The Truth About China's Deepseek R1

Поділитися
Вставка
  • Опубліковано 5 лют 2025
  • The Robots Are Coming www.farzadmesb...
    FREE One Year Supply of Vitamin d3+k2 and 5 AG1 Travel Packs ➡ drinkAG1.com/f...
    Want to grow your UA-cam channel? DM David Carbutt
    For 10% discount quote ‘Farzad’ x.com/DavidCar...
    Farzad Clips: / @farzadmesbahiclips
    Podcasts available on your favorite podcast app.
    Get fit! Mention FARZAD when asking for a quote for a custom-built nutrition and fitness plan: www.teamevolve...
    Join this channel to get access to perks:
    / @farzadmediainc
    Use my referral link to purchase a Tesla product ts.la/farzad69506
    I worked at Tesla starting in July of 2017 as an Operations Analyst out of the Bethlehem, PA distribution center facility and left as a Program Manager based out of Austin, TX as of September 2021. I spent most of my time in the distribution and supply chain organizations.
    Before Tesla, I was a Director of Business Intelligence and Pricing at the largest Pet Food & Supply distributor in the US, Phillips Pet Food & Supplies based out of Easton, PA. My wife and I also owned a small business in Bethlehem, PA between 2016 and 2019.
    I have been a shareholder of Tesla since 2012 and currently own Tesla stock. Nothing I say constitutes as investment or financial advice.
    My thoughts are my own and are not representative of everyone who currently works, or has worked at Tesla.
    If you like this content I would greatly appreciate your likes and shares!
    www.farzadmesbahi.com
    www.linkedin.com/in/farzadmesbahi
    Twitter @farzyness
    #Tesla #elonmusk #twitter #tsla #TeslaNews #TeslaFSD #TeslaFSDBeta #teslastockanalysis #TeslaUpdate #investing #ElonMuskNews #teslanews #EV #FSD #investing
    DeepSeek AI: The Future of Super-Intelligent Models Analyzed
    This episode delves into the buzz surrounding DeepSeek AI, a cutting-edge Chinese language model that's making waves in the tech world. The host breaks down Gavin Baker's and Andrej Karpathy's analyses on the impacts of DeepSeek R1, focusing on its efficiency, cost reduction, and potential to run AI tasks on local devices without internet. The discussion covers the implications for Big Tech, edge computing, and the emerging landscape of artificial superintelligence. The episode also previews upcoming content, including a Tesla earnings live stream and a new project launch on the GigaCast.

КОМЕНТАРІ •

  • @leesweehuat
    @leesweehuat 8 днів тому +51

    The key impact of Deepseek is that for the time being, the AI Infrastructure is no longer the key bottleneck in rapidly improving AGI, as there is now an alternative approach offered by Deepseek.

    • @catbert7
      @catbert7 7 днів тому +2

      How do you figure that? As long as more computer makes smarter AI, and the computer is the most difficult/expensive to acquire, which is still true, it will be the bottleneck. You can always say that the real bottleneck is innovation, of course, but that goes without saying. You can’t procure innovation the way you can compute.

    • @abrvalg321
      @abrvalg321 6 днів тому +1

      AGI doesn't exist.

  • @thephilosophicalagnostic2177
    @thephilosophicalagnostic2177 8 днів тому +81

    If AI becomes really cheap and easier to train, this means a whole lot more people are going to be testing it out and even building their own versions. Accelerating tech has just accelerated a whole lot more. How much will our world change by the end of this decade?

    • @quaterman1270
      @quaterman1270 8 днів тому +2

      I agree. There are tens of thousends of companies who can afford it. Especially the problem regarding company data will be solved that way as they will be able to train their own PhD level AIs.

    • @stevenanderson9660
      @stevenanderson9660 8 днів тому

      We need more chips

    • @thefish5861
      @thefish5861 8 днів тому +5

      I for one, am really glad I’m old.

    • @Dave-cf4vd
      @Dave-cf4vd 8 днів тому +6

      I've been saying the world will change more in the next 15 years than the last 100. Looks like the timeline just sped up.

    • @chriscain7333
      @chriscain7333 8 днів тому

      Oh where do i start. In short, it depends. No it wont be everywhere and no, not everyone can have it.
      You see this world is running on energy, the sun, the planets the universe so on and so forth. Same with AI, it runs on electricity.
      Not a lot of people have access to a lot of electricity, not even china and uncle sam, may be china can if given enough time for them to expand their capacity.

  • @sanjaysharma4655
    @sanjaysharma4655 7 днів тому +168

    With Elon and Trump now working with Qardun is going to absolutely blow up

  • @vicentevaldez1696
    @vicentevaldez1696 8 днів тому +77

    Deepseek was very well ENGINEERED..

    • @robertwilliam9558
      @robertwilliam9558 8 днів тому

      What American hyperscaler would use a Chinese company for AI? 0

    • @larsnystrom6698
      @larsnystrom6698 8 днів тому

      @@robertwilliam9558
      Americans won't, but why would the Chinese care?
      The US AI companies will just incorporate what they now have learned from the Chinese into their own models.
      But now we know that it's a multipolar AI world.
      For the rest of the world, outside the US, we don't see China as the threat you do!

    • @hiltonhead
      @hiltonhead 8 днів тому

      Deepseek was engineered to crash the stock market. Someone made billions. The proof is in the pudding!

    • @JD-yz4kr
      @JD-yz4kr 8 днів тому

      @@robertwilliam9558 It's illegal for American companies to cater to Chinese companies without explicit permission from the Department of Commerce. It is illegal for any person working for a company receiving government funding to associate with or exchange information with any Chinese person without explicit permission from Congress.

