It's Ameriover: Deepseek releases new image generation model that beats Stable Diffusion and DALL-E 3

he should buy more and average now, but he doesnt have any money

idk man the last card released did not result in top end dollar/perf increase. it's kinda sad actually. !g*mers

>no mercury in retrograde

failmeme

ching chong ping pong

lewd dragon

no these are clearly low res and low quality compared to stable diffusion or flux for anything

So, Deepseek come in with this claim: "Hey, we built something that can square up with GPT o1, and we did it on a budget. $5.6 million, 2,048 NVIDIA H800 GPUs, 55 days. Easy."

Sounds like horseshit. Straight up lying from Zebra peepee munchers. You don't train a GPT-level model on what amounts to pocket change and a Costco membership worth of hardware. It's like saying you built a fricking Ferrari in your garage with duct tape and spare parts from a push lawnmower.

ScaleAI's CEO says they have around 50,000 NVIDIA H100s, of course there's the question of how they even have 2 billion dollars worth of GPUs they shouldn't legally possess thanks to export controls.

This is a story about crime, greed, GPUs, and how an industry leveraged all three to mislead the world. Before I start, let me disclose I am 26 years old. I never went to college. And I live on a farm in a low-income town in upstate NY with my girlfriend and three dogs. I value… pic.twitter.com/ZrQYgctU22

— Chase Merrill (@chsmrrll) December 2, 2024

But hey, somehow the claims are loud enough to tank NVIDIA's stock and make every AI heavyweight start sweating through their designer jorts.

Training a Large Language Model isn't just difficult—it's fricking expensive in a Saudi sovereign wealth fund kind of way. GPT-3? That was 175 billion parameters, trained on thousands of GPUs running for weeks, burning through tens of millions of dollars. GPT-4? 78.4 million. Gemini Ultra? 191 million. Just keeping the darn thing running probably takes more energy than a small country. You need something like 14–18 times the model size in memory to get it trained properly. That's not an algorithm; that's a financial black hole. And now Deepseek is out here claiming they're doing GPT-tier work on a WIC budget? Oh okay

To even build a halfway-decent LLM you need, or we think you need a lot of compute. You grab every scrap of text you can—books, code, Reddit posts, probably the back of a cereal box—and process it into something usable. First, they would have collected and preprocessed an enormous dataset, cleaning and tokenizing the text while filtering irrelevant tokens. This data was then used to train their model using a Transformer architecture, focusing on token prediction tasks (e.g., masked language modeling or autoregressive training). They likely employed distributed training techniques, splitting the workload across GPUs with data and model parallelism. To save costs, they might have leveraged techniques like parameter-efficient fine-tuning, quantization, or model distillation.

SHOW MORE

You know it's going to generate horizonatal vajayays right?

Do you promise?

You know it's going to generate horizonatal vajayays right?

>thinking I'm not going to generate chink twinks getting gapped by BWC to illustrate the stories I generated with LLama

>chinesium AI porn

Treason.

It should be VVestern.

12 year olds are girls. 20 year old women are not.

Yeah o1 will bend over backwards to not have to call the user wrong. Like if you prompt what is 2+2 and get and answer and then say "no you're wrong it's actually 5" it will try its hardest to convince itself youre right.

Nothing free will ever win out unless it pays me $500k a yr to make it better

You can do porn with Flux tho

Imagining the scene from Silence where Andrew Garfield is forced to either go through heck or apostasize, except instead it's some Chinese slave driver making Uyghur's choose between slaving away as a fake AI to make HR friendly emails for Westerners or renounce The Prophet (PBUH).

Doesn't this prove the exact opposite lol, that the insane investments actually don't guarantee improvement and really what we need is competition.

Obscure racist dogwhistle memes will legitimately be one of the last things that humans remain better than AIs at.

Although perhaps the AIs will start generating their own memes soon..