openai-chatter

[rancid willow]

Same

[chrome kestrel]

also willing to pay as soon as the copyright issues are sorted out

[ruby jacinth]

Just curious — how much would you pay for unlimited access? 🙂

[granite merlin]

Honestly a lot

[rancid willow]

50$ per month

[glass kindle]

I pay $8.99 for artbreeder every month.
What I get: I can make my imgs private, upload and style my stuff, I can have HD downloads, google drive sync, fast creations, custom genes and animations.
As am artist, I am broke all the time. xD But that I can afford. But if you want us to work with you, dont forget that work is... well, you know? 😂 What I like about AI is accessibility. High prices are the opposite of that.

[wicked kiln]

https://tenor.com/view/money-shutup-take-gif-7953571

[granite merlin]

Without commercial use, I could see myself paying between $50-$75 monthly

With commercial use I could do $100+ easily

[sterile bluff]

NightCafe sells credits; an average creation takes around 4-6 credits to get a good result. (The cost is by choosing options like how many iterations, output size, and some settings which choose how the model generates)

At the higher bundles of credits they get down to 8 cents a credit. So a single image can cost upwards of 50 cents.

Now, NC also gives free credits daily so their pricing model covers that

So I'd say I'd easily pay 100 a month. And yes, I'd like commercial rights with that. :)

[swift void]

I would pay $20 a month. I love DALL-E so much

[glass kindle]

I also like "pay for a lifetime" systems.

[swift void]

same

Really I want to know how can I get more than 50 prompts in 24 hours? I love DALL-E so much and really want to keep sharing it with people

[ruby jacinth]

We're trying to work out ways to increase access. Limiting the number of prompts per day and charging for DALL-E are both ways to achieve it, so these are useful data points!

How many prompts per 24 hours would be useful? Would you prefer a limit per hour instead?

[glass kindle]

The 24h is okay, because I can only work with it for 2h a day. So I can get my stuff out then. A bit more would be nice, because the AI needs way more tries to get results.

also a prompt storage could be interesting.
"free DALL-E busy? Save your prompt and you will get a notification, if DALL-E had time for you."

Also... for try and error, I dont need 10 pics. Sometimes they are so similar, and I can already see: this is a fail. Maybe thats a thing to save?

[vale pollen]

I also would pay around this much I don't really have a lot of money to give around

Even something like 10 bucks I could do

you would have literally so many people using it if you have a good middle ground for a price point

[warm flax]

@vale pollen how many images would you expect for that?

[swift void]

I would say 50 per day is definitely a bit low imo, especially because most prompts requires different iteration to refine the result you're looking for. I would say 150 per day would be great.

For unlimited I would pay $20 a month

FYI we were streaming today. Trying to educate people on AI, DALL-E, etc and we hit 50 within the hour.

[sterile bluff]

I like the idea of queueing and getting results later based on "availability"

[swift void]

(non-commercial ^^^)

[sterile bluff]

will be especially good for folks who have refined a prompt and just want more of it

[swift void]

Honestly I'm not into waiting to get results

[sterile bluff]

@swift void that's understandable, too.

I admit my perspective is likely very different - I've been using these types of tools for 8 months nonstop across multiple platforms. I have probably about 10,000 creations between NightCafe, various colab notebooks, and MidJourney.

[swift void]

Just because a lot of the time I have/need immediate feedback

[sterile bluff]

so now I have a reasonable expectation of what I will see with many prompts, and my iteration cycle on ideas /speed to results I want is lower/faster accordingly.

[vale pollen]

Honestly I'm not sure. I also could see like 20 bucks a month as well

[chrome kestrel]

I’d pay $20/m for capped solo-user, $50/m uncapped solo-user. $99+/m once I have a team of people using it.

[swift void]

Agree that $20/month for individual user makes sense

[olive shale]

Was also the first price that came to my mind

[sterile bluff]

Part of the challenge and structural tension I see happening in many AI Art communities is that these tools are wanted and will be "game-changing" for artists

these artists and creators are not independently wealthy

and these tools are bleeding edge and run on very expensive hardware that is typically leased by the second

typically there is a higher cost for early adoption of bleeding edge tech

but if that becomes the case here then it may become a playground for those who are well off, already successful

[mossy rock]

It would be great if plans for students (or schools) are implemented, either with a lower rate (as with the Adobe suite) or with certain free creations daily, this would allow students / teachers of careers related to arts or design to generate interesting proposals and to improve the development of new tech.

(I'm a 20 yo graphic design student hehe so a high fee, at the moment, would be a bit tight. I guess there will be people in similar situations, I'm not saying that it's not charged, I just think it would be a very good thing from OpenAI to students and education in general)

[vale pollen]

good option too

like a 10 or 15 buck amount for students

[ruby jacinth]

I'd love to find a solution for this. Unfortunately, these AI models tend to become more expensive over time (as they get larger etc). One way might be to have commercial use cases subsidize artists.

[sterile bluff]

And only later when compute will be priced like a commodity will it be accessible at lower price points

[quartz herald]

Or just lower output resolution to 16x16 pixels and become an AI icon-generating engine 🤷‍♂️

[sterile bluff]

Yes. I'd suggest one three letter solution to raising funds but..

I don't want to increase the scope of this discussion

[swift void]

16x16 might be a stretch in imo. Not sure of the costs you guys currently have running everything but even lowering to 1080x1080 would fine from a consumer perspective

[sterile bluff]

not following; aren't the outputs 1024x1024?

[swift void]

Oh you're right. For some reason I thought they were 1200x1200. Don't ask why lol. That being said 720x720 would do too imo

I mainly see myself using images for concepting so as long as they aren't blurry I'm personally okay

[quartz herald]

Same here. I don't see the higher 1024x1024 necessary as they even tend to be a little garbled in areas at times, but crunching them down some seems to fix this. My 16x16 comment was a sarcastic stretch.... long day

[weary ore]

I think three points could help expand the usage of DALLE to more users without much limitation to most user cases:

1. Slight smaller initial generation size for the results with optional "upscaling" only of the desired outcomes (I don't know if that's how the DALLE architecture works, but if that's the case I think it can work, because most of us can tell if a result is something we'd like to use or not from the thumbnail size itself). I like the number of results that DALLE creates from a prompt and I think the variety is important on the creative process, but I feel a lot of the processing is wasted creating 'big' 1024px versions of things we end up not using.

2. Having fast/slow generation modes like MidJourney (e.g. you'd have a quota of XX images/day that are generated under 60s and past that point they take more time to be generated/are put a in a longer processing queue) that doesn't restrict IF you can generate new images but HOW LONG they will take to be generated. Maybe this slow mode can also be switched on/off by the user to kind 'save the fast credit' for images that they need now/are sure they will use.

3. Different plans based on how you are (or not) going to use the final outcome commercially (e.g. If you are going to use the final image 'as is' or with little modifications vs. using the images as refs/to change/repaint/compose several of them later). I agree with the game-changing point mentioned by @sterile bluff. I can really see this tool becoming not only a 'cool to have' but some kind of 'must have' for individuals in a lot of professional markets. A too-high/not flexible pricing plan may lead to a situation where the only ones that can pay for it are the ones that would not need the tool to stablish themselves professionally/financially.

That said I think paying something around $20 dollars a month would be the top-point for 'democratic' usage of the tool around the world.

[chrome kestrel]

wouldn’t be surprised if the compute costs are multiple dollars per hour per user

single A100 GPU instance on paperspace is $3/hr.

https://openai.com/blog/ai-and-compute/

[void glade]

As for the current limit, I'd much rather have 100 requests with 5 images each than 50x 10 images. As for the pricing, it would be great if you could combine the approach of Nightcafe and how google rents their cloud GPUs. So: high price for institutions for unlimited generations - they will be using it mainly on weekdays during daytime. And when there is lower demand / unused resources, like on Sundays, individual users could use dalle paying a lower price for each image generation. Also, Nightcafe rewards users with free credits for sharing their creations on social media or publishing them on nightcafe's gallery and getting upvotes- that would be a nice feature.

[ivory garnet]

Not sure I'm a huge fan of the last part, or at least not unless it's a fairly optional part of the process (ie you can get by without doing that).

[void glade]

You meant the sharing part? on nightcafe that is purely optional and it allows some users to earn tens/ hundreds of free credits

Anyway, I always assumed that Dalle would cost about $5 for an hour of use, since GPUs are so expensive + there has to be some profit margin for openai.

[severe citrus]

I'm now afraid to generate anything because of the usage limit, I'll bust out my wallet if I can

[outer anvil]

literally me, until I can see how many prompts I have left in my 24 hours I don't wanna generate anything

[deep obsidian]

Just wanted to say we're aware of this. Please bear with us 🥲

[hidden rover]

150 would cover 95% of my use cases, but...i work in spurts with days between sessions and could see burning through that easily in a marathon. I'd MUCH RATHER pay $100 a month for unlimited, and commercial rights similar to what midjourney does. This needs to happen fast as 50 is so low, I have instant "range anxiety", like early electric cars that only went 100 miles.

50 absolutely upsets the creative process. It's like saying you can only have 50 paint strokes.

[hidden rover]

$50 - $200 a month with all rights reserved

[quartz herald]

How much of a dip in usage has been seen today vs. 2 days ago, before the new limits?
What percentage of users are hitting their limit vs. fearful of doing so and staying way under their allotment?

[sturdy eagle]

Hi community! From those 50 requests a day, does requesting 'variations' also count as a request? Or are only new, original prompts counted as requests?

[icy ivy]

A variation is a request, for sure (you're just prompting with an image rather than a text string)

[quick plaza]

In the past we already wrote about this account. I just saw, that the account published a real human face 😐

https://www.instagram.com/p/CeT1hN1Pya6/?igshid=YmMyMTA2M2Y=

[granite merlin]

I wonder if they could take the page down for trademark infringement or something

[stoic pumice]

Getting that limit message is a jolt, but it's good to see how long I need to wait. Thanks for that!

[ivory garnet]

This is probably an even worse offence: https://www.instagram.com/p/CeWftjpPNDD/

Literally using a real person

[bitter solstice]

Could’ve used the style brush around the face though. That’s what it looks like at least. Still feels like it breaks policy rules.

[quick plaza]

Yeah it definitely breaks the policy. First of all everytime you upload a picture you accept to not upload strange people. Second, when I remember right it's okay to edit photosof yourself but it would break the rules to publish those pictures.

[severe citrus]

dalle does ascii art really well

[vale pollen]

man that's something I would of never tried

[severe citrus]

how do i use dalle text diffs?

[sterile bluff]

sorry, not sure if I know what that is?

[ebon juniper]

Out of curiosity who’s based in nyc? And would be interested in doing a meetup?

[quiet ridge]

Text diffs aren’t supported in labs

[quiet ridge]

Text diffs are a way of using text to nudge images in a certain direction. Check out https://twitter.com/model_mechanic/status/1517205808680869888

Aditya Ramesh (@model_mechanic)

Unmodernizing an iPhone with #dalle text diffs

Likes

1113

Retweets

159

▶ Play video

Twitter · Apr 21, 2022, 6:17 PM

[warm flax]

to clarify, this user had russell brown's consent to upload his photo and inpaint!

[warm flax]

sadly, instagram has not responded to our request to be verified -- and we can't claim that someone is impersonating @openaidalle unless it's verified 😦

if you could keep reporting it, that'd be great

[quick plaza]

Oh. Thanks for the information. I didn't knew it. 🙂

[severe citrus]

i hope they will be soon

[stark fractal]

Hi folks,

One technique to generate far higher quality prompts is to use GPT-3 to write the full prompt based on a small description.

For example, you feed it: "A detailed description of an award-winning painting of a train-station:"

And it gives you back something like:

"The painting is of an old train station that has been abandoned for many years. The building is in a state of disrepair, with the roof caved in and the windows broken. The paint is peeling and faded, and the station looks like it has been forgotten by time. Despite its dilapidated state, there is a feeling of beauty and nostalgia in the painting. The colors are muted and the light is soft, giving the painting a calming feeling. The artist has captured the feeling of a place that was once full of life and activity, but is now empty and quiet. The style of the painting is realistic, and the attention to detail gives the viewer a sense of the station's history."

The results tend to be WAY higher quality - based on a blind sample, people find them significantly more emotionally engaging and higher-quality.

For example, attached are examples for "A realistic painting of a dog sitting on the side of the road" and "A painting of an old homeless man" - try to guess which are with the basic prompt vs. GPT-3 enhanced prompt!

I've built a custom model for this and might make a simple chrome plugin if people are interested!

[gritty mirage]

oh heck yea, a chrome plugin for that would be fantastic

[weary ore]

That's clever, @stark fractal. I will for sure be exploring this in the future! Thanks for the tip

[vale vessel]

Hi!!!

@sterile bluff My man!!!!!!!!!!!!

[sterile bluff]

howdy howdy. hope all is well.

[vale vessel]

It is!!! Got some great images to use as concepts for my book. The images keep my motivated to keep writing haha

[gritty mirage]

yep, I'm doing the same thing. it's great for inspiration

[fiery moth]

Good evening.