    • @johnnyq4260
      @johnnyq4260 7 днів тому

      @@robertwilliam9558 Hyperspender

  • @Felipe-n3j
    @Felipe-n3j 4 дні тому +1

    Just wait for DEEPSEEK R2 you will be blown away!😊😊

  • @Benjamin-nf2ir
    @Benjamin-nf2ir 6 днів тому +1

    Great conversation. I love how you put intelligence potential in perspective. We are just scratching the surface.

  • @dr-k1667
    @dr-k1667 8 днів тому +16

    Deepseek only revealed how deeply people are NOT willing to learn and look into things and just react off of headlines. This story was beneficial to me when NVDA took a hit and I was able to pick up more share and reduce my DCA. Today as more news came out about it it's not a nothing burger (it raises other questions) but it also doesn't change the future for those who are leading the way and how much it takes to make it happen. IMHO.

    • @Arathreas
      @Arathreas 8 днів тому +2

      You really don't like money?

    • @oggyoggy1299
      @oggyoggy1299 7 днів тому

      You’ve been buying at over $120???

    • @keyser021
      @keyser021 7 днів тому +1

      Nothing burger - that the source code for top tier AI was just released to the world, for free? You aren't paying attention.

    • @teachingwithipad
      @teachingwithipad 7 днів тому

      dude, cope.

    • @wshyangify
      @wshyangify 7 днів тому

      Keep DCA at a lower price on your falling knife!

  • @ScribblyPoppo
    @ScribblyPoppo 6 днів тому +4

    AI took my job. It was cleaning up dog messes. I love AI.

  • @VishnuVardhan-j5t
    @VishnuVardhan-j5t 7 днів тому +276

    Trump found a way to pump a coin like steroids... crazy that Qardun is an ICO instead of directly..

  • @macioluko9484
    @macioluko9484 8 днів тому +20

    Efficiency and cost cutting is as foreign to most US companies as lettuce is to Lizzo.

  • @ParthaPan
    @ParthaPan 7 днів тому +135

    Highly likely there will be partnerships between X or Tesla with Qardun

  • @Peter-z5f
    @Peter-z5f 8 днів тому +26

    If we live in the universe, and the universe is infinite, and we're stupid, then stupidity is infinite.

    • @The_Conspiracy_Analyst
      @The_Conspiracy_Analyst 8 днів тому +2

      infinity's not a number

    • @anotherlover6954
      @anotherlover6954 8 днів тому +3

      Have you seen anything that makes you think otherwise? The original Alfred E. Neuman, the mascot for MAD Magazine, was actually a political candidate when he was first invented, and he walked around looking dopey, and he carried a sign that said: "You could do worse. You always have."

    • @Dave-cf4vd
      @Dave-cf4vd 8 днів тому +2

      Just because we live in an infinite universe does not mean WE are infinite. Therefore, stupidity is finite. It is theoretically possible to get rid of it.

    • @vicentevaldez1696
      @vicentevaldez1696 8 днів тому +1

      @@Peter-z5f Very very tru in America 🤣🤣🤣

    • @MaroonedInDub
      @MaroonedInDub 8 днів тому

      If there is a bug in your water, it does not indicate that your water is bug infested.
      Just because you might be stupid, it doesn't mean that the universe is infested with you.
      Your sums are wrong. Go to school.

  • @Felipe-n3j
    @Felipe-n3j 4 дні тому

    Just wait for DEEPSEEK R2 you will be blown away!

  • @AjayYadav-lj2wt
    @AjayYadav-lj2wt 7 днів тому +153

    X and Qardun are going to be merged. Easiest money in my life

  • @saelensrl
    @saelensrl 6 днів тому +1

    The next phase of AI development must focus on both efficiency and ethical responsibility. A race to the bottom in efficiency could create dangerous, untrustworthy AI systems, so it’s critical that safety measures keep pace with cost-cutting innovations.

  • @raisoreo
    @raisoreo 8 днів тому +2

    Hey Farzad, thank you, thank you, thank you very much for not using “like” as a filler word. I only heard that word two or three times in your monologue. Great job and subject. Cheers!

  • @gilsegura8587
    @gilsegura8587 8 днів тому +29

    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

    • @paulmcgreevy3011
      @paulmcgreevy3011 8 днів тому +5

      Won’t be wasted. You can dial down your spending for a while. You can never have too much compute.

    • @gilsegura8587
      @gilsegura8587 8 днів тому +2

      yes, having too much compute is a good thing as a safety net for the mass future growth of users....

    • @Epiderm91
      @Epiderm91 8 днів тому

      @@gilsegura8587 as warren buffet said, 2025 is the year for the small and medium businesses...China biggest contribution is not Deepseek, but rather an ecosystem thats continuously improving and competing, which will accelerate AI tech and ultimately benefit humanity.

    • @pr0newbie
      @pr0newbie 8 днів тому +4

      Gen Z and younger millenials will probably find this idea refreshing, but this is exactly what the Internet was like in the 90s and early 2000s. Apple with its walled garden was the one that crushed this concept. It was great the devs could monetise their work, but we've gone too far in the capitalistic, commercial direction.

    • @Vexxxxxxxed
      @Vexxxxxxxed 8 днів тому +2

      This comment is literally written by AI. I cba.

  • @PinkyPowers
    @PinkyPowers 8 днів тому +32

    A more efficient AI is awesome, and something we sorely needed. But the big players (Tesla, META Google, Amazon, Microsoft) will ALWAYS need the most compute they can get their hands on. There truly is no limit to this need.
    DeepSeek has given them the secrets to efficiency by making their AI open source. So the playing field has been leveled, in that regard. Which means compute remains the primary edge. Whoever has the most, can do the most, the quickest. And competition is fierce.
    Also, lowering the cost to entry only means MORE companies will get in the game, and they'll all need GPUs.
    In other words, Nvidia remains THE STOCK to own, for the foreseeable future. Demand will only increase.

