Latest / Elon Musk Podcast / Chinese DeepSeek Shakes Silicon Valley

Transcript

Chinese DeepSeek Shakes Silicon Valley · Elon Musk Podcast

• 0:01
  Hey everybody. Welcome back to the Elon Musk
• 0:04
  Podcast. This is a show where we discuss
• 0:07
  the critical crossroads that shape SpaceX, Tesla X, The
• 0:11
  Boring Company, and Neurolink. I'm your host, Will Walden.
• 0:18
  So how did a Chinese AI company suddenly cause so many ripples
• 0:23
  in the tech world? And why didn't NVIDIA, the
• 0:25
  leading AI chip maker, lose $600 billion in market value in one
• 0:32
  day? And is deep seeks low cost, high
• 0:35
  performing AI model a sign of an emerging global AI race, or
• 0:41
  something a little less threatening?
• 0:44
  So on Monday, the NASDAQ fell 3.1%, sending shock waves across
• 0:50
  the stock market. And NVIDIA bore the brunt of the
• 0:52
  losses, with its stock price plummeting 17% and wiping out
• 0:56
  $600 billion in its market value.
• 1:01
  The cause of this dramatic drop? Something called Deepseek, a
• 1:04
  relatively unknown Chinese AI firm, unveiled a cutting edge AI
• 1:09
  model that deliver comparable performance to industry leaders,
• 1:13
  yet require significantly less compute power and cost far less
• 1:18
  to train. Now the implications of this
• 1:21
  announcement spread very quickly.
• 1:23
  For some, Deep Seeks achievement reaffirmed concerns about
• 1:26
  Silicon Valley's costly AI development practices.
• 1:29
  Others, particularly national security advocates, saw as a
• 1:33
  wake up call in the race for artificial intelligence
• 1:35
  dominance, particularly between the United States and China.
• 1:41
  And for those worried about AI safety, DEEP Seeks Open Access
• 1:43
  model represents a step toward uncontrolled proliferation of an
• 1:48
  advanced AI systems. Yet as transformative as Deep
• 1:52
  Seeks rise might appear, its actual significance remains a
• 1:57
  topic of debate. While some see it as a moment of
• 2:00
  reckoning for Western AI labs, others question whether it truly
• 2:05
  changes the competitive dynamics of this industry.
• 2:07
  And one of the biggest concerns about Deep Seek is the
• 2:11
  uncertainty surrounding its long term goals.
• 2:14
  The company is a spin off of High Flyer, a decade old Chinese
• 2:18
  quantitative hedge fund Co founded by Deep Seeks CEO.
• 2:23
  Originally developed for trading algorithms, Deep Seek eventually
• 2:27
  shifted to building large languid models and committed to
• 2:30
  open sourcing its tech. Its stated mission is ambitious
• 2:34
  yet pretty vague to unravel the mystery of AGI with curiosity.
• 2:41
  And for now, Deep Seeks business model seems almost altruistic
• 2:46
  and offers its EI tools to developers at minimal cost
• 2:49
  without clear profit driven incentives.
• 2:51
  But that could change at any moment.
• 2:54
  The company could pivot to monetization by introducing
• 2:57
  subscription fees or restricting its EPI access.
• 3:01
  Some theorize the Deep seat could also funnel user data into
• 3:04
  high Flyers hedge fund operations to extract financial
• 3:07
  insights. Now another layer of uncertainty
• 3:11
  comes from the Chinese government, which might step in
• 3:14
  to regulate how Deep Seek distributes its AI.
• 3:17
  This prospect of a Chinese company providing advanced AI
• 3:20
  tools to developers worldwide, including in nations considered
• 3:24
  adversarial to China, is unlikely to go unnoticed by
• 3:28
  Beijing. Adding to the concern is how
• 3:31
  Deep Seeks managed to build its breakthrough models.
• 3:34
  It didn't invent a fundamentally new AI architecture, though.
• 3:37
  Instead, it optimized existing techniques pioneered by American
• 3:41
  firms like Open AI and Google. Using older hardware and
• 3:46
  streamlined processes, Deep Seek achieve performance levels
• 3:49
  comparable to some of the best AI systems available today at a
• 3:53
  much lower cost. Now this approach shows the
• 3:57
  vulnerability for USAI labs. The technological leads are
• 4:02
  becoming increasingly short lived.
