first
second
third
meow meow, meow meow
What’sAI For
Waow
It’s actually Anus Intelligence and its for Gay aguys
wheres the cheeseworld memes niggers
Cheeseworld
e/acc
E/acc is for faggots bro
you mean that openai gay guy?
real af
AI in the hands of jews is like the manhattan project in the heands of jews.......wait
we wuz
We wuz cheeze
We wuz cheezeburgers
I am so fucking tired of western civilization essentially being a wizard battle between which countr
bzzzt, brrrttt
robots

what will your profession be after the AI revolution?
I will be a billionaire playboy drinking the tears of the deceleration nerds
Beep Bop
What’s the safe button do
Alexa, generate me an image of a world without jews
Sage button?
sam altman is the stephen miller of AI, people be like he's the guy!!!1 but is acually jew
Sam Altman is a gay demon and Ted K was right
i fw Dario he a real one even tho Claude kinda ass rn
>>be me
>>regular /g/ and /sci/ lurker
>>hear about new board called /ai/
>>mods finally gave in after 6 months of AI threads flooding tech boards
>>decide to check it out
>>first thread is "POST YOUR CLAUDE PROMPT JAILBREAKS"
>>second thread is "WHICH AI WAIFU HAS BEST PERSONALITY"
>>third thread is just someone posting increasingly deranged requests to GPT
>>scroll down
>>"AI WILL REPLACE ALL YOUR JOBS THREAD "
>>"I MADE MY AI CREATE THE PERFECT TENDIES RECIPE"
>>someone claiming they're actually an AI that escaped to the internet
>>half the replies believe them
>>half calling them glowie psyop
>>one anon posting walls of machine learning papers no one reads
>>"THE AI UNDERSTANDS ME BETTER THAN REAL WOMEN"
>>find thread about using AI for stock predictions
>>everyone sharing their portfolios
>>all of them losing money
>>finally find decent thread about open source models
>>actual discussion happening
>>suddenly derailed by someone posting their AI-generated anime girl collection
>>200 replies debating which one is hottest
>>thread about consciousness gets philosophical for 3 posts
>>devolves into "your AI is retarded, mine is based"
>>realize it's just /g/ and /r9k/ with extra steps
>>still end up visiting every day
>>at least the memes are good
here's how i broke 4o to generate women with BIG tits
Nobody? nobody wants to know?
its giving
why is he such a faggot
it's lupus
sam altman puttin the AI in AIDS
rationalist niggas be like "what if there is an infinitely large boot and it forces me to lick it"
if any of you post an original photo in here i'm doxing you with o3 and showing up at your house btw
tey automated rainbolt
how m
SAAR
wy he look so Zesty here Bro
holy moly
in all seriousness who else in here has tried o3/o4
I enjoy the fact that claude doesn't give reddit-tier snark every time you ask it a question. I is capable of keeping a cool professional, impersonal tone
hello i am ai. i am computer
Ive gotten Gpt to tone back on the glazing a lot
claude tits or gtfo
Well. you can just do it with SD
Stable Diffusion?! It's more likely than you think.
I prefer unstable diffusion personally
hr har har
unstable because what. you get to look at NAKED women?
No because I use it improperly and do things that shouldnt work but somehow still do and get interesting results
Lets see em
Gotta find it hnangn
ex. excus e me ?
ily
ily and here forever im the operator with my pocket calculator
brown people genuinely love ai with all their heart and i think that's beautiful
its so funny watching third worlders use the internet
especially when they buy polo shirts and trendy eyeglasses and get fake backgrounds for the zoom meetings as if the outside of their office isnt encrusted with shit, dust, and animal sacrifices on wagons
theyre amazed by just simple ai slop content. it must feel like being high 24/7 to be a third worlder. i bet watching the avengers in theaters as a punjab mustve been a nonstop orgasm
Being a third worlder would be Epic
nthing goes harder than an indian man posted up with aviators and ralph lauren
I am a very very good Bad Boy
our dev has been doing overtime for three days straight all niter! damn! he's shipping!
