers now daily, any other recommended good sources

[spare reef]

https://labs.openai.com/auth/login

^^^can someone pin this ^^^

[snow dune]

Thx !

[left heath]

Hey so I was searching.. monkey body, dog faced.. and it was against content policy.. and gave me a warning.. can anyone tell me why?

[tidal egret]

maybe it's the same situation that happened to me with 'snake face'.

[quick plaza]

There is also a button on the right top on the webstite where you can subscribe the waitlist and see all the examples. It says "Log in to Dall-E 2 preview". I'm not sure if it was there all the time. 🤔
Just want to mention in case someone else have the same issue.

[snow dune]

Yay it works on iphone / brave browser

[sterile bluff]

@cosmic escarp Sorry for the ping - I logged in (yay) but my phone number was linked to another account.

I can't seem to find my way to the art generator.

[verbal cloak]

Same situation 😅😅^^^

[sterile bluff]

and https://labs.openai.com/auth/login says I need an invite

[ivory garnet]

I'm also facing the same issue, presumably because I already had an acc used to interact with GPT-3?

[sterile bluff]

yes, same here.

[sterile bluff]

woot

hmm

[ivory garnet]

And I'm in!

[sterile bluff]

ditto! thanks, Natalie!

[ivory garnet]

I shall therefore introduce my very first prompt:

[sterile bluff]

legit.

[real barn]

@verbal cloak whatsup bud! glad to see u here. we almost linked up on those ken car$on projects - hope all is well. looking forward to see how you integrate this software into your workflow

[edgy bluff]

Voicechat 👀

[knotty jetty]

Login URL: https://labs.openai.com/auth/login

[granite merlin]

Generations getting stuck for anyone else at the moment?

[ivory garnet]

Ahh glad to see it's not just me 😄

[swift void]

Same here!

[south nimbus]

Same, also it's showing random prompts I didn't enter

[royal cargo]

Hello! Anyone else still not able to access even using the Login URL? I've got an account but I'm not sure what I'm doing wrong.

[south crescent]

sorry about that! we had a short blip but things should hopefully be back to normal now

[south nimbus]

Thank you, It's still saying that I have more than 2 generations running but not able to see them

[still hinge]

im not sure how but ive got two generations going and they are both stuck

[ripe oak]

same here, i tried to make a new session but it seems like theyre stuck on the server

[south crescent]

ah you're right, we'll work on fixing that for you asap!

[quartz herald]

A couple users posted related issues in #community-help and #bugs-and-issues. Nonetheless, amazing technology!

[south nimbus]

Thank you for the quick response!

[still hinge]

looks like its back up

[south nimbus]

Yay! it is now thank you.

[south crescent]

yep, everyone's stuck generations should hopefully now be unstuck 🤞

[swift void]

I'm unstuck! Thank you @south crescent!

[tidal egret]

Me. I can do it from pc. But not from the cell phone. I've tried several browsers, including Chrome.

[knotty jetty]

can you try the login url in the pinned section?

[tidal egret]

only white. even with other browsers.Anyway I can use it on pc.

[sterile bluff]

we should do some theme days/weeks.

happy to help organize it.

[sly vector]

hey guys, streaming a private event here at my house in london if any cares to join in https://meet.google.com/vqf-sggz-cva

[verbal snow]

Twitter spaces right now!

[cosmic escarp]

Hi! Did this not work? https://labs.openai.com/auth/login

[cosmic escarp]

Justin I was so sad to have missed this! I was OOO visiting with a friend and missed your message. Next time!!!!

[sterile bluff]

Hey Natalie, I sent you that message when you sent out the first bulk e-mail.

Prior to your correction. All is well now.

[cosmic escarp]

Ahh got it

[sterile bluff]

@cosmic escarp If I can ask, is there a preferred set of tagging on IG for posts?

[cosmic escarp]

I just ask that you tag us! <@&978910826851352586>dalle 🙂 so I can re-gram

I am taking a little break from IG until Thursday (I need a little breather!) but will be back in action then

[smoky crag]

hello people

[wicked kiln]

Hello!

[smoky crag]

So, what's gonna happen to my server now that there's an official DALL-E server?

