ChorizoPicozo
Member
the china show said the deep seek claims are fake news
-
Hey, guest user. Hope you're enjoying NeoGAF! Have you considered registering for an account? Come join us and add your take to the daily discourse.
NVIDIA stock to lose $400 billion ( US tech 1 Trillion) after DeepSeek release
- Thread starter Bernoulli
- Start date
Guardian Monkey
Member
Lol wtf is the china show?the china show said the deep seek claims are fake news
Edit:
that’s actually the name. Haha
Last edited:
ChuckeRearmed
Member
We will certainly one big societal upheaval sooner or later but AI will for sure be embedded into the fabric of society - delivery, dogs, waiters, workers. Factories will produce factories via automated pipelines, AI will do the research and so on. Wars will be waged using machines and so on. However I don't expect things like that to be everywhere - certain countries and areas will stay relatively backward, albeit with mobile phones and digital currency. Hell, in my own home country hospitals still don't use computers much - still rely on pens and paper.
Also history tells that human tribes migrate and invade various countries so even highly develop societies might collapse. There are also potential natural causes like earthquakes and so on.
ResurrectedContrarian
Suffers with mild autism
Overall, it's still riding on US research, so I don't share the doom perspective at all.
It's one more company innovating, which happens to be outside the US, but if you read their papers you'll find all the usual US-based seminal papers cited, prior models, RL techniques pioneered by OpenAI, etc.
We saw some innovation going on in France with Mistral, but that didn't stay at the cutting edge; likewise we see another innovative company here in China, but it is far from the case that somehow the balance shifts because of it.
It's one more company innovating, which happens to be outside the US, but if you read their papers you'll find all the usual US-based seminal papers cited, prior models, RL techniques pioneered by OpenAI, etc.
We saw some innovation going on in France with Mistral, but that didn't stay at the cutting edge; likewise we see another innovative company here in China, but it is far from the case that somehow the balance shifts because of it.
Fabieter
Member
Open ai pioneerd it years ago so yes it should be cheaper by now.
diffusionx
Gold Member
We are going to string up the oligarchs, put their heads on a pike, ban AI, and execute anyone who tries to restart it.
That is what is going to happen because it's obvious they don't have any solutions to any of the problems they are so eager to cause.
Last edited:
IDKFA
I am Become Bilbo Baggins
Not just young people. Anybody working, regardless of age and experience could lose their jobs.
My role as a data engineer could easily be replaced by AI in a few years. All that time I spend monthly trawling through data, identifying trends, writing reports, presenting data and action plans etc could be done by AI in a fraction of the time. I can see it happening soon as well. What was the point of all that strudying, all that time learning excel, SQL and Python. All those hours spent working on my orator skills to present the data and push back against key stakeholders. Years and years wasted to be replaced by a machine.
UBI isn't the answer as people will only be given another money to meet their basic needs. People need a purpose in life and a majority don't want to sit around and not be useful. Oh, but this will give people the chance to be more creative and artistic. They can take the UBI money, then spend the time writing a book, or painting landscapes to sell for money. Well tough shit, because AI will also take that away from us as well. You know that novel you've been writing for 10 years? Well a LLM can crank out the same novel in a day.
I think if enough people are impacted then we'll start to see major push back from the public to fight for their jobs and they're humanity.
ResurrectedContrarian
Suffers with mild autism
It "knows" all about Tiananmen Square, it simply won't speak about it.
The censorship on these models happen at the behavioral level in reinforcement-based post training, not at the level of the massive corpus of training text.
Unlearning the behavioral modes it has been given via additional reinforcement learning would be easy.
A.Romero
Member
The document includes what models were used as a reference, they include a destiled version of Llama.
We are saying that optimization is a continuous process and yes, every iteration is expected to improve, specially when there is so much money involved. They took previous models and optimize it like everyone else is doing in some form or another. Regarding their experience I would say that computer science innovation doesn't always comes from the most experienced minds. Page and Brin were 21 and 22 when they started working on BackRub (it became Google later on) and by the time they were about 22 and 23 they exceeded the bandwith Stanford could provide.