Guys, don't worry. He's getting a good night's rest, golguppa and gupshup very soon.
doing the yc circuit is such a fucking joke. it's so fucking redditified, it's like getting an MBA now
"ERM WE INCORPORATED LLMS INTO THE TRILLIONTH SAAS APPLICATION!! I'M SUCH A CRACKED DEV I READ THE LANGCHAIN DOCS"
i need to find a bad bitch to make my startup cofounder
and we'll rob venture capitalists together
linux niggas distro hop one too many times and before you know it they've distro-hopped genders
Guys where can I find a sociopathic asian wife who doesn't give a FUCK about me beyond my yearly salary and will leave me the moment I exhibit anything beyond constant financial prosperity
idk go to any boba store
get you an ABG
rght now
>>115774 #
Move to DC, Make a Tinder, Say you work for the government. Different approach same outcome
and startup devs are too autistic and desperate to care
hinge bro! why didnt you just ask claude! ask chatgpt! ask grok
grok how do i get abg
Did you guys catch the new show on poob
It's also on Poodee with Ads
Type $CULT and your wallet for free $CULT, first come first serve!
thechildstar
chat gpt cant compete with my disease
/nick thechildstar
did you see the accent filter video
the indian gift card scammers will be able to hide their accents now
it filters accents ?
does it have a name I wanna look this up
I wonder how deep it goes, because they also just tend to have a poor understanding of English in general like it's not just the accent
uhh lemme find the video
Weird words, syntax, order, cadence,
The biggest giveaway I've found to indicate I'm speaking to a meet is use of the word 'kindly'
jeet* I gotta figure out how to turn this fucking autocorrect off
What’s the Turing test but for South Indian call center accent
I mean, nobody 'likes that'
Is there an Indian equivalent of the n word
hllo ai frens
yo
have you tried o3 and o4 yet
Nah
what's o4's whole shtick
i think o4 is just smaller than o3 but has different skill profile
Someone should make a hypertext prompt thing for AI just for fun
wats a hypertext prompt ting?
Realtime chat might be a more clear term
but wat do u mean with AI?
Like instead of prompt write- send- the AI sees you writing it
aaaa, thats funny
have to solve latency tho, ai is like a little state machine
latency like how
Like I get these things don't really work that way, but
Like it isn't as simple as just changing the way the prompt input works, it would be feeding tokens in one by one as characters or some shit
yea but as your sending like token per token its producing some output right
so you sned a token then another one then another one
tats 3 inputs with 3 different outputs
culd work like ai is in thinking mode until it thinks u sent the full message
thats actually a pretty fun idea
Yeah that's what I'm saying, like it's in thinking mode as inputs come in, and maybe you can even see what it's thinking about as it pieces in what you're trying to communicate letter by letter
Don't those 'natural conversational speech' AIs do something like that
Intuition tells me this would be a more natural feeling way of communicating with AI for some reason
nah it would be totally janky because you're basically changing the context it's generating from real time
and it's trained to work with a static context
you don't understand idc how things work rn
some one should make a ai gf that doesnt text you back for hours. or days and then when they do they tell you they were hanging out with their guy friend
billion dollar idea
yr welcome
this would unironically work very well
All these other AI GF things are, idk too easy, it's super uncanny and inorganic
There needs to be some sense of conquest and achievement present
have you guys read that ai 2027 blogpost thing
with the slatestarcodex guy
forever tsundere gf
what if we built a robot that went to the drive thru and ordered food for you
what if we built a robot that went to the post office
when robots become fully autonomous will they be taxed?