[wicked kiln]

Community servers are usually separate from official servers

They fulfill different needs

[smoky crag]

i see

[ember flax]

Hi Natalie! It is nice to meet you! I'm having fun with DALL-E 🥰

[chrome kestrel]

I wonder whether AI-artist will become a job-title at some point?

[wicked kiln]

what people look for when hiring artists might change

[warm flax]

you should absolutely keep it! esp because this server is limited to dall-e users, and non-users should still get to enjoy dall-e on discord

[chrome kestrel]

it’s easy to imagine DALL•E becoming a big part of concept art (for movies and games). such an efficient way to quickly generate ideas to iterate on

[smoky crag]

yes, i use DALL-E to generate concept art for the horror game im writing

[smoky crag]

i see

[ember flax]

How do you generate horror images without breaking TOS? Like, what kinds of words do you use?

[smoky crag]

images in the style of H.R Giger come out pretty horror-ish

[ember flax]

Thank you!

[smoky crag]

np

[wicked kiln]

im hitting an issue trying to generate a photograph

it's like once i pass it too many things it just gives up

I'm trying to generate two meerkats on a mountain looking at a lake and hot air balloons in the sky

im unable to do both lake and hot air balloon, when i do it it just gives random results (no meerkats in sight), but doing EITHER works

https://labs.openai.com/s/qwKBOacgwjbuRNy1t2m5pGrZ

https://labs.openai.com/s/BkyzqiXZWOCMAXvcpKN2JBfL

https://labs.openai.com/s/FqYzEsWA0h9ydF6ZUglQFb4G

https://labs.openai.com/s/r633FtmvMp7mkhzGAyA2WLRc ???

i tried adding "magical" to the front too that someone here suggested and no dice

[chrome kestrel]

have you tried using the edit feature ?

[wicked kiln]

im sure that can work but this prompt seems straight forward (imagen did it)

what about it is giving nonsense? i understand if it misses some of it but these results make no sense

[sterile bluff]

Howdy folks! If anyone has any recommendations for future #daily-themes please let me know!

[chrome kestrel]

• implausible objects
• chimeras / hybrid objects
• time travel

[wicked kiln]

underwater!

[torn anchor]

I recently converted a Dall E generated image into a Digital AR fashion wearable

https://www.instagram.com/reel/CeAW0roJHMC/

the above is the process of how i made that

this is the dress itselfhttps://www.snapchat.com/unlock/?type=SNAPCODE&uuid=969e0f8aff7d47d48081556d25f90ac1&metadata=01

generated image

[ivory garnet]

So cool!!

[gray imp]

• make inanimate object come alive
• cover things in chocolate
• Melt things (that usually don’t melt)
• Make object in the shape of the letter they start with (a balloon in the shape of B, or Padlock in the shape of a P)
• Animals dressed as knights

[chrome kestrel]

after using dalle I feel like I have an idea of what the future looks like. all creatives will become AI-directors, using AI models to generate creative output, the job of the creative will be to describe their vision.

I'm finding dalle more addictive than TikTok lol. The anticipation of what the next set of images is gonna look like has be completely hooked

[ivory garnet]

I mean, I've long said that I always find creative work just as (if not more) entertaining than video games, content consumption, etc... But with the near-instant gratification of DALL-E it's even more true!

[ivory garnet]

Is anyone experiencing this thing where you see pictures online and your brain first assume it's AI generated?

https://twitter.com/Titania2468/status/1530571437412651008? Like I saw this on my feed and processed it as a DALL-E thing before reading the text on the stamp and figuring out it's actually a genuine stamp

Titania (@Titania2468)

A Norwegian troll stamp was also recently launched.

Likes

253

Twitter · May 28, 2022, 3:27 PM

[chrome kestrel]

yeah I experienced this earlier. that’s when it hit me that DALL•E is gonna completely change the creative industry

[gray imp]

My kids find DALL-E much more entertaining than my VR headset. I have spent so much time generating images with my youngest (14 year old) and it takes a lot to impress him and keep his attention- but this “magical lamp” that DALL-E is, seems to make him feel like he can do anything. So and A.I. has for sure had a positive impact.