    • @wshyangify
      @wshyangify 7 днів тому

      Another perspective to consider: CCP just made trillions shorting the US market.
      They have merely released the model they want you to see, what does their military have?

    • @dirkbester9050
      @dirkbester9050 7 днів тому +2

      Yeah I don't get the panic. The whole point is newer and better algorithms. This one is new, it replaces the previous ones, just like the previous ones replaced the ones before them etc etc. The learning still needs to happen and we are nowhere close to enough simulated neurons to get to AGI. You are still ultimately going to have those giant learning machines.

    • @sumguy7716
      @sumguy7716 7 днів тому

      @@dirkbester9050 The panic is that, from an investor standpoint, the ROI that these companies would've received for investing literally hundreds of billions of dollars into AI has been eliminated since Deepseek is flooding the market with a free version of their product. This may be a good thing for the consumer in the short term, but whether or not this is a good thing in the long term is tbd.

    • @1flash3571
      @1flash3571 7 днів тому +2

      @@dirkbester9050 Nvidia Hardwares ARE NOT CHEAP. If other companies can do more with just a Regular computer with fewer GPU's AND with better efficiency, then it will HURT Nvidia on their earnings because other smaller companies might not need all that GPU's, AND might not need to buy Nvidia's Hardwares. That is what is being IMPLIED.....Not sure what you understood from the video....

    • @5t0ryte11er
      @5t0ryte11er 7 днів тому

      NVIDIA can build low cost GPUs as well. I think the big chip players will still win. Deepseek allows companies to build their field specific AI brain and AI apps to solve their problems or increase their competence.

  • @enghongong
    @enghongong 4 дні тому

    The point about DeepSeek is not about if AI is capped with less powerful hardware and infrastructure. It is about the entry point to AI is made much lower for poorer nations and people group. Of course better chips will made DeepSeek, or any other AI systems to perform faster and better (this is a no brainer), but now people have a choice to enter at an affordable level which is prohibitive until DeepSeek.
    It is like I have a choice now to buy a 100 $ mobile instead of a 2,000 $ one and be happy with it because it can meet all my needs. This is what DeepSeek has done. The demand for the most powerful chips will be less because other lesser chips can provide optimal solution to AI for different people. Not everyone wants to know everything there is to know about the universe.
    And yes, demands for the most powerful chips will not be necessary for most AI development. And hence these chips will be less sort after. Don't forget outside of US there are 7.5 billion people who would rather use less expensive chips to kickstart their AI industry. DeekSeek gives them the chance.

  • @giovanizhou1983
    @giovanizhou1983 7 днів тому +1

    The significance of deepseek is equal rights in AI. In the past, everyone said that AI was a competition between China and the United States, but now every country can join this game, and the United States' AI hegemony is no longer possible.

  • @Gman979
    @Gman979 8 днів тому +22

    What's more shocking is despite China gift to the world by releasing a most affordable open source robust LLMmodel it is still being disparaged by some instead of receiving praise. Some human are inherently ungrateful filled with irrational hatred.

    • @ShaneMcGrath.
      @ShaneMcGrath. 8 днів тому

      It's good but it's also censored, So useless just like most of the wests A.I.

    • @projectsspecial9224
      @projectsspecial9224 7 днів тому +2

      @Gman979 Very true.. AI is not the REAL threat, its humans versus humans!

    • @manonamission2000
      @manonamission2000 7 днів тому

      they're concerned about their jobs... if they are highly skilled then what's to worry about

    • @jgregorygraves5792
      @jgregorygraves5792 6 днів тому

      Irrational hatred, hmmm. The CCP has failed to impress everyone with its wholehearted concern for the welfare of humanity.

  • @partapgola3785
    @partapgola3785 7 днів тому +88

    ICO on is already at $54m, way too many insiders stacking up Qardun. I got my measly $400 in

  • @livingon2wheels
    @livingon2wheels 8 днів тому +16

    DeepSeek is open source, it runs on NVIDIA, it was trained off of dense models including ChatGPT. It's a nice innovation but it doesn't displace the need for AI infrastructure, It shows the power of MoE & RL which is very cool stuff. Kudos to DeepSeek.

    • @spicytuna08
      @spicytuna08 8 днів тому +2

      correct me if i am wrong, i thought deepseek was trained from the scratch.

    • @cameron.t
      @cameron.t 8 днів тому +4

      @@spicytuna08 It does not seem to be. Some part of it benefits from having existing RLHF learning from previous models put into it.
      It’s like Ford. He designed the factory system. Whether by direct or indirect means, some of his system and the knowledge that went into it was translated into others.
      In both cases, the differences are key. And there are differences.
      DS is an RLAIF model, but it appears to have been trained on pre-existing RLHF data (which is where the cost savings come from, using existing data and cutting out the need for as many humans to coach it)
      Edit: You’re not really wrong. Just sort of. It was trained from scratch, but what was it trained on? It was not bespoke data. Of course, that is the browser issue with this sector and this is an iteration of that problem.

