Latest / AI Fire Daily / 🎙️ EP 355: Jensen Huang & Trump Phone Call & AI Improves Fetal Brain Anomaly Detection

Transcript

🎙️ EP 355: Jensen Huang & Trump Phone Call & AI Improves Fetal Brain Anomaly Detection · AI Fire Daily

• 0:00
  Imagine a piece of software so undeniably powerful
• 0:04
  that it's essentially being treated like a nuclear
• 0:07
  asset. I mean, politicians are actively calling
• 0:10
  it the new oil. Yeah, the rhetoric is intense
• 0:12
  right now. It really is. They see it as this
• 0:14
  massive geopolitical weapon, a tool that will
• 0:17
  completely define global dominance. But then
• 0:21
  I want you to imagine that exact same underlying
• 0:23
  math. It is quietly sitting in a hospital room
• 0:26
  in China right now. Wow. Yeah. The contrast there
• 0:29
  is huge. It is actively scanning a pregnant mother's
• 0:32
  ultrasound. It is literally learning to detect
• 0:35
  fetal brain anomalies in real time. That is just
• 0:38
  incredible to think about. The contrast between
• 0:40
  those two realities is absolutely staggering.
• 0:43
  Well, welcome to today's deep dive, everyone.
• 0:45
  We're really thrilled to have you with us. We
• 0:47
  really are. Today, we are opening up your curated...
• 0:50
  stack of AI sources. And we are unpacking a massive
• 0:53
  collision happening right before our eyes. A
• 0:55
  collision that touches almost every industry.
• 0:58
  Exactly. We're looking at the geopolitical sprint
• 1:00
  for AI supremacy. We're looking at the bitter.
• 1:04
  radioactive fight over safety regulations. The
• 1:07
  stakes are just incredibly high there. They are.
• 1:10
  Plus, we'll cover the mind bending scale of the
• 1:12
  newest language models. And finally, how humans
• 1:16
  are actually using this stuff on the ground to
• 1:18
  save lives. It really is a profound collision.
• 1:21
  I mean, we are witnessing the raw mechanics of
• 1:23
  a technological revolution. We really are in
• 1:26
  the thick of it. But, you know, to truly understand
• 1:28
  where AI is heading. We cannot just look at the
• 1:32
  software. Right. You have to look at the hardware.
• 1:35
  Exactly. We have to start physically. We have
• 1:36
  to look at the massive infrastructure being built
• 1:39
  in the real world to power it. Yeah. The scope
• 1:41
  of your sources here is honestly breathtaking
• 1:44
  because, well, before you get a smart chatbot,
• 1:47
  you need to pour concrete. Which brings us to
• 1:49
  our first major shift in the landscape. Let's
• 1:52
  look at this geopolitical sprint. Let's do it.
• 1:55
  Recently, there was a completely surprise phone
• 1:57
  call. Donald Trump actually called NVIDIA CEO
• 2:01
  Jensen Huang. Yeah, that call really set the
• 2:04
  tech world buzzing. And this happened live on
• 2:06
  stage, right, at the All In Summit. It was totally
• 2:08
  unexpected and it was highly public. Trump described
• 2:12
  AI data centers in very specific, highly strategic
• 2:15
  terms. He didn't hold back at all. No, he didn't.
• 2:18
  He literally called them the oil of the next
• 2:21
  20 to 25 years. The new oil. That is a heavy
• 2:25
  label. He highlighted their immense economic
• 2:28
  importance for the country. He framed them as
• 2:30
  the ultimate strategic resource. He also made
• 2:33
  a very stark argument about global competition
• 2:36
  there. He argued that slowing down development
• 2:39
  actively hurts U .S. competitiveness. Specifically,
• 2:43
  he pointed to the broader technology race against
• 2:47
  China. To him, speed is basically the only metric
• 2:50
  that matters right now. The economic stakes here
• 2:52
  are incredibly high for everyone involved. It's
• 2:55
  not just tech companies spending money anymore.
• 2:57
  Oh, absolutely not. The government is getting
• 2:59
  involved. The Pentagon is stepping into this
• 3:01
  space directly. They are actually considering
• 3:03
  a $5 billion loan for a company called Fluidstack.