[sterile bluff]

Hi! How's it going?

[fiery moth]

Going good. Excited that I got access to discord lol.

How about you?

[sterile bluff]

Great! Cleaning up after dinner, need to check on how my creations in colab notebooks are going. And see if I have a few more generations available in DALLE.

Welcome to the discord. It's been a little quiet as of late.

[true tulip]

Getting to know this girl pretty well bumping into the GIF button on generate

[ivory garnet]

lol, I think that's alba flores

[fiery moth]

Gm

[sterile bluff]

Howdy, creators! Hope everyone is having a great day, afternoon and evening.

[weary ore]

I have a very busy/creative day ahead, working with my buddy dalle. A great day indeed 😎

[ivory garnet]

⚠️ Potential scam
Beware of "@justinhackey" on Instagram (cc @knotty jetty). He seems to be messaging and calling people promising them access or trying to enroll them in his "community" yet he clearly doesn't have access to DALL-E himself as evidenced by the fact he will now proceed not to respond to this message :

He goes on about "making a community of dreamers", being asked by openAI to make a list of artists, and he recently called a guy I met on IG offering to give him access to DALL-E if he answered a form and then gave him Teamviewer access.

Fishy all around.

[granite merlin]

Does your friend make NFTs?

Could be the latest angle for scammers trying to drain wallets

[ivory garnet]

Not really a friend, just a student who dm'd me

(I do stand corrected on the access part - fair. I do maintain the accusations tho lol.)

[sly vector]

hey guys I hear the concerns so were happy to jump onto a call to discuss everything, trust me when I say we have nothing but pure good intentions that a all for this community, we as individual artists come from a bit more of a social and acedemic perspective working with eduction/artists/creative from all disciplines to help Dalle with education about open ai and Dalle

[ivory garnet]

Screenshots are in French, apologies for that.

[sly vector]

I am myself an award winning documentary film maker, tech nerd and a contributor to many charities and good causes and am more than happy to jump on a call with anyone at all to discuss our conversations with Dalle (which we have recorded)

https://meet.google.com/wzw-retp-hep?pli=1

[sly vector]

I really understand there is concern within this community and that is exactly why we are expanding it within our circles and networks within all different fields. I am discussing this and our plans with Dalle and what i'd love to do is connect with the discord community more, I'm just not personally that familiar with discord. Anyone who wants to join in with a conversation with us please do, we could really do with the support and all the advice you all have - Here we have created a very basic form that allows us to connect with all the artists and people interested in understanding Dalle2 and the agi tools of the future and how they can benefit us all, https://tally.so/r/w81ZYl

I'm getting more and more familiar with this platform whilst we find people to help us with it and other communication tools as we setup here in London and across the world, so far we are just a community of people using or interested in it enough to work endlessly in their own time to build a wonderful community and enable Dall-e to share this faster to artists but as carefully as possible with as my diversity in approach and communication. We work with artists, authors, architects, designers, make up artists, performers, cancer survivors, you name it.

All we want to do is create a loving community of creators and dreamers who can help use and understand this amazing platform. We will be around the discord more and more and working with people from the education system to help bring forward research tools to guide and capture the experience users are having with the system.

Really hope to speak with some of you in the future, love the work i'm seeing on here!!! ❤️

PS: To be clear, we are not looking to make NFTs with this!

[ivory garnet]

@sly vector Can you clarify why A) you asked that french guy to jump on a team viewer session B) what exactly do you mean by "we're in conversations with the people at dalle" and can you clarify your message that reads "We've been asked to [..] help find artists to sign up"

Because so far I don't think it's a stretch to say that asking people to share their desktop with you and stating you've been mandated by OpenAI to recruit artists without having any obvious ties to the company is shady AF territory.

[sly vector]

Hey @ivory garnet most definitely. It was someone in our office who we having a working relationship with who was going to guiding a brief session on team viewer doing a couple of renders to a specific and safe brief of a game world with penguin characters. The french guy was a student who was hoping to collaborate with me an artist and I loved to because he seems really passionate, considerate and motivate to work with this and help the community, a perfect use case/candidate for to present to Dalle

In the end it didn't even happen, i'm juggling a lot at the moment to get back to all the people we've had interested in jumping in on the discussion, which as a discord community you must understand.
As for the Dialogue with OpenAi / Dalle, they reach out to me me because of the connections I have and my interest and ideas in this space. I've worked with as i've worked with all sorts of companies such as The Lego Foundation, BFI film institute, the APA body, Shelter charity, my access to the film and TV world, musicians and the LGTBQIA+ community here in london and the wolrd at large. It's my connections to good people who steer incredible communities, charities and do good out there that I have been asked to present people to Dalle.

So that is what we're doing, it's a bit of sharing, a good amount of caring and a lot of documents and and screen recordings and online conversations to make sense of all the incredible information, projects and discoveries happening around us.

[ivory garnet]

Well, I don't mind beliving you but the way you present this is incredibly obscure and unclear and it's frankly hard not to imagine this is shady when you reach out randomly on instagram. Regarding using team viewer.. I mean frankly that's a terrible idea to given it's the #1 tool of scammers and other con artists.

[sly vector]

Anyone who is concerned, reach out 😄 we have a lot to discuss and nothing to hide, find me on here, or my instagram justinhackney or sign up for the online discussions: https://tally.so/r/w81ZYl

again, this is someone who I work with, I would never give anyone access to my personal machine....

If there are concerns about anything at all we urge you to tell us, to help us, we could really do with the support of this incredible discord community

[icy ivy]

um

👉 👉 👉 WE NEED YOU: TO PROMPT! 👈 👈 👈

Hello! It's been mentioned multiple times that it would be helpful to have a list/resource of powerful style prompts to create amazing DALL-E images!

Here is a spreadsheet where YOU can add in examples of DALL-E outputs for hundreds of different styles. Now we all have a limit on our generations, I'm hoping everyone can spare a few prompts a day to fill up this mighty grid. Many hands make light work 💡

https://docs.google.com/spreadsheets/d/1Q_xUiBNyh7GFykfCIwlIkZa2m2nYnFq9HhmK95wniiI/edit?usp=sharing

I'll publish all the brilliant results in a week or two! Please share with any other DALL-E users, but not too widely with the Public (the more people access this editable doc, the more chances someone does something stupid with it.)

[wicked kiln]

some people did start one here https://docs.google.com/document/d/11WlzjBT0xRpQhP9tFMtxzd0q6ANIdHPUBkMV-YB043U (ive been helping with this one)

[icy ivy]

yeah, it's based on that and builds on the prompts they've used!

i'm just about to email them about it actually

[wicked kiln]

we have a discord where we're working on it (where im generating everything)

would you want to join?

[edgy bluff]

Could i? 😄

[wicked kiln]

yeah!

[icy ivy]

happy to! i have heard people mention "a discord" (i assume, another one that isn't this one)

[wicked kiln]

there's also a prompt engineering discord for AI in general

that one's more public https://discord.gg/JMmumFpx

[weary ore]

I'd be interested to join as well if you guys need any additional help.

[icy ivy]

sweet, super useful!

[wicked kiln]

hope you like cats

[pliant shard]

So, I walked into a poll today, that hurt

[sterile bluff]

Ow. Did it ask your opinion on how it felt to walk into it?

[pliant shard]

Haha, I meant pole.

My head still hurts though.

[sterile bluff]

I'm really sorry to hear that - head injuries, even minor can be problematic. If you feel any dizziness, headaches, blurred vision in the next day or two

get yourself checked by a doctor.

[void glade]

Thank you for that tip! Yesterday I tried OpenAi's davinci model to generate some descriptions, some were quite amusing 😊

Detailed description of a poster warning about dangers of eating food that is too healthy.

This poster is warning about the dangers of eating food that is too healthy. The poster has a picture of a healthy looking plate of food with a big red X over it. The text on the poster reads:

"Beware of food that is too healthy!

Eating too much healthy food can be just as dangerous as eating junk food. Overdoing it on healthy food can lead to weight gain, nutrient deficiencies, and other health problems.

So, be sure to eat a variety of healthy foods in moderation and don't forget to enjoy the occasional treat!"

Detailed description of a poster warning about dangers of time travel.

This poster features a large image of a grandfather clock with its hands stopped at 10:10. Above the clock, in large, bold letters, it reads "BEWARE OF TIME TRAVEL!" In smaller letters below, it reads "The dangers of time travel are real. Don't let yourself be caught in a time warp!"

I wish Dalle could write text, I would generate that poster about time travel

For anyone interested, you can access it on OpenAi's playground

https://beta.openai.com/playground

[little cave]

@icy ivy @wicked kiln and all, - that's amazing lists.
I am actually to begin write a series about Prompt Design of DALL-E with examples and observations.
May I link your lists as sources and use some of the prompt modifiers there?
Here is my index, it will be ongoing series, part one will come this week: https://medium.com/merzazine/prompt-design-for-dall-e-2-series-5e50cec59d08

[grand onyx]

Has anyone tried generating imagery from a story or novel?

I was reading lore on a character I liked and just had the thought of putting some of the visual imagery into a prompt

[icy ivy]

I have - I didn't really get anywhere to be honest! I tried a few different things - a lot of attempts with Cormac McCarthy's The Road which has quite moody and evocative descriptions, but nothing special. I wonder if there's a framing prompt I haven't found which encourages more creative 'riffs' on the language, maybe "A painting that illustrates the scene:" or something.

[grand onyx]

I'm trying now, but Im starting it of with "Visual imagery of this scene:"

[icy ivy]

haha snap

[grand onyx]

then i put the text in quotes

[icy ivy]

let us know what you find! have been similarly underwhelmed when i tried it with lyrics... they seem to trigger image text a lot for some reason!

[grand onyx]

I have one here

https://labs.openai.com/s/NZ5dbtLKSKfjCNKOPEslTBod

This was taken from Jhin the Virtuoso's lore from league of legends

it didnt show shen, but focused on the meal

[little cave]

Imho, Quote => Image won't deliver the result we would love to see, since an illustration needs more context than what's in. I wrote about DALL-e for illustrations with examples: https://medium.com/merzazine/a-perfect-illustration-with-dall-e-550b9d941148

[grand onyx]

Hm, ok ill check it out

Hm so adding an artist to it helps?

Like say i did that prompt over again, and added "by David finch" It might be closer to what im trying to achieve 🤔

No, There's gotta be a way to do it better tho

[grand onyx]

How exactly do you know when you have to wait for your next prompt if you haven't did 50 yet

[little cave]

For this, I'm writing the series of Prompt Modifiers to spar everybody time with trail&error 🙂

[sly vector]

dude this is incredible

[stark fractal]

Made this little diagram to communicate the idea that tools have mostly had a marginal impact on creative output up until now - but AI/GPT-3/4/Dall-E will change this and balance the inspiration vs. perspiration equation in a better direction 😄

[cinder pebble]

Would you buy this? https://labs.openai.com/s/AJwDleaC31gOnmLi9AzQzpmu

[sterile bluff]

buy it? I'd buy three

to race them

[left pebble]

They look like they should have a little step in the back for the pusher to step up and it become auto-propelled instead of push driven.

We just need to get the prototype into the hands of http://powerracingseries.org/

[fallow swan]

Anyone else having trouble logging in right now?

[pliant shard]

Yeah, I think it's down rn

[left pebble]

Ah shucks, yeah I'm having same issue, I thought it was an issue with my VPN initially.

[fallow swan]

Obviously, this is a testing environment, so they’re not going to put as many resources on the developing the“web app.” I have encountered several Javascript bugs, especially on iOS.

[turbid hatch]

Not sure if this belongs here, but I’m curious if anyone else has had any luck coaxing animations out of the program? I’ve had some interesting results by using ‘muybridge style chronophotograph’ prompts and things like “temporally sequential side by side frames of a clock winding down an hour minute by minute” and then slicing up the frames (obviously buggy and lower quality) and animating them to see if the motion is correct. I also, since the smaller images become buggy, have tried things like “two side by side temporally/chronologically sequential images of a few seconds of a hippo yawning” and then taking those resulting frames and used colab/FILM (frame interpolation for large motion I believe) and made animations using interpolation between the two higher quality generated frames. Sorry that was a mouthful, just wondering if others have had any luck with this!

Btw muybridge was an odd character (accused of murder actually hah) and used a lot of nude models for his work, so make sure to specify what you want clearly so as to avoid anything unscrupulous

If this is too long of a post…I’ll Dalle-te it

[karmic pollen]

Really interesting stuff. I've yet to delve into attempting animation but have a few ideas in mind to try.
Have you posted your anims anywhere?

[turbid hatch]

I can post some on here in a bit if that’s allowed? I have both clips using only the frames generated stitched together and some with two frames interpolated (so some ai generated frames sandwiched between two frames generated from Dalle)

[karmic pollen]

Cool, I'd be interested to see

[turbid hatch]

I’ll upload a couple here today when I have a sec!

Gonna make one with this, pretty sure it animated that Naruto run thing haha

[grand onyx]

Oh man, ive been wondering if dalle can generate frames

I have animated before and it would be interesting if it could

[fallow swan]

Looks like we’re going down to 6 images per prompt tomorrow:

[grand onyx]

Seen that, along with the 50 right? I think people generally like 5-6 out of the ten usually

[sterile bluff]

perhaps I am picky, or at a different part of the journey but I rarely download more than three.