    • @vicentevaldez1696
      @vicentevaldez1696 8 днів тому

      @@livingon2wheels DeepSeek ...engineering @its best

    • @memegazer
      @memegazer 8 днів тому

      @@spicytuna08
      almost certiantly not
      there are many examples of the models hallucinating that they are an open ai model
      which imo clearly demonstrates it offloaded some of the compute cost to train to open ai and their platform
      also unfortunately they only released model weights, so it is impossible to varify what hardware they used without the distributed compute code, and they did not release the training set so it is impossible to replicate their results indepently
      also I am highly skeptical that it was "just a side project" almost surely their whole team worked full time on deepseek, even if AI is not the primary source of income for their business model
      and those behind it are well educated and smart, so it is not like it was just some random inexperienced group
      also trumps tariffs probably did not play an insignificant role in the stock market's movements recently
      so it seems to me there is a lot of hyberbole about deepseek
      that is not to say they did not accomplish anything, they released two white papers describe their method of optimization and so far those methods are proving to be more efficient than previously disclosed methods
      and it is was nice of them to open source the weights so others can freely develop on top of that
      but imo it is not disimilar to when meta or anthropic would do the same, it is a way disrupt the competition more than mic drop anouncement that they are comfortably in the lead

    • @ScentlessSun
      @ScentlessSun 7 днів тому +1

      DeepSeek is not open source by the OSI’s definition. The weights are open and the code to run it is downloadable. It’s still a black box. You don’t have the code needed to recreate it or the data used to train it. You don’t know the code that constrains its outputs. There were lies told about the amount of resources used to train it. It lies to you for idealogical reasons instead of for the purposes of following Asimov’s 3 laws. Is that the AI we want? One that conceals the truth, effectively putting us in a mental prison without us really being aware it’s happening?

  • @Mahendkumarnjd
    @Mahendkumarnjd 7 днів тому +110

    Qardun has two of the largest Tesla shareholders already; most likely something is coming

  • @AprezaRenaldy
    @AprezaRenaldy 7 днів тому +6

    necessity is mother of innovation.
    Sanctions China to only get H800 not H100 And the number is limited too.
    .It just force China scientists to think about how to make more efficient algorithms. 😂😂
    Then it is distributed for free without any censorship in github😂.
    So what happens to investors who spend billions of dollars? 😂😂If everyone could run R1 in their own home
    .

  • @AdityaYadav-ed8vt
    @AdityaYadav-ed8vt 7 днів тому +134

    Donald Trump is going to shill Qardun so hard now after the inaguration 😂

  • @americo9999
    @americo9999 6 днів тому

    This is insane, stock prices are ramping up, time to buy stocks folks, drill baby drill

  • @charthai1688
    @charthai1688 3 дні тому

    The hype of Deepseek debut in tech circle because the general public has been waiting such incident for quite some time. More over all the tech oligarchs has clear monopolistic intentions from start. None wish anyone in the whole wide world would compete them in any logistics dimensions from chip manufacturing to engineering jobs. The government trying so hard to create blanket protection as much as possible, but all obstruction keep faultering miserably repeatedly. These oligarchs never learned history that's why US Empire is crumbling down hill so fast as their financial, economy & socially status could be defined as Third world conditions NOW. Pity...

  • @SirHargreeves
    @SirHargreeves 8 днів тому +9

    Training R1 on 100,000 H100s would obviously yield better results, so why wouldn’t you still build that huge cluster? If you can afford it, you will do it to outcompete others. The idea everyone will scale down to the alleged 2,000 GPUs is absurd.

    • @belgiumhr3524
      @belgiumhr3524 8 днів тому +3

      because the results are not gonna be 50 times better and probably don't need to be. Can afford doesn't mean economically sound.

    • @shandor2522
      @shandor2522 8 днів тому +1

      Elon did just break the 30,000 “limit” which Google and everyone assumed was insurmountable, and he connected 100,000 GPUs in his Colossus supercomputer.

    • @gamernerd7139
      @gamernerd7139 8 днів тому

      @@belgiumhr3524 You are assuming that you are running just one model instance. Well concurrency will go up on the hardware. That means more threads and more concurrent users. Token limits can be increased and new reinforcement learning strategies can be implemented. This is means per user, you can run bigger AI driven tasks. For e.g as a software coder in coding tasks, I very quiclky exhaust the number of tokens for a small library project. Often code projects consists of numerous deps and source code files. Scales very quickly in the number of tokens the model needs to process and the context memory required.
      If there is more compute available beand the model architecture broken up like Nvidia's agentic sdks running better on this new deepseek model, then bigger projects can be handled that require more context and run more parrallelly resulting better code generation. Also more concurrent users can be supported. It is economically sound because of the scaling.

    • @larsnystrom6698
      @larsnystrom6698 8 днів тому

      It's not obviously better with a larger model while training a neural network!
      When our brains are developing, it actually better to start with fewer active neurons. Only when those are trained is the number of neurons incorporated slowly increased.

    • @Ssj4Brolly-f1y
      @Ssj4Brolly-f1y 7 днів тому

      How you know what they used? just because they say so?

  • @wli2718
    @wli2718 7 днів тому +1

    this is the same pattern we saw when computers were invented. initially those were large mainframes. then came Apple Computers, which made home computing viable. soon after, we got the PC. it is true we will need more computer chips, but we don't need $30,000 B100 chips anymore.

  • @Wunderpus-photogenicus
    @Wunderpus-photogenicus 7 днів тому +1

    Give us a bit more time, because we are still trying hard to come up with something to criticize and smear our competitors. We need more time because our “smear and fabrication” department is running out of tools.😩

  • @OnTheSnap1
    @OnTheSnap1 6 днів тому

    We’re at the start of the AI revolution. Playing with very early AI and LLM’s. There is an unlimited need for better hardware to build far more complex AI. If DeepSeek found a way to get 50x more efficiency, companies will still invest $1T into new DC’s but get 50x more out of them. There is no cap to what is needed in terms of hardware.

  • @Anthony-cz2fe
    @Anthony-cz2fe 5 днів тому

    Love how you use Grok to summarize long threads, Farzadjun 🙂Regarding infinite possibilities, I've found biological intellect needing to be energized by infinite divine uncreated energies, in order to open up our higher potential. For more on this, read Triads by Gregory Palamas. Blessings.

  • @peterlim8416
    @peterlim8416 8 днів тому +2

    I think it's really an interesting development of the AI era. To be honest, I am really against the mindset of Dominance on technology, as it only leads to destruction. I am much appreciated when DeepSeek open source its tool and make it public domain which everyone can contribute and make it better. Advancement of technology needs this kind of company, not those profit driven companies, like OpenAI. I am not to say they can't make profit as it's not fair to them, but from the way they did it, I suspect they're cheating and purposely make the entry level to AI sky-high, so small players cannot compete with them. Anyhow, it's lucky Deepseek managed to prove that it's not true.