• 3:08
  Fluidstack, right? And Fluidstack is heavily
• 3:11
  involved in pushing physical AI infrastructure
• 3:13
  forward. I mean, $5 billion is a staggering amount
• 3:18
  of money. It is massive. If that gets approved,
• 3:21
  it would become the Pentagon's largest loan ever.
• 3:24
  Wow. It basically proves that data centers are
• 3:26
  now a matter of strict national security. You
• 3:29
  know, think about the massive steel mills of
• 3:31
  the Industrial Revolution. That's a good comparison.
• 3:33
  They literally forged the physical backbone of
• 3:36
  an entire century. Today, these massive data
• 3:40
  centers are our new steel mills. Yeah, they are
• 3:43
  the foundries of the future. But they require
• 3:45
  immense amounts of land to build. They draw billions
• 3:48
  of gallons of water just for cooling. The resource
• 3:51
  drain is incredible. They require dedicated power
• 3:54
  grids just to keep the servers from melting.
• 3:56
  Is this frantic building pace actually sustainable
• 4:00
  for us? Right. And the energy demands alone are
• 4:03
  completely unprecedented. Totally unprecedented.
• 4:05
  We are talking about single facilities requiring
• 4:08
  the electrical output of a small city. It's hard
• 4:11
  to wrap your head around. You need the physical
• 4:13
  hardware to train the smartest models. Without
• 4:16
  that silicon foundation, your software simply
• 4:18
  cannot compete globally. Yeah. The economics
• 4:21
  point directly toward a massive land grab right
• 4:24
  now. Whoever builds the biggest computational
• 4:26
  infrastructure controls the future market. So
• 4:30
  let me ask you this, then. Are we treating AI
• 4:33
  as a tool to solve human problems, or are we
• 4:36
  fundamentally treating it as a territorial land
• 4:39
  grab? Right now, it is absolutely a land grab.
• 4:41
  The companies and countries hoarding the most
• 4:43
  compute are securing a monopoly on future intelligence.
• 4:47
  You cannot build a smarter AI without the physical
• 4:50
  real estate to process the math. So whoever pours
• 4:53
  the most concrete... ultimately controls the
• 4:56
  global intelligence market. Exactly. And the
• 4:59
  scale of this new oil rush is just massive. It
• 5:01
  really is. Which explains perfectly why any talk
• 5:04
  of slowing down has become so intensely controversial.
• 5:07
  If whoever builds the biggest data center wins
• 5:09
  the economic war, you can see why politicians
• 5:11
  are panicking. Oh, absolutely. The pressure is
• 5:13
  on. It perfectly explains why the debate over
• 5:15
  AI safety has turned completely radioactive.
• 5:18
  Yeah, it is a very tense subject. We are caught
• 5:20
  in a fierce ideological war right now between
• 5:23
  safety and raw speed. The rhetoric surrounding
• 5:26
  this debate has become incredibly fiery recently.
• 5:29
  We are seeing major political figures drawing
• 5:32
  hard lines in the sand. And just to be clear
• 5:34
  for you listening, we are neutrally reporting
• 5:36
  the viewpoints from your sources here. Right.
• 5:38
  We aren't taking any sides. But the language
• 5:40
  used in these sources is intense. Trump recently
• 5:43
  dismissed AI existential threat fears entirely.
• 5:47
  He publicly called those fears a hoax. He went
• 5:51
  further than just calling it a hoax, actually.
• 5:53
  He specifically attacked prominent safety advocates
• 5:56
  in the space. Yeah, he named names. He singled
• 5:58
  out Dario Emo Day. the CEO of Anthropic. And
• 6:02
  Anthropic is one of the leading AI companies
• 6:04
  heavily focused on safety guardrails. Exactly.
• 6:07
  Trump's quote was intense. He wrote, That is
• 6:12
  really strong language. He is actively framing
• 6:15
  safety regulations as a direct threat to American
• 6:18
  dominance. It is a profound escalation in tone.
• 6:22
  Jensen Huang, on the other hand, offered a slightly
• 6:25
  different, more nuanced perspective during that
• 6:27
  summit. Yeah, Huang took a slightly different
• 6:30
  approach. He criticized the extreme Doomer predictions
• 6:33
  that AI will inevitably destroy humanity. Right.