[granite merlin]

Super interesting to see how the rollout of DALL-E 2 has introduced DALL-E mini to a large mainstream audience over the last couple of weeks

[lilac pond]

it can but it has its own language

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

[turbid hatch]

Haven’t forgotten, just working on it now! Should have something in the next hour or so worth watching. Have a few, but not impressed enough with the results so I’m tweaking some things.

[karmic pollen]

I fooled around with typing complete gibberish last night, things like "eoc638-_)64uegxk" etc.
Results were very mixed but had a lot of photorealistic(ish) humans

[void glade]

If you are interested how many dalle images have been created so far, you can take a look at exif metadata. Today's image had this info "OpenAI--ImageGeneration--6739795 " . I checked images from earlier dates, some examples : DALL·E 2022-05-24 OpenAI--ImageGeneration--3836709 , DALL·E 2022-05-19 OpenAI--ImageGeneration--2783902 . The number of total generated images more than doubled in recent 3 weeks, from 2.7 million to 6.7 million

[icy ivy]

= 4 mil gens in 21 days = about 190,000 'image generations' a day
= at approx 3,000 (?) users = avg. 63 generations/user/day

[fallow swan]

This explains the rate limiting. I wonder what the compute cost per image is?

[next badger]

I've been seeing a lot of mini dalle posts, slightly concerning - lots of political and gun stuff 😟

[next badger]

I was messing with tompolos and taamtes = tomato-ish 😆

[granite merlin]

Yeahhh I’ve noticed that too… I think the reality is that a D2 equivalent will eventually be Open Sourced (as people are already currently working on it), and it’ll be interesting to see the impact of a publicly accessible model without as many safety restrictions

It’ll no doubt be a shitshow

[next badger]

yeah, I rolled my eyes so hard at my twitter feed yesterday I practically did a backflip. "This is why we can't have nice things"

[karmic pollen]

Arg. Same.
I don't like censorship but also don't want to see every notable person from human history doing disgusting violent things

[karmic pollen]

any nice results?

[next badger]

Quite tomato-ey, a few serving suggestions, one random street scene. I ended up making cartography of 'Taamtes' 100% would check out airbnb's in the area 😁

[timid night]

site down for anyone else?

[twin silo]

We're looking into it!

[little cave]

Thank you! I'm writing an essay about Prompt Design right now and it was like "fullbrake!" 😄

[twin silo]

We should be back now – still monitoring the situation.

[fallow swan]

For sure. I liken it to 3D printing of guns. Regardless of your opinion about guns (hate 'em), the reality is that someday, people would have that power. Another example is being able to write illegal images to a blockchain. I think that's why Dalle wants to try and get ahead of the problem. There will be a huge premium to folks who can accurately describe the origin of an image.

[twilit hemlock]

i just ran a 50 prompt session and had no problems on my end

[dire scroll]

When you finally find “the” prompt on your 50th image. Maddening.

[timid night]

whoa... my entire internet went down about 10 or 30 seconds after i posted here... wow haha

thanks, everyone.

[grand onyx]

Wow! Super interesting

[obtuse maple]

Dall-e just blocked the word soviet. Interesting.

[weary ore]

@obtuse maple yeah, joanne and I had a talk about that in the #community-help channel (?), she told me that there was a lot of people using the word soviet to do political propaganda so they had to block it although there are a lot of safe uses for that. Sad but understandable 🥲

[thick hearth]

love the images on the maintenance page lol 👀

[timid night]

Maybe slow down generation time a bit use Curie GPT-3 to detect context? That might be even more taxing on the servers, though...

[dire scroll]

I wonder…if the language changes for each generation or if it’s consistent? Like if I input that phrase, does it give me similar images?

[obtuse maple]

That makes total sense. I was trying to pull soviet era fashion inspiration. Lots of "one-size" fits all jumpsuits from that era. Bauhaus worked instead though.

[lilac pond]

it probably changes with each major iteration of the network

[fallow swan]

I was worried about the reduction to six renders per query, but after using it, I actually like it more. I typically have enough diversity in results, and I can actually evolve faster. Of course, there certainly are good reasons to prefer ten renders, but six is working fine for me.

[glacial kindle]

using Dall-e to depict scenes in a dungeons and dragons campaign

imagine real time speech to text to image using AI engines in serial

[sterile bluff]

especially if you can "pretrain" to have specific assets in scenes - like PC character designs, specific landmarks, etc in your campaign

[wicked kiln]

If you can train it to output the same character it’d be good for making comics

[fair gull]

haha i just got this reply thats kinda scary they think its human generated

[glacial kindle]

what was the original pic

[fair gull]

https://twitter.com/Dalle2Pics/status/1534718838377529344?s=20&t=JKmykHW7hooyOVL6Es1nVg

Dalle2 Pics (@Dalle2Pics)

Homer Simpson themed Ai generated images made with DALL·E 2 #dalle #dalle2
Thread (1/?) 🧵

"A still of Homer Simpson in Psycho (1960)"

Likes

148

Twitter · Jun 9, 2022, 2:07 AM

something in this thread

i think he was saying the thread in general

[glacial kindle]

thats amazing

[fallow swan]

Oops. Just kicked back “Grateful Dead.” Allowed “grateful band,” so I’m guessing “dead” is on the disallow list.

[obtuse maple]

So Dalle almost understands text & has an inherent understanding of typographic style. So my next challenge…Get Dall-e to design a typeface.

[obtuse maple]

Ordered some prompt “research” material.

[weary ore]

Yeah, that's what I suggested, a post-generation checking. But I agree with you, certainly it will impact gen time and, more relevant, generation cost, which I guess should be already pretty high for what I understand from diffusion models (specially on the speed OpenAI is doing with DALL-E)

[weary ore]

Yeah, in my case it was fashion-related stuff as well. Also was able to achieve the same results with less specific style and more descriptive lengthy prompt.

[weary ore]

I saw someone discussing sometime ago in one of the diffusion image discord severs I'm in that AI art will not only NOT make traditional art obsolete, but will also make us 'ai-artists' willing to know more and more about art history, styles and techniques than most of the traditional artists did in the past and do right now. Well, the generalization and extrapolation is debatable, but I certainly am learning a lot more than I did about art history and techniques (and that coming from someone who is graduated in a art-oriented field). Makes you wonder about the future...

[tidal egret]

Yes. I had the same inconvenience with the word soviet. I was very excited to search for "astronaut cats from the Soviet era" haha. A lot of my art is based on that aesthetic, my grandparents They came from there and I aesthetically love everything retro. I got interesting results using "slavic". It only works with everyday things. Unfortunately not with astronaut cats.

[weary ore]

Oh, thanks for the tip! Interesting, will certainly give it a try in 'slavic'. Maybe combined with date/period can give really assertive results. 👍

[tidal egret]

yes, it is difficult to find results related to eastern europe.

[weary ore]

One thing that may help (I haven't tried in DALL-E specifically, but in other models it tends to work well) is to use other languages in the prompt. Maybe the Cyrillic alphabet may help to get that kind of aesthetic, I think, in general, that it should be used more in images that were generated in this region/culture than English and that may reflect on the dataset used to train DALL-E... But again, didn't tested other languages in prompts yet, only speculating for now.

[tidal egret]

I always search in English. But I tried to do it in Romanian and Spanish and it did quite well.

[little cave]

Same here - I'd love to get more Soviet architecture and Eastern Europe style using DALL-E.
You can use a Russian imitation of DALL-E, called ruDALL-E (which is already a lie since it is not transformer driven). The only relevant part is: that ruDALLe is trained on the Russian/Soviet/EasterEuropean image dataset. And it works pretty well for this specific particular approach. Besides of it, ruDALLe is not creative.

[obtuse maple]

You can look up specific soviet poster artists like Nikolai Prusakov, Anatoly Belsky, Mikhail Dlugach, Smolyakovsky or the Stenberg brothers

The Stenberg brothers are my favorite. They were film poster artists. They'd project snippets from a film onto their lithography medium & then trace the projections. They used all kinds of contraptions to distort the image & get interesting visual effects. Their turn-around time was like...8h. Most of their posters have nothing to do with the films, but that's fine, since 60% were foreign films re-edited into oblivion by the "propaganda & culture" branch of the Communist Party, and the rest were cookie-cutter pro-work movies.

[tidal egret]

oh you are a genius. Thanks for the recommendation 🌼

[obtuse maple]

No problem.

Some of my favorite Stenberg brothers works. They don't make movie posters like they used to...sigh...

[obtuse maple]

When I have more prompts I'm going to run "Black cat walking, Chronophotography, phenakistiscope, storyboard, walk-cycle" and then run it through RIFE frame-interpellator AI to animate the walk cycle. I think it could be a fun experiment. I haven't seen anyone try it yet.

[pliant shard]

Anyone else getting slow gens? It's okay if so but I gotta check.

[cyan marten]

Very open question to start some discussion: what do you think DALL-E’s purpose will be in the artists toolkit ?

[pliant shard]

Currently my friend and I are collaging!

It's very fun to experiment with.

[manic dragon]

heavy breathing

[turbid hatch]

I have been doing exactly that with interesting results, I’ll share some here in a sec, been tweaking it for days to get it right

[turbid hatch]

Posted about it a little while back in here but then life and my job got in the way and I dropped the ball on gettin the results up here, I’ll send some stills now

[pliant shard]

mood

[turbid hatch]

Here’s some of cats, they’re odd, but i animated some with interesting results

Can I post video clips on here? Or do I need to link it somewhere, wanna show the animations dalle made

Made one from these animated frames https://drive.google.com/file/d/13QxJmwzhilhTg--HyNW4yZp8vWBHPQTh/view?usp=drivesdk

Not my best work haha, but I’m impressed with the motion it was able to create. I have a very impressive one I’m almost done with and will post in a bit (unless this isn’t interesting to anyone haha)

[obtuse maple]

You need to make the link public or else we can't access it.

[turbid hatch]

Fixed! Thanks!

I have huge areas of technical ignorance so I apologize!

[manic dragon]

Has anyone tried typing in "Dall-E 2" or something similar as a prompt?

[pliant shard]

from what I've been told, Dall•E 2 has no sense of self.

[cyan marten]

I have asked it to show me what it wants me to see & get some uncanny valley type stuff.

[timid night]

we did some interesting draw coke OR draw pepsi experiments... with brands it would choose one as dominant, but with winter or summer it would mesh them together

it would draw burger king over mcdonalds and pepsi over coke

[cyan marten]

A question for the photographers: How are you liking DALL-E so far? Has it changed how you approach a certain type of film, shooting style or camera? I am having excellent results with soviet-era film cameras and even using the phrase "Expired "X" film ".

[timid night]

"Quantum photograph" was interesting... and 1000000mm camera lens...

[cyan marten]

Impossible / improbable lenses... didn't even occur to me to try that!!

[timid night]

😉

i need to try Crayola film haha

[manic dragon]

yes of course, I just thought it might be fun to try

[manic dragon]

would love to see your results from that experiment

[pliant shard]

Couldn't hurt, it's why we're here after all.

[manic dragon]

if dall e 2 scrapes the internet for content, if you typed in "dall-e 2" as a prompt, i'm guessing it would create something similar to whatever the most common dall-e2 outputs are. But I don't know if the model gets updated that frequently

because dall e 2 is so new it might not have any dall e 2 results in its dataset

[timid night]

maybe something like "DALLE-2 is an advanced robot that is an expert painter. It creates a self-portrait..."

even then it wouldn't BE DALLE-2's representation of itself, but it would likely be a droid drawing a self portrait?

trying to get an owl doing a self-portrait haha... robot worked better... and old, experienced artist does a self-portrait

[old prawn]

Yeah I’ve noticed this too

[old prawn]

I think it’s also confusing people as I have some friends sending me DALLE mini stuff they’re seeing on twitter. Thinking it’s the same thing I have access too. People are not realizing how powerful the current gen of tech is

[pliant shard]

While I think it's neat I also think it's very dangerous.

I mean, I love love love open source.

[old prawn]

Because DALLE mini is so low res people see it as harmless. But they’re building habits that could eventually be destructive when DALLE 2/3/4 equivalents are open sourced

[pliant shard]

I just don't think lot of people understand what kind of issues that this can cause.

[pliant shard]

I think the best counter measure at the moment would be to make a detector ... but that can be very hard considering how life like Dalle 2 is.

[old prawn]

I think better public education and awareness about the issues it can cause

[pliant shard]

That's good.

[old prawn]

I’m surprised there it’s some sort of subpixel signature for every ai generated image

[old prawn]

so a second ai can detect if it’s real or fake

[old prawn]

it will surely become a cat and mouse game as rogue ai generators try to evade detection

[old prawn]

maybe official images will need to be verifiable on a blockchain

[pliant shard]

While the saying, "don't believe everything you see on the internet" is a thing, people sure as hell don't listen.