• 4:04
  In early years, it might have taken months or years for
• 4:08
  competitors to replicate innovations like Open AI, GBT
• 4:11
  models and deep seeks. Rapid optimization shows that
• 4:15
  the time lag between cutting edge development and global
• 4:18
  adaptation shrinking is anthropic Co founder Jack Clark
• 4:23
  succinctly put it deep seek means AI proliferation is
• 4:28
  guaranteed now. Deep seeks rise also exacerbates
• 4:34
  the perception of AI development as a geopolitical race.
• 4:38
  Some venture capitalists and tech advocates have long framed
• 4:42
  AI progress as a competitive moment between the United States
• 4:47
  and China. Their argument rests on fears,
• 4:50
  then, an authoritarian regime armed with advanced AI could
• 4:53
  outpace democratic nations and use AI technologies for
• 4:57
  strategic or military dominance. And critics warned that this
• 5:01
  narrative serves self-interest goals.
• 5:03
  By casting AI as an arms race, proponents can argue against
• 5:08
  regulation that might slow innovation or reduce returns on
• 5:11
  investments in AI startups. Additionally, it channels more
• 5:16
  funding into military technology products, which often benefit
• 5:19
  the same venture capitalists advocating for this framing.
• 5:23
  Yet there's a genuine danger in accelerating AI development
• 5:26
  under the banner of competition. When safety concerns are
• 5:29
  sidelined, the risk of releasing powerful, unaligned AI systems
• 5:34
  grows. While US labs like Open AI and
• 5:37
  Anthropic have attempted to build safety measures into their
• 5:40
  AI models, these efforts remain incomplete.
• 5:44
  Deep Seek, on the other hand, appears to have no public facing
• 5:47
  strategy for ensuring its AI systems are safe or aligned with
• 5:51
  human values. And if you have any human values
• 5:54
  of your own, you take a look at the subscribe button or the
• 5:57
  follow button on whatever podcast platform you're
• 5:59
  listening on right now. Because I've noticed that only
• 6:02
  10% of you, around 10% of you are actually subscribed to the
• 6:06
  show. So if you could do me a favor
• 6:08
  and hit the subscribe button, I would love that.
• 6:11
  Now the people really think the best thing that humanity can do
• 6:15
  is create things smarter than ourselves, though, and then
• 6:18
  we'll see what happens after that.
• 6:20
  The lack of a clear plan for managing superhuman AI
• 6:23
  intelligence only heightens concerns that framing AI as a
• 6:26
  race will lead to reckless outcomes.
• 6:29
  But despite these concerns, there are reasons to believe
• 6:32
  that Deep Seeks emergence doesn't fundamentally alter the
• 6:35
  AI industry's trajectory, at least not yet.
• 6:40
  For one, much of what Deep Seek has achieved was already
• 6:43
  anticipated. Open source models have been
• 6:45
  reverse engineering breakthroughs from major labs
• 6:47
  for years, and the cost of training AI models has been
• 6:50
  declining steadily. The cost of producing GPT 4
• 6:55
  level intelligence has dropped by a factor of 1000 in just 18
• 6:59
  months. The further price drops, while
• 7:03
  disruptive, were expected and won't necessarily destabilize
• 7:08
  major players right now. Additionally, U.S. companies
• 7:12
  continue to invest heavily in building infrastructure for more
• 7:14
  powerful AI systems. And while Deep seeks cost
• 7:18
  effective approach may be ideal for current AI tech, major firms
• 7:22
  like Open AI and Google remain focused on training increasingly
• 7:26
  advanced models that require state-of-the-art hardware.
• 7:29
  These investments in AI infrastructure are aimed at
• 7:32
  pushing the boundaries of what AI can do, not just optimizing
• 7:36
  existing technologies. Another factor working in favor
• 7:40
  of US labs is the Biden administration's export controls
• 7:43
  and advanced ships to China. Although Deepseek has
• 7:46
  demonstrated its ability to innovate with less powerful
• 7:49
  hardware, restrictions are still relatively new and may limit
• 7:53
  China's ability to scale its AI efforts over the long term.
• 7:57
  The lack of access to cutting edge chips could make it harder
• 8:01
  for Chinese companies to deploy their models as widely as US
• 8:04
  firms can. And while Deepseek has made
• 8:07
  headlines for its optimization, American companies still lead in
• 8:11
  conceptual breakthroughs. Much of their industry's current
• 8:14
  focus has shifted toward developing AI agents capable of
• 8:18
  more autonomous and complex, complex human like tasks.
