We'll have more on the audio feed later this week. Tune in on your podcast app of choice: Spotify: spoti.fi/32aZGZx Apple: apple.co/3AebxCK Etc. pod.link/1522960417/
I agree. From my perspective whether I take what China says with a handful of salt or take what billionaires in the USA say with a handful of salt, it’s the same to me. I already read information from other sources. - east and west. ASAHI Shimbun, the Staits times, the economist and others. Then try to figure out what makes sense to me. It’s going to be difficult for people with little to no basic knowledge of science, economics and politics and desire how to figure it out. People are going to have to suspend judgement, be aware that there are no binary right wrong answers. I see more cults forming. Sad.
DeepSeek was just a side project to use some of the leftover GPUs they had from the main business. The company does have H100's but it was a point for them to use H800. It was also important for them to use only home-grown engineers, and no engineers returning from work abroad. It's just a flex to you Americans, and probably a celebration of the new year and completion of their China 2025 project: "Made in China 2025" (MIC 2025) is a strategic industrial policy initiative launched by the Chinese government in 2015 to transform China into a global leader in advanced manufacturing and high-tech industries by 2025. The plan aims to reduce China’s reliance on foreign technology, boost innovation, and upgrade its manufacturing base from low-cost labor to cutting-edge, value-added production. Mission complete! Congrats to China and humanity!
Yes The word genius comes to mind yes it does matter where you went to school like friends etc. No, I'm not that smart. I'll admit it I'm an ex-marine and yes I've been investing why CEO BYD Wang a genius It's where he did. It makes it all the harder with all the sweeter xpeng CEO undervalue let's see who wins autonomous driving. First in China
Has it occurred to anyone that nothing could be this good…just like FTX….and the founder is a Chinese Hedge Fund manager…Has anyone checked who has big short positions in Nvidia….. when the US AI bubble finally bursts I hope it’s not because of a bluff like this….give us a real implosion, we deserve at least that….
@@finbarrmcgrath1686I like your angle. I’m also backing off to see the bigger picture. The proverbial bone has been thrown and all the dogs has given chase but pay attention to who threw the bone.
Chinese scientists are absolutely the best, innovative and hard working. Thanks for the free open source. Great for humanity not a threat to anyone ❤❤❤
“Because we believe the most important thing now is to participate in the global innovation wave. For many years, Chinese companies are used to others doing technological innovation, while we focused on application monetization - but this isn’t inevitable. In this wave, our starting point is not to take advantage of the opportunity to make a quick profit, but rather to reach the technical frontier and drive the development of the entire ecosystem.”
Deepseek has chinese government sensorship (e.g., not respond to political questions), however its open source code so you can remove this sensirship codes.
Trump: " Sam and Larry will need $1T" but seriously, it's bigger than AI; China has more and smarter technologists. We can only win by getting over fear and paranoia by collaborating together
Unfortunately that’s not going to happen. History repeats and like Rome, the US will fall hard and fast... and this cycle will continue with other rising powers as the world continues
im not a tech person but i often wonder what the world will be like in 20-30 years. to me the big 7 US tech companies are just monopolies. it benefits everyone when its open source. common sense tells me, competition is good for the market, drives the cost down and benefits the consumers.
Startups are often 100x more efficient than large companies. Large companies flowing with cash always spend like drunk sailors. Keep in mind that your brain operates in 20 Watts vs gpt using a small city worth of power. So lots of room for optimization beyong deepseek effuciency.
true, but id imagine the way brain and computer operates/conputes is very very different and not well understood, so its not really apples to apples comparison between the two in terms of power consumption but interesting point nevertheless
That was actually in the original script of the matrix in which the machines use human brains to run the simulation. That battery stuff came later and was a dumb idea.😂
That's what I've been comparing it to as well. We know what something close to peak efficiency (running brains is expensive, and every advanced organism has been using one for hundreds of millions of years) looks like, and what the big guys burn to "copy" it, and there's a huge gulf. Now, suddenly, instead of a huge warehouse and a nuclear reactor, you can run an advanced, fully functioning language model off something the size of a suitcase, powered by a wall outlet. It's not a fully functioning brain substitute, YET, and it's not as efficient as an actual brain, YET, but it's so many orders of magnitude closer it's ridiculous.
You have no idea what you ar e talking about. The initial reaserch took a lot of brains and white papers to accomplish. China is doing what it does best , Leveraging all the R&D and producing from that, regardless of cutting costs. Without those five-fingered discount white papers and R&D, they would not know what costs to even cut! Where China scores, is that they work in unity and have much less red-tape and barriers to progress through. While The wes have to deal the the dumbass democrat and DEI and all the woke, illegal immigration isses, hitting costs, just to progress. China doesnt have that issue.
In the 1950's Toyota was forced to develop a manufacturing system that used less resources because they did not have the capital to compete with the US automakers. The resulting system produced the most reliable cars at the lowest cost. Toyota is now more valuable than the entire US auto industry. Having the most money is not the most efficient
Similar to Japanese Drum manufacturing Arrogant American manufacturers like Ludwig Roger’s and Slingerland dominated the world drum market. But Tama and Pearls Drums through innovation and hard work destroyed them. Ludwig does OK of course but Pearl Tama and Yamaha now are the leaders
It's open weights, but not open "code", as LLMs do not run code in the traditional computer sense. If the censorship is baked in to the LLM then it would take a large Lab to re-train the model to realign it with [your government/culture].
@@howardlandman6121 NO! That was all due to globalist tossers, who engineered the conversion of most western countries from production base to service. They used the emerging economy and cheap labour as bait, while destroying much production businesses. They did it in Europe too, capitalising on cheap labour and next to no human rights. Now those eastern economies have well skilled and trained staff with decades of experience, that the west cannot currently match. What's infuriating is that this situation was obvious and did not suddenly fall out the sky. Many people and politicians were against it when it was happening. India is another country, that currently gets sympathetic money, yet has a space race. Unfortunately it's difficult to fix stupidity and greed. They make good bed-fellows.
@@howardlandman6121 They trusted the Western advocated globalism. They thought they could rely on the West for chip supply while making manufacturing a backbone of their economy. Obviously they were wrong, but they are closing in quickly.