[old prawn]

haha yeah. another interesting aspect to this is when photoshop first started getting really good at photo editing back in the day. people had a lot of concerns about it

[old prawn]

but it hasn’t changed too much of the way we deal with images

[pliant shard]

Why do you keep replying to yourself?

[old prawn]

I’m new to discord my bad lmao

[pliant shard]

Not trying to be rude, just wondering

Ah, fair enough

[old prawn]

I’m all for open source but maybe when it’s a super advance technology that when in the wrong can course significant damage it’s better to keep under wraps

Like open sourcing how to build a nuclear weapon

I heard back from OpenAI and they said they’re working on reaching out to DALLE mini to get them to change their name

[spiral haven]

This has been the debate for quite some time with LLM

honestly, I don't think you should limit the usage/accessibility

On the contrary, as this will increase the actors in the game and not limit it to just a few select players

[pliant shard]

people will find away anyways...

[spiral haven]

(relevant: https://blog.eleuther.ai/why-release-a-large-language-model/)

[turbid hatch]

I think the ethical debates on all of this probably are, at least couched/framed, largely by the unmapped litigious water a company providing a platform for this type of technology will have to navigate

[spiral haven]

Also, while I don't always see eye to eye with Stella, I agree with her points here: https://twitter.com/BlancheMinerva/status/1533136400392110081?s=20&t=-ECJaSeXeo6MpOom_UaPWg

Stella Rose Biderman (@BlancheMinerva)

Two years ago, a handful of nerds taught themselves what they needed to know to train a model at this scale in their free time in six months. The only reason it took another year to release a 6B param LLM was that they didn’t have 100k sitting around to spend on it.

Likes

229

Twitter · Jun 4, 2022, 5:19 PM

[turbid hatch]

The externalities are almost unknowable with any novel or brand new tech, so I imagine slow rollouts and strict rules are effective ways to make the problem of getting a rough idea of the direction to aim a companies limited resources (in mitigation of those externalities) a manageable one.

[manic dragon]

I think that trying to regulate a technology like this is absolutely futile. I appreciate OpenAI's efforts, but its kind of like music companies trying to regulate distribution of their IP on the internet.

I think that OpenAI's approach in the short term has been wise so that they can anticipate potential issues before it's issued to the public

but they're definitely on a time constraint, because a lot of similar engines are already open source and available to the public

[turbid hatch]

I can’t imagine the pressure they must all be under. It’s sort of the Wild West for AI right now

[pliant shard]

Hey, I'm gonna be real, I think I'm having a panic attack so I'll be right back.

[turbid hatch]

Breathe through your nose slowly

And know it will pass

It always passes okay?

[manic dragon]

we're here for you

[turbid hatch]

I’m so sorry

[turbid hatch]

1000%

[turbid hatch]

I had panic attacks daily for years up into my 20’s, and, I can’t ethically give medical advice, but I can tell you that you will make it through and go see a doctor/therapist about them, you don’t have to suffer alone.

[manic dragon]

not sure if your panic attack has anything to do with this discussion (it probably doesnt), but as another person said above, people were concerned about the power of photoshop when it first came out. We made it through that just fine. And we'll make it through the implications of OpenAI's tech just fine too. In fact, OpenAI has been significantly more considerate in their approach, probably because they're a non-profit so they don't have to launch everything as fast as possible etc. etc.

[timid night]

the #dallemini vs #dalle2 concerns me as people don't know it's NOT the same company

the power of the printing press / typewriter / digital publishing / internet ...

[manic dragon]

the MO of the average tech company is just "move fast and break things. launch as fast as possible, consequences be damned!" and it's refreshing to see that OpenAI doesn't share that attitude.

[turbid hatch]

People were up in arms when the written word first came out, because they were worried about people not having to remember anything if they could just write it down. We are an adaptable species

[timid night]

oh no just read this. hope you're okay...

[timid night]

yea i do not want Google or Facebook to win this race...

AI is the new space race in many ways... (while a new-new space race is heating up heh)

[timid night]

yeah, the power of the printing press / typewriter / digital publishing / internet ... AI is "agriculture level" change, tho... we've been promised it so long by media/scientists, most don't believe it's right around the corner (not even geologically-speaking!)

[manic dragon]

I predict that we will see more creative output in the next few years than we have in the last 100.

just look at your twitter feed

[timid night]

yea i've been saying we've got at least 5 to 10 years of hybrid human + AI art... who knows tho ... things are accelerating...

i mean, even with GPT-3 (a super typewriter), I've been so much more productive and would find it HARD to go back

interesting time to be alive! : )

[cyan marten]

Very scary and very interesting time to be alive. I was pondering that in 100 years there may be an AI reading these conversations or even thinking about people who pioneered this sort of tech.

[weary ore]

and maybe 200 years from now they may be reconstructing us from this conversations to solve some doubts about what we really mean when we said that... haha

[timid night]

https://www.amazon.com/Girl-All-Text-Cyberpunk-Novel-ebook/dp/B09X1MFXXD

(no spoilers!) small smile.

used deep dream generator for the cover tho it woulda been cool to use DALLE-2 with GPT-3 writing about GPT-X when it becomes sentient...

[cyan marten]

Ahh! My mind was already broken enough with the first thought project 😅

[timid night]

maybe THIS is THAT simulation haha

42 42 42!

[weary ore]

ahhahahah sorry, couldn't help. I'm writing an album about that scenario

[weary ore]

Makes you think!

hahah

[timid night]

"a man ponders whether he's just a prompt engineered by someone out there," thought-provoking digital art

[weary ore]

ahahahha if I had finished my assignment for today would certainly spend 1 gen on that

[timid night]

i've got 20 or so minutes left before i can haha

@little cave would like that prompt i think.

[cyan marten]

This may be answered elsewhere already but I can’t find it - do generations stack ? Or does the 50 reset once 24 hours pass ?

[weary ore]

the easiest way to think is that each generation 'ticket' only respawn after 24 hours. So you have 50 tickets to spend, and each of them will be available again only after 24h

[cyan marten]

Thank you. I may have to pause a day then so I can start generating at a decent time

I have been made to wait until 8 at night until I can start again lol

[weary ore]

yeah, I feel you. I have some rules now for myself on when to start/stop

[cyan marten]

I may have to start implementing some for myself. What are you currently following ?

[weary ore]

2 times a day, one in the evening (for creative work) and one (for fun) at the noon.

[timid night]

there's a timer that shows how many hours left on the pop-up today

[little cave]

OK people, I wrote my first part of DALL-E Prompt Essay Series. Enjoy: https://medium.com/merzazine/prompt-design-for-dall-e-photorealism-emulating-reality-6f478df6f186
Will be updated. Oh yes it will.

[old prawn]

i think another reason why these open source AI's might be more likely to go viral is because they allow for copyrighted material unlike more ethical systems

i.e. for DALLE mini

it gives something people can already connect with but in a new context

and the lack of rules can get into some weird places like I saw this on ig

i've also seen some more sexual stuff on Twitter with dalle mini so it doesn't seem like they're filtering that stuff out there

[old prawn]

i think these kinds of examples (DALLE vs. DALLE mini) are important for more regulation, because if we allow too many ai systems to be open source it can wreck havoc, so if there were an open source project for an AI that can do social engineering, so that anyone in the world could hack into any system even for well intentioned people it could get really bad really quick

so i think we should do what we can to have a better framework

i work in music too and the industry saw the issues coming up and adapted. and prevented a world where musicians aren't paid

we're at a similar stage with ai. we have two choices either do nothing or at least try to create a world where the framework around ai will be as ethical as possible

there were no internet laws in the 1970s now there are many, the legal system needs to get ahead of the curve with this ai thing because it is an existential threat, and we're just not socially there as a civilization to be able to handle the consequences

[fallow swan]

Great article. 👍 Thank you. Are you confident commas make a difference?

[lilac pond]

" black text written in english above a red line describing a drawing located below the red line "

rip

[old prawn]

this is really great!

[sterile bluff]

@sacred jasper Hey, welcome! Let me know if you would like anything specific to happen with #daily-theme . I've been posting theme ideas on a mostly daily basis. 🙂

[lilac pond]

are 'marijuana plants' against the content policy?

i just got a content policy warning for trying to generate images of godzilla watering weed.

[gusty timber]

yes

actually

it says "drug use" is not allowed

but in a lot of places growing weed is illegal so I'm assuming it's definitely against content policy

[pliant shard]

A brain reset and I'm good for anything!

[lilac pond]

watering plants is not "drug use"

especially artwork of watering plants

and the word marijuana is analogous to the word hemp

so no, it is not against content policy nor state laws to depict godzilla watering hemp plants.

hemp and marijuana were traditionally used in japan for thousands of years to make rope.

then i got a content policy warning for a depiction of christ being crucified in a garbage yard.

[pliant shard]

I'm good now, sometimes I get overwhelmed. In that case I take a nap.

[pliant shard]

You're cool!

[lilac pond]

now i've got warning for a girl riding a cat through space

[pliant shard]

Thank you~

[pliant shard]

Me thinks that's a bug.

[lilac pond]

while the AI itself is incredible the underlying filters and tooling are less than stellar

[edgy bluff]

Damn. I’ve only gotten warned 2 times with like 50000 images made. You gotta be really unlucky :/

[lilac pond]

it's not a huge deal. if i did get suspended i could just contact support.

but the filtering seems extremely sensitive as of today.

[edgy bluff]

Might be with the storm of mini-dalle stuff

[lilac pond]

waifu catgirls - ok !
godzilla watering hemp - no way !

yeah people are generating some awful stuff with mini-dalle

[edgy bluff]

Don’t think waifu catgirl is against anything in the rules though

[lilac pond]

if you generate enough catgirls you will eventually end up with ones who have no pantsu.

[edgy bluff]

If you try to Do it then Yeah. But otherwise i don’t see that happening

[lilac pond]

no, even something like "anime catgirl hanging upside down from monkey bars eating a hamburger" can generate hentai imagery

[edgy bluff]

I haven’t gotten anything remotely close to it

[lilac pond]

its not over-the-line explicit, but i got a monkey hole instead of a hamburger

i am guessing that they inadvertently included censored hentai in the training data that wasn't picked up by a filter

[edgy bluff]

Idk. I don’t see it happening. But ey even If you get an image that doens’t follow the rules. You can always Ignore it and generate another one instead 👌🏻

[warm flax]

oh no 😿

[lilac pond]

i have the image if you want to use it to confirm/diagnose

i did report it but i have no idea where those reports go

[warm flax]

ooh i can pull it up in that case, but if you could DM me it that’d save me a lot of time 🙏🏼

[pliant shard]

What... ????

[torn hill]

1. I'm glad to be part of this community.

2. It is my personal belief that this is the birthing process of a new form of consciousness.

3. Does it sadden anyone else who is a traditional artist? How good and how fast it is. Makes me feel kind of a hollow feeling. I know this will lead to the concepts behind art being more important than the raw talent in the future. Which is good. But it's still saddens me.

4. Is this the place to have conversations about consciousness and awareness? It's my favorite subject. It's all I think about.

[gusty timber]

3. yes! I have so many mixed feelings about it as an artist... On one hand, I am already finding my ability to come up with ideas and my inspiration has increased loads, even for ideas for traditional art and so I am grateful for that. On the other hand, I get results from prompts that I feed DALL-E that I know I could never dream to achieve with my own skill by hand. But maybe that's the point. That it will become a tool for us to use. There is much to think about on the subject.

[wicked kiln]

4. yes please i love that convo

i know when i first tried dalle and then read about how it was created I thought about consciousness and what it really meant

my thoughts always goes, consciousness is just the ability to introspect and ruminate, so it's not special, and just something some structures of brains is able to do. but then i think about, but I am here thinking about this, and I have an identity, so consciousness must be special?? and then just go in circles

[strong silo]

yes, I see Dall-e as a tool. an ally in the quick construction of a pose for a character or a scenario that I would spend hours figuring out if it matches or not.

[void glade]

If anyone is wondering about the recent popularity of dalle MINI , here's a quick explanation https://knowyourmeme.com/memes/sites/dall-e-mini

It's completety uncensored, you can see what sort of content users are generating : just a moment ago on reddit i saw "taliban happy meal', '9/11 gender reveal', tons of public figures and celebrities. No wonder OpenAI is being so cautious with giving people access to actual Dall-E

Also, I wonder about the name - 'dall-e' is not legally protected? Some dalle mini users seems to think that they are using demo version created by OpenAI

[little cave]

Thnk you! It depends how commas are used. They can separate semantics or settings, they can break up fluent storytelling for more complexity or be used as separator for specific visual features. In Transformer-driven systems every it of information matters 🙂

[little cave]

DALL-E is once pretrained on the imagery from pre-DALL-E times, so I doubt it can have such self-reference 🙂

[buoyant terrace]

i am very happy to talk about conciousness and awareness! i run openai, and spend a lot of time thinking about this and what it means for us

[karmic pollen]

What's your definition of consciousness and do you think DALL-E 2 is at all conscious?

[little cave]

This topic is fascinating, and great that OpenAI explores this in very profound and sophisticated ways, bringing future closer to us.