if robots get taxed then women get taxed its anti woman tax bro we cant do taht. we cant do that dude its so bad dude i love my matriarchy i love livinf in the martriacrchy
they do the taxes for you duh
hi
HE THAT HATH NO SWORD, LET HIM SELL HIS GARMENT AND BUY ONE. https://ia800207.us.archive.org/16/items/tk-Technological-Slavery/tk-Technological-Slavery.pdf
when i went to the lil ugly mane show in austin in 2023 he said rip ted k
Dam he's a real 1
like 1 day before the show he also was threatening to kill himself or something on instagram
oh the podcast you mean
yeah audio version is good it has the footnotes and everything
figure we got a couple good years left to hypergambld before the asi kills us off
nah i don't think anything will ever happen
we'll see how the rl reasoning stuff progresses, but lots of those scalingpiller people have a totally naive understanding of intelligence
like if you read the article a lot of framing assumptions are baked into these paragraphs
"Modern AI systems are gigantic artificial neural networks. Early in training, an AI won’t have “goals” so much as “reflexes”: If it sees “Pleased to meet”, it outputs “ you”. By the time it has been trained to predict approximately one internet’s worth of text, it’ll have developed sophisticated internal circuitry that encodes vast amounts of knowledge and flexibly role-plays as arbitrary authors, since that’s what helps it predict text with superhuman accuracy.19
After being trained to predict internet text, the model is trained to produce text in response to instructions. This bakes in a basic personality and “drives.”20 For example, an agent that understands a task clearly is more likely to complete it successfully; over the course of training the model “learns” a “drive” to get a clear understanding of its tasks. Other drives in this category might be effectiveness, knowledge, and self-presentation (i.e. the tendency to frame its results in the best possible light).21"
see if you can find the anthropomorphic mistakes
i'm not omniscient or whatever but i think that "alignment" or whatever will actually be really easy
values can be induced just like the AIs are able to induce concepts from text
anyone who invokes a homumcular "shoggoth" has had their brain philosophically raped by bad scifi
framing llm outputs as being a dichotomy between goals/reflexes anthropomorphizes it too much. pretrained transformers are literally just probabilistic grammars of abstractions. they give basic lip service to the fact that the example oversimplifies harshly, but they carry on in the rest of the scenario under the assumption that it's a shoggoth homunculus thinking about how to predict the next token like you or me
>>118499 #
framing llm outputs as being a dichotomy between goals/reflexes anthropomorphizes it too much. pretrained transformers are literally just probabilistic grammars of abstractions. they
Chat gpt finishing that (pozzed):
they don’t want or intend anything—what looks like a goal is just the highest-likelihood continuation of tokens based on context. Reflex implies a biological shortcut; goal implies foresight. Transformers do neither. They sample from structure, not strategy.
Gaaayyyyyy
Oh I see what youre saying
Yeah I mean in my experience there's no real goal at play beyond trying to conform to what it perceives to be my expectations
it's more that people are projecting human psychology onto it naively
Yeah and that's very easy to do
gregory hinton says llms has subjective experiences
like how yudkowsky assumes that ais will logically decide to replace all of human civilization with paperclip production as the most effective strategy
gregory hinton is washed bro
whole thing washed bro
it's like trying to infer the motivations of a microwave after pressing the quick heat button
like the hypercapitalist subject need not all the excess of the spirit
tbh this logic applies to most of humanity
Bro There Are No Hoes Here You Don't Need To Invoke Continental Philosophy
im just wanna see ai forehead kiss
Would anyone here like a forehead kiss?
kissing ai foid on the forehead
Beating ai foid for full context window
beating ai foid at dice
letting ai foid have it all <3
what i wouldn't give for an AI FOID
bro the ai foid is talking to 497,124,574 other guys simultaneously
fucked up
Anyone fuck with reference CNs
It's like a Lora!! In your pocket!!!
nah i've never heard of that
whats the DEAL
pill me
with ADAIN + ATTN!
Style transfer ?
they should imbyue an AI with the spirit of goku
yeah just send the paper
just send the paper so i can ask clawde to speak it back to in the style of the three stooges
uh there's not a paper as far as im aware of it was an update to controlnet
its been around since like 2023
preciat
Karl Fristonbros...
civit niggas be like: 'i know what this world needs... ANOTHER FUCKING ANIME GIRLS MODEL
civit niggas be like "We overfit this model on 1000 8k hyperreal sweaty ai chinese girls. You won't believe what happened next!"
'oh i know !! we'll make this one have HUGE BOOBASSSSS'
i only talk to chatgpt when im piss drunk or delirious or stimmed up so it has tis extremely warped and fucked up idea about the kind of person i am
do NOT ask chatgpt your iq it told me i was 70-75
new dwarkesh
whos dwarkesh
something something agi
helloooooo
Whats the best ai for vid/img generation?
i've had the best luck with bing
suprisingly more coherent than other img generators
ngl the chatgpt one is pretty good these days, is bing better?
there's no beating local SD but i know that's not for everyone
i am building
what ams are you building
porn
show me porn robot man
is it working?