• 6:37
  He stated those apocalyptic predictions actually
• 6:40
  lack hard scientific evidence. But he also clearly
• 6:44
  noted that AI safety research still matters immensely.
• 6:48
  Right. Huang is really trying to walk the middle
• 6:50
  path there. But while the public shouting match
• 6:54
  happens. We are seeing a quiet counter movement
• 6:57
  within the industry itself. The companies are
• 7:00
  moving on their own? Microsoft just released
• 7:01
  a brand new AI code of conduct. They are drawing
• 7:04
  their own internal lines. What exactly does that
• 7:06
  code of conduct look like in practice? Well,
• 7:09
  they are explicitly programming their models
• 7:10
  not to hack external systems. Okay, that seems
• 7:13
  like a good baseline. And they are strictly banning
• 7:15
  models from evading human control. They are trying
• 7:19
  to self -regulate before governments step in
• 7:21
  and do it for them. The big three are also trying
• 7:23
  to find some common ground here. OpenAI, Anthropic,
• 7:27
  and Google are currently discussing shared AI
• 7:30
  safety standards. Yeah, a unified front. They
• 7:33
  are looking at collaborative model testing and
• 7:35
  extensive risk reviews before releasing new systems.
• 7:38
  But as your sources note, the really hard part
• 7:41
  comes later. Agreeing on standards in a boardroom
• 7:45
  is one thing entirely. Right. Actually enforcing
• 7:48
  them globally is a totally different beast. Other
• 7:51
  leaders like Elon Musk and Sam Altman also see
• 7:53
  the risks. They talk about the risks all the
• 7:55
  time. However, they completely disagree on how
• 7:57
  to actually regulate them without killing innovation
• 8:00
  entirely. You know, I have to admit something
• 8:02
  to you. Two sec silence. I sometimes wrestle
• 8:06
  with prompt drift myself. Oh, yeah. Prompt drift
• 8:09
  is incredibly frustrating. It really is. Prompt
• 8:11
  drift is when AI subtly changes how it answers
• 8:14
  your familiar queries over time. It just slowly
• 8:16
  shifts its behavior. I feel overwhelmed just
• 8:19
  trying to keep my own daily AI tools behaving
• 8:21
  consistently. I totally get that. Trusting these
• 8:24
  massive sweeping global safety agreements feels
• 8:28
  almost impossible from down here. It is a very
• 8:31
  valid anxiety to have. The technology shifts
• 8:35
  under our feet faster than the ink can dry on
• 8:37
  any new regulations. It really does. You get
• 8:39
  used to a tool and suddenly it behaves completely
• 8:41
  differently. Which brings up a critical question
• 8:43
  regarding these shared industry agreements. How
• 8:47
  can anyone realistically enforce these big three
• 8:50
  safety agreements on a global scale? Honestly,
• 8:52
  enforcement is practically a nightmare in this
• 8:55
  current landscape. Yeah. Policing proprietary
• 8:57
  models behind closed doors at Google or open
• 9:00
  AIs, difficult enough. But policing decentralized
• 9:03
  open source technology globally is nearly impossible
• 9:07
  right now. Because anyone can access it. Exactly.
• 9:10
  Anyone can download an open source model and
• 9:12
  just remove the safety filters. Safety agreements
• 9:14
  only work internally. Open source technology
• 9:17
  simply cannot be policed. Yeah. That is the harsh
• 9:20
  reality we are facing. And while the politicians
• 9:23
  and CEOs argue over the rulebook, the engineers
• 9:26
  are basically ignoring the noise. They are just
• 9:28
  working. They are just down in the labs building
• 9:30
  bigger, much faster engines. The bleeding edge
• 9:32
  capabilities are expanding at a terrifying rate.
• 9:35
  Let's look at what is actually happening under
• 9:37
  the hood. Let's do it. Elon Musk announced that
• 9:40
  Grok 4 .8 is currently in training and it is
• 9:43
  being built entirely on a brand new C++ stack.