They trusted the Western advocated globalism. They thought they could rely on the West for chip supply while making manufacturing a backbone of their economy. Obviously they were wrong, but they are closing in quickly.
Didn’t they train Deepseek R1 on the outputs of other LLMs? If that is the case, then shouldn’t the total cost of training Deepseek include at least some of the cost of training and querying the antecedents?
@@Alex.kantrowitz please what is AGI? i asked chatgpt and deepseek the same question of how they are different, they gave me different answers. deepseek said they do AGI and chatgpt does narrow AI. is this correct? I then asked the question on human control on AGI, deepseek said AGI is goal driven so there is risk to conflict with human interests, e.g. if we set improve climate risk as a goal, they potentially can wipe out human if we are not careful 😅
Oh crap, we Americans are having to face real life now, competition is hell but healthy. Time to get out of our dream world of being truly educated. Our students here in America have mostly became indentured slaves of the Banks with terrible, unrelenting academic charges which burden them for a life time. The very rich have fed on their blood and spirit far to long.
Feels like the afterhour activity on Nvidia stock is a bait. They are slightly increasing it giving retailers an idea that it may bounce back and when they fomo in they will have the liquidity to exit.
Your guest needs to see an old video by Tony Saba. The disrupter of technology comes from the outside. This quant company is for investment modeling. Their guys are not glued to the business model of building a very expensive complicated LLM to make more money.
When dedicated models to solve Accounting or Finance or Marketing or any niche area that allows for cheaper labor start happening, then we'll see significant impact much more than you see with Microsoft's 360 or Apple's Intelligence usage. Those applications of AI will be game changers. Open Source Deepseek just changed the game by lowering costs.
help me out here as someone not up on the tech part. I understand they use a process that takes in information but then streamlines it or reduces so it does not use everything it learned to fine tune itself to save or create more space for its function. but how do we know it is not missing key information in this process? not sure I explained that well, but I have a problem seeing how that could create problems later.
Big $$ Appliaction is industry. The capability to analyze all the process data which is currently collected in a typical plant, 100% of the time, is worth mega $$$$.
Looking forward to a Video Generating AI from Deep Seek, that can generate 2 h movies from one prompt locally on my desktop in 30 minutes. 🤔😀 What size is the Context window with R1? 🤔 NVIDIA mini computer might be a cool option, soon.
Por favor esto que acaba de pasar beneficia a toda la industria de inteligencia artificial. Otra cosa hubiese sido si deepseek hubiese sido tecnología cerrada.
How else a third world country riddled with constant wars holding it back supposed to develop? It has to copy the first world countries. Give China at least 20 more years and they will catch up to the US In technology.
While DeepSeek’s innovation is groundbreaking, by no means has it established a commanding market lead. Because it published its research, other model companies will learn from it, and adapt. Meta and Mistral, the French open-source model company, may be a beat behind, but it will probably be only a few months before they catch up. As Meta’s lead researcher Yann Lecun put it: “The idea is that everyone profits from everyone else’s ideas. No one ‘outpaces’ anyone and no country ‘loses’ to another. No one has a monopoly on good ideas. Everyone’s learning from everyone else.” So it’s execution that matters. Ultimately, it’s the consumers, startups and other users who will win the most, because DeepSeek’s offerings will continue to drive the price of using these models to near zero (again aside from cost of running models at inference). This rapid commoditization could pose challenges - indeed, massive pain - for leading AI providers that have invested heavily in proprietary infrastructure. As many commentators have put it, including Chamath Palihapitiya, an investor and former executive at Meta, this could mean that years of OpEx and CapEx by OpenAI and others will be wasted
Scale will still be needed for deployment. Energy will be transferred into compute as a basic resource for much of future economic activity. More people and systems will use and call on these systems, calling for chips capable of embodying this.
Sell cloud compute and closed AI model and hold on chipmaker. Unclear if the potential increase chip need in inference will offset potential decrease in training.
What happens in a few days when 03 is released and blows R1 out of the water? Could it be that compute/parameter scaling AND R1-type Reinforcement Learning is the key to superintelligence? The R1 paper specifically said that it's RL does not work for smaller models.
A major open question is how much DeepSeek relied on data-copying & data-massaging of training data from OpenAI & other top AI labs. Also, did they train using true outputs of GPT 4 o3 & other systems? A key fact that hints to this is, if you ask DeepSeek R1 who it is, it replies that it's OpenAI's chatbot. Still, DeepSeek made major efficiency improvements in both design & deployment.. This does not weaken NVidia's cost benefits. It actually makes the hardware super capable, if they're used with DeepSeek's hyper-efficient model..
They are spooked because Deep Seek algorithm isn't relying on hardware scaling to get the same State of the Art outcome. Meaning more GPUs isn't going to improve the End result. US Tech companies as of today is still in Brute forcing stage but as it turns out its Not about doubling or tripling Compute Power but getting the Algorithm right. For example for Math its Not how much compute Power to get the Right answer but Right method used to get the Right answer. More Compute Power is useless and does not improve anything. Its in chaos right now because Everybody in US thought this is a long way off. The business model was about milking this time period until it actually happens. Now all that is in jeopardy. Who is going to buy more GPUs now? AI now is entering a stage where the game has changed to Better Algorithms to get improvement overnight. Nobody is prepared for this. Now its about getting Profit short term to reap the Benefits from it Today. Startups can now already start to come up with Products for Industries specific use. And the Price is Cheap. So Cheap that AI companies who had spent all those Billions with Trillion in market cap could not Reap the benefits of it. Charging how much? Since Deep Seek is selling it Dirt cheap. Smartphone companies will put it into use, EVs will put it into use , Maps, engineering companies, electrical components companies etc etc.
@@weiskl887 I guess eventually then these fools will come to understand it about knowledge representation, and not algorithms. That’s the deepest problem that nobody ever thinks about is the knowledge representation system... they all have static/fragile designs, Hawkins' products are really bad even in the sorta neural-net form he has is fucking self-crippling in the number of associations that can be built up. I've never seen a fucking neural-network capable of self-reflection and differentiation, among many other conceptual paradoxical forms humans have no problems thinking about. This is one of the reasons I hate a lot of CS and AI people, they brute force everything when they can work from a post processing standpoint and at that point you only have rudimentary functions that can easily be parallelized in a GPU/APU even on commodity hardware 😂
@@reptilexcq2 Looks like you are oblivious that most top U.S. AI researchers asked the same questions. I've read the paper. It doesn't provide enough training details, of *_exactly_* what they did.