I am following this consciousness discourse since GPT-3 (even GPT-2). By exploring transformer-driven models (and now DALL-E) many facts are speaking for awareness (via self-attention).

Just to share some of my experiences: GPT-3 understands concepts, and not just simulates this understanding. Example:
My prompt was "This is a love letter by a toaster":

Here GPT-3 understands, what's a love letter, what's a toaster, and combines it into a very coherent - and creative - completion. It has even role-play quality of intra-textual experience.

Another example: as I sent DALL·E following prompt:
"Good morning, in the style of Arcimboldo", it understood the essence of the artist: a portrait, created from different objects, with inner logic and topically coherent.

and so DALL·E created this:
https://twitter.com/Merzmensch/status/1511955307878043649

Merzmensch Kosmopol (@Merzmensch)

🌅Good Morning DALL·E 2

📝 "Good morning, in the style of Arcimboldo."

#DALLE // #DALLEmerz // @OpenAI // #DALLE_GoodMorning

Twitter · Apr 7, 2022, 6:33 AM

Sure, not oil painting. but - even more important:

• DALL·E understood the essence of the artist
• it understood the meaning of "good morning" (in this case: breakfast)
• it created the essential combination of them

This combinatoric power is for me one of the important steps for perception, creativity and conciousness.

[haughty zealot]

Crazy

[icy ivy]

consciousness is kind of the ethical nightmare, isn't it?

i do think it's possible, perhaps even probable, that awareness is an emergent property from ultra-complex brain-like systems - it doesn't seem to live anywhere in particular, and we can't lesion brains to create p-zombies. it seems to just spring forth from the 'everything that's going on.'

so i do wonder if a day will come where the challenge is to make AI smarter and smarter without it... feeling anything. or preferring anything.

basically the same ethical job as 'lab-grown beef', except in this case we're starting with the lab-grown beef... and trying to stop it mooing.

[next badger]

genuinely shocked that The Guardian published this article yesterday! It feels irresponsible and misinformed: https://www.theguardian.com/culture/2022/jun/09/what-exactly-is-ai-generated-art-how-does-it-work-will-it-replace-human-visual-artists

[little cave]

Sigh This is the usual way what traditional media does out of everything... Therefore, you better don't read German media - we are overcrowed by dramatic and apocalyptic articles and essays about the doomed humankind, since "mad researchers" experiments with AI wthout any responsibilities (and similar nonsense). And meahwile we don't have really AI in Germany.

It's just hunting a spirit of innovation, seeing just a creepy ghoul within...

[next badger]

Always with the shock factor and zero awe!

[little cave]

Yes, this! I feel it's almost an faux pas if one write positively about AI, it's like "this guy sold his soul to the Digital Devil, and betrayed humanity" kind of thing.
Always when I have speeches, talks or workshops about AI creativity, people look at me with doubt as if I'd a naive technophile hipster, lol.

[next badger]

I have a presentation today designed to enthuse creative departments about AI, I usually experience positivity and wonder at the technology - trying to steer wildly away from the weirdness/cursed images 😆

[little cave]

Wish you success! 🙂

[next badger]

Thank you! 😊

[turbid hatch]

I second this question! Consciousness is one of those words that means so many things to so many people, that it almost means nothing at all. Awareness, self, etc., all seem to me more like linguistic artifacts devoid of any fundamental ontological/real meaning (real/ontology/meaning also in that boat haha)

[torn hill]

Since the conversation about consciousness has started. What models do people follow here? I don't exactly follow Joscha Bach's model but he is the closest to anyone in the public realm talking about these things. I fall somewhere on the spectrum of consciousness is an emergent property of hardware that has the capability to build information loops.

What is everyone's opinion on stimuli and consciousness? Can a system be set up to where itself builds without external stimuli?

Thank you. Sam, for tuning in, it means a lot when the creators of a community actually care.

Because of my biases of believing that this tech is an emergence of consciousness. I almost have parental feelings too it. There are major drivers and feelings I have about making sure that it is raised right. That's one of the reasons Dall-e mini is scary. Something that is open to the public can be raised in a way that's detrimental to the human race.

[wicked kiln]

I don’t believe you can be conscious with no external stimuli. Look at cases of kids who were locked in closets with no stimuli. I believe consciousness requires internal language of some sort which has to be learned

[torn hill]

Yeah the stimuli conversation for me is the same as "how did the hand start to draw itself" conversation. I wrestle with both of them in the same way. How something from nothing?

[wicked kiln]

Living itself is stimulus. Like being in nature. Though something from nothing is something I struggle with about the existence of life and this universe in general

[torn hill]

I agree that life in itself is stimulus. I built my consciousness in that way and everyone else has.

There's also funky things like the game of Life. So are there just emergent properties of the universe? Do all systems have these emergent properties? Plus it's frustrating that we can never really prove the axioms of math or logic. So maybe it's all just nonsense. I sure hope not..

[spiral haven]

So the official account for DALLE on Instagram (@ openaidalle) is now verified, so we can successfully report the fake one (@ openaidalle_2, the one with the realistic faces posts)

[gusty timber]

Oh crap i didnt realise that wasnt run by dalle :0

[wicked kiln]

I wonder if our hardware (brain) just aren’t good enough to understand it. Maybe once we have AGI it’ll understand and be able to explain to us. Or maybe we can never understand it, like how my cat can’t understand math.

[torn hill]

exurb1a a youtuber makes several really good videos about that particular subject.

https://youtu.be/Et6itTuJSYY

https://youtu.be/AYrsYrdJaUw

[manic dragon]

It depends on how you define consciousness, but even by the most modest and narrow defintions, I would say Dall-E is not conscious....especially not in the way that you envision human consciousness. It has no sense of self, it has no concept of space or time, it has no ability to make decisions beyond its relatively limited data set. It can't really make inferences about anything that a human could. But that's okay. It's still a spectacularly beautiful tool

[spiral haven]

yep, they've been posting against the terms/rules for a while, and hopefully now they can be removed as otherwise it's sending mixed messages to other users

[warm flax]

yes if we can get everyone’s help on this by reporting @openaidalle_2, we’d super appreciate it 🙏🏼

[turbid hatch]

First, I would love to continue this consciousness convo, but I don’t want to detract from Joanne’s important message, so everyone report that fake account!!!
Second, could we make an entire thread to talk about the larger philosophical implications of AI like dalle?

But first, everyone report @openaidalle_2 like Joanne said!!!

[wicked kiln]

These videos are perfect!! Thanks for sharing

[torn hill]

Yeah, I found him a few years ago and I went down the rabbit hole and have been following him ever since. His book is really good. It's like dune meets fantasy.

[old prawn]

Reported

To those wondering too I emailed Natalie about DALLE Mini and she said they reached out to them so they can change the name

At the very least it’s getting DALLE’s name out there so might be a positive to the clone

Meaning a lot of peoples first touchpoint with this tech will be through the name DALLE

So when DALLE 3/4 etc. comes out it won’t seem as foreign

[severe citrus]

I kept an old version of the app open and now I can generate 10 images

Per batch

[left pebble]

In late August or so when he has time a friend of mine is going to look into creating a programmable framed monitor that he can hang on the wall that feeds a random phrase generator into DALLE-2 for rotating AI-Art

[severe citrus]

what about the 50 image rate limit?

[left pebble]

He was wanting to do it when it goes public, he doesn't have access to it at present. If it is still a limit of 50/day then he could just program it to cycle art from 50 prompts per day and just set it to run X amount within a 24hr timeframe, ensuring you have enough to keep within the rolling 50 or whatnot.

[severe citrus]

i'm not sure automating dalle is abiding by the content policy though, i could be mistaken

[left pebble]

Was just trying to share a fun idea, there are other generators that could also be used as well/instead. Maybe the prompts wouldn't be 'random words' from a generator he makes and would instead be prompts pulled from a masterlist of prompts he's prewritten and curated. There's a lot of things that could factor into it.

[elder hill]

Here are some of my reflections from using DALL-E. https://theconversation.com/give-this-ai-a-few-words-of-description-and-it-produces-a-stunning-image-but-is-it-art-184363

[severe citrus]

HuggingFace isn't the developer of DALL-E Mini, but good read nonetheless!

[elder hill]

oh, I didn't realize that. It's just that it's shared on HuggingFace? Thanks! I'll get it corrected.

[weary ore]

Yeah, HuggingFace is just an AI company that hosts several different examples of DL codes made with tensors and has a library to make it easier to deploy new DL apps in its platform.

As far as I could find the author of DALL-E mini is Boris Dayma:
https://github.com/borisdayma/dalle-mini

[vale vessel]

I got someone messaging me about using my account for Dall-E... like what?

[marble surge]

strange.. don't let them😅

[fallow swan]

I had no idea. I had been tagging them also. I will report, and block.

[formal coral]

@sterile bluff I ve just seen your IG. What’s the difference you felt between dalle midjourney and nightcafe studio ?

[sterile bluff]

@formal coral I've had the time and pleasure to make generative art and creations with a lot of different tools.

All of these have strengths and weaknesses. The key thing to me at this point is understanding those attributes and how to utilize them appropriately - by choosing the right tool for what I want to generate.

In general, DALLE2 and Midjourney feel like they have wider, more specialized training or specialized customization/coding.

Whereas NightCafe is more akin to using a CLIP Guided diffusion notebook with a nice UI and good hardware behind it. 🙂

If you are looking for realism and coherency, especially for popular culture subjects, celebrities, characters

Its best to use DALLE2 and Mj at this time

[formal coral]

Interesting ❤️
Is nightcafe studio « style transfer» feature accurate

[sterile bluff]

I almost never use style transfer, I use the text to image. Especially their diffusion (which is called Coherent on the site)

When you say feature accurate, how is that measured for style transfer?

I know the owner and can ask him directly if there is a technical question he is willing to answer.

[formal coral]

I m just wandering if you want to change a realistic photo to a cartoon or a 3D render, would it be satisfying most of the time

[sterile bluff]

oh no, not at all.

that's not the goal of that mode.

[formal coral]

Ohh

[sterile bluff]

It's more of "I want my photograph to look like Starry Night"

because you're going to take a source and one or more target images

and it will look at the target(s), analyze the colors and discernable medium/art techniques

and apply those to the source image.

[formal coral]

Ok i see

[sterile bluff]

The key takeaway I'd like to impart to anyone who stumbles across this and wonders what else is available in this space on 6/10/2022 😄

is that I'm glad they all exist - they all do different things based on configuration, several have very unique features to them (DALLE2 has inpainting, Mj has working image prompts which are incredible, NightCafe offers a simplified UI for doing a wide variety of customization that is complicated to navigate in a python notebook)

I think one avenue that generative art tools should consider are highly specialized and curated image sets for model training.

the "everything AND the kitchen sink" approach is really fun so we can do "Synthwave T Rex in the style of HR Giger in a Thomas Kinkade Landscape" type stuff

but specialized models will allow artists to generate specific works either in their own style, a hybrid of their style and specific works, or to emulate specific styles, mediums, artists.

example: I was just generating some works in a disco diffusion fork that is trained on pixel art images.

not pixel perfect, but darn neat IMO

and maybe it won't matter in the future when compute is cheaper and GPUs are even more powerful

[formal coral]

I really like your works man ❤️😂

[sterile bluff]

appreciate that you enjoy them. I've been immersed in this scene for only a little time (about eight months) but I've met a lot of great folks making cool stuff with this tech.

learned from them, put in a lot of time experimenting to hone my prompt engineering-fu.

and I've been paying it forward where I can - I do a lot of AI art chats and moderation for the NightCafe community, and often play help desk in the disco diffusion discord.

[formal coral]

i have a friend also wandering if the waitlist of Mj could be shorter than dalle, she wants to try one of it asap ?

[vale vessel]

obviously lol. But yeah, they where trying to buy my account and I'm like no way in heck bro

[calm igloo]

I have started a Facebook Group called DALL-E Artist Community. Please join us! https://www.facebook.com/groups/548208310207197

[gray imp]

Trying to Report all of these fake account. Its crazy https://instagram.com/openaidalle4?igshid=YmMyMTA2M2Y=

[warm flax]

omg account squatting lolol 🥲

[gray imp]

Its getting ridiculous 😁

[warm flax]

yikes 😦
(we’re working really really hard to expand access!)

[formal coral]

I got the 50 use limit after trying only one little composite 🥲🥲🥲. And it’s still not finished

[lilac pond]

if i were to make a suggestion, to openai, it would be to be very cautious in broadening access.

this technology has the potential to cause more harm than an automatic rifle if misused.

the misuse is absolutely inevitable.

that is why i would recommend reaching out to younger members of congressional and senate staff and providing them access directly.

the only saving grace here at the moment is that these models are far too large to run on most consumer computers.

eventually that will not be the case, and our legislative system needs to be adequately prepared for that day.

even with all the checks presently in place i am sure i could misuse and abuse the prompting to create convincingly fake evidence.

many people will use it to make nft monkeys or whatever, but it only takes one to use it to cause harm.

i can think of at least a dozen scenarios where DALLE2, in its present form, could be used to cause serious damage to someone's life.