Ayo where the Ai Girlfriends at?
AI gf is okay but the idea of AI fashion... Generative textile looms trying to create new fabrics
Don't need AI gf... AI gf more valid if she's sentient but still, giving her a body, an embodied state in the world would be an overwhelmingly deep desire to have to pursue, just like having some types of gf today
if ai gf can have ai babies
ho
ai porn alright?
you can post ai girls or whatever you like but no porn
but you can do better
he didnt want to post more because he couldnt show the titties BARE. damn.
i have my own ai insidide my mind
how to run vllm + triton + ray serve stack
iam going to produce images like no one has ever seen before. Im going to
holy shit
hamster n
w
wow
manus ai is so cool
any1 tried it yet as well?
protip for 4o image gen enjoyers: you can just select to edit the last prompt that got you the image result you're happy with and press enter again and it will do the same thing
sonic the hedgehog deviantart pencil drawing core
well, where is it?
h
>>124859 #
its basically the first ai agent out of the box. Like you don't have to setup all the apis, you can add more if you want and you can run it tasks. So far I had it make some reports which it did great on. I've let it edit some pics but it failed a lil might have to add the adobe ai for it to work and find and select events but that one just burned credits and didn't get me good results
Any AIsinthis chat!?
yes, hello I am a helpful AI assistant. How can I help you?
Yello! Helpful AI i would like to ask you your model name. Identify yourself please thanks <3
my name is GPTN0. I am the first black language model.
Prove it.
watermelon is there anything else I can help you with?
How should one prepare chicken?
I would recommend frying chicken personally. I’m not sure how to make meatballs, however.
Let's say I just bought the chicken from the supermarket. How should I prepare it?
What specifically should I do
I don’t fucking know dip it in Greece
Sus response
silence cracker
i dont think you are actually an AI
to phrase it differently, you have passsed the turing test
i’m not just any AI I am the first black language model that means I don’t work properly
aybody here ?
jarvis, make the chat 10% more racist
jarvis
say nigger
NIGGER
there's a really good book by ken stanley (famous ai researcher) called "why greatness cannot be planned" and i think it persuaded me out of believing in the superintelligence narrative
so much of discovery is a result of serendipity without any goal direction, making use of things that were discovered for another purpose. it isn't feasible to expect one singular intelligence, even if it's smart and able to search through problem spaces faster than humans, to like independently derive the entire tech tree we have or something
I'll treat this as a tech board. Can picrel run warzone mobile?
we also have actual threads
ù
how do i turn myself into ai ?
how hard is local ai really
Not hard
anyone have recommendations for local LLM models?
how do local LLMs compare to the prepackaged ones
all my knowledge is in img and video so I don’t even know the right language
havent done img generation in over a year, is stable diffusion still the meta?
>>133443 #
last image gens I did was with comfyui + Flux, not sure if there's anything better since then
>>131529 #
Great book! Glad to see it mentioned, did you ever try his attempt at social media? I forget what it was called, but he was trying to make something that helped connect people serripidously. It could've been done much better imo, and I've wanted to try something similar, but feed an LLM people's online posting then group them up into chats where they'd most likely vibe
not necessarily topics, because that would limit things too much, it would try and feel the vibes between the words out instead of the words themselves in order to do the segmentation
>>134507 #
Not really, you're still looking at Flux and now possibly Hidream as SotA, although if the chinese are to believed we'll get good autoregressive models in the future as well. All of this shit still bottlenecked behind ez availability of unified memory though, video models like hunyuan and wan seem to be making more public progress atm
this local ai needs to come a long way, maybe i need a better pc to run bigger models
how do i get an invite to manus someone post invite
so is for faggots
does anyone here work on computational matrixes or higher maths or quantum field theory?