@@weiskl887 Your comment is full of vague nonsense. For example you claim "for Math its Not how much compute Power to get the Right answer but Right method used to get the Right answer." Your statement confuses & mixing a lot of things. GPU and (what you call) "method" are BOTH important. Compute power is definitely critical and will still be needed to optimize ANY model/network, to make it give the right answer. Amount of GPU/compute power for training is what determines development cost, and is an advantage DeepSeek exploited. But you strangely ignore this. DeepSeek made a major improvement in their model's architecture. This efficiency improvement is usable by everyone, and it'll most help anyone having massive compute/GPU clusters for training/optimization. So this doesn't remove or lessen the advantage that powerful GPUs provide (during training and runtime). In fact, it supercharges already powerful GPUs even more.
Trouble for Open AI, Perplexity, Mistal, etc, because they have competition I think Nvidia is safe. There has never been a time in history where you don't need more compute. More companies needing AI compute means data centers still need to be updated.
I just dont see how deep sink breaks the scaling law . Maybe if we were already at agi or asi then maybe, but we still dont have enough compute for like video to used by the public let alone truly complex tasks.
Stock prices of big cloud providers and chip makers shouldn't fall. Now that everybody can run and even train their own state of the art models by following the recipe provided by deepseek, the need for compute is going to peak like crazy. The only ones concerned should be closedAI and other big corps that already invested fortunes training inefficient models. Clearly many people freak out because they don't understand that open source is good for everybody except for closed source fanboys.
I want a language model that runs on my smartphone to help build an open source and decentralized global platform for collective terrestrial intelligence, CTI.
This interview is seriously lacking in technical knowledge. Deepseek R1 is 670B parameters... that's far larger than anyone can realistically just run locally on their home machines. He's likely referring to the "distilled" R1 models that are actually other architectures like Qwen and LLama finetuned on synthetic data created by the actual R1. Their performance is much worse than the real R1, but many people are confusing those models for the real R1 and saying "R1 can be run on your laptop." No, no one's laptop can run a 670B parameter LLM.
It's a nice model but it's not a better model. The price comparison is stunning from a use perspective but this shouldn't be having any impact on NVidia. Their capacity will continue to be maxed out. The real pressure is on OpenAI at the free cheap level.
if deepseek's inference AI benefits token for token are 50 times cheaper, then surely replication would make o3 models chain of thought scale 50 times token wise and allow their models to blow through ARC challenge and frontier math even more significantly? I mean it would literally make o3 o4 level AI with barely any effort thanks to deepseek. why are there no ARC or frontier maths results for deepseek r1 with it's western chain of thought doing the heavy lifting like o3, given AGI is what matters? I mean AGI is what bootstraps exponential AI development, not cheap non AGI models and OpenAI say they are onto ASI after besting the ARC AGI to reap those exponential gains which will produce the cheapest models of all. So would be cool to see how r1 compares with o3 in that regard.
100% does not matter who's got the best software. It only matters who has most and best hardware. Computing power and data storage is key. All software gets copied and very quickly becomes obsolete. Obsolete software very quickly becomes cheap and then becomes free.Computing power and data storage will always grow in demand. Whoever has the most wins the game.
deepseek isn't quite there yet it still gets stuck, openai is just going to take the deepseek efficiency gains and integrate them into the higher compute gpt-5 to actually make the ai agents work 100%
It seems to me the Deepseek is fundamentally adding a gearbox or filter in the compute process thus reducing the resources and cost of achieving the result. Its a process innovation for AI operations. Basic disruption. Well done Chinese or not.
Text based LLM is just a party trick compared to all the areas it is going into, AI that operates physical and virtual entities in the physical and AR/meta world is processor and energy thirsty on another level.
How deepseek is better than chat GPT is ,Deepseek tells you how its arriving at the answer so that one can know the limits and restrictions of the arrived answer.
what make DS threat, is make other nation catch up with US like EU and others, no true US full ai domination , and trumps new jobs of stargate r likly f****ed
Life is linear and not linear at the same time.. I would bet on both, upscaling, the billions, the fan fare, the hype and the celebrity events along with the frugality... Each to its own devices... Westinghouse versus Nicola Tesla, the story continues on and on...
If OpenAI or Google or Elon said, "We've just reduced our cost per token by 97%" the AI World would be very excited. Too much focus is on Nvidia. AI just got a massive win and acceleration to ASI... I think AGI is 2025. ASI is the question. When?
All I've seen can only conclude that the Chinese company is on par with U.S. companies AI models. How do we know they created this on the cheap? How do know they haven't simply improved from existing models? Let's see if they take the lead by more than a hair.
are we sure about how much did the training cost or what they used for this? Only from what they said? Maybe they lied....how can we so sure about this?
that seems like a moot point given the quality of model they produced with second rate chips at hand. it wrecks the current business model of us labs imo.
Even if we assume they had higher training costs, it’s a fraction of the cost to run it compared to OpenAI, $0.15 compared to $15 per million tokens, so your point doesn’t really matter
@@lotzewing Chinese cost has nothing to do with usa or other western cost ...it's like making a car in china vs Germany. Of course china will make it a lot cheaper.
To answer your question, they do have a lot of gpus, and they dont need to lied. Their main business is quant trading. They are one of the biggest quant trading fund in china, and its well known they have a lot of gpus, because theyre one of the first to use ai for trading in china and its disclosed to their investors. Also, the released models are indeed disruptive but doubt that was what the team proud of. Because if you read their paper in which they described how they did it, you know its a break through and they can twait to share. First of all the name is deepseek, its a direct reference to googles deepmind that created alphago that ignited all the ai craze today. Deepmind had a product called alphago zero went for a route called reinforced learning which has minimal human intervention except the basic input of the rule of go. And its unbeatable. Deepseek went for THAT, except not for playing go, but making the model reason for general problems by itself, while most of other models are just glorified autocompletes. Openai possibly went for similar route but they are guiding the reasoning process instead of letting it go, thats why it’s expensive. What it entails is that, first, reinforcement learning is capable of inducing reasoning, and we do not need much data for it. Which means potentially the road to AGI is NOT data constrained like a lot of us used to think.