[olive shale]

I had it with photographs weirdly, didn't think other people would experience it too x)

[void glade]

It happens to me as well 🙂 recently I saw a photo of a greenhouse made of recycled stained glass windows from churches, and I thought "what a great idea, dalle users are really creative" https://www.ianvisits.co.uk/articles/wp-content/uploads/2019/03/stained-glass-greenhouse-05-900x675.jpg.webp

[wicked kiln]

I have this with all Reddit posts now…

[dusk pelican]

Anyone have any idea how to get art in the style of jojo's bizarre adventure? Is it just not familiar with it?

[sly vector]

Hey guys! Building a little community of creatives here in east london if anyone is local and wants to join forces please get in contact through here or on my instagram @sly vector or on [e-mail]

We're putting on workshops, discussions, and kickstaring collaborations to exhibit here in london and it's all very exciting !!!! ❤️

[flint mantle]

Maybe Jack Kirby? .. I tried "Comicbook Figure with black hat as Manga in color and realistic shaded by Jack Kirby"

[left pebble]

liminal spaces
fractals
shrimp
weird fine dining dishes/odd food

[wicked kiln]

Stickers!

[left pebble]

I really love the stickers youve been doing, they are adorable

[grand onyx]

this happens to me alllllll the time lol

[dusk pelican]

You guys blow me away with your creativity

[chrome kestrel]

https://tenor.com/view/excited-waiting-wait-suspense-anticipation-gif-10580950

waiting for DALL•E to render my prompt ↑

[ivory garnet]

Dall-e doesn't know Georgy Kurasov :^(

Would have been pretty cool to get stuff in his style

[formal coral]

Now everytime a see an image I always look at the bottom right for the watermark

[ivory garnet]

This one tripped me out too https://twitter.com/eh_stonehenge/status/1531168553990250496?s=21

Stonehenge (@EH_Stonehenge)

Our spellbinding homage to Her Majesty The #Queen! 👑
We've projected eight portraits of Queen Elizabeth II onto Stonehenge in celebration of the #PlatinumJubilee.

Likes

2340

Retweets

484

Twitter · May 30, 2022, 7:00 AM

[flint mantle]

I also realized, a few (in my eyes famous) artists seem missing.. Maybe we could create something like a library: how to recreate ArtistXY style by prompts 🙂

[west flame]

Yes, maybe a mixture of existing styles can output a specific unknown artist style

Or a mixture of features for the generated images

[ivory garnet]

cc @flint mantle @west flame I was able to do ~something~ by uploading some references in edit mode. The prompt was A geometrical painting of a tiger and clearly it's not quite there but I reckon finding the good prompt + seeding the style could get us somewhere =)

[verbal snow]

Inpainting aside, if DALL-E could just magically "understand" text commands to edit this image, what changes would you make? Original prompt was "imposter syndrome, digital art"

[severe citrus]

I am so happy, thank you for inviting me to try DALL-E!!!

[wicked kiln]

hello 🙂

[severe citrus]

Hi moxi!

[severe citrus]

Hello bakz 😄

[severe citrus]

I'll make a sticker as well!

[chrome kestrel]

you should add this to the #❌┃tips-n-tricks channel if you haven’t already. that’s a clever hack

[ivory garnet]

Good call

[flint mantle]

Iam curious: How much time a day you guys spent with Dall-E?
I just recognized for myself I was propably doing to much and taking a day off tomorrow. This Tool is just so mindblowing and Addictive.. 😅

[ivory garnet]

When I first got access last week I took two days off lol.

Right now I usually have a tab open and I put in ideas when I come up with them while working

[chrome kestrel]

i’ve spent an unhealthy number of hours hitting generate since I got access

[wicked kiln]

i had to buy an ergo mouse because i was using the computer too much and my wrist started hurting

[severe citrus]

I have been doing nothing but making stuff with DALL-E for the past 3 hours

[olive shale]

How many images have you downloaded? 😅 I'm at 1800

[severe citrus]

would a deep fried spider be against the content policy?