• 9:47
  Building on a C++ stack is a huge deal for overall
• 9:51
  performance. Why is that so significant? It means
• 9:54
  they are stripping away the heavy, clunky code
• 9:56
  and writing much closer to the metal. It makes
• 9:59
  the AI process information incredibly fast. And
• 10:02
  the scale they are aiming for is just massive.
• 10:04
  The Grok... 4 .8 model will reportedly run on
• 10:08
  2 .5 trillion parameters. That number is absolutely
• 10:11
  huge. It is. Before we move on, let us quickly
• 10:14
  define parameters for everyone. The internal
• 10:16
  settings an AI uses to make its decisions. That
• 10:19
  makes sense. It's basically like stacking Lego
• 10:21
  blocks of data to build intelligence. The more
• 10:23
  blocks you have, the more complex and nuanced
• 10:26
  the final structure becomes. Whoa. I mean, imagine
• 10:29
  scaling to 2 .5 trillion parameters. It is a
• 10:31
  lot of blocks. The sheer computational math happening
• 10:35
  every single second is staggering to think about.
• 10:38
  It is practically incomprehensible. It really
• 10:40
  is. It requires massive clusters of specialized
• 10:43
  chips just to run the calculations. But don't
• 10:46
  refresh your apps looking for 4 .8 just yet.
• 10:49
  Not yet. The launch for that version does not
• 10:51
  look close at all. Grok 4 .7 is scheduled to
• 10:54
  come out first. OpenAI is pushing their own boundaries
• 10:57
  right now with GPT -6 Astra agents. Oh, the agent
• 11:01
  models. They recently shared an internal guide
• 11:03
  for getting better results from these autonomous
• 11:06
  systems. Right. They are telling users they need
• 11:08
  to meticulously clean up their skills, their
• 11:11
  workflows, and their task goals. Because these
• 11:14
  agents are not just answering simple questions
• 11:16
  anymore. Right. They are taking action. They're
• 11:18
  taking multi -step actions on your behalf. There
• 11:21
  is a major privacy catch with OpenAI right now.
• 11:24
  This is important. They launched something internally
• 11:26
  called Project Lily. This is something everyone
• 11:28
  listening really needs to pay attention to. What
• 11:31
  exactly is Project Lily? Well, OpenAI hired hundreds
• 11:35
  of human contractors to read users' private chats.
• 11:38
  Wow. These contractors are manually grading ChatGPT's
• 11:42
  answers to improve the system's empathy and accuracy.
• 11:45
  So real human eyes might actually be reading
• 11:49
  your sensitive prompts to train the next model.
• 11:51
  That is a crucial reminder for you to check your
• 11:53
  settings today. If you use these tools for personal
• 11:56
  or medical questions, go check what privacy settings
• 11:58
  you have enabled. Definitely check them. Apple
• 12:01
  is also making massive moves in the daily AI
• 12:03
  space to keep up. iOS 27 is officially rolling
• 12:07
  out to users today. This is the biggest Siri
• 12:09
  upgrade people have waited years for. Finally.
• 12:12
  If you have an iPhone 15 Pro or newer sitting
• 12:15
  in your pocket right now, check your updates.
• 12:17
  Siri is finally getting the deep contextual AI
• 12:20
  integration we were promised. It can actually
• 12:22
  understand what is on your screen now. Exactly.
• 12:25
  We are seeing AI acting as automated agents everywhere
• 12:28
  now. It is stepping out of the chat box and straight
• 12:30
  into the real world. Yeah, the real world applications
• 12:33
  are exploding. A system called Nioma is now running
• 12:35
  live product demos for enterprise companies.
• 12:38
  Live demo. Live demos. They have already run
• 12:40
  over 50 ,000 automated demos. The AI agent actively
• 12:44
  talks to website visitors and turns them into...
• 12:47
  to qualified sales meetings. Web search agents
• 12:49
  are also completely automating deep web research
• 12:52
  for users. How does that work? Instead of you
• 12:55
  clicking through Google links, they self -learn
• 12:57
  your specific use case. They scour the web, read
• 13:00
  the articles, and deliver much richer context
• 13:02
  directly to you. Even email is being completely
• 13:05
  taken over by AI agents. Hello Inbox is a new
• 13:08
  tool that helps test marketing campaigns and
• 13:11
  spots hidden deliverability issues. Inbox management
• 13:14
  is huge. And Slashy Assistant actually handles
• 13:16
  your daily inbox for you entirely. It tracks
• 13:19
  your follow -ups, writes responses, and automates
• 13:21
  recurring email work perfectly in your own voice.