I am at the beginning of my "investment journey", planning to put 385K into dividend stocks so that I will be making up to 30% annually in dividend returns. any good stock recommendation on great performing stocks or Crypto will be appreciated
As a newbie investor, it’s essential for you to have a mentor to keep you accountable. Ruth Ann Tsakonas is my trade analyst, she has guided me to identify key market trends, pinpointed strategic entry points, and provided risk assessments, ensuring my trades decisions align with market dynamics for optimal returns.
I managed to grow a nest egg of around 120k to over a Million. I'm especially grateful to Adviser Ruth Ann Tsakonas, for her expertise and exposure to different areas of the market.
I don't really blame people who panic. Lack of information can be a big hurdle. I've been making more than $200k passively by just investing through an advisor, and I don't have to do much work. Inflation or no inflation, my finances remain secure.. So I really don't blame people who panic.
Without a doubt! Ruth Ann Tsakonas is a trader who goes above and beyond. she has an exceptional skill for analysing market movements and spotting profitable opportunities. Her strategies are meticulously crafted on thorough research and years of practical experience.
how would you recommend i enter the crypto market? I am also looking at studying some traders and copying their strategy rather than investing myself and losing money emotionally. What's your take on this approach?? and How can i reach her, if you don't mind me asking??
if the chinesses have developoed a better cheaper model this could be good for nividie if they apply the same model with nividea more poweregful technology its good for everyone all round the marcket should not be crashing but going up
Deepseek is a word/token model, not a picture or continuous visual solution. Those solutions require significant compute and remain untouched. Example: Tesla FSD and robotics are not competitors with Deepseek
We'll have more on the audio feed later this week. Tune in on your podcast app of choice:
Spotify: spoti.fi/32aZGZx
Apple: apple.co/3AebxCK
Etc. pod.link/1522960417/
i think it's getting a bit rich talking about censorship in china when we're going through the biggest control of information in history in the west
You are ATYPICAL guy grew in information cocoon createdby the Western mainstream media.
@@zhoma982 ? Seems hypocritical to me
Right-wing billionaires turning UA-cam into an extension of Fox News is 99.99% of the Western censorship that matters.
I agree.
From my perspective whether I take what China says with a handful of salt or take what billionaires in the USA say with a handful of salt, it’s the same to me.
I already read information from other sources. - east and west. ASAHI Shimbun, the Staits times, the economist and others.
Then try to figure out what makes sense to me.
It’s going to be difficult for people with little to no basic knowledge of science, economics and politics and desire how to figure it out.
People are going to have to suspend judgement, be aware that there are no binary right wrong answers.
I see more cults forming.
Sad.
yes, plenty of western LLM censorship, say, on israel/gaza and other taboo subjects....
DeepSeek was just a side project to use some of the leftover GPUs they had from the main business. The company does have H100's but it was a point for them to use H800. It was also important for them to use only home-grown engineers, and no engineers returning from work abroad. It's just a flex to you Americans, and probably a celebration of the new year and completion of their China 2025 project:
"Made in China 2025" (MIC 2025) is a strategic industrial policy initiative launched by the Chinese government in 2015 to transform China into a global leader in advanced manufacturing and high-tech industries by 2025. The plan aims to reduce China’s reliance on foreign technology, boost innovation, and upgrade its manufacturing base from low-cost labor to cutting-edge, value-added production.
Mission complete! Congrats to China and humanity!
can you imagine a 10 year project in USA,,,,errrrrrrrrrrr NO!
Timing of this news was extra flex coming on top of the $500 billion Trump/Stargate/Broligarch talk.
Yes The word genius comes to mind yes it does matter where you went to school like friends etc. No, I'm not that smart. I'll admit it I'm an ex-marine and yes I've been investing why CEO BYD Wang a genius It's where he did. It makes it all the harder with all the sweeter xpeng CEO undervalue let's see who wins autonomous driving. First in China
Has it occurred to anyone that nothing could be this good…just like FTX….and the founder is a Chinese Hedge Fund manager…Has anyone checked who has big short positions in Nvidia….. when the US AI bubble finally bursts I hope it’s not because of a bluff like this….give us a real implosion, we deserve at least that….
@@finbarrmcgrath1686I like your angle. I’m also backing off to see the bigger picture.
The proverbial bone has been thrown and all the dogs has given chase but pay attention to who threw the bone.
This is what we neeeeed, competition that truly innovates
but it's unfair competition, DS gives the world a short cut way to the frontier, but China still in sanction, they well never surpass US, said to say.
Power to opening source just like the PC windows 🪟 hardware was opening source
Necessity is the mother of invention, not competition
The US would collapse...
What’s impressive about Deepseek is that they use 100% homegrown engineers. No H1B, 100% Chinese
中国的聪明人还是很多的。比较实干、谦虚。
@@pass3d可是被无脑粉红破坏了。。。如果中国不是共产主义 而是古典中国多美好
And most of them are graduates and 1-2 years new people can you imagine the power the team has…
It only took 3 years after someone else did all the work for them to roll out their first version.
It's not like they have a choice
Chinese scientists are absolutely the best, innovative and hard working. Thanks for the free open source. Great for humanity not a threat to anyone ❤❤❤
Well, if you exclude the biologists working in Wuhan. Then they’re literally a threat to everyone. 😬
and very humble. no rah-rah-rah, no bragging from the Chinese.
No one has a problem with the Chinese. People have a problem with the CCP. This includes many Chinese
“Because we believe the most important thing now is to participate in the global innovation wave. For many years, Chinese companies are used to others doing technological innovation, while we focused on application monetization - but this isn’t inevitable. In this wave, our starting point is not to take advantage of the opportunity to make a quick profit, but rather to reach the technical frontier and drive the development of the entire ecosystem.”
DEEPSEEK is CHINA …Chinese New Year gift to the world for …FREE! ❤❤❤❤
Nothing is free. Not even those working on Deepseek would say that! There is always profit. Not necessarialy monetary!
Deepseek has chinese government sensorship (e.g., not respond to political questions), however its open source code so you can remove this sensirship codes.