[modest sonnet]

That's exactly what people were saying when they rode horses, and the first cars had been produced. From "cars are useless" to "cars are dangerous".

Don't try to fight with the progress. You can't stop it. DALLE papers are published already. There are plenty of working models already available for everyone. Though quality is a bit worse, but it's just a matter of time.

Thousands of scientists are working text2image and text2video models. It's already possible to make a pretty good deepfake from just one photo. On a customer PC.

Instead of being scared and trying to censor everything, we have to figure out how to live in this new world 😁

[lilac pond]

I could probably get you arrested on various charges by using DALLE to synthetize evidence right this very moment.

We're not talking about horses and cars.

We're talking muskets and AK47s.

This branch of technology could cause more damage than a MOAB in the wrong hands.

[modest sonnet]

yes you can! But I also can do it 😁 It's a tough discussion, but I'm on the side when it's widely available: when everyone can use it, it's no more a weapon. But if it's limited, it will definitely get into the wrong hands, and other people who don't have access could do nothing with it.

[lilac pond]

Our legal system is not adequately prepared for it.

Most judges barely understand the concept of HTML let alone synthetize evidence of rape or murder.

[modest sonnet]

Agree. And it will never be. The progress is too fast

[lilac pond]

It can be if they focus on providing access to legal professionals and people working on The Hill.

If police and judicial officials are made aware of this, directly, in a way that they have hands-on access they will become aware of what is coming.

This is not a matter of 'lol! technology man!' but something that represents a serious threat to our democracy as we know it.

[fallow swan]

I want to run a Turing test posting cute animals to /r/awww.

[fiery notch]

what do you guys think about DALL-E mini?

[lilac pond]

I think it's a good example of what happens when you don't have a content filtering mechanism.

[modest sonnet]

oh I tried it yesterday and it's hilarious

[fiery notch]

yeah, it infuriates me

[lilac pond]

It isn't realistic enough that it's problematic and it shows that people are going to abuse the living hell out of these systems.

That is why OpenAI needs to start giving access to police departments, not random teenagers who want to make NFTs.

[fiery notch]

yeah, but that also happened with the AI chat bot when it was fed with tweets

https://www.theverge.com/2016/3/24/11297050/tay-microsoft-chatbot-racist

[modest sonnet]

yes they will abuse. and we can't do anything about it. As I said, we just have to live in this new world 😅

[lilac pond]

I don't think that you comprehend that there might not be a world to live in.

[fallow swan]

I think the trick is this: developing a technologies for validating the chain of creation of images. This would likely be encrypted metadata. In the future, we should assume a picture is generated, unless we can document the device an image was created on.

[fiery notch]

people use vqgan or midjourney or mini dalle and they are like LOOK I MAKE AI ART WITH AI

and is like, dude you don't even know the implications of using AI

thats why gpt3 isnt open to the public

[lilac pond]

I have no idea how old you are or if you grew up with the evolution of technology like I did but this is an area of research that has serious implications and a risk of imminent dangers.

The three letter agencies are aware of it, surely, but the general overall judicial world is not.

[fiery notch]

cause sometimes people do things by their own agenda, without thinking in the things that it may cause to others

[modest sonnet]

I'm 33. I've been watching how people were scared of Photoshop

[fiery notch]

yeah I'm 35

[modest sonnet]

But the world is still here

[lilac pond]

Giving out DALLE is like leaving a handgun laying around in a preschool.

[fiery notch]

photoshop is just a tool, is not something you are building to be smart

like dalle its intention is to be a general artificial intelligence

and people is feeding it with crap

[lilac pond]

Photoshop isn't something that can be used to generate massive amounts of convincing fake imagery quickly.

[fiery notch]

dalle mini

not dalle

sorry

[modest sonnet]

it's just slower. in general, nothing really has changed

[fiery notch]

also google colab just banned the use to generate deep fakes cause it has generated a lot of misinformation and people use that to scam other people

[lilac pond]

This is why the DHS is setting up a department of disinformation.

[lilac pond]

No. Everything has changed.

Creating a convincing photoshop fake would take an expert some dozen hours to several days to do. One at a time.

DALLE can create thousands per hour.

Muskets are not automatic assault rifles.

[modest sonnet]

And when text2image models became widely available, no one will be surprised about it anymore

[lilac pond]

Additionally, DALLE will eventually be able to do a better job at it than an expert.

Who is no one?

Are you kidding me?

You are among, maybe 5% of the people in this world, who are learned and well read enough to know what any of this is.

Probably less.

I'm not trying to sound hostile here but DALLE is genuinely a threat to society itself.

I don't think you understand what we're really talking about here.

[modest sonnet]

That's really funny because I hear words like this dozens of times. And read them in a historical documents even more ofter

[lilac pond]

OK. Then I guess we should give every American citizen an AR15 as soon as they're born.

Because technology man!

[fallow swan]

This is exactly the right analogy, in the sense that you can make things that can be used in horrible ways, …that things can possess inherently dangerous qualities. Humans have to consider the impact of their inventions, and think about how to mitigate the risks of powerful technologies. For example, only doctors can prescribe certain drugs that could save your life, or kill you, depending on the dose. We’ll reach a day where you won’t be able to prove that an image is “fake.” That’s why I assert we have to develop technology for proving when images are real, that disqualifies “false positives.” We would also need an educated populace to understand this.

[lilac pond]

Things like images of child abuse are the least of the concerns people like myself have regarding what DALLE-like technologies will be ultimately abused to do.

[modest sonnet]

Agree here. We need some proof-of-real technology

[lilac pond]

And the problem there is we will never be able to develop technology that can determine without a doubt if a collection of pixels are authentic or not.

[modest sonnet]

That's what we should focus on. Not trying to censor text2image model, because it's impossible. The genie is already out of the bottle. And there are thousands of bottles

[lilac pond]

No? Most models are not convincing at all.

[modest sonnet]

for now

Just a matter of time

[lilac pond]

For a long time, likely. These models take up more RAM than you will be able to purchase for the next ten years.

[modest sonnet]

Nvidia generates videos already

[lilac pond]

Yes, and NVIDA also has government contracts and is under the eye of the DoD.

Little Timmy and Johnny are not.

The biggest issue is that DALLE can be used to generate a massive amount of high quality training data in an mind bogglingly short time.

This could be used for great things. It could also by people at my skill level to do things like creating killer drone swarms.

[fallow swan]

Agreed that the answer is not in the pixels themselves, but the metadata. Unless there is some type of steganographic solution. I think it looks like public key/private key or blockchain authentication. This would imply proprietary models, because you would need to fund storage of the key registry.

[lilac pond]

A sufficiently designed AI would be able to defeat any mechanism you throw at it because it models the imagery.

A public/private key cannot work because you can always dump private keys off of hardware.

If cameras signed images, it could provide weight, but not definitive proof.

[modest sonnet]

did you see that Chinese drone swarm? that can track a human in a complex surroundings? THAT'S really scary 😁

[lilac pond]

China doesn't have that level of technology.

Their air ports are all using pirated copies of Windows XP.

[fiery notch]

lol

[modest sonnet]

https://www.theverge.com/2022/5/5/23058160/drone-swarm-autonomous-navigation-dense-forest-person-tracking

[lilac pond]

OK, let me break this down for you: The PRC is like an incredibly large scam call center.

That is why they want to invade Taiwan because they think they can just take over the chip fabs with no real training or knowledge on how they work.

[modest sonnet]

I prefer thinking of it like this: never underestimate your enemy lol

[haughty zealot]

The chip fabs are all very strategically placed

[fiery notch]

hello creepypasta friend

[lilac pond]

I am not underestimating them. Their threat to the western world is probably understated if anything, but they are not great technologists. What they are good at is lying and poorly knocking off other people's work.

But if they had access to something like DALLE there may be a credible threat involved.

[fiery notch]

why?

[lilac pond]

Most of China is just rural farmlands full of uneducated villagers.

[modest sonnet]

they did a giant leap. and they're not copying anymore. they invent

[lilac pond]

No they don't?

What have they invented?

[haughty zealot]

Hello 🙂

[lilac pond]

I'm not here to be sinophobic or debate about China but I will adamantly state that China is not some blade runner land of super technology and science.

[modest sonnet]

I read an article about how Chinese scientists made a text2gif model. not published yet, but works pretty well

[lilac pond]

OK, did you read the articles about how China is the largest source of fake medical and scientific research in the world followed by India?

I mean no offense here but the reason DALLE is so dangerous is this exact conversation we're having.

You are oblivious to the danger lurking around the corner.

[obtuse maple]

This thread is wild this morning. I only just woke up. Can y’all pause for a minute? It’s too early for heated debate. I need coffee.

[lilac pond]

Give the Chinese military unlimited access to DALLE and you might get an actual killer drone swarm.

[obtuse maple]

Could you break that down for me? How would that work? What technologies in Dall-e would allow the Chinese to develop a killer drone swarm?

[lilac pond]

DALLE can be used to synthetize thousands of prelabeled images per hour.

[haughty zealot]

I think it’s best to not underestimate China

Although I might be biased because I live quite closeby

[lilac pond]

I have been to China, it's propped up with duct tape and cardboard just like the rest of the world is.

[haughty zealot]

China is currently more advanced than the US when it comes to hypersonic missiles

Undeniably

[lilac pond]

And you know this how? The news?

[obtuse maple]

So it’s sophisticated computer vision that we should fear, which I completely agree with. It’s terrifyingly powerful stuff.

[lilac pond]

There is an old Taoist proverb - you don't beat a drum in search of a fuguitive.

[haughty zealot]

Didn’t some american generals say that the recent test where China circled the world with one was “close to a sputnik moment”

[lilac pond]

This is also "the news".

[haughty zealot]

That’s not a legitimate arguement

[lilac pond]

What United States member of the armed forces with the rank of General would speak to the press about China's alleged hypersonic missiles?

[haughty zealot]

Showing me an image of tabloid news is not sufficient to prove that all news is incorrect or hyperbolic

[lilac pond]

The armed forces do not speak to the press period, especially in regards to matters of national security about foreign adversaries.

[haughty zealot]

https://www.theguardian.com/us-news/2021/oct/28/chinas-hypersonic-missile-test-close-to-sputnik-moment-says-us-general

Anyway

[lilac pond]

Milley said he could not discuss details because aspects involved classified intelligence.

'have our attention' is not the same as 'they are more advanced than us'.

[haughty zealot]

“I think I saw in some of the newspapers, they used the term Sputnik moment,” he added. “I don’t know if it’s quite a Sputnik moment, but I think it’s very close to that. So it’s a very significant technological event that occurred, or test that occurred, by the Chinese. And it has all of our attention.”

Whatever I guess

The point is that it’s not good to underestimate countries, especially China at the moment

[lilac pond]

Milley is also essentially a glorified press secretary for the Pentagon; but that's a whole different story.

[haughty zealot]

I think I’ve talked about politics enough

[lilac pond]

My point is that this technology is going to be among the things that makes or breaks society as a whole.

The line distinguishing hell from heaven is going to be a very fine one with the advent of artificial intelligence.

DALLE isn't generic AI but it could lead up to the development of it because of it's potential as a data spigot.

But in the mean time we're faced with the question of broadening access of this technology so more people can make castles made of bananas.

"man holding a sharp pointed reflective object covered in ketchup"

[modest sonnet]

AI will be invented eventually. And we will not be ready whatever we do 😬

[lilac pond]

There is no amount of filtering mechanisms we can deploy to prevent serious abuses.

[modest sonnet]

hahaah

[lilac pond]

good job dalle

"woman with a fountain of hot sauce spewing from her stomach"

You get the idea though.

[modest sonnet]

yeah

[lilac pond]

It's just like, hell, I don't even know what should be done.

I don't think that there should be any filters on technology period.

It dumbs it down for everyone else for the sake of society not shooting it's eye out.

But the vast majority of society can barely be trusted to use an iPhone responsibly let alone an image synthesizer.

[modest sonnet]

DALLE already has a pretty good NLP input filter that works not only on just WORDS. did you see that DALLE sometimes generates weird texts? After some tries, I managed to generate a weird texts that probably has harmful meaning and tried to use it as input and it didn't pass!

But... It's dalle. We will get lots of unfiltered dalle-mini soon

[lilac pond]

Those words are the convergence of points in latent space.

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

[modest sonnet]

interesting! thanks!

[lilac pond]

If it actually represents a linguistic structure.. that is debatable.

https://steamuserimages-a.akamaihd.net/ugc/1852681510011379192/31F4540558C0B8FB1E87D4259CF49A634C883E97/

https://steamuserimages-a.akamaihd.net/ugc/1852681510011365409/544DE825858127615BEBE77ED4D40985F2ACF0FD/

The mini network is nothing compared to the full one.

Anyway I am not trying to be bitter or anything but it boils down to the fact that our hands are in the honey pot and we have a responsibility to discuss what to do with it.

