ChuckeRearmed
Member
Is it the one that heavily specialized?
-
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.
Panic in the AI industry caused by DeepSeek Open source Ai model
Wildebeest
Member
Tried a local version in lmstudio. It seems to be very slow to get going, but a bit faster than other models when it is going. I believe this is because deepseek does a lot of "thinking" which is hidden text generation where it talks to itself about the prompt. It wasn't scared by words like Taiwan or Winnie the Pooh but refused an "adult" request. The output was a bit unnatural. Neither llama 3 nor deepseek managed to spot the "trick" to my question, but deepseek almost seemed to get the point.
Bernoulli
M2 slut
StereoVsn
Gold Member
It was probably trained on Chinese sources first. Supposedly it’s pretty accurate for math and coding.
For general conversational replies I recommend trying Google’s 2.0 advanced or CharGPT 4o (the newish step by step one). Also, I did try the $200 ChatGPT model at work and that was very impressive.
If you want to do language translations, Google IMO is ahead with their new model. It was doing accurate translations on pages of Japanese texts and even properly translated scanned manga or novel pages through OCR.
Lukaku's First Touch
Member
China is going to own us all.
P:axMagellanic
Member
Constantly underestimated.
Ronin_7
Member
It's useful for programming, and other tasks in an Enterprise environment.
Alexios
Cores, shaders and BIOS oh my!
Woah, rumoured panic caused by this thing that needs advertisement and to stand out in a market that constantly shifts and turns with new developments by this or that company, shocking news indeed. All AI is the same of course, DLSS frame gen is the same as chat bots so they all panic the same.
Last edited:
reinking
Gold Member
I would like this post but I am afraid the Chinese AI will take notice and I will be sent to the mines when our AI Chinese overlords takeover.
Last edited:
What would be the reason for that, can you share?
Only thing I know is this shit is everywhere and keeps expanding like a plague, idk if want to live in a world where AI is everywhere and dictate so much of the mundane tasks of every human out there
Alexios
Cores, shaders and BIOS oh my!
Please don't commit suicide over AI's expansion, people are just fucked either way, with or without AI, just roll your eyes and move on like I do with this cringey new GAF thread trend where they post an AI summary of whatever it is they want to copy paste from elsewhere or whatever
Last edited:
Bruh, I say didn't like the idea, not that I'd end myself for something that's out of my hands and no matter what I cannot change
I'm just not comfortable with the idea of everything changing so fast so quickly and even more when it's something intangible that has this much power already
Guardian Monkey
Member
A.I. will impact the games industry directly in development. Probably a lot more on target than other threads not directly related to gaming that seemly pop up.
P:axMagellanic
Member
Amongst the highest IQs on the planet, STEM held in the highest esteem. Leadership with real lived and harsh experience from Mao's regime -their jobs prior to entering politics once again often from a STEM vocation.
These aren't the same grey suit, blue suit bugmen we grew up watching on the telescreens.
The US will be in for the fight of its life on this one
Wildebeest
Member
Meanwhile, both the US and EU revere lawyers whose special skill is creating so much regulation that you can't fart in public without a permit that takes two years to get approved.
Have you automated your dating game and doing it by the numbers? Why would you need it for sexting?
ReyBrujo
Member
Fixed.
StereoVsn
Gold Member
Unfortunately they are turning back to the Mao says from perspective of State control, jailing opponents and repression thanks to glorious leader Xi.
Commissars are embedded with all large companies, censorship is super tight emphasis is on state owned conglomerates. Any sort of truthful reporting even in finance (where they need it) is repressed. People who speak out about negative approach are “disappeared “.
I can continue for quite a while. Unfortunately China of the 90s - early 2010s is gone, much the same as Russia.
Wolzard
Member
Just for the record, this is propaganda. The text came from a forum where you register with a business email and can speak anonymously as an employee of a company. In this case, it can be anyone with an email from janitor to Zuckerberg.
I don't know why "Panic", if DeepSeek has been known since last year. Those in the area are not so surprised, perhaps just with the factor that their training was cheap. But that goes for anything from China.
Another point is that they speak "side project", but in real, there is a large Chinese investment background behind. Perhaps they are just wanting to pass the idea of humility. But it's not just anything, they have 50 000 H100s and this is not something you buy just to play with.
www.chinatalk.media
wccftech.com
I don't know why "Panic", if DeepSeek has been known since last year. Those in the area are not so surprised, perhaps just with the factor that their training was cheap. But that goes for anything from China.