• 13:24
  It really feels like we are handing over the
• 13:25
  keys to our digital lives. We are. You're trusting
• 13:28
  these autonomous agents with our most personal,
• 13:30
  sensitive workflows. But looking back at OpenAI
• 13:33
  for a second, does a program like Project Lily
• 13:35
  fundamentally compromise the trust we place in
• 13:38
  these intimate chatbots? You know, it is a classic,
• 13:42
  painful friction point in tech development right
• 13:45
  now. How so? Companies desperately need real
• 13:48
  user data to make the models smarter and safer.
• 13:51
  But users sacrifice their private thoughts to
• 13:54
  feed that machine. There is simply no free lunch
• 13:56
  in AI training. We are quietly trading our deepest
• 14:00
  privacy just to get a smarter digital assistant.
• 14:02
  Exactly. It is a subtle tradeoff we are all quietly
• 14:05
  accepting without reading the terms of service.
• 14:08
  We'll take a quick break here, but when we come
• 14:10
  back, we're going to talk about how this tech
• 14:11
  is hitting the healthcare sector. Stick around.
• 14:14
  Welcome back. We have talked extensively about
• 14:17
  the macro level tug of war happening today. We
• 14:20
  really covered a lot of ground. We discussed
• 14:22
  geopolitical land grabs, trillion parameter math
• 14:26
  and the privacy tradeoffs of the bleeding edge.
• 14:29
  But all of this only really matters when it is
• 14:31
  applied to the physical world. Absolutely. That
• 14:34
  brings us to the human body itself. This is where
• 14:37
  the technology gets profoundly personal. And
• 14:40
  honestly, this is where it gets incredibly hopeful.
• 14:43
  Yeah, it's a positive shift. We are looking at
• 14:45
  a major AI breakthrough in clinical health care
• 14:48
  right now. It involves a deep learning system
• 14:50
  known as PAICS. Let us define deep learning quickly.
• 14:55
  Software that mimics the human brain to recognize
• 14:57
  complex patterns. Perfect. This PAICS system
• 15:00
  was recently tested in a massive clinical trial.
• 15:03
  It was deployed across five different hospitals
• 15:06
  in China. The setup for this trial was truly
• 15:09
  fascinating to read about. It was. The AI was
• 15:12
  integrated directly into the ultrasound machines.
• 15:14
  It helps doctors detect fetal brain abnormalities
• 15:17
  much earlier in the process. And it does this
• 15:20
  during live, real -time, prenatal ultrasound
• 15:22
  scans. Right, live scans. Imagine being in that
• 15:25
  quiet ultrasound room. The gel, the wand, the
• 15:29
  anxious parents looking at the screen. It's a
• 15:31
  very tense moment. And right there, the AI is
• 15:33
  actively drawing bounding boxes on the monitor.
• 15:36
  It is scanning the pixels faster than the human
• 15:38
  eye can process them. The scope of the trial
• 15:40
  was highly specific and tightly controlled. The
• 15:43
  system was trained specifically to identify 10
• 15:46
  different types of complex malformations. And
• 15:48
  the results of the trial are actually quite remarkable.
• 15:51
  What did they find? AI -assisted sonographers
• 15:53
  saw a massive jump in their overall detection
• 15:56
  sensitivity. It actually increased from 78 .6
• 16:00
  % accuracy up to 87 .3 % accuracy. That is a
• 16:04
  nearly 10 % improvement in catching severe life
• 16:08
  -altering conditions early on. That is huge.
• 16:10
  And importantly, the false positive rates remain
• 16:12
  completely flat. That is the vital step. This
• 16:15
  is crucial to understand. It means the system
• 16:18
  did not simply flag more cases incorrectly out
• 16:21
  of caution. Right. It was genuinely finding microscopic
• 16:24
  anomalies that the human eye was simply missing
• 16:26
  on the screen. The AI -assisted scans did take
• 16:29
  about 40 seconds longer overall. But in the context
• 16:32
  of a prenatal scan, that is absolutely nothing.