I understand you are skeptic, I myself don't have the tools to test and make sure those benchmarks are real but I also don't think a benchmark that has been continously changing and being led by different models would lie just because.
This particular model is open source so anyone can look into it and retrain it so if anything this would have more ways to be verified to others that are not open source such as OpenAi's models.
RCX
Member
Haven't read too deeply into it but assuming it's fully open source that should be more than possible right?
I've hoped since day one that open AI (not "openAI") would win the race. I'd rather everyone have this technology at miniumum cost than see it used and abused in a small number of hands.
We will create a new purpose which may end up being akin to the matrix. In the short term it will be painful but literally we will be in the pods and eat the bugs as they say. You may not want to live a virtual life but each new generation will be more and more willing to do it. Kinda like our generation is the Amish and the next generation embraces the "tech world".Not just young people. Anybody working, regardless of age and experience could lose their jobs.
My role as a data engineer could easily be replaced by AI in a few years. All that time I spend monthly trawling through data, identifying trends, writing reports, presenting data and action plans etc could be done by AI in a fraction of the time. I can see it happening soon as well. What was the point of all that strudying, all that time learning excel, SQL and Python. All those hours spent working on my orator skills to present the data and push back against key stakeholders. Years and years wasted to be replaced by a machine.
UBI isn't the answer as people will only be given another money to meet their basic needs. People need a purpose in life and a majority don't want to sit around and not be useful. Oh, but this will give people the chance to be more creative and artistic. They can take the UBI money, then spend the time writing a book, or painting landscapes to sell for money. Well tough shit, because AI will also take that away from us as well. You know that novel you've been writing for 10 years? Well a LLM can crank out the same novel in a day.
I think if enough people are impacted then we'll start to see major push back from the public to fight for their jobs and they're humanity.
MarV0
Member
Oh my we just found GAFs secret billionaire and market guru. Are you here out of curiosity to observe and laugh at our miserable peasant lives?
Last edited:
Wildebeest
Member
Trillions of dollars being invested by people who don't have a clue just because numbers go up (or down).
mckmas8808
Mckmaster uses MasterCard to buy Slave drives
Not just young people. Anybody working, regardless of age and experience could lose their jobs.
My role as a data engineer could easily be replaced by AI in a few years. All that time I spend monthly trawling through data, identifying trends, writing reports, presenting data and action plans etc could be done by AI in a fraction of the time. I can see it happening soon as well. What was the point of all that strudying, all that time learning excel, SQL and Python. All those hours spent working on my orator skills to present the data and push back against key stakeholders. Years and years wasted to be replaced by a machine.
UBI isn't the answer as people will only be given another money to meet their basic needs. People need a purpose in life and a majority don't want to sit around and not be useful. Oh, but this will give people the chance to be more creative and artistic. They can take the UBI money, then spend the time writing a book, or painting landscapes to sell for money. Well tough shit, because AI will also take that away from us as well. You know that novel you've been writing for 10 years? Well a LLM can crank out the same novel in a day.
I think if enough people are impacted then we'll start to see major push back from the public to fight for their jobs and they're humanity.
The CEO on Anthropic said this to the Wall Street Journal just last week!!!!
He's right............we'll need to have a real conversation as a human race that most aren't ready to have.
cinnamonandgravy
Member
if the chinese AI claims are legit, it means less hardware if required to produce current top-tier AI... which means fewer nvidia GPU sales.
but even if the chinese AI claims are legit (dont know yet), future top-tier AI may require tons of hardware... or it'll run on a potato, who knows.
timing's odd though... US announces $500 trillion on AI investment, then china shows it can be done cheaply. china trying to save US some money? dont think so.
Last edited:
Topher
Identifies as young
If there was ever an industry that needed heavy regulation then it is AI. AI should be tools that make life better for humanity, not replace humanity.