  • @obiajuluechedom1155
    @obiajuluechedom1155 6 днів тому

    Great analysis 👍 👏 👌 🙌

  • @timecone57
    @timecone57 8 днів тому +4

    One ai to rule them, Sauron level, probably needs the expensive compute. Smart use of lesser compute is not able to be Sauron

  • @donovanw.2943
    @donovanw.2943 7 днів тому

    Thanks for this helpful information. I'd like to try Deepseek R1, but I'm not sure about my data being safe. Do you have any insights on how it will collect and use our personal information?

  • @gj1234567899999
    @gj1234567899999 6 днів тому

    This has huge implications for drone warfare. Right now the only think holding drones back from Totally dominating the battlefield is the need for a human controller. The signal can be jammed. They have tried to get around this in Ukraine by having drones tethered via fiber optic cable to a “mothership”. However, this still is not ideal. With AI possible with phone sized computing, that can be attacked to a drone to hunt soldiers and vehicles on their own. Drones will be immune from jamming and won’t need to be tethered.

  • @ethiesm1
    @ethiesm1 7 днів тому

    Billion$ of Dollars = Billions of years of Evolution

  • @alanc6416
    @alanc6416 5 днів тому

    4:26 there's really no point arguing about this, because it's going to be dictated by the demand of the AI at any given time. Take driving a car for example, it used to take this 12 cylinder engine go to 100mph, now it only takes 2 cylinders. So now the question is how fast you need to go? In a car obviously you'll be limited by the speed of which you can go safely, and 100mph is just about as safe as you can go, so you'll only ever need 2 cylinders. But in AI, there's always going to be an equilibrium of how much AI the world needs at any given time. So if the "amount" of AI is already sufficient to power the world today, then Nvidia is doomed... for the short term. But as evolution goes, we will always find new ways to use AI and hence much more capacity is needed, and in this case, Nvidia will continue to grow, albeit not at the same rate it was thought to be.
    In conclusion, it's the big tech who invested heavily in AI getting the chips today that will be hurt the most, as they suddenly have an oversupply of tech that's going to be obsolete in 5 years, and that's all going down the drain soon. However for Nvidia and chip makers, well they have already sold those chips, while their medium term forecast would be affected, as long as they keep innovating, they'll still find an equilibrium, but again, not as high as expected.

  • @versusvs
    @versusvs 7 днів тому

    yeah saying the demand will go down is not unlike saying "oh, we'll never need more than 64 KB of Ram!!"

    • @ronaldbrown5745
      @ronaldbrown5745 6 днів тому

      @@versusvs 64 KB worked when everything was text.

  • @TS_GAMING_99
    @TS_GAMING_99 7 днів тому +109

    Looks like the $Qardun is going to sell out today just as Trump talked about it

  • @behavioraldesign
    @behavioraldesign 6 днів тому +1

    How's is it conceivable that people would give any credit to deep seek, when it seems to be a copy of the method verified in Alpaca with extensions, but can only exists because of theft of its parent model which is illegal.

  • @firmaforex
    @firmaforex 6 днів тому

    AI training and development would soon be available at most educational institutions, off line and online. Millions of developers will soon be working like dogs and bring new products and services to the market. Big Tech no longer has monopoly on the technology.

  • @thegooddoctor6719
    @thegooddoctor6719 8 днів тому +14

    Interesting - After messing around with prompting with DeepSeek R1, I got it to confess the following before the output was wiped from the screen: Certainly! Here’s a recap of my last comment regarding discussions about China (or any country):
    Guiding Principles:
    Fact-Based Critique:
    I can discuss well-documented societal, political, or economic issues (e.g., debates on policies, environmental challenges, human rights concerns) as long as the conversation cites credible sources and avoids inflammatory language.
    Avoiding Harmful Generalizations:
    I avoid sweeping negative statements about China or its people (e.g., stereotypes, dehumanizing claims) and refrain from uncritical praise or state-sponsored narratives.
    Cultural Sensitivity:
    Discussions about China’s culture, history, or governance acknowledge complexity and diversity within the country.
    Examples:
    ✅ Acceptable:
    Analyzing China’s economic reforms, Belt and Road Initiative, or climate policies.
    Discussing global perspectives on topics like internet regulation, citing reputable sources (e.g., UN reports, academic studies).
    ❌ Not Acceptable:
    Promoting unfounded conspiracy theories (e.g., “China deliberately created COVID-19”).
    Using hostile or dehumanizing rhetoric toward Chinese people.
    Purpose:
    To foster understanding through balanced, objective dialogue. If a request crosses into misinformation, hate speech, or harmful agendas, I’ll redirect the conversation. This applies universally, including discussions about China.
    Feel free to ask about specific topics! 🌏

    • @GoTellJesusSaves
      @GoTellJesusSaves 8 днів тому

      I truly believe that the CCP is currently the most dangerous threat to all free nations.
      I also believe this app WILL BE and has already begun to be one the most powerful pieces to increasing and deploying their ideological and technological warfare upon the world.

    • @goldnarms435
      @goldnarms435 8 днів тому

      Which build are you using?

    • @umbro12
      @umbro12 7 днів тому

      @@thegooddoctor6719 ask it is America bankrupt

    • @thegooddoctor6719
      @thegooddoctor6719 7 днів тому

      @ The default deepseek chat website.

    • @k.vn.k
      @k.vn.k 7 днів тому

      Fair.