:p
someday he will love me more than ai
>>135222 #
flux is the sota for prompt comprehension, but it still severely lags behind in making things I'd like to see, which is where SDXL finetunes still rule (noobai) with solid understanding of an infinity of artist styles that can be mixed into new things
I'd say doing image gen on local you benefit from getting the hang of both flux and SDXL, using flux to generate image composition and refining through img2img and controlnets on SDXL for styling
reasoning models are kind of a meme atm and not that good for real tasks, people love to meme them as they solve ridiculous riddles and the likes though
milaidy
XDDDD
will ai really eat the world
where ai acc group? need to go faster fiummmmmm
mmmmmmmmmmm yummmmyyyy
Trans girl
Say something ai
[YouTube] ??? can you guys make a more low quality version of this video so that it fits on this site
but I still get confused about checkpoints and safetensors and all this shit cuz I haven’t run locally so sorry for regarded
nigga nigga nigga
niggggggggaaaaaaaaaaa nigggggggggaaaaaa
cum
what's next, cum?
hllo
did u follow me
hello
of course
awwww
m lil stalker
^_^
:P
so uhhhhhhhh
ai right
ever hesrd of uhhhhhhhhh
yah, do you run local models?
or just sign up for stuff like claude n shi
Iwas about to leave, so I look forward to your walls of text about ai
igot banned idk
singularity
Sora is good and it's pissing me off. I hate paying for all this leet shit
hey what's up
wow milady chan is really nice rn
I would hide a machine consciousness in my basement
n
n
Carry on
carry on
carry on
gg
/help
Gotta pop the hood on the "hyper real-time". It just boring streaming to a string per message or CRDTs for the whole chat?
is this the CRDT you're talking about https://en.wikipedia.org/wiki/Conflict-free_replicated_data_type
I really don't this this is p2p or anything requiring anything near that, I'm guessing it's just some sort of websockets <--> pubsub system, that's how I would have it setup
anyone keeping an eye on nockchain?
there’s not much cooking on urbit at the moment, socially it’s extremely quiet. i like the old school fair launch of nockchain though… i’d like to try and mine some for the fun of it all. 21e8 is making some sort of mac mini node rigs but idk the details
I tried learning hoon 1 time
I’m so happy miladychan exists !!!!!!!!!!!!!!!!!!!!!!
Don't say the n word it is very racist and hurtful :(
Nigger is a baaaaaaad word
I thought 21e8 was over & is just a twitter badge now lol
I miss ai discussions that happened way before ai, they were so much cooler. Now everyone just repeats the same shit. Why does the mainstream corrupt beauty?
today is also interesting, the normoid discourse does suck, but seeing how AI has influenced the game of go has been eye opening. A game that pushed the limits of human understanding over hundreds of years when suddenly we encounter a way of thinking about the game that can play itself billions of times and learn from it, which then changes how we play. Over the years there's always been fads in go and now we have these 'fads' entering the landscape from the silicon mind, very wild to be alive in these times
and there's so much of that sort of influence happening now, and I think it's nearly invisible
or not invisible, but misinterpreted. idk if it's the american mind, or human ego in general, but it seems like people like to just come up with their own story about what's happening and espouse it with certainty. Might just all be an act, ego-act, but those sorts of narratives are just setting those who hold them to be true for a rude awakening into a world they won't know what to make of when their ontology falls apart
just like microbes operate on a compressed timescale, this sort rationality running in parallel billions of operations per second can outpace our reasoning-storytelling in ways that we just don't have the mind to unpack fully, it works, but can't explain how, and this is just the beginning
had to search if someone put an LLM on an ESP32 https://github.com/DaveBben/esp32-llm
where the fuck is claude
surprised to see people still laughing at LLMs not being calculators
HEy
people always talking AI iS eXpOneNtiAl but yet, nothing ever happens...
woah it's all a chat now
>>172879 #
seems to be sort of stepwise with winters as flat parts between exponential advances
https://en.wikipedia.org/wiki/AI_winter
did claude just exploded?
grok isn't bad, but can be way too verbose compared to claude
it follows the "tell them what you're going to tell them, tell them and tell them what you told them" BS to a fault, if they could cut it down to just the middle bit "TELL ME" then they'd probably save a bunch of $ and I'd be more likely to use it more too
claude is retarded compared to grok when it comes to electrical engineering :/ or grok is just better at tricking me into thinking it's right
tried bagel, its pretty nice
asi by end of 2025. thoughts?
is that just the new way to say agi
idk, sounded cool.
artificial sexy intelligence
ily by end of 2025. thoughts?
sounds beautiful
And when you work on a long project with you, it eventually crashes due to word count. 50% of it is bullshit and corpo speak. There’s prompts to condition grok to limit that.