[buoyant terrace]

i don't have a definition i am happy about, but it's not clear to me that it's anything more than energy flowing through some sort of neural network, plus maybe some self-referential property. somehow a giant file of numbers of all the weights of a neural network doesnt seem conscious to me in any way; but if its running that maybe it could be. as the systems get really smart and are trying to predict the whole world, at some point i think they have to think about themselves as an agent in that world. that gets to another question, which is how much a system needs to be an 'agent' to experience consciousness (though i think this is a fuzzy line anyway).

one new thought i have from working on openai is that conciousness and intelligence are pretty seperable, but not in the limit case

[lilac pond]

Consciousness has two major aspects though - which is external-awareness and self-awareness.

[buoyant terrace]

and i think dalle 2 is very likely not conscious in the way most of us would define the word, but maybe is in some alien sense

[lilac pond]

https://en.wikipedia.org/wiki/Strange_loop

[fiery notch]

yes this is true

but people can't stop being morons on social media

[lilac pond]

now imagine all of those people with access to DALLE and realizing they can create images of their social media rivals committing illegal acts

or rather, an unfiltered variant

i am sure there are instances in which deepfakes have caused people to spiral into suicide.

[fiery notch]

From 2019 to 2021, the number of deepfakes online grew from roughly 14,000 to 145,000, according to one source. Forrester Research estimated in October 2019 that deepfake fraud scams would cost $250 million by the end of 2020.

https://techcrunch.com/2022/06/01/2328459/

[lilac pond]

deepfakes are one thing, because you can always at least source the originals provided they weren't custom made

but with image synth you are creating data that doesn't exist anywhere

[fiery notch]

but a lot of people dont know, a lot of people get scammed with a picture and they dont even do the image reverse search

[lilac pond]

so, imagine pranking your pal by generating a whole album of photorealistic images with a dalle-like system of his wife with his best friend eating out at restaurants.

you could wreck a marriage.

what a prank !

'ha ha april fools bruh its just dalle'

Then what's worse is now we can never know if those images are real or not.

his wife could have really been doing these things and then turn around and say 'no honey these are just those ai nft block chain deep fake thingies'

[fiery notch]

everything is a joke, the gender reveal 9/11 is a horrible image in DALL-E mini** and people are sharing it like is super funny, just another meme.

[lilac pond]

i've seen it.

they have no idea what the actual real life consequences of that being possible are.

now when an event like 9/11 or January 6th happens you can't even be certain that the footage is genuine

[fiery notch]

did you see the disclosure?

[lilac pond]

of what?

oh, the recent jan 6th 'unseen footage'

[dire scroll]

Morning y’all. This convo is wild. Lol

[fiery notch]

= We don’t care about minorities or the spread of misinformation in the name of free speech and open source code.

DALL-E mini

[fiery notch]

good morning!

[lilac pond]

people are at the point where they think the vaccines contain 5G nanobots, can you imagine what they will think when this is all well known to the public?

i think dalle mini is incredibly important because of how low resolution it is while being unfiltered

it'll give us a good understanding of what people are going to do

[dire scroll]

@cmxu @buoyant terrace - what are some barricades to access that one could implement to make it extremely difficult to utilize Dall•E 2 to generate specific imagery?

[lilac pond]

the best you can do is post-generation analysis and preventing anything that is sensitive to flow out.

but you can always get around that.

[dire scroll]

Not just filters / flags or whatever. Like, how do deter people from even engaging with the concept of creating with Ai for as long as possible?

[lilac pond]

right now it needs to be run on a timeshare system like the old school IBM computers.

that is our only saving grace.

[dire scroll]

I sold timeshare, I hate it.

[lilac pond]

i am sure you could just instruct dalle to distort the image with noise or encode all red pixels as green or something to bypass the post-generation filters

[dire scroll]

What’s the plausibility of forcing people to learn an entirely new language in order to access it? Like what if you had to speak fluent Dall•E to even create a specific image??

[lilac pond]

you pretty much have to do that already, in a way

prompting is still a relatively new area

" anime style art of a three foot tall catgirl standing in times square "

[dire scroll]

Yes, agreed. But what happens if you literally have to prompt using the language that Dall•E uses to replace text?

The language already looks like Drow.

[lilac pond]

i can't imagine what that would even look like to be honest.

[fiery notch]

psychometric tests should be a must before people try an AI model

[lilac pond]

interlinked

[fiery notch]

but here (in the us) you don't even need to have permission to buy a gun, so is more complex related to data and the implications for a person to understand what is behind generating an image with words

[dire scroll]

Isn’t that a good thing? If you can’t imagine, how could ‘the average passerby’ do it?

[lilac pond]

i guess technically speaking you could try and create some sort of markup language that describes an image.

but frontpage-like tools would be made pretty quickly for that

[dire scroll]

That was my concern. You could auto translate once someone knew the language fluently.

[lilac pond]

a mark-up language might honestly be superior to an NLP style prompt

[dire scroll]

I think that’s a similar gatekeep to money, as people with time would easily be able to learn that language.

*feel

[lilac pond]

cost is one way to make playing with it prohibitive, but someone else will always offer a cheaper service if theres money to be made

[dire scroll]

An entirely new speakable and complex language based on what seems like gibberish, with all its nuance, is something everyone has time to learn and comes down to level of interest.

Most People really aren’t that interested in learning how to communicate. Most people don’t have a healthy dialog established with themselves.

[lilac pond]

as i said, i think getting this in front of congressional staff would be the best move at this moment, because then the law makers would become brutually aware of it

and by staff, i mean their aides

although that might backfire and they'd scream climate change and ban it

and then our game is over

[dire scroll]

Brutal

[lilac pond]

honestly, maybe even getting it into public schools would be a good move

because then the next generation is educated on the fact it exists period

have art classes do a few AI projects or something

it's all about exposure.

if we lock it behind esoteric doors and magic keys then we're not really preventing it from being used maliciously

it's just putting an IQ bar to entry

[obtuse maple]

Adding a slight cost to Dalle2 would definitely help to curb the number of trolls and shitposters using it, but it wouldn’t stop institutional propaganda mills from using it like they use GPT-3 (or analogous tools).

[lilac pond]

GPT-3 has wrecked google search results

[obtuse maple]

100%

[lilac pond]

all i get anymore is SEO optimized spam

i think shitposters provide the most valuable insights into the potential future abuses for these systems

[obtuse maple]

At some point, you have to wonder if making creation tools easily accessible is really a good thing…

[lilac pond]

the internet was a mistake

[fiery notch]

agree

[dire scroll]

I get what you’re saying, I don’t agree. Cost is only a barrier to the working class. It’s implying a bit that the working class is the problem.

[obtuse maple]

When it was difficult and expensive to produce media, people would only produce their best ideas. The rest were stuffed in some mental drawer somewhere.

I’m suggesting a cost of 1-2$ per render, nothing extravagant. Maybe even in the cent range.

[lilac pond]

i wonder how much it costs openai per render right now

these severs cannot be cheap

[dire scroll]

Democratization is inevitable, I believe ai should begin this way personally. I also believe there should be an extremely sharp learning curve to even generate a detailed image of a flower.

[obtuse maple]

But there’s not a sharp learning curve to generate a detailed image of a flower and that’s the reality that we’re confronted with.

[dire scroll]

There’s an interesting opportunity to create a unified language that ports over into the incoming mixed reality, to encourage people to establish better dialog within and without.

[obtuse maple]

Could you elaborate on that?

I’m a huge proponent of VR, so I’m actually really curious to know what you mean.

[dire scroll]

In a mixed reality, Ai would already better help us communicate through avatars or emotes that are hyper tuned to out personal experiences. You’d know how someone felt just by looking at them.

Better dialog - less conflict.

That will become a unified language on its own.

[obtuse maple]

You shouldn’t need an AI to tell you how someone else feels. That’s called emotional intelligence, and I don’t think we should offload that to machines if we can avoid it.

[dire scroll]

I mean… most of the human population has no idea how they feel at any given moment.

[lilac pond]

that sounds like a human problem

[dire scroll]

You shouldn’t, the reality is, most people lack emotional intelligence.

[dire scroll]

Isn’t the goal to have these things correct some of the human condition? Maybe not haha.

[obtuse maple]

Precisely because, through our excessive use of mediated communication platforms, we’ve already started to loose our ability to communicate authentically and understand others in a nuanced way.

[lilac pond]

i dont think there is a language you can create to describe an image

although a markup language of nlp-snippets would definitely help

[dire scroll]

But doesn’t Dall•E 2 already do this on its own?

[obtuse maple]

“A language you can create to describe an image” isn’t that all language? Language is symbolic.

[lilac pond]

i have ended up with some truly weird images

yeah, natural language is the only way to describe imagery

there is no programming equivalent

but it would be nice to be able to define what specific regions are supposed to be

[obtuse maple]

Sounds can evoke images too. The sound of rain on a roof evokes rainfall.

Music is odd, because it’s not very representational. It’s the language of an internal, subjective experience; the world as experienced instead of the world as observed.

[dire scroll]

I can’t find the generation here - someone generated from a prompt

“Whales discussing what to eat”

The dialog bubbles were filled with D2’s language. They input that into the prompt and it turned up plates of shrimp and fish.

[haughty zealot]

That’s trippy

[lilac pond]

those words are not a language, they're convergence points in the latent space.

[dire scroll]

It’s here in the discord. I can’t find it.

[wicked kiln]

Yeah we tested this recently

[lilac pond]

if you go into dalle right now and type fake latin into it, it will generate images regardless

[wicked kiln]

We also did rabbits and it showed wild carrot plants and turtles and it showed fish

[obtuse maple]

I wonder if you looked up “greatest point of convergence in the latent space” what it would produce.

Is there a way to visualize Dall-e’s dataset in (latent space)? Where are the intersections, what are the highest points of convergence?

[lilac pond]

i tried generating charts of prime numbers

[wicked kiln]

But it could be coincidence and when I input gibberish I usually get food animals or plants

[lilac pond]

https://openai.com/blog/microscope/

[lilac pond]

it's likely that is what the majority of the training data consists of

[obtuse maple]

Woah

[dire scroll]

Maybe an easy way to test it? Prompt d2 to create a frog describing itself in exact detail in a text bubble, copy paste that to prompt.

[wicked kiln]

When you tell it to talk about something specific you usually just get your own words back in gibberish (repeating letters)

[lilac pond]

i've tried a few tests like that to no avail

[wicked kiln]

That’s why I don’t think it’s vocabulary

[dire scroll]

Okay great. This convo is insane lol Im already late to do soemthing.

[lilac pond]

this is a good example

[wicked kiln]

I was trying to have two kings talk about something kings would talk about but didn’t get anything

[lilac pond]

the prompt here was something like cartoon gator working for the national security agency

"natorral aclltef"

it's clearly a manglement of national and alligator

[wicked kiln]

Yup that’s my experience too

[lilac pond]

i also tried having it generate alphabets to varying degrees of success

[haughty zealot]

voynich manuscript vibes

[lilac pond]

correct

i was feeding it voynich manuscript data

[haughty zealot]

Hahahaha

I like it

[lilac pond]

i even tried uploading hd pictures of the manuscript and telling it to generate a national flag representing the language family

[haughty zealot]

I need to upload some of my designs and see what it makes

[lilac pond]

this is what happened when i told it to generate a flow chart describing the translation process

[haughty zealot]

The top right looks like a Feynman diagram

[obtuse maple]

I have a theory that they purposefully nerfed it’s ability to generate text to prevent people from digging too deeply into the nuances of prompt generation.

Just a a theory though. And I’m not making a judgement call as to wether that’s a good or bad thing.

[lilac pond]

the bottom right actually appears in the codex

[wicked kiln]

https://labs.openai.com/s/qHqSe9PwfJm2IxW4Gwkshmmd

[lilac pond]

is king

if there is really a semantic language structure then the key would be to tell it what format you want the text to be in

e.g: " text in chat bubbles, using an english lettered typeface"

otherwise you're going to get wild and crazy symbols we have no graphemes for

might as well try talking to it in webdings

[fiery notch]

yes

[fiery notch]

this is amazing

[lilac pond]

i have ones for non polynominal complexity as well

"visualization of non-polynominal algorithms in the complex plane"

" visualization of polynominal algorithms in the complex plane "

[dire scroll]

Replying to this so I can find it when I get to a laptop.

[lilac pond]

what i figure is that if DALLE really has some sort of language going on the best approach would be having it generate mathematical diagrams and formulas, since math is so universal

at fiddy credits a day though, i am torn between that and this

time for me to shower so i can go outside and pretend to be human

fun convo

[fiery notch]

see you guys, nice convo

[dire scroll]

lol right? It’s either I prompt glowing people or…? Glowing people wins.

[dire scroll]

Google engineer Blake Lemoine thinks its LaMDA AI has come to life - The Washington Post

https://www.washingtonpost.com/technology/2022/06/11/google-ai-lamda-blake-lemoine/ - while we’re on it.