You are pretty naive if you think “Open”AI doesn’t have US government censorship built in. Try ask it about Gaza and see what you get.
Open ai has censorship of US government unable to take question about israel gnocide, native American gnocide
why do you want to do that?
@@mjlho . He is prolly suffering from existential allergy to products made in chyna chyna chynaa (credit to drump).
Open source is the way to go. We will need to keep an eye on even the US government sponsorship. This is huge good news
11k subs! Let’s go buddy, you deserve it!
Thank you so much!!!! Let's go!!
Trump: " Sam and Larry will need $1T" but seriously, it's bigger than AI; China has more and smarter technologists. We can only win by getting over fear and paranoia by collaborating together
Sure the Chinese are better at math, but we crush them in Gender Studies...
Unfortunately that’s not going to happen. History repeats and like Rome, the US will fall hard and fast... and this cycle will continue with other rising powers as the world continues
1000% tariff for Chinese export to the US.
Cool insights on DeepSeek! Curious to hear more about its impact on smaller startups. Great episode! 😊
Thanks so much for watching!
im not a tech person but i often wonder what the world will be like in 20-30 years. to me the big 7 US tech companies are just monopolies. it benefits everyone when its open source. common sense tells me, competition is good for the market, drives the cost down and benefits the consumers.
Startups are often 100x more efficient than large companies. Large companies flowing with cash always spend like drunk sailors.
Keep in mind that your brain operates in 20 Watts vs gpt using a small city worth of power. So lots of room for optimization beyong deepseek effuciency.
true, but id imagine the way brain and computer operates/conputes is very very different and not well understood, so its not really apples to apples comparison between the two in terms of power consumption but interesting point nevertheless
That was actually in the original script of the matrix in which the machines use human brains to run the simulation. That battery stuff came later and was a dumb idea.😂
That's what I've been comparing it to as well. We know what something close to peak efficiency (running brains is expensive, and every advanced organism has been using one for hundreds of millions of years) looks like, and what the big guys burn to "copy" it, and there's a huge gulf.
Now, suddenly, instead of a huge warehouse and a nuclear reactor, you can run an advanced, fully functioning language model off something the size of a suitcase, powered by a wall outlet.
It's not a fully functioning brain substitute, YET, and it's not as efficient as an actual brain, YET, but it's so many orders of magnitude closer it's ridiculous.
Agree, it’s like turning on a yacht rather than a heavy cargo ship. Smaller companies are far more nimble.
You have no idea what you ar e talking about. The initial reaserch took a lot of brains and white papers to accomplish. China is doing what it does best , Leveraging all the R&D and producing from that, regardless of cutting costs. Without those five-fingered discount white papers and R&D, they would not know what costs to even cut!
Where China scores, is that they work in unity and have much less red-tape and barriers to progress through. While The wes have to deal the the dumbass democrat and DEI and all the woke, illegal immigration isses, hitting costs, just to progress. China doesnt have that issue.
In the 1950's Toyota was forced to develop a manufacturing system that used less resources because they did not have the capital to compete with the US automakers. The resulting system produced the most reliable cars at the lowest cost. Toyota is now more valuable than the entire US auto industry. Having the most money is not the most efficient
Similar to Japanese Drum manufacturing Arrogant American manufacturers like Ludwig Roger’s and Slingerland dominated the world drum market. But Tama and Pearls Drums through innovation and hard work destroyed them. Ludwig does OK of course but Pearl Tama and Yamaha now are the leaders
Good point.
It’s opening source and you can modify and update the code to filters based in your need and running it on your own space
It's open weights, but not open "code", as LLMs do not run code in the traditional computer sense. If the censorship is baked in to the LLM then it would take a large Lab to re-train the model to realign it with [your government/culture].
Glad to see your channel getting traction. Congrats on 10K subscribers!
Thank you so much, means a lot! Glad to be catching some steam :)
China has shown the US that purpose drives innovation much more effectively than profiteering
No. Quite simply The west did the hard work. China just stood on its shoulders.
Don't get it twisted. Pioneers perish. Settlers flourish!
That must explain why China's chip manufacturing is so far ahead of the US, Europe, and Taiwan. :-P
@@howardlandman6121 NO! That was all due to globalist tossers, who engineered the conversion of most western countries from production base to service. They used the emerging economy and cheap labour as bait, while destroying much production businesses. They did it in Europe too, capitalising on cheap labour and next to no human rights. Now those eastern economies have well skilled and trained staff with decades of experience, that the west cannot currently match. What's infuriating is that this situation was obvious and did not suddenly fall out the sky. Many people and politicians were against it when it was happening. India is another country, that currently gets sympathetic money, yet has a space race. Unfortunately it's difficult to fix stupidity and greed. They make good bed-fellows.
@@howardlandman6121 They trusted the Western advocated globalism. They thought they could rely on the West for chip supply while making manufacturing a backbone of their economy. Obviously they were wrong, but they are closing in quickly.
They trusted the Western advocated globalism. They thought they could rely on the West for chip supply while making manufacturing a backbone of their economy. Obviously they were wrong, but they are closing in quickly.
Didn’t they train Deepseek R1 on the outputs of other LLMs? If that is the case, then shouldn’t the total cost of training Deepseek include at least some of the cost of training and querying the antecedents?
great episode Alex. Looking forward to the Hoffman interview this week!
Thanks so much, Dan :)
Thanks Alex for jumping in there and knocking out this video. Lot happening this week! 🤖🤖🤖
My pleasure! And we have even more on the Big Technology Podcast audio feed
@@Alex.kantrowitz please what is AGI? i asked chatgpt and deepseek the same question of how they are different, they gave me different answers. deepseek said they do AGI and chatgpt does narrow AI. is this correct? I then asked the question on human control on AGI, deepseek said AGI is goal driven so there is risk to conflict with human interests, e.g. if we set improve climate risk as a goal, they potentially can wipe out human if we are not careful 😅
ebooksbyai AI fixes this. DeepSeek's Impact on AI Research
People are hyped because it is next level
Oh crap, we Americans are having to face real life now, competition is hell but healthy. Time to get out of our dream world of being truly educated. Our students here in America have mostly became indentured slaves of the Banks with terrible, unrelenting academic charges which burden them for a life time. The very rich have fed on their blood and spirit far to long.