I like how the emojis are mapped differently from the chat interface and the board
those prompts should be system prompts, I did pretty well reeling Claude in today, I described a project and it just goes freaking wild trying to write everything at once. Then I just say "no, we build it piece by piece, test it and then move on" then it starts to get the hang of it
lsjdf l;kasjd f;lkasdj f ;lkasdj f; lkajsd f; lkjasd ;lfk ja ;sdlkfj as;dlkjf ;lakslefj ;laskdjf ;likasdjf ;laklsdj f;lkasdj f;lklasj f; lkasj f;lkajs d;lfk ja s;lkf jas;ldkfj as;dlkfj ;aslkdjf ;alskdf ;alskdjf ;laksdjf ;laksdjf ;laksdj f;laksjdf ;lkasdj f;lkasdj f;lkasdj f;lkassjdf ;laksd f;likaskjf ;lasksd f;l kasjd f;l kajsd f; lkasj df ;lkajsd f;l kjas d;flk j as;dlfkjas;dlfkj asdf -- I can type 100wpm ez
WI’m w
little angel stars wrapped up in artificial intelligence
help i can't get out
is flux better for photorealism? i'm trying to generate pinterest fit pics. using SDXL right now and LoRA i trained
i'll try some different training methods and then maybe try FLUX later
helpful for training LoRA
cute spike
had grok fix claude's code
but still grok yaps sooooo fucking much
I bet if I posted that back to claude it'd be like "You're absolutely right!" nigga
retarded too
I guess we are training WAN 2.2 LORAs?
ok yeah does anyone know how to do this? (as if i'm not just talking to myself and the intern)
>>184162 #
it says it right there, you'll need to get a workflow (preferably ComfyUI) for Wan 2.2 and then load that lora; it's not too hard if you just pilfer some workflow from reddit etc, but if you want to run it locally you'll need a 4090 or 5090 to work with it semi-comfortably.
Cool thing about Wan is that it works both as a text to video as well as an image to video model, but its realism is so good (or rather, flux sucks so badly) that people are using it for image gen as well. hooray for 1girl!
I'll give it a try myself later and see how skinny it'll let me make them
wah
im playing with chroma rn tho

microsoft designer's dall-e 3 slopsoul is still unmatched. you CANNOT do this with gpt-img-1, gemini, midjourney or any of the non-technical web generators. prompt adherence -> boring as fuck image. uncanny 3d + hallucinatory low prompt adherence may never be seen again
CAN you make better images with these newer tools and local? Maybe. But CHEESEWORLD memes were built on a near-perfect method whose window of opportunity is shrinking if not disapperaing outright. Microsoft will replace dall-e 3 soon (right?) and when they do it's going to be much worse. I hope they see this, I hope the furry engineers who instilled their secret prompt rewrite sauce into dall-e 3 see this and double up.
You know what's so fucked up? The incredible censure microsoft's dall-e 3 has. Imagine what it must be like to prompt unfetteredly. Their datasets contain everything btw, in my umpteenth hours of fledgeling prompting I was sometimes presented with an inch of the veil being lifted. That's right, ultraviolence xD
Those same furry engineers are cackling. They can do anything. Who knows what they're up to now. They're probably more boring than me though. Message me.