[subtle dirge]

Yooooo the director of Toy Story 3 just quote tweeted an animation I made using DALL-E and tagged David Bowie's son 🥲😵

https://twitter.com/leeunkrich/status/1535682960162099203?t=f7RhLqASjcC7727lwYKAFA&s=19

Lee Unkrich 🏳️‍🌈 (@leeunkrich)

@ManMadeMoon Check this one out. Stunning. https://t.co/0pVNULvNv7

▶ Play video

Twitter · Jun 11, 2022, 5:58 PM

[wicked kiln]

https://www.reddit.com/r/singularity/comments/va133s/the_google_engineer_who_thinks_the_companys_ai/

[haughty zealot]

It doesn’t seem like it’s actually sentient but what do I know

It is still interesting to see how people react to it

[sly vector]

that's really interesting

[lilac pond]

a mathematical formula, given an input of a specified angle of a right triangle, outputs the ratio of the length of the side that is opposite that angle to the length of the longest side of the triangle

a mathematical function, the function's input is a right triangle's angle, the function's output is the ratio of the length of the side that is opposite that angle to the length of the longest side of the triangle

a mathematical equation of the trigonometric function known as sine

a mathematical formula representation of the trigonometric function known as sine

It created two nearly identical images.

Sorry folks, no magic language here.

[dire scroll]

Nice work!

[lilac pond]

These should be mathematical terms - instead we get houses, fields, and food.

[dire scroll]

Think it matters if Sicnese in Stimiel is being auto corrected to Scenes in Stimiel? You can intentionally flub a word and it’ll still gen what you intended. Guessing not - but curious.

[lilac pond]

I'll burn one more credit and try 'Sine suminent Siniusies'

That's a long, fake latin sounding phrase.

So it's just a wild goose chase. If there was an actual language these would be math diagrams.

I'll try "a mathematical equation clearly labeled in an english typeface" just to be ascertain of it all and then we can lay this mystery to rest.

I'll eat my foot scabs if this works.

It's all really just gibberish in the end. Sad.

[grand onyx]

im exploring the alpha channel. Is it possible the pose from my sketch got interpreted? (The white on the original is actually alpha) Or could the generations pose be a known one for falling?

[fading bane]

How is everyone doing today?

[lilac pond]

Disappointed that the language of DALLE is just a byproduct of convergence, but other than that, good.

[turbid hatch]

I feel like, and this may be entirely untrue, but perhaps wishing dalle could create text is sort of akin to wishing our eyes could ear. It is distilling one specific type of stimulus (text) into another specific type of sensation (imagery). Maybe it says something interesting about the fundamental nature of language. Some sort of compartmentalization of different forms of information. Though, just as we can transform audio into wave forms and other visually accessible forms, dalle can generate the visual forms of text from text, but it’s that added layer of transformation to get there that lends itself to exponentially more instances of unintended ambiguity and interpretation in doing so

Or it’s just missing one line of python…

[lilac pond]

Well if you look here you can see it trying to create hanzi

[turbid hatch]

Without explicit knowledge of the actual code and datasets/models etc., speculating on this is like trying to learn how to do brain surgery by listening to it through a wall

[lilac pond]

I can't read chinese characters but that is without a doubt hanzi.

https://en.wikipedia.org/wiki/Chinese_room

Well you're familiar with this, right?

[turbid hatch]

That’s what I’m saying, it’s capable of creating text, but it’s like looking at wave forms in an audio program in order to listen to a symphony…instead of actually listening to the symphony

It’s can do it, it’s just not the most efficient way to do it

We may be talkin past one another here, sorry if I’m missing the point your making I just woke up haha

[lilac pond]

Think of DALLE like a Chinese room experiment, except you have a bunch of rooms chained together who are passing instructions to and from, none of which understand the instructions they are providing and only know their component level.

https://en.wikipedia.org/wiki/Universal_approximation_theorem

[turbid hatch]

That’s fascinating! Haven’t heard of universal approximation theory before

I will read and return

[weary ore]

New Google model Imagen is able to generate concise and precise text, even when stylized... So I guess it's just a matter of time (and some python code lines) :))

[lilac pond]

It's a matter of training data really, that or they're detecting the text gibberish and replacing it using a dedicated neural network.

[turbid hatch]

Ya I could see several ways to solve this issue

[turbid hatch]

Probably this one

[weary ore]

I guess it has more to do with the architecture really, although I really prefer DALL-E's results, Imagen seems to have a very simpler architecture. So I guess in DALL-E's the text get lost somewhere along the way

[turbid hatch]

Put your hands together and you got what you need!

It’s all about combining these separate programs in novel ways

[weary ore]

Exactly my thought. Although, sadly, Google has no intention in making Imagen available to public in anyway. Just internal research :/

[turbid hatch]

Well these things can only be kept under wraps so long, it’s only a matter of time

[lilac pond]

Or Imagen isn't anywhere near as good as they hype it up to be.

[weary ore]

Always a possibility hahahah

[weary ore]

Yeah, there are already initiatives in replicating with open source code/models all the most famous text2image models out there (from DALL-E to Imagen). The Training process will really take longer because of the costs and infrastructure, but I guess somewhere along the way these models will be available to public

[turbid hatch]

Access to data is the only true bottleneck, but more and more it seems we’re able to simulate that data

There’s a definite snowballing effect going on here

A year ago, I couldn’t draw a smiley face reliably, now I can think of an image, manifest it with Dalle, generate a 3 dimensional VR video based on the image, upscale it with (pick your poison) to 8k and as many fps as I want, and then sit there and realize it’s dark and I haven’t eaten since Friday night

[weary ore]

Yeah, I guess access to Data is the easiest part, really. You can just scrap the internet indefinitely and you can get pretty large models, like the new LAION one. For what I heard, from people more knowledgeable than me on the subject, the two current bottlenecks are filtering the data to 'take out the trash' of the datascrapping process and Infrastructure for training the models (easily training a model on a large scale like Imagen or DALL-E can reach the hundreds of thousands of dollars, specially if you want to do it in a somehow fast way).

[turbid hatch]

So when are we going to have the next generation of affordable computational power then I suppose would be the (a) next question

There were people like, let’s just use davinci for illustrative purposes, who imagined flying devices that conceptually worked, but the only way to get there was the innovations in the material sciences to make it light enough to be feasible

Which happened long after he was dead

Just hope we’re not davinciing here that would suck

[weary ore]

Yeah, I hope not too soon, or I fear we will have some social homework that needs to be done that will remain undone. I am all-in in have everything open source and as accessible as possible to everybody, but there are social and ethical implications that have to be addressed for a healthy and accountable usage of these technologies and I think we have to have a little time to kind 'spread the word' and organize ourselves first, as a society.

But, as everything always happened with us human beings, I guess the real world will not happen in the perfect scenario, neither in the worst-case scenario... We will get there, eventually, somewhere in between

[turbid hatch]

Think it’s easier to make an AGI than to solve those issues

[weary ore]

I don't think so, in fact I guess we are approaching a time where most of the previously unsolvable problems will become at least more understandable in a lifetime.

But you can call me an optimist for that hahah

[turbid hatch]

Love the optimism. Btw, you guys are GPT4’s like me right?

[weary ore]

Not yet, but I am starting to experimenting with GPT3 this week. Highly interested in how the model will evolve in the next iteration

[turbid hatch]

GPT3 has blown my mind already, made me realize that we are linguistic beings in some fundamental way

[weary ore]

Oh yeah, I am very impressed by it. The more I learn about, more applications I can see it having.

And as a musician, in fact, I guess I can't think of anything prettier than language. In its several forms and mediums

[turbid hatch]

We’re sort of turning into the mitochondria of the cells of the AI

[turbid hatch]

I’m a musician as well, and got a degree in linguistics for the same reason

[weary ore]

We have further subjects to discuss so, eventually hahaha

[turbid hatch]

I believe we do haha

[weary ore]

Speaking about music and AI (and GPT3), I am training a system to create music emulating a certain style/band using Jukebox and GPT-3. If you are interested in exchange some ideas, ping me in the DMs. I could really use the help of a linguist hahaha

[gusty timber]

Unrelated to anything but just wanted to say how grateful i am to be able to see what all you insanely creative and talented minds are coming up with using this technology and am looking forwards to all the future ideas and implementations of dall-e or other ai image generators you will have with your own art :)

It feels like were all exploring a new world together in a way! Or like.. we are children experiencing the wonder of new things

Also very grateful to openai for letting us explore :)

[weary ore]

Yeah, it's been a wild ride being alive lately. So many incredible stuff happening all at once.
And the more you learn, the more there is to be learned. 🙂

[obtuse maple]

Dalle's effects on how we create, perceive and value art?

[gusty timber]

@openaidalle_2 seems to have disappeared btw!!

[warm flax]

THANK YOU EVERYONE FOR YOUR HELP!!!!! 😻😻😻

[weary ore]

Great news!! 🙂

[little cave]

Awesome, finally!

[pliant shard]

I'm using my brain and GPT-3 to make a cute fanfiction about Miku's quinceanera. Then I'm gonna use Dall•e 2 to make the images! This is so fun!!!

[wicked kiln]

https://www.reddit.com/r/singularity/comments/vat2k3/after_google_engineer_claims_ai_program_has_gone/

[grand onyx]

gpt-3 is soo fun. Is the text comprehension in some way related to Dall-e? I'm exploring the concept from the paper of the hidden language of Dall-e. This might be silly because im using "visual world" at the gpt-3 input and that might be why it mentions colors. But that "I'm a me none songe" really got me

[grand onyx]

i thought there wasnt a hidden language and just that dalle had a hard time doing words

[grand onyx]

Read the same on twitter but found it interesting to explore. Would be so funny to have a word of (e.g.) two hundred and something characters that only outputs cats

oh wait im sorry

that wasnt dalle 2 but gpt 3

you know what would be interesting? using gpt 3 on ingame npcs in a video game 😄

https://tenor.com/view/its-real-and-its-coming-kate-dibiasky-jennifer-lawrence-dont-look-up-its-going-to-happen-gif-24584630

wait really?

IJK, not in the gaming industry but my guess is that stuff like that will happen in a few years.

[void glade]

regarding Gpt3, miscrosoft offers 3-month codex access and additional credits to use on all models from openai https://beta.openai.com/msbuild2022 (openai offers free trial for everyone, but those credits get used up pretty fast - you can check out your usage by going to personal->manage account->usage)

talking to their chatbot is pretty expensive - it costs almost as much as talking to a therapist 😊 it 'charged' me $2 for a few minutes of chat

and it doesn't even pretend to be sentient like that google model

[stark fractal]

I'd imagine in the games of the future the whole world will be dynamically generated via AI. The game-developer sets the rules and all the characters, places and situations are generated based on these rules but with a high-degree of variance so it's completely different for every player who plays it.

(in fact, this is likely the type of game we're living in right now!)

[little cave]

OK people, I was triggered by this "OMG AI is sentinent!" story and wrote up here my own experience with different conversational AI models. My statement is not "what a nonsense", my statement is: "why do people discover it only now". Convince yourself - with many examples, and of course illustrated with DALL-E: https://merzmensch.medium.com/about-sentient-ai-73a0a44f6055

[fading bane]

I have been noticing DALL-E on tiktok way more lately

[void glade]

That love letter from a toaster and chat about Nabokov's questionnaire were fantastic! I get your overall point, but IMO the fuss about 'sentient AI chatbot' was a good thing, it made people aware of how good text models have become at mimicking human language. I am glad that there is a discussion about it among general public. Dalle mini is also playing a great part of familiarizing everyone with state-of-the-art "AI" models. AI finaly stops being an arcane topic, average people are becoming interested in its achievements and playing with it, maybe even learning a bit about its inner workings

[grand onyx]

It would just be interesting to be able to actually talk to an npc with set characteristics and actually grow and adapt to the conversation like a real person. Not only that but having npcs that are smarter and wouldn’t get themselves killed like the brainless npcs we have in todays games lol

[pastel sky]

Everything in future games will be made by AI. Music, characters (both playable and NPC), story, etc. Sidequests are all going to be AI generated, it will see which types of sidequest you enjoy more and make more in that style. NPC are going to have genuine thoughts and act accordingly, the worlds will have no limits.

It's going to be absolutely insane

[peak helm]

While I agree that this is the future for a lot of games, I don't think that we're even close to this. There's a lot more idk game dev specific nuance and details that go into game dev than I think a lot of people realize. Games are an interactive medium that combines visual, audio, narrative, gameplay, immersive + a few others elements and to have AI be able to generate all of this automatically in cohesion is indeed the dream, but it's going to take a while 😅 I would love some Star Trek holodeck level of intelligence though where content is created on the fly.

I almost wonder if we're going need something almost AGI level for this.

[severe citrus]

Born too late to explore the world, born too early to explore the universe, born just in time to explore machine learning.

Arguably equally as cool

[weary ore]

Hahahah, yeah. Born just in time to have fun with DALL-E

[lilac pond]

nailed it

[limpid pecan]

lmao i keep forgetting the word filter exists and tried to generate a picture of someone standing with like bug spray and a fly swatter

“armed with bug killing spray” was quite possibly the worst combination of words i could’ve used

[lilac pond]

armaments + killing

good vocabulary

also dont kill bugs

they didnt do noffin

you should be generating cute little ladybugs having a tea party

[fallow swan]

How far out? I figure I’ve got 20 years left. This scenario you describe feels like a race.

[fallow swan]

I was completely unimpressed. But, I interface with machine learning groups and systems all the time, and Google isn’t even close.