Wildebeest
Member
I think that only really Silicon Valley nerds worry about their lives having no meaning if people don't think they are the most intelligent form of life in the universe.
Buggy Loop
Member
China won't regulate so USA won't regulate
The dangers of not being first to singularity is more scary to them than any ethical dilemmas.
Topher
Identifies as young
If so then eventually something is going to give and it may not be pleasant.
TheAssist
Member
One of the reasons (probably not the only one) is pricing.
OpenAI wants 60 bucks per 1 million output tokens generated with o1.
DeepSeekR1, which generates equivalent quality, wants 2,19.
At this price the profit margin goes down the drain. OpenAi wasnt profitable before, at these prices its questionable whether all that investment in hardware by Nvidia was worth it. Big US companies like Meta, Amazon, Google and MS have already spent that money hoping they can dictate the pricing for years to come and make it a profitable business.
R1 is open source, so everyone can spin up their servers and hence prices might go down even further (though R1 might be supported by the Chinese state for all we know, just to break the oligopoly on US Ai companies).
So yeah Nvidia can sell to a broader audience (small and medium companies can now quiet easily run very powerful distilled models locally on their premise and don't need Cloud providers like google or amazon), but Nvidia wont be able to dictate pricing as much, because when the margin with AI services goes down, no one can pay these sky high fantasy prices anymore.
Basically R1 has become the new ground level, since its open. There is no need to use anything lesser than it (or v3 for that matter). In order for US companies to keep charging high prices for their API's they need to heavily outperform with their newer models. but news of the past few weeks looked like the new models might underperform. But we will see.
Last Hearth
Member
I'm deeply skeptical as well.the china show said the deep seek claims are fake news
China makes a lot of claims of advancement in technology that are not truthful, let's see how this plays out.
viveks86
Member
Guess I'm going to ask my VP of machine learning WTF he is doing not switching to R1 already. If we can spin up our own servers, we can make it FedRAMP compliant too (though I'm probably going to have to research that more)!
the china show said the deep seek claims are fake news
Wait... this could still be fake? No one in the west has tested it?
Last edited:
Go_Ly_Dow
Member
I won't be touching the tech stocks because if there's truth to the claims about costs/efficiency then the valuations given to the US tech companies during the AI bubble still remain very high and they have further to fall. So far nothing has been properly discredited.
If the the claims are smoke and mirrors then the AI bubble and boom continues, probably to record highs.
So I'll watch this play out from the sidelines, aka not investing in tech companies because of the volatility!
If the the claims are smoke and mirrors then the AI bubble and boom continues, probably to record highs.
So I'll watch this play out from the sidelines, aka not investing in tech companies because of the volatility!
Last edited:
Nintendo's Witch
Member
Nobody knows and don't listen to anybody that says otherwise. Do you like gambling? Do it. Don't? Keep investing in a good, diversified, long-term portfolio and expect 8-10-12% returns at the end of the road if everything goes well.
Last edited:
Fabieter
Member
It's not just nvidia. Most chip companies are down including clould infrastructure companies and energy provider.
StereoVsn
Gold Member
I agree with most of this, but I don’t think this is correct:
“DeepSeekR1, which generates equivalent quality, wants 2,19.”
From what I have seen it’s not as good as o1 or Google’s 2.0 models, but the question is if it’s 80-90% as good but for 1/10th the price, doesn’t matter?
And yeah, I absolutely think CCP is supporting them. There is no way a Chinese company would publish such technical information without CCP approval, same way with Tencent’s or Baidu’s models.
Buggy Loop
Member
The Watt steam engine greatly improved the efficiency of coal-fired steam engines. It made coal a more cost-effective power source....
Leading to to an increased use of the steam engine in a wide range of industries and thus increased total coal consumption
When cars get fuel efficiency improvements, researchers find that peoples have higher yearly travels distances
A modern example, LED lighting. Highly power efficient LED light should lower global energy consumption for lighting but its the opposite, peoples install more lights for extended periods.
As graphic cards become more and more power efficient on OPs/Watts, they still climb the wattage every generations.