OpenAI trying to convince us that AI cost Billions to develop. Deepseek: hold my chopsticks
Great name and fun logo, haven't used the interface -yet- but I bet it has thought. "Ju-Yang" The Mayfly build
I have utilised ai with medical scans I have made, it makes complex questions easy to make into real time scans to see if ideas are useful
Sunk costs are sunk costs. People tend to forget that. Discovering new ideas isn’t the challenge, escaping from old ideas is. ~ Keynes
Feels like the afterhour activity on Nvidia stock is a bait. They are slightly increasing it giving retailers an idea that it may bounce back and when they fomo in they will have the liquidity to exit.
Your guest needs to see an old video by Tony Saba. The disrupter of technology comes from the outside. This quant company is for investment modeling. Their guys are not glued to the business model of building a very expensive complicated LLM to make more money.
When dedicated models to solve Accounting or Finance or Marketing or any niche area that allows for cheaper labor start happening, then we'll see significant impact much more than you see with Microsoft's 360 or Apple's Intelligence usage. Those applications of AI will be game changers. Open Source Deepseek just changed the game by lowering costs.
This is a very, very smart market manipulation/intelligence operation going on here.
What’s your full hypothesis for how that’s happening?
How do I get in touch?
They are also bootstrapping themselves to even better models. .
Messaged you on linkedin.
Yes, its open source, experts have confirmed the numbers.
help me out here as someone not up on the tech part. I understand they use a process that takes in information but then streamlines it or reduces so it does not use everything it learned to fine tune itself to save or create more space for its function. but how do we know it is not missing key information in this process? not sure I explained that well, but I have a problem seeing how that could create problems later.
Democratization access to a AI tool creeps the shit out of Wall Street, the game has changed and DeepSeek will be attacked in every way possible.
Big $$ Appliaction is industry. The capability to analyze all the process data which is currently collected in a typical plant, 100% of the time, is worth mega $$$$.
謝謝！
Thank you so much!!
Alex love your pod!
Thanks for being here!
Great discussion, thanks for having it. 🙏🏼
Nice Vision Pro in the background 👍
Looking forward to a Video Generating AI from Deep Seek, that can generate 2 h movies from one prompt locally on my desktop in 30 minutes. 🤔😀 What size is the Context window with R1? 🤔 NVIDIA mini computer might be a cool option, soon.
Por favor esto que acaba de pasar beneficia a toda la industria de inteligencia artificial.
Otra cosa hubiese sido si deepseek hubiese sido tecnología cerrada.
Hmmm, there might be less hardware needed for training, but there will be an increase in hardware for inference, i think.
Once again they've taken US innovation and improved on it. I give them props though and if OpenAI don't respond, they're in trouble. Great video!
Also crazy how they've innovated on the hardware front undermining NVIDEA.
0 to 1 is American 🇺🇸 1-10000000 is Chinese
it wasn’t always that way however- China was the 0-1 country for centuries
How else a third world country riddled with constant wars holding it back supposed to develop? It has to copy the first world countries. Give China at least 20 more years and they will catch up to the US In technology.
No, Chinese scientists broke the box of the scaling paradigm and fundamentally changed AI. You just cannot overcome your bias and prejudice
They were dropping a new mode this month, that was better than o1. OpenAI still has a lead but it’s only 2 months of one.
While DeepSeek’s innovation is groundbreaking, by no means has it established a commanding market lead. Because it published its research, other model companies will learn from it, and adapt. Meta and Mistral, the French open-source model company, may be a beat behind, but it will probably be only a few months before they catch up. As Meta’s lead researcher Yann Lecun put it: “The idea is that everyone profits from everyone else’s ideas. No one ‘outpaces’ anyone and no country ‘loses’ to another. No one has a monopoly on good ideas. Everyone’s learning from everyone else.” So it’s execution that matters.
Ultimately, it’s the consumers, startups and other users who will win the most, because DeepSeek’s offerings will continue to drive the price of using these models to near zero (again aside from cost of running models at inference). This rapid commoditization could pose challenges - indeed, massive pain - for leading AI providers that have invested heavily in proprietary infrastructure. As many commentators have put it, including Chamath Palihapitiya, an investor and former executive at Meta, this could mean that years of OpEx and CapEx by OpenAI and others will be wasted
Scale will still be needed for deployment. Energy will be transferred into compute as a basic resource for much of future economic activity. More people and systems will use and call on these systems, calling for chips capable of embodying this.
10:08 That was mentioned by some important people at the time those particular sanctions were put in place.
Off the wall question, if you own stock in Google, Microsoft, Nvidia would you sell?
YES 😮 SELL, HOLD CASH, REVIEW INFO, BUY BACK IF ANSWERS ISSUES COMPETIVELY...IF.😮
Sell cloud compute and closed AI model and hold on chipmaker. Unclear if the potential increase chip need in inference will offset potential decrease in training.
What happens in a few days when 03 is released and blows R1 out of the water? Could it be that compute/parameter scaling AND R1-type Reinforcement Learning is the key to superintelligence? The R1 paper specifically said that it's RL does not work for smaller models.
A major open question is how much DeepSeek relied on data-copying & data-massaging of training data from OpenAI & other top AI labs. Also, did they train using true outputs of GPT 4 o3 & other systems? A key fact that hints to this is, if you ask DeepSeek R1 who it is, it replies that it's OpenAI's chatbot. Still, DeepSeek made major efficiency improvements in both design & deployment.. This does not weaken NVidia's cost benefits. It actually makes the hardware super capable, if they're used with DeepSeek's hyper-efficient model..
They are spooked because Deep Seek algorithm isn't relying on hardware scaling to get the same State of the Art outcome. Meaning more GPUs isn't going to improve the End result. US Tech companies as of today is still in Brute forcing stage but as it turns out its Not about doubling or tripling Compute Power but getting the Algorithm right.
For example for Math its Not how much compute Power to get the Right answer but Right method used to get the Right answer. More Compute Power is useless and does not improve anything. Its in chaos right now because Everybody in US thought this is a long way off. The business model was about milking this time period until it actually happens. Now all that is in jeopardy. Who is going to buy more GPUs now?