• 8:22
  And Open A is new offering, such as tools that can operate
• 8:25
  independently to perform tasks, hint at where the next wave of
• 8:29
  competition will unfold. Deep Seek has yet to demonstrate
• 8:32
  similar ambitions, and there's little evidence that it can help
• 8:35
  peace US firms in this area. But the sudden rise of Deep Seek
• 8:41
  is kind of concerning for people.
• 8:43
  Its success has revealed vulnerabilities in the business
• 8:46
  models of major AI labs in the United States, and it's raised
• 8:50
  questions about how AI technologies should be shared or
• 8:53
  regulated internationally. However, it's still challenging
• 8:58
  for all parties involved. The AI industry is shaping the
• 9:04
  way that we do business in the way that we live, in the way
• 9:08
  that we communicate. Now there's declining costs of
• 9:12
  open source technologies. I'm a huge proponent of open
• 9:15
  source. I built my whole web development
• 9:18
  career on open source. I built one of the most popular
• 9:23
  frameworks for for a software to launch me into my career on open
• 9:30
  source technology, on Bootstrap. So I understand that open source
• 9:36
  is a huge, huge win for AI. Do you want the huge tech
• 9:42
  companies building everything and controlling everything?
• 9:45
  Or do you think this is free speech?
• 9:47
  I want to know in the comments, do you think code is speech?
• 9:51
  Because if I build something and I keep it to myself and that I
• 9:54
  sell it to you. Capitalism, right?
• 9:57
  But if I build it and I give it away and I let other people
• 10:02
  build on top of it, not only does that ensure that that
• 10:06
  software continues to grow and become better, but it can
• 10:10
  transform into something completely different.
• 10:14
  And the language of that model could change as well.
• 10:18
  So if you believe that open source is a bad thing, please
• 10:22
  leave it in the comments below. And if you think it's a good
• 10:25
  thing, also tell me why. And if you want open AI to win,
• 10:30
  if you want Google to win, you know, that's capitalism at work
• 10:34
  right there. But for now, the emergence of
• 10:37
  low cost, high performance Chinese AI models is as much a
• 10:41
  moment for reflection as it is a cause for concern.
• 10:45
  Should we be concerned about this model?
• 10:48
  This time? We don't have any evidence that
• 10:49
  we should be. So use the open source model.
• 10:54
  Check it out. Grab its GitHub repo, test it
• 10:59
  out, tinker with it. If you're a software developer,
• 11:01
  it's a great way to check it out and get inside that code.
• 11:06
  Dig in. I'm a software developer at
• 11:08
  heart. I build websites for businesses,
• 11:11
  Will walden.com, WILWAL, don.com.
• 11:16
  I'm an engineer and I have been for about 20 years.
• 11:19
  So I'm very versed in this sector, and I believe that the
• 11:23
  safety that can come with open models like this is paramount
• 11:29
  collaboration, a huge, huge deal for open source software.
• 11:34
  And the innovation that could come from this could change the
• 11:38
  landscape and take the power away from those that are in
• 11:43
  charge right now, like Sam Altman at Open AI and Google
• 11:47
  and, you know, XAI, things like that could take some of that
• 11:50
  power away. And I think we need some of that
• 11:52
  power in our hands as a community, as an open source
• 11:55
  community. Think of WordPress.
• 11:59
  Most of the web is powered by WordPress.
• 12:03
  I don't know the exact stats, but it's somewhere it's over 50%
• 12:07
  is powered by WordPress. WordPress is an open source
• 12:10
  technology that anybody can change and fork can make into
• 12:13
  anything they want to make it into according to their terms.
• 12:17
  So something like that for AI that could power so many LLMS,
• 12:24
  could power so many different chat bots, can power so many
• 12:28
  different companies for free and open, I think that's a winner
• 12:32
  for everybody. Hey, thank you so much for
• 12:39
  listening today. I really do appreciate your
• 12:41
  support. If you could take a second and
• 12:43
  hit this subscribe or the follow button on whatever podcast
• 12:46
  platform that you're listening on right now, I greatly
• 12:49
  appreciate it. It helps out the show
• 12:50
  tremendously and you'll never miss an episode.
• 12:53
  And each episode is about 10 minutes or less to get you
• 12:56
  caught up quickly. And please, if you want to
• 12:59
  support the show even more, go to patreon.com/stagezero and
• 13:05
  please take care of yourselves and each other and I'll see you
• 13:08
  tomorrow.