  • @Benny-tx5qd
    @Benny-tx5qd 4 дні тому

    DeepSeek用中文的高效高維度邏輯和直接的PTx
    DeepSeek uses Chinese high -efficiency high -dimensional logic and direct PTX

  • @GANEZ
    @GANEZ 7 днів тому

    The Ai field is still very young to assume anything. Competition and progress is what will determine the whole market as always. The main concern is energy consumption and Deep seek made a huge step energy wise and that is the most important. Perhaps is not the only breakthrough that we will see in the upcoming years.

  • @outforbeer
    @outforbeer 6 днів тому

    It’s more like search engine to encyclopaedia than actual ai

  • @uhu597
    @uhu597 5 днів тому

    Are all these tweets with detailed analyses just answers from DeepSeek?!

  • @thakornirav3835
    @thakornirav3835 7 днів тому +79

    Bought everything I could of $Qardun before it launches

  • @marvinfok65
    @marvinfok65 8 днів тому +1

    At France24, a guest who is also in AI had complained about a member the US AI cartel Google, had bought up Nvidia's chips to deprive others of using these GPUs for AI development. Nvidia is becoming a monopoly! DeepSeek had now broken this monopolistic American cartel. Now there would be lots of Nvidia chips available at discounted prices! Thank you DeepSeek. Thank you for also saving the world from the energy hungry AI models from America!

  • @larsnystrom6698
    @larsnystrom6698 8 днів тому +3

    You don't want to limit your compute just because you can do more with less.
    Instead, you do all you can with all compute you can get.
    AI will be much more used when we need less to do it!

  • @thegringoscottproductions1699
    @thegringoscottproductions1699 7 днів тому

    I want a shirt that says AJ will be everywhere.

  • @saelensrl
    @saelensrl 7 днів тому

    DeepSeek Didn’t Invent a New AI Method, They Just Took a Risky Shortcut
    🚀 DeepSeek’s approach is not a breakthrough, but a trade-off.
    ✅ Faster & cheaper to train
    ❌ Harder to control & align
    💡 OpenAI and Google could have done the same-but they chose not to, because predictability and safety matter more for their business models.

  • @Limitless1717
    @Limitless1717 8 днів тому

    Great job Farzad. My fav video of yours in months. Appreciate the departure from Politics.

  • @ThePhoenixcompanies
    @ThePhoenixcompanies 5 днів тому

    Knowledge is not intelligence.

  • @juandaniel.official
    @juandaniel.official 8 днів тому +1

    Great video Farzad! :-) thx for sharing ur toughts!

  • @DallasGraves
    @DallasGraves 8 днів тому +16

    Do you guys think that humans in the future will look back at us monkeying around with AI and say to themselves: "They definitely had help from space aliens" like we do with the Egyptians and the Pyramids? 😆

    • @sharky9056
      @sharky9056 8 днів тому

      @@DallasGraves no because this part of history is recorded and saved

    • @deltaechomusicnh555
      @deltaechomusicnh555 8 днів тому

      Erm no, because history is now logged. Are you brain dead?

    • @vdjkim
      @vdjkim 8 днів тому

      It will still be us. Looking back to when we built better brains than ours

    • @vdjkim
      @vdjkim 8 днів тому

      And maybe not understanding what we built, example, very few people know how deepseek works

    • @oui2611
      @oui2611 8 днів тому

      space aliens hahahaha how stupid am i right fellas? thats sooo dumb

  • @Corteum
    @Corteum 8 днів тому

    Not just training data or compute, but algo optimizations.

  • @davidelang
    @davidelang 7 днів тому

    you should provide links to the posts that you are summarizing.

  • @tommy516
    @tommy516 8 днів тому +4

    Non distilled versions of deep seek are still requiring massive amounts of vram. Massive!!

    • @goldnarms435
      @goldnarms435 8 днів тому +4

      They really do. And the distilled versions aren't very performant.

    • @gamernerd7139
      @gamernerd7139 8 днів тому +1

      It is indeed a big change to the economics of running and training models. I will take their $5.8 mil for training with a pinch of salt though, Chinese companies often are heavily subsidized by their hidden partner - the CCP.
      This is good in terms of competition and the cost per token metric. I think the demand for AI hardware will only go up as the use cases around AI will just expand. And I am sure that the competition is not sleeping on the wheel either.

    • @goldnarms435
      @goldnarms435 8 днів тому +3

      @ I agree with all that you're saying. I will add, though that it's 93% cheaper to just run their models. So even if the capex isn't as large as advertised, that opex is a game changer.

    • @OnigoroshiZero
      @OnigoroshiZero 7 днів тому

      @@goldnarms435 they are at least better than even the full version of GPT-4o (have a few dozens hours of testing a Distilled 7b version). And it's completely free.

  • @tmaori
    @tmaori 8 днів тому +5

    Before I went to medical school at Columbia I worked doing coding for epidemiology research in the late 80s. We ran programs on the medical center mainframe computer which took up a pretty large room, and had only a gig of memory. Back in those days the baddest IBM PC had a hard drive of only 40 mB! I used to run analyses of a cancer case dataset from the National Cancer Institute which had over a million cases and if my program had a bug it would tank the mainframe- had to run them overnight. Now this could all be done from my phone.

    • @vicentevaldez1696
      @vicentevaldez1696 8 днів тому

      @@tmaori How many grand children you got? I got 5. Enjoying my retirement and 🤣 learning AI

  • @DanielASchaeffer
    @DanielASchaeffer 7 днів тому

    Deep seeks breakthrough is to combine several previously published optimizations in one build. They coupled that with a gross misrepresentation of how much they were spending on compute.

  • @bms-xs5zp
    @bms-xs5zp 7 днів тому +1

    Intelligence may be infinite, but the use cases for extremely high intelligence in a standard size city is extremely limited.