• 16:35
  That's just a few extra breaths. It's the tiny
• 16:37
  fraction of time for a significantly safer medical
• 16:39
  outcome. Now, the trial only focused on high
• 16:42
  -risk pregnancies for now, so broader testing
• 16:44
  across general healthy populations is still definitely
• 16:47
  needed. But there is one critical illuminating
• 16:50
  detail in this study that we have to talk about.
• 16:53
  The AI was not perfect. No, it wasn't. Human
• 16:55
  experts still played an absolutely vital role
• 16:58
  in the diagnostic loop. The human doctors correctly
• 17:01
  overrode about 60 % of the AI's mistakes. That
• 17:05
  number is absolutely fascinating. 60%. More than
• 17:07
  half the time, the AI got confused by a sh**.
• 17:14
  This is the perfect encapsulation of our current
• 17:17
  AI era. The technology acts as an incredibly
• 17:21
  powerful second set of eyes. Yeah, a second set
• 17:24
  of eyes. But the human is absolutely still in
• 17:26
  control of the room. We are applying vital clinical
• 17:29
  judgment where the machine inevitably fails.
• 17:32
  It really highlights the synergy between raw
• 17:34
  silicon and human biology. It really does. The
• 17:37
  AI never gets tired. It processes millions of
• 17:40
  pixels perfectly every single second of the day.
• 17:42
  But the human understands the broader context,
• 17:45
  the physics of the ultrasound wand and the actual
• 17:47
  patient lying on the table. But look at that
• 17:50
  60 % human override rate again. With an override
• 17:53
  rate that high, is the machine actually failing
• 17:55
  or is it functioning exactly as a co -pilot should?
• 17:58
  It is absolutely functioning as a successful,
• 18:00
  safe co -pilot model. You think so? The machine
• 18:04
  acts as an ultra -sensitive safety net meant
• 18:07
  to catch every possible edge case. The doctor
• 18:10
  acts as the ultimate filter for clinical reality.
• 18:13
  Right. In high -stakes medicine, an oversensitive
• 18:16
  co -pilot is exactly what you want. You want
• 18:18
  it to point out everything suspicious. It acts
• 18:20
  as an oversensitive alarm, relying on human wisdom
• 18:23
  to filter the noise. Yeah. For now, the human
• 18:26
  remains the final vital authority in the loop.
• 18:29
  Let us take a step back and look at the big picture
• 18:32
  of your sources today. Let's do a quick recap.
• 18:34
  We are witnessing a massive ongoing tug of war
• 18:38
  globally. At the macro level, AI is explicitly
• 18:42
  the new oil. The new oil. It is a multibillion
• 18:45
  dollar geopolitical weapon being forged in massive
• 18:48
  data centers. Yeah. It is triggering fierce ideological
• 18:51
  battles over regulation, open source tech and
• 18:54
  global safety. But down at the micro level, the
• 18:57
  story changes completely. It really does. Whether
• 18:59
  it is Siri on iOS 27 managing your personal texts
• 19:03
  or a doctor looking deeply at a fetal ultrasound
• 19:05
  scan, AI is currently acting as a highly capable
• 19:08
  yet fallible assistant. It is a profound tool
• 19:12
  that desperately still needs our human intuition.
• 19:14
  It entirely relies on us to guide its immense
• 19:17
  computational power. But I want to leave you
• 19:21
  with one final thought to mull over today. Right
• 19:23
  now, doctors are correcting 60 % of the AI's
• 19:26
  mistakes. They are doing this safely inside those
• 19:29
  quiet ultrasound rooms today. But as Grok and
• 19:32
  GPT -6 push into the trillions of parameters,
• 19:35
  as they start training on data sets we cannot
• 19:37
  even comprehend, what happens to human oversight
• 19:41
  when the balance eventually tips? That is the
• 19:44
  big question. What happens the day the machine
• 19:46
  starts catching things doctors cannot even see?
• 19:49
  Two sec silence. At what point does the human
• 19:51
  safety net become the bottleneck? Thank you for
• 19:53
  taking this deep dive with us today. Keep asking
• 19:56
  hard questions and please stay deeply curious
• 19:59
  out there.