AI now is entering a stage where the game has changed to Better Algorithms to get improvement overnight. Nobody is prepared for this. Now its about getting Profit short term to reap the Benefits from it Today. Startups can now already start to come up with Products for Industries specific use. And the Price is Cheap. So Cheap that AI companies who had spent all those Billions with Trillion in market cap could not Reap the benefits of it. Charging how much? Since Deep Seek is selling it Dirt cheap.
Smartphone companies will put it into use, EVs will put it into use , Maps, engineering companies, electrical components companies etc etc.
Your questions suggest you have no clue as to how advance DeepSeek is. Go read the damm paper and stop speculating with stupid questions.
@@weiskl887 I guess eventually then these fools will come to understand it about knowledge representation, and not algorithms. That’s the deepest problem that nobody ever thinks about is the knowledge representation system... they all have static/fragile designs, Hawkins' products are really bad even in the sorta neural-net form he has is fucking self-crippling in the number of associations that can be built up. I've never seen a fucking neural-network capable of self-reflection and differentiation, among many other conceptual paradoxical forms humans have no problems thinking about. This is one of the reasons I hate a lot of CS and AI people, they brute force everything when they can work from a post processing standpoint and at that point you only have rudimentary functions that can easily be parallelized in a GPU/APU even on commodity hardware 😂
@@reptilexcq2 Looks like you are oblivious that most top U.S. AI researchers asked the same questions. I've read the paper. It doesn't provide enough training details, of *_exactly_* what they did.
@@weiskl887 Your comment is full of vague nonsense. For example you claim "for Math its Not how much compute Power to get the Right answer but Right method used to get the Right answer." Your statement confuses & mixing a lot of things. GPU and (what you call) "method" are BOTH important. Compute power is definitely critical and will still be needed to optimize ANY model/network, to make it give the right answer. Amount of GPU/compute power for training is what determines development cost, and is an advantage DeepSeek exploited. But you strangely ignore this.
DeepSeek made a major improvement in their model's architecture. This efficiency improvement is usable by everyone, and it'll most help anyone having massive compute/GPU clusters for training/optimization. So this doesn't remove or lessen the advantage that powerful GPUs provide (during training and runtime). In fact, it supercharges already powerful GPUs even more.
Alex speaks and sounds just like Sam Altman ? Btw. Subscribed, and great topic and discussion.
Trouble for Open AI, Perplexity, Mistal, etc, because they have competition I think Nvidia is safe. There has never been a time in history where you don't need more compute. More companies needing AI compute means data centers still need to be updated.
Until deepseek designs GPUs s at 10% of the cost.😂
Long term, yes.
Short term, demand will disappear for a while.
Data centers will be replaced by blockchains. Nvidia is walking dead.
The important here is the market, because you can do the same task but with low budget.
I just dont see how deep sink breaks the scaling law . Maybe if we were already at agi or asi then maybe, but we still dont have enough compute for like video to used by the public let alone truly complex tasks.
The scan code to install the app tries to force mirroring.
Stock prices of big cloud providers and chip makers shouldn't fall. Now that everybody can run and even train their own state of the art models by following the recipe provided by deepseek, the need for compute is going to peak like crazy. The only ones concerned should be closedAI and other big corps that already invested fortunes training inefficient models. Clearly many people freak out because they don't understand that open source is good for everybody except for closed source fanboys.
The censorship is only on the online version, if you run it locally there is no such restriction.
I want a language model that runs on my smartphone to help build an open source and decentralized global platform for collective terrestrial intelligence, CTI.
This interview is seriously lacking in technical knowledge. Deepseek R1 is 670B parameters... that's far larger than anyone can realistically just run locally on their home machines. He's likely referring to the "distilled" R1 models that are actually other architectures like Qwen and LLama finetuned on synthetic data created by the actual R1. Their performance is much worse than the real R1, but many people are confusing those models for the real R1 and saying "R1 can be run on your laptop." No, no one's laptop can run a 670B parameter LLM.
Any concerns about data security of data?
Lol. I love this ?. Like The pirates of silly-con valley haven't been spying on you for 50 years.😮😮😮
On the eve of the New Year of the Snake, a Chinese intellectual donates his invention to humanity, to make it a level playing field for all.
Wow, you’re making an impact!
The r1 model which outperforms o1 is not a distilled model.
Condensing knowledge is a winner!
It's a nice model but it's not a better model. The price comparison is stunning from a use perspective but this shouldn't be having any impact on NVidia. Their capacity will continue to be maxed out. The real pressure is on OpenAI at the free cheap level.
The value of the chips depends on how much money you can make from them and Deepseek has taken a lot of that profit potential away
Sounded as if you know what you're talking about, lol.
@@votebrian66 Nvidia is over max capacity and preparing the models has nothing to do with their operating costs.
Correction. I think high end gpu h800 is banned from China, not h100
H100 is better than h800
faster chips will always be needed as the data files grow.
will everyone's data files start growing faster, soon? cool 😎
I like to hear what Prof. Geoffrey Hinton thinks about DeepSeek
if deepseek's inference AI benefits token for token are 50 times cheaper, then surely replication would make o3 models chain of thought scale 50 times token wise and allow their models to blow through ARC challenge and frontier math even more significantly? I mean it would literally make o3 o4 level AI with barely any effort thanks to deepseek.
why are there no ARC or frontier maths results for deepseek r1 with it's western chain of thought doing the heavy lifting like o3, given AGI is what matters?
I mean AGI is what bootstraps exponential AI development, not cheap non AGI models and OpenAI say they are onto ASI after besting the ARC AGI to reap those exponential gains which will produce the cheapest models of all. So would be cool to see how r1 compares with o3 in that regard.
100% does not matter who's got the best software. It only matters who has most and best hardware. Computing power and data storage is key. All software gets copied and very quickly becomes obsolete. Obsolete software very quickly becomes cheap and then becomes free.Computing power and data storage will always grow in demand. Whoever has the most wins the game.