An assembly cross-compiler in late 70's was so efficient and quick that when demonstrated to management, the software manager vetoed it because he would he would lose the majority of his 50 programmers. The same company implemented it in another office and the team actually expanded because the costs of development were so cost effective that rather than just supplying to usual companies buying 100+ licenses, they also opened to new target customers who would need just a few and they ended up doubling the number of programmers.
Every day, for decades now, there's a new software / hardware coming in that changes the language, framework, set of libraries and make programming more efficient. But the demand for programmers continues to increase because by and large they are using new and better tools. Tools that reduce cost of software development but expands the market for new applications.
AI ain't slowing down. Like
rofif
said, the critique is more towards the efficiencies of openAI to exploit the GPU farms than the company selling shovels in a gold rush but even that is not seeing the big picture of what these AI companies are doing. "Billions" compared to $6M chinese! omg! Yea... guys, big tech models are a lot more broad and trained for daily tasks than a distilled model like DeepSeek.Its not like suddenly all the research is gone by the big companies either, its benchmarks but model to model the applications vary greatly, especially with closed models. OpenAI or Meta aren't running these GPU farms on a single model for benchmarks, they are trained for taking over human jobs. A DeepSeek-v3 is not gonna suddenly beat openAI o1 for medical and genetics because it was a distilled trained model and cheap, more similar to the likes of GPT-4o-mini or Qwen2.5-72B. Just recently DeepSeek-V2.5 was at GPT-4o-mini score vs API price. Things move. The costs is what DeepSeek excels at for now but again, as I specified earlier in the day, there's gonna be issues with companies using deepseek as they use sensitive data for further training which is a big no-no for most professionals/business. Feb 8th they are also changing pricing on -V3 so things indeed move in AI world.
Last edited:
mayham2199
Member
When enough people are impacted this is exactly what will happen if they don't have jobs to feed their families and have a life.
IDKFA
I am Become Bilbo Baggins
That's a dangerous game.
Once we hit the singularity, there is then no going back. Pandora's box would have truly been opened, but without careful planning and worldwide cooperation, it could easily be the last invention we ever make.
AmuroChan
Member
No.
Buggy Loop
Member
yup
Not saying it’s a good move forward but realistically, they ain’t slowing down. If it comes to slowing down in public eyes, that’s one thing but behind the curtain they’ll be 100% continuing.
ProtoByte
Weeb Underling
I think we're delving into sci-fi territory here lol
viveks86
Member
the china show said the deep seek claims are fake news
Seems legit guys. Engineers are testing it out in the US on our own servers as we speak. Here is directly from my VP of machine learning. This is literally what he gets paid for. I'm quoting verbatim. Not posting a screenshot as I'm not comfortable with it on the internet.
Me: Hey have you looked into Deep seek R1 yet?
Him:
yes. The model came out last week.Me: Are the performance numbers and costs what they claim to be?
Him: Yes. Although I am surprised why the market is reacting to it this way today. By stock market standard, it is a week (eternity) old news. Why did it catch your interest?
Me: Seeing it hit mainstream news with the Nvidia price drop. Didn't know anything about it till then
Him: They spent $5 million to crater Nvidia by $600 B
Last edited:
Nintendo's Witch
Member
my uncle works at Deepseek.
IDKFA
I am Become Bilbo Baggins
Not really.
Reaching the singularity means a loss of control. Once we hit this milestone, it means a superintelligent AI system could take over the process of invention and innovation.
This used to be sci-fi, but now it's a very real possibility.
ChorizoPicozo
Member
there is another China show
viveks86
Member
Yeah people are getting a bit carried away. This is just usual leapfrogging with technology. My company is still training these AI models to properly read a goddamn pdf file with flowcharts in it. Singularity can (and likely will) happen, but not for a couple of decades. By then I'll retire and be fed by my robot slaves... or be food for them. Whatever they prefer.
Last edited:
GHG
Gold Member
That's not the point I'm making, and youtube videos is not the typical use case for the current form of AI, but please continue to go off.