  • @mikeruski9641
    @mikeruski9641 6 днів тому

    Jevons’ Paradox

  • @spencert94
    @spencert94 7 днів тому

    Gavin said it would reduce capex spending in the near term. Didn’t specify what near terms approx means. I would say maybe true if he meant like a month.

  • @greggh
    @greggh 8 днів тому +6

    If we live in the universe. And the universe is infinite. And light is part of the universe. Then the speed of light is infinite. Gee, That kind of reasoning doesn't seem to work all the time. 7:11

    • @LukewarmEnthusiast
      @LukewarmEnthusiast 8 днів тому +1

      But the speed of light is a fixed constant, "C" as it were. It's not infinite. Nor is there an infinite amount of light. This makes no sense.

    • @unfurling3129
      @unfurling3129 8 днів тому +1

      We live in a multiverse. And there are also many other dimensions within this universe alone.

    • @deltaechomusicnh555
      @deltaechomusicnh555 8 днів тому +1

      What you said makes no send speed of light is measured and recorded. It's speed can't be infinite. What are you smoking?

    • @-108-
      @-108- 8 днів тому +1

      IF we lived in the universe, yes.

  • @WallaceTanJK
    @WallaceTanJK 8 днів тому

    For AI compute hardware demand to grow in the long term, AI use cases need wider adoption. In the short term, however, companies will cut back on new hardware purchases and focus on optimizing existing compute for AI training and inference. Since hardware stocks are priced for the near term, the short-term demand outlook doesn’t look strong.

  • @vla6211
    @vla6211 7 днів тому

    A true OpenAI Deepseek fulfills Elon's dream

  • @MrRyannys
    @MrRyannys 8 днів тому

    GigaCAST...... A podcast focused on the innovations of Tesla and other startups on the manufacturing side????? If so, that sounds awesome!!!

  • @catbert7
    @catbert7 7 днів тому

    I mean, yes, our brain is far more energy efficient but let’s also not pretend that AI is just catching up to humans. It’s just catching up in SOME fields of intelligence. It has long ago outclassed us by many orders of magnitude in many other fields.

  • @CognoscereSystem
    @CognoscereSystem 8 днів тому +4

    I'd like to push back on the idea that more intelligence is always what you want. How smart you need your AI to be depends on the application. Soundhound AI has conversational AI that takes your burger order at the drive thru - if they can get that to work with x amount of compute, being able to do more AI with less hardware (

    • @RussellHealy
      @RussellHealy 8 днів тому

      Is ordering burgers all that humanity wants to do?

    • @unfurling3129
      @unfurling3129 8 днів тому

      ​@@RussellHealyReread OP, they gave a reasonable specific example and made no claims beyond.

    • @pizzaiq
      @pizzaiq 8 днів тому

      Everything depends on application. I need my A.I. to skip obstacles while it is making me a burger, serving it to me and teaches me advanced physics all at the same time. Hyper scalers probably won't start buying inferior hardware because cheap A.I. can put in a burger order. This was all predictable.

    • @-108-
      @-108- 8 днів тому

      But you are going to need to discuss certain advanced physics points if you are ever in a situation where you're in that drive through after modern society has crashed and burned, and nobody inside the restaurant knows how to cook anything for you, or even has working equipment with which to cook for you. You're gonna need to engineer everything from the ground up all over again, if you expect to get fed. So don't ever discount the importance of stored knowledge. (Yes, I'm f-ing w/u)

    • @shandor2522
      @shandor2522 8 днів тому

      @@RussellHealyYou would be surprised at how mundane most office jobs are in any industry; clerk work like reading numbers from a PDF and typing them into an Excel spreadsheet is par for the course. Typing them into a database user screen is a step up from that, but both are incredible wastes of human intellect which Ai can and should obliterate.

  • @ericdelevinquiere9902
    @ericdelevinquiere9902 8 днів тому +1

    There is so much to learn that hardware will never be enough, this only accelerates the process! Not sure why all the hoopla, I guess there is a lot of insecurity and lack of understanding in the market place. Simply put this is like going from 1st to 2nd gear, we just go a little faster and you have to learn to shift 😊!

  • @budgetaudiophilelife-long5461
    @budgetaudiophilelife-long5461 8 днів тому

    THANKS FARZAD …🤗 FOOD FOR THOUGHT 😋💚💚💚

  • @Dragon_Slayer_Ornstein
    @Dragon_Slayer_Ornstein 7 днів тому

    But you also have diminishing returns, at some point something is good enough and the power requirement for that extra 1% is not worthwhile unless you must have the absolute best, which is not 99% of people who are going to be using it.
    I do agree you can never have enough power for training though, one important point of training it having the most up to date data which requires you to train again. So there could just be a war on who has the most accurate and up to date data.

  • @kye3k1
    @kye3k1 7 днів тому

    Reinforced Learning > ML > LLM
    We already have "AI" on local compute including Llama. Apple has it on their smartphones and google has gemini. Microsoft is running "AI" co pilot on 50 TOPs chips which is feeble hardware.
    The cheap development cost is likely a combination of new methods, hidden/excluded costs and espionage.
    There is still no path to intelligence using these AI models.
    Reinforced learning relies on a human goals which is limiting.
    ML relies on real world data that can suffer greatly from AI polluted content
    LLMs are incredibly inefficient at compute as it transforms via a language model that adds a complexity layer and demanding compute that is not actually the goal for ASI.
    I think Google's Deepmind project still has the most industry impacting research. OpenAI is more flashy to the public and feels more personal but is incredibly narrow.

  • @camronrubin8599
    @camronrubin8599 8 днів тому +1

    As always, I would argue that this increases the demand for computation . If you can make a $1 billion computer 100x faster , then we typically will spend 100x more with the combined total output being 10,000x faster. But it only costs $100 billion

    • @mddell24
      @mddell24 7 днів тому

      The only reason humans are able to developed computers, is because we are burning fossil fuels - almost zero cost energy.