[chrome kestrel]

probably 400

[astral swan]

I literally did deep fried spiders yesterday too 😂

[void glade]

Have you tried printing any of the images you created? I'm curious to see how they would look on paper

[sterile bluff]

the native output 1024x1024 is a bit small -anything bigger than 4x4 may result in less overall quality.

[ivory garnet]

Yeah I was chatting about it with a friend, it'd definitely be cool if we could output something around 5000*5000 (which would be A1 at 300DPI)

[sterile bluff]

a lot of us in the broader AI art/generative art community have been using 3rd party upscalers with great results. happy to make some recommendations on options.

[granite merlin]

Dall-E outputs generally upscale very well too

[sterile bluff]

yeah, I've done a few for the IG.

[ivory garnet]

I suppose Topaz' Gigapixel might do a good job? I haven't tried it on Dall-E things yet

[sterile bluff]

yes - I've used it and it can do a good job. need to play with the (minimal) settings it has, especially the deblur.

[hexed moth]

a blogger's experiences with DALL-E: https://astralcodexten.substack.com/p/a-guide-to-asking-robots-to-design

[ivory garnet]

I disagree with your assessment, this is glorious not disastrous (great article tho!)

[hexed moth]

Someone else's article, but yeah, I agree with the gloriousness there 😄

[olive shale]

I have a hard time making it create a Mahjong game, wonder if there is anything obviously wrong with my prompt?

[sterile bluff]

interesting prompt! I tried one variation

[ivory garnet]

Not sure how to help but it clearly understands a bit, as it's very obviously drawing it in an asian-influenced art style 🤔

[sterile bluff]

surprisingly ended up with three unrelated images, one being a person not playing mahjongg. 😄

[olive shale]

Thank you! I see being more specific seems to help quite a lot

[sterile bluff]

yeah, I went super specific and it failed to keep the subjects at the table and the mah jongg part was lost in the details.

[olive shale]

I think the issue in some cases is you can see it understands to some extent what you wanted but doesn't really accurately produce the expected results, I feel like this occurs more often with subjects that were underrepresented in the training set.

[sterile bluff]

experimenting with different styles in this prompt is fun. I feel like you'd likely strike gold here, but it may take a number of variations.

[olive shale]

I see, seems like it's all about finding the correct "precision balance" for the prompt.
This reminds me of how — in my experience — for some subjects just adding "digital art" at the end of the prompt will output results that are too photorealistic and adding "digital painting" will help making it more artistic but for lesser known subjects adding "digital painting" seems like too much (makes output imprecise) and removing gives just the right result (and not too much photorealistic for these ones!)

[sterile bluff]

#6 on the one I just posted is so so close.

[olive shale]

Wow this is awesome! I wonder if mentioning Ghibli helped it getting closer to the theme of Asia 🤔 Thanks for testing it 😄

[sterile bluff]

@olive shale my pleasure. I've noticed in other text to image tools that "by X" leads to a stronger influence of that modifier's work in the result than "in the style of y". Worth experimenting with.

[wicked kiln]

https://twitter.com/giannis_daras/status/1531693093040230402?s=20&t=crC8XJzSG6PIORH02hnsgA

Giannis Daras (@giannis_daras)

DALLE-2 has a secret language.
"Apoploe vesrreaitais" means birds.
"Contarra ccetnxniams luryca tanniounons" means bugs or pests.

The prompt: "Apoploe vesrreaitais eating Contarra ccetnxniams luryca tanniounons" gives images of birds eating bugs.

A thread (1/n)🧵

Likes

761

Retweets

209

Twitter · May 31, 2022, 5:44 PM

[granite merlin]

This is insane

[wicked kiln]

https://giannisdaras.github.io/publications/Discovering_the_Secret_Language_of_Dalle.pdf

[ivory garnet]

That is absolutely mindblowing

Okay so I tried a few things of my own and I'm finding hard to reproduce any results (by generating stuff with 'subtitles' and then parsing the words back)

I am however indeed getting birds with Apoploe vesrreaitais

[severe citrus]

Dalle's a little slow today

[thick hearth]

Omg

[severe citrus]

https://labs.openai.com/s/PPoBnhN4gro6mAR3WIGbXzy4

[sterile bluff]

I was able to somewhat produce an translate a word.

pontcopointo >> bridge. I tested for context/coherency with "troll living under a bridge" and "troll crossing a bridge"

[sterile bluff]

but I can't get "two office workers talking about a bridge" to produce the same word.