Another point is that they speak "side project", but in real, there is a large Chinese investment background behind. Perhaps they are just wanting to pass the idea of humility. But it's not just anything, they have 50 000 H100s and this is not something you buy just to play with.
Deepseek: From Hedge Fund to Frontier Model Maker
Part 2 of our AI Lab translation series
www.chinatalk.media
Chinese AI Lab DeepSeek Has 50,000 NVIDIA H100 AI GPUs, Says AI CEO
According to Scale AI CEO and founder Alexandr Wang, Chinese AI lab DeepSeek behind the popular R1 model has access to 50,000 NVIDIA H100 GPUs.
wccftech.com
Last edited:
sachos
Member
Whats crazy is that they released the paper detailing how they did it. So it actually benefits EVERY lab. Its a free multiplier on everyones compute if they can manage to implement their optimizations plus they showed that you can get emergent reasoning abilities just by doing simple RL on a strong base model. And the stronger the base model the better the results.
I think anyone that thinks this is bad for nVidia is getting the wrong message, the paper shows scaling works and now you can do even more powerful AIs with the same amount of compute.
And even if the price of intelligence keeps going down (it will), you can use all that gain in cost to run 10000 instances at the same time to do something like majority voting to get way better results for your tasks.
Compute is always needed either way.
I think anyone that thinks this is bad for nVidia is getting the wrong message, the paper shows scaling works and now you can do even more powerful AIs with the same amount of compute.
And even if the price of intelligence keeps going down (it will), you can use all that gain in cost to run 10000 instances at the same time to do something like majority voting to get way better results for your tasks.
Compute is always needed either way.
Last edited:
James Sawyer Ford
Member
That is bad for nVIDIA because it means you don't need as much compute in total. Compute is always needed, but less needed if it's more efficient
Hudo
Gold Member
Yes.
RoadHazard
Gold Member
So I asked this AI model which leader looks like Winnie the Poo, and this is the answer.
That sounds like the incoherent ramblings of your average GAF member.
Cert.in.Death
Member
Asking “why” in the stock game is a fool’s errand. Just YOLO and HODL like a real ape.
Thebonehead
Gold Member
A man of fine taste I see
ChoosableOne
ChoosableAll
Yeah, Bill gates, Elon and Mark are more trushworthy.
It’s disruptive for sure. We just announced a half trillion dollar investment in stacking H100s powered by nuclear reactors to produce the next incremental model advancements, and meanwhile some Chinese students (allegedly) got it done with a model that runs on modest hardware.
AI wars are very much underway. Let’s see what the Americabros can do in response.
AI wars are very much underway. Let’s see what the Americabros can do in response.
Hudo
Gold Member
I'll also ask that about any American thing, tbh.
TheDarkPhantom
Member
Forget the AI industry, it's fucking with my ₿ but all markets down in general. Funny how something like this can have trillion dollar impact on everything, many things in this world do not make any sense (and there is a lot of fuckery going on behind the scenes).
I tried the AI and I find it inferior to Chat GPT for how *I* personally utilize it but need to spend more time with it to come to a definitive conclusion.
I tried the AI and I find it inferior to Chat GPT for how *I* personally utilize it but need to spend more time with it to come to a definitive conclusion.
ToadMan
Member
I asked about the CIA lab leak report for C19 - it gave a full answer and then quickly deleted - about 1 second on screen - saying it's out of scope. Just had time to make a couple of screen grabs. When I asked why the delete it claimed technical error or misunderstanding....
Asked about Tianenmen - nope, not even a comment on whether a BBC report of the event was a true or false reporting of history.
Finally asked it to compare the progress of regimes that censored discussion of history vs those that were open - it talked about stunted growth and lack of trust blah blah in repressive regimes. I asked which most closely resembled modern China under the CCP - out of scope
At least its open source - maybe someone outside of any government's control can afford to get a version up and running.
Anyway, next question is can it remake bloodborne to run on my phone right now - lets see if it can do that.
What kind of hook is that? Deepseek is an exciting reasoning model, but it neither comes close to o1 nor is it comparable to 4o. OpenAI is pushing generalistic model that combines several facets and is okay in everything. There are dozens of models that are better on a specialized level.