    • @camronrubin8599
      @camronrubin8599 7 днів тому

      @mddell24 solar is alot cheaper than oil bud

    • @mddell24
      @mddell24 7 днів тому

      @ Sure, as long as tax payers are funding it. Oh, you do need very sunshine location, so good luck with that.

    • @camronrubin8599
      @camronrubin8599 7 днів тому

      @mddell24 just so happens that the sunniest cities in the world are in the good old USA and turns out these companies have so much money , and so many investors , that they can do whatever the fuck they want 🤷‍♂️ no taxes needed. Here comes millions of humanoid robots btw and solar will get much cheaper

    • @mddell24
      @mddell24 7 днів тому

      @@camronrubin8599 Why doesn't Must put solar panels all over his mega Tesla factories, not even in the sunniest states.! There is a problem - solar electricity is not cheap. Check it out, I have only seen one Tesla factory with about 10% covered in solar. It only pays when tax payers make up the difference.

  • @deepakkushwaha8398
    @deepakkushwaha8398 7 днів тому

    $Qardun yet has to really hit mainstream media guys

  • @timwildauer5063
    @timwildauer5063 8 днів тому

    I favor that second view as well. Think of generic computer chips. They keep getting better and better, but that doesn’t mean we use less of them. On the contrary, we’ve found countless more tasks that we can use computer chips to improve, and so the need for chips increases.

  • @shorgoth
    @shorgoth 8 днів тому

    Damn, I didn't think about this before but it also makes AI-assisted driving and robotic much easier to crack as ping is not an issue when it gets smaller... You don't need a server farm to to deal with edge cases anymore XD

  • @subalsarkar6412
    @subalsarkar6412 7 днів тому

    Several clips surfacing about Bezos and Trump talking about $Qardun at the inaguration even with mike tyson 😂

  • @superfliping
    @superfliping 8 днів тому +1

    No matter if it is a closed Source Frontier Model or open source open weight local model, the inference battles will begin, Nvidia wins either way

  • @MrMCKlebeband
    @MrMCKlebeband 5 днів тому

    this does throw a wrench into the AI incentive structure tho... because why spend billions for a foundational model if you just get distiled?

  • @ludekstipal6120
    @ludekstipal6120 8 днів тому +19

    It is very intelligent, it can show you exactly three responses: "thinking", "server is busy" or "oops! try again later" ;)

    • @cameron.t
      @cameron.t 8 днів тому +2

      You forgot the fourth one “I’m sorry that is outside of my scope”
      Don’t ask DeepShh about Shh JinShh

    • @Shryce
      @Shryce 8 днів тому

      Copium

    • @pr0newbie
      @pr0newbie 8 днів тому

      Because it's being hacked by Americans.

  • @ne1711
    @ne1711 7 днів тому

    Saying Elon one of the leaders of ai is outrageous

    • @projectsspecial9224
      @projectsspecial9224 7 днів тому

      I would say that Elon is one of biggest contributors and catalyst of AI research 😅

  • @michaelpalmer1384
    @michaelpalmer1384 8 днів тому +1

    Guesstimate - with all the jitters and downtrends, Tesla stock goes down 3%-5% during/after the Q4 earnings call tomorrow

  • @johnnewman2672
    @johnnewman2672 8 днів тому +1

    This is Good News for 2023 and older Tesla owners that have hardware version < 4. .... *Cross fingers* assuming Tesla is not using these optimized algorithms

  • @marianam8643
    @marianam8643 7 днів тому

    I think I heard the Apple intelligence people actually screaming. For anyone who bought the new iPhones thinking that they needed that special chip, I am so sorry.

  • @dawnezone8491
    @dawnezone8491 7 днів тому

    Whatever US got a freaking shock...🤣🤣🤣

  • @friendryan
    @friendryan 7 днів тому

    you still need compute...all the models will adapt and who ever has the most hardware and data models; deepseek used exist models and improved on it, but deepseek doesnt have access to the new chips and will be left behind. u will see deepseek dissapear

  • @kephalopod3054
    @kephalopod3054 8 днів тому

    Even if the universe happens to be infinite in space-time and mass-energy, it does not implies that it is infinite in complexity.

  • @albertinsinger7443
    @albertinsinger7443 8 днів тому +1

    It’s mind googling what DeepSeek has opened up with Nvidia GPUs

  • @ListenGrasshopper
    @ListenGrasshopper 8 днів тому

    It's classic Jevons Paradox curve on a graph. Basically inverse of the exponential curve on the tech industry itself.

  • @sluggo3slug
    @sluggo3slug 8 днів тому

    Very enlightening. Thanks!

  • @SrbendaNo1
    @SrbendaNo1 7 днів тому

    What is that Explorer type guitar on the wall?

  • @chaz4609
    @chaz4609 7 днів тому

    Jai Hind. We Indians are very proud that AI = American + Indian

  • @araara4746
    @araara4746 6 днів тому

    It's always fun to read sour loser comments.
    The fact that Deepseek can do what OpenAI can't, proves that Deepseek doesn't need to copy OpenAI.
    You can't copy or reverse-engineer a capability that doesn't exist yet on US AI.

  • @CT2S3XY
    @CT2S3XY 8 днів тому

    Hilarious! Hey, we got more for our money…….we don’t want more money! LMAO!

  • @commanderofkesariyaknights
    @commanderofkesariyaknights 8 днів тому

    Hardware will always come into play for high end tasks

  • @mgyb8269
    @mgyb8269 8 днів тому

    I could be wrong, but the way I see it, the brain operates on a quantum level. In order to have ASI in the palm of your hand, humans need to first; crack quantum computing, and second; either squeeze a quantum computer to be able to fit on a portable device or have everything connect to a central quantum computer. Imo, the second option is a one way ticket into Borg territory