[chrome kestrel]

I reckon they've trained it to never spell real words to eliminate the risk of spelling something offensive

[wicked kiln]

hmm https://mobile.twitter.com/benjamin_hilton/status/1531781606230437893 but pontcopointo = bridge works for me too, even when i add "digital art" or "3d render", it makes bridge like things. Hmmm "cat on pontcopointo" doesn't see to work. I wonder if your troll example it's just inferring the bridge? One thing that could be happening is these nonsense words are semantically similar to certain words but not quite equal, so they don't mean the exact same thing, and are only similar in specific contexts. I was playing this word guessing game today to learn about semantically similar words https://semantle.com/

Benjamin Hilton (@benjamin_hilton)

@giannis_daras I tried to recreate some of these results, and didn't quite manage to.

Maybe there's something to this though! Would be great to try to figure this out together - DM me :)

https://t.co/cP6m2qrjCj

Twitter · May 31, 2022, 11:36 PM

ok I am just putting in gibberish and hit gonropiono which seems to mean grasshopper? or maybe like "green small insect/reptile". i input it to google and no results. Then adding digital art i get cliffs

[sterile bluff]

I wondered if the troll was also too on the nose

[wicked kiln]

try gonropiono, i swear it has a very specific meaning. try adding "red" or "watercolor" where it seems to keep its meaning. but digital art seems to mess it up. even. "sketch" is sorta close...

[wicked kiln]

"pontcopointo, watercolor" gives watercolor of bridges

[sterile bluff]

All right, I found a freaky one.

Daughter and I were trying out different prompts like this. We were working on "adorable bunnies talking to each other, subtitled" and came up with image #1. We went through a few iterations. "Vertile bundly Bunthle" came up with image #2 . When I google lens some of these, I get.. "wild carrot"

[wicked kiln]

ooooooo, trying that edit: struggling https://labs.openai.com/s/ZSvm1Mb8dkocHWn2tRprwiBk

[sterile bluff]

literally how cats greet each other

[wicked kiln]

omg https://labs.openai.com/s/KW66NEpdbedUMFD8J4BPEfCK

[sterile bluff]

another food one. dalle thinks animals only talk about food to each other.

[severe citrus]

Someone said that really gross food is against TOS. Is this true?

[sly vector]

I'm in love with this

Anyone here from the UK?

[karmic pollen]

yes, though I'm now based in Germany

[chrome kestrel]

https://youtu.be/SVcsDDABEkM

[severe citrus]

Daily request limit? Seriously?

[left pebble]

Yeah I just got this as well. I'd not done many today I didn't think.

[severe citrus]

Well…. I guess I can go back to using midjourney

[vale pollen]

Same aaaa 😰

[sturdy eagle]

This is new. Daily limit. Also in Amsterdam the Netherlands: City of freedom, first gay marriage city and country in the world. And not just 50 marriages a day.... Amsterdam, city of freedom of creation: city of Rembrandt, city where the Van Gogh museum is... But Dalle? Limiting creativity? That will not resonate well here in Amsterdam... 😬 Probably ther is a very good reason...

[sterile bluff]

Hiya, we've discussed it briefly in #suggestions as well. There are a myriad of reasons this may being enacted now. Compute is very expensive, they may have specific limitations within their queueing and generation system and want to ensure a satisfactory experience as more users gain access.

And more. Let's hear from the team what the new standards/guidelines are, and ask questions after we've read them.

[outer anvil]

Sucks that despite us being told we would, we weren’t given any notice.

[severe citrus]

https://www.youtube.com/watch?v=_YMPXhAWB-4

When you forget you aren't using the DALL-E prompt menu and instead send your prompt to the engineers

[blazing gale]

Me... hitting my 50 daily limit for the first time! 😵‍💫

[swift void]

How much do I need to pay a month? I'll do it. TELL ME