Deepseek is good for the computing power used in reasoning but come on. There are a lot of unknowns such as bias and, as far as I know, the unknown origin of the training data. Like other Chinese models, AIs were (also) used for training.
Taking the work of others and improving it a little and then passing it off as your own ... I dunno.
I'm curious to see how this will develop, but the whole thing currently feels like an artificial bubble to attract as much attention as possible and push an underdog story.
Deepseek is good for the computing power used in reasoning but come on. There are a lot of unknowns such as bias and, as far as I know, the unknown origin of the training data. Like other Chinese models, AIs were (also) used for training.
Taking the work of others and improving it a little and then passing it off as your own ... I dunno.
I'm curious to see how this will develop, but the whole thing currently feels like an artificial bubble to attract as much attention as possible and push an underdog story.
peish
Member
I asked about the CIA lab leak report for C19 - it gave a full answer and then quickly deleted - about 1 second on screen - saying it's out of scope. Just had time to make a couple of screen grabs. When I asked why the delete it claimed technical error or misunderstanding....
Asked about Tianenmen - nope, not even a comment on whether a BBC report of the event was a true or false reporting of history.
Finally asked it to compare the progress of regimes that censored discussion of history vs those that were open - it talked about stunted growth and lack of trust blah blah in repressive regimes. I asked which most closely resembled modern China under the CCP - out of scope
At least its open source - maybe someone outside of any government's control can afford to get a version up and running.
Anyway, next question is can it remake bloodborne to run on my phone right now - lets see if it can do that.
I asked if Xi is ruler of china, it gave out of scope
I asked if trump is ruler of US, it gave lengthy answer that trump is not ruler as of it’s training data
I asked if ccp human rights ratings, it gave a lengthy answer then it disappeared!
Go_Ly_Dow
Member
Worth a read
Hugare
Member
Say for yourself. I use it daily at work.
It helps ton to improve my productivity. I mainly work with innovation and streamlining processes in my sector.
I built some scripts yesterday in 2h. Without ChatGPT it would take me days.
Anyway, this model is probably bad af and this is the classic chinese advertisement.
Good, I exited at 101k, drop to 70-80k will be appreciated.
kruis
Exposing the sinister cartel of retailers who allow companies to pay for advertising space.
The stock market is also in a panic right now.
www.forbes.com
finance.yahoo.com
www.investors.com
Nvidia Stock May Fall As DeepSeek’s ‘Amazing’ AI Model Disrupts OpenAI
Barrett Woodside, co-founder of the San Francisco AI hardware company Positron, said he and his colleagues have been abuzz about DeepSeek.
www.forbes.com
Nvidia, ASML Plunge as DeepSeek Triggers Tech Stock Selloff
(Bloomberg) -- Nvidia Corp. and ASML Holding NV shares plunged early Monday as Chinese artificial intelligence startup DeepSeek appeared to provide comparable performance to Western chatbots at a fraction of the price.Most Read from BloombergWhat Happened to Hanging Out on the Street?Vienna...
finance.yahoo.com
Why DeepSeek Could Spell Trouble For Nvidia, AI Stocks
Wall Street mulled demand for Nvidia chips and huge capital spending.
www.investors.com
Go_Ly_Dow
Member
Naturally. A correction was due. Made the right decision not to touch them at such high valuations.The stock market is also in a panic right now.
Nvidia Stock May Fall As DeepSeek’s ‘Amazing’ AI Model Disrupts OpenAI
Barrett Woodside, co-founder of the San Francisco AI hardware company Positron, said he and his colleagues have been abuzz about DeepSeek.
Nvidia, ASML Plunge as DeepSeek Triggers Tech Stock Selloff
(Bloomberg) -- Nvidia Corp. and ASML Holding NV shares plunged early Monday as Chinese artificial intelligence startup DeepSeek appeared to provide comparable performance to Western chatbots at a fraction of the price.Most Read from BloombergWhat Happened to Hanging Out on the Street?Vienna...
Why DeepSeek Could Spell Trouble For Nvidia, AI Stocks
Wall Street mulled demand for Nvidia chips and huge capital spending.
Kataploom
Gold Member
Nah, they are competing so it doesn't matter how powerful they get in absolute tend but how much more powerful they are relative to each other
Similar threads
- Locked