Anstatt Atomkraftwerke für den Energiebedarf reichen nun PV-Anlagen!
deepseek isn't quite there yet it still gets stuck, openai is just going to take the deepseek efficiency gains and integrate them into the higher compute gpt-5 to actually make the ai agents work 100%
And they'll have to make new changes open source as well because of the MIT license. No one will pay a high price for their products anymore.
remember deepseek is not multimodal, gpt-5 will have o3's reasoning and gpt-4o's multimodal capabilities as well as chatgpt's conversational dialogue
It seems to me the Deepseek is fundamentally adding a gearbox or filter in the compute process thus reducing the resources and cost of achieving the result. Its a process innovation for AI operations. Basic disruption. Well done Chinese or not.
Text based LLM is just a party trick compared to all the areas it is going into, AI that operates physical and virtual entities in the physical and AR/meta world is processor and energy thirsty on another level.
How deepseek is better than chat GPT is ,Deepseek tells you how its arriving at the answer so that one can know the limits and restrictions of the arrived answer.
Whilst chat gpt powered by billion dollars, Deepseek his powered by sustainable bamboo investment
If DeepSeek can do the same with AI video - then there must be opportunities in business creating ai video tools to make movies, games.
bro they already have ai image know
what make DS threat, is make other nation catch up with US like EU and others, no true US full ai domination , and trumps new jobs of stargate r likly f****ed
The problem it meeting demand and a way of getting paid for further development and to pay for services
BIG TECH OBVIOUSLY BILKED TRILLIONS OF INVESTORS $$$$$$$$$$
Life is linear and not linear at the same time..
I would bet on both, upscaling, the billions, the fan fare, the hype and the celebrity events along with the frugality...
Each to its own devices...
Westinghouse versus Nicola Tesla, the story continues on and on...
this is just overblown, i mean is there a reason that e.g. Arista Networks is down 23% ?
If OpenAI or Google or Elon said, "We've just reduced our cost per token by 97%" the AI World would be very excited. Too much focus is on Nvidia. AI just got a massive win and acceleration to ASI... I think AGI is 2025. ASI is the question. When?
All I've seen can only conclude that the Chinese company is on par with U.S. companies AI models. How do we know they created this on the cheap? How do know they haven't simply improved from existing models? Let's see if they take the lead by more than a hair.
thx alex
Now all we need in America is the Chinese to develop Ozempic at 2% of the cost!!
The US: can we invade China under the name of freedom?
This is how AI will be used in USA. Investigating avenues for war, etc.
If DeepSeek can do it, no body in the U.S. can do it?
While we discuss the subject, let's trot out the usual propaganda clichés about China.
You not only can't register, you can't even chat for assistance. The narrative should shift to getting your app to work.
are we sure about how much did the training cost or what they used for this?
Only from what they said?
Maybe they lied....how can we so sure about this?
that seems like a moot point given the quality of model they produced with second rate chips at hand. it wrecks the current business model of us labs imo.
Even if we assume they had higher training costs, it’s a fraction of the cost to run it compared to OpenAI, $0.15 compared to $15 per million tokens, so your point doesn’t really matter
@@lotzewing Chinese cost has nothing to do with usa or other western cost ...it's like making a car in china vs Germany. Of course china will make it a lot cheaper.
To answer your question, they do have a lot of gpus, and they dont need to lied.
Their main business is quant trading. They are one of the biggest quant trading fund in china, and its well known they have a lot of gpus, because theyre one of the first to use ai for trading in china and its disclosed to their investors.
Also, the released models are indeed disruptive but doubt that was what the team proud of. Because if you read their paper in which they described how they did it, you know its a break through and they can twait to share.
First of all the name is deepseek, its a direct reference to googles deepmind that created alphago that ignited all the ai craze today.
Deepmind had a product called alphago zero went for a route called reinforced learning which has minimal human intervention except the basic input of the rule of go. And its unbeatable.
Deepseek went for THAT, except not for playing go, but making the model reason for general problems by itself, while most of other models are just glorified autocompletes. Openai possibly went for similar route but they are guiding the reasoning process instead of letting it go, thats why it’s expensive.
What it entails is that, first, reinforcement learning is capable of inducing reasoning, and we do not need much data for it. Which means potentially the road to AGI is NOT data constrained like a lot of us used to think.
Lol. On the basis that every usa 🇺🇸 company lies day and night since 1776...this is nowt 😅
palantir is doing more than proof of concept with its enterprise customers, to be fair.
"we all benefit in the long run"? Uh, who is "we"?
theres such poetic justice when silcon valley gets 'disrupted' before theve even made a profit 🤣😂
American should support their AI and pay for their service
My poor portfolio has DeepSixed yet again, this time by DeepSeek.
I am at the beginning of my "investment journey", planning to put 385K into dividend stocks so that I will be making up to 30% annually in dividend returns. any good stock recommendation on great performing stocks or Crypto will be appreciated
As a newbie investor, it’s essential for you to have a mentor to keep you accountable. Ruth Ann Tsakonas is my trade analyst, she has guided me to identify key market trends, pinpointed strategic entry points, and provided risk assessments, ensuring my trades decisions align with market dynamics for optimal returns.
I managed to grow a nest egg of around 120k to over a Million. I'm especially grateful to Adviser Ruth Ann Tsakonas, for her expertise and exposure to different areas of the market.
I don't really blame people who panic. Lack of information can be a big hurdle. I've been making more than $200k passively by just investing through an advisor, and I don't have to do much work. Inflation or no inflation, my finances remain secure.. So I really don't blame people who panic.
Without a doubt! Ruth Ann Tsakonas is a trader who goes above and beyond. she has an exceptional skill for analysing market movements and spotting profitable opportunities. Her strategies are meticulously crafted on thorough research and years of practical experience.
how would you recommend i enter the crypto market? I am also looking at studying some traders and copying their strategy rather than investing myself and losing money emotionally. What's your take on this approach?? and How can i reach her, if you don't mind me asking??
if the chinesses have developoed a better cheaper model this could be good for nividie if they apply the same model with nividea more poweregful technology its good for everyone all round the marcket should not be crashing but going up
Deepseek should learn to spell “American exceptionalism” with lower case letters. My auto-correct won’t let me do so easily.
Deepseek is a word/token model, not a picture or continuous visual solution. Those solutions require significant compute and remain untouched. Example: Tesla FSD and robotics are not competitors with Deepseek
Just imagine the cheap models we'll create once someone cracks "AGI"
High level talk about low level optimization 😊 Well it should have been.