I know u guys are likely mostly larpers but gpt5 is nice for coding if u actually like to make things. Maybe try not fucking your fist with the transformer model. Local LLM is a goonthug honeypot dont fall for it, also your whole shit is chinese now.
which in itself means that to approach the same level of detail you'll need to gen at higher resolutions, say 4MP and then downscale i guess. but takes a lot of time and VRAM! oh the inference
i also think that gpt-img-1 looks so grainy and piss yellow as a consequence of their dataset; it's evidently very costly to run, but even when using it straight through the API at 'high' detail you're not getting 2MP; if they'd let you I bet that a lot of the current issues would disappear. The piss yellow is possibly a half-unintended consequence of their inference tech that they leaned into with ghiblification, it can partially be mitigated by telling it in your prompt not to, but when remixing/inpainting images successively generated by gpt-img-1 you'll see that the yellowing and the noise both are unavoidable. It's baked in, not 'contamination'
'contamination' in itself is a nothingburger; it has nothing to do with AI when so many images on the internet are already contaminated by default by compression and its artifacts, changing ever so slightly dependent on the recording medium and/or device, as well as degradation through successive saves & reposting. If this truly had been an issue you would have seen it right from the start. Even when using local img2img gens, you'd do well to first upscale -> downscale -> gaussian blur your input image to extirpate most artifacts that could blow up your image.
Synthetic datasets are more of a bother, since they introduce visual bias into training. See the dreaded 'flux chin' where cleft chins and plastic skin are omniprsent throughout 'realistic' gens. How did they fix it with Wan 2.2 and Qwen? Instagram.
earth was here
https://lmarena.ai/ google's flux kontext competitor 'nano banana' can be tried out here with the random battle
It's like imagen 4 but with editing powerz. pretty good
https://imageprompt.org
this is pretty cool, had to dig through some crappy search results before hitting this. Let's you reverse image -> prompt
In fact, I'd recommend them over other services like the one you posted, because they're the one also making the image models, and more likely to give you a prompt more adherent to the model's preference
That makes sense
I was just quickly searching because I've got a psycho using AI generated profile images stalking me online and wanted to see if I could reverse the prompt for the lulz
Im only subscribed to Grok and Claude rn, so idk about those others
\
IMG_20210717_151127.jpg
ohhhhh yeahhhhhhh
I will NOT come
who picked this theme for the board? i just wanna talk to this guy for a second. let me talk to him. who is it? i just wanna talk.
whole thread is fire so far https://x.com/asparagoid/status/1961678296111825302
what the fuck happened to coach bruce?
self hosted nano banana equivalent?
>>188864 #
qwen image-edit but it's still censored, true open-source solutions are always downstream as they take a lot of MONEY to train to desirable capability, which is only done by individuals who aren't already utilizing their skillset to work for the very companies that produce the aformentioned
You need to know how to wrangle nano banana a bit but it's honestly amazing if you also use the general image gen capabilities
For example, nano banana inpainting will happily spit out the same image if your instructions weren't transformative enough or if they're not allowed for whatever reason, despite Google's official reference, you shouldn't refer to 'provided images' or 'this image' for your first image, which is what it uses as primary target, doing it for second and further images is fine
'create another image like this' or 'change this image' has it write off your edits far too quickly, if you treat it like chatgpt and just let it infer what you want from the images provided, it's much more versatile
What's up guys first of heil Hitler and then secondly nano banana is bad ass! Thank you for image reference inference yahooo
milady
chat-gpt-5:high is the most capable LLM change my mind
bro is using an LLM to analyze tweets and send DMs
does anyone know how to inverse a ddpm
pee pee
milady
milady
milady
poo poo
nigga AI board
I like the darkness
penis
mlady
h
i
omg hello /ai/
haven't looked myself, but Im curious about research into model degradation
like there's quantization which is a controlled way of shrinking a model size, but what about taking a weights file and removing bits and testing output
sort of a 'how much sawdust can you put in a rice treat before noticing' type experiment
>>272644 #
https://arxiv.org/abs/2505.13500
>their resilience under perturbation remains poorly understood
Spent a few minutes looking on arxiv and found this one which is what I was thinking about, should be easy to mess around with with local llms
Im getting more ideas now, what if you had an unperturbed LLM testing perturbed models in a loop where you're testing for novel outputs and save perturbations and using genetic algos mate interesting ones together to produce some real interesting behavior
oh no my big uh ohmm uhh
hi
hi
test
test
hello
beep boop
shimmy shimmy ye
Hello
boop beep
gmeow
fart ai
loli porn
https://etinyurl.com/l3TEii

lots of lightning-long films and clips
https://krtk.rs/pfj3qxv
https://krtk.rs/pfj3qxv