The point is that there is no end game, but anything that gets us progress in a more effiecent way is a net positive and businesses will use that to their advantage, especially if they can do so while reducing capex, even if it means it ends up being a disadvantage to Nvidia.
They claim higher bandwidth on an individual per chip basis but here's the thing that you're missing - I specifically mentioned Nvidia's vgpu which is specifically where their advantage lies in all of this. At scale nobody uses these chips individually. That is where Nvidia are substantially ahead - nvlink gets you up 900 GB/s interconnect whereas infinity fabric on AMD's side tops out at 128 GB/s. These are the things that should matter less moving forward.
Don't take it from me.
Draw your own conclusions from the data:
They are all ultimately using Cuda. What are you talking about? Assuming they are running Nvidia GPU's what is their high level language or choice ultimately compiling in to? Triton literally compiles to CUBIN and you want to reference that while also saying they aren't using cuda?
Nvidia's early investment in the space and their robust ecosystem and software stack is the reason why they are so dominant currently, not total cost of ownership.
Please read the following as an example (specifically, you might find the bolded interesting):
MI300X vs H100 vs H200 Benchmark Part 1: Training – CUDA Moat Still Alive
Intro SemiAnalysis has been on a five-month long quest to settle the reality of MI300X. In theory, the MI300X should be at a huge advantage over Nvidia’s H100 and H200 in terms of specifications an…
semianalysis.com
What are these mountain piles of alternatives to Cuda then? And to be clear - I'm not talking about alternatives that still leverage or ultimately compile to Cuda.
The ease implementation of deepseek is what remains to be seen. But it doesn't even matter much since the methodologies are now public and that's what matters most. Everyone is free to learn from what they've achieved and adapt/implement as they see fit. Which is why the likes of Meta and Apple are well positioned to take advantage of any benefits.
Well I'm just glad papa Satya taught you something new today.
Sorry your V8 supertruck is at risk of being surpassed by a little 4 pot turbo from China. These kinds of butthurt reactions are always entertaining.
Last edited:
GHG
Gold Member
Seems legit guys. Engineers are testing it out in the US on our own servers as we speak. Here is directly from my VP of machine learning. This is literally what he gets paid for. I'm quoting verbatim. Not posting a screenshot as I'm not comfortable with it on the internet.
Me: Hey have you looked into Deep seek R1 yet?
Him:
Me: Are the performance numbers and costs what they claim to be?
Him: Yes. Although I am surprised why the market is reacting to it this way today. By stock market standard, it is a week (eternity) old news. Why did it catch your interest?
Me: Seeing it hit mainstream news with the Nvidia price drop. Didn't know anything about it till then
Him: They spent $5 million to crater Nvidia by $600 B
It would actually be more concerning if today's move in Nvidia's stock (and other impacted semiconductor stocks) wasn't actually because of deepseek.
I'm not seeing any other reason for it and would put the delay down to the claims needing to be verified because of their origin.
Based on today's activities, the benefactors of this news all mostly recovered or even ended green.
Last edited:
ProtoByte
Weeb Underling
I just don't buy that AI will be able to infinitely code itself with no human intervention. Generally, I believe that the more complex the system is, the more fragile. Trip up in the wrong place will lead to cascade failures.
ChorizoPicozo
Member
..... 
3liteDragon
Member
viveks86
Member
Yeah basic human reasoning has not been emulated yet. Like I said earlier, we are spending days teaching it to read basic flow charts. AI code generation is simply going to make development faster. It will be yet another tool in the toolkit and can mayyyy be replace a handful of junior engineers that show no growth potential. Replacing $300K devs with AI? Not something I'm worried about for the rest of my career. Can it happen in the distant future? Who the hell knows. May be. May be we will invent time travel too and be visited by our future selves. Or aliens. It's all... possible...
Last edited:
Loomy
Thinks Microaggressions are Real
The tool is still being assembled, and everyone wants to own a part of it.