Latest / AI Fire Daily / 🎙️ EP 351: OpenAI to Navier–Stokes Problem & Insilico AI Drug Shows Anti-Aging Signal

Transcript

🎙️ EP 351: OpenAI to Navier–Stokes Problem & Insilico AI Drug Shows Anti-Aging Signal · AI Fire Daily

• 0:00
  Intro music. Think about the absolute hardest
• 0:04
  problem, you know? Right. Did a swarm of like
• 0:06
  10 ,000 AI agents just solve math's hardest problem?
• 0:11
  Maybe. And, you know, that same technology might
• 0:14
  be figuring out how to reverse human aging. Yeah,
• 0:16
  it is a completely wild day for science. Welcome
• 0:19
  to today's Deep Dive. We have a massive stack
• 0:22
  of paradigm shifting news today. It's September
• 0:24
  8th, 2026. Our mission is to unpack these major
• 0:29
  AI breakthroughs. We want to understand what
• 0:31
  they actually mean for you. We have a pretty
• 0:33
  incredible roadmap for today. First, we are tackling
• 0:35
  a highly controversial claim from OpenAI. They
• 0:39
  might have solved the Navier -Stokes Millennium
• 0:41
  Prize problem. Yeah. Second, we're looking at
• 0:43
  the rapid pace of progress. We have a rapid -fire
• 0:45
  look at major AI model updates. We'll see how
• 0:48
  autonomous agents are dominating human benchmarks.
• 0:51
  Finally, we examine a truly mind -bending medical
• 0:54
  trial. An AI -designed lung drug accidentally
• 0:56
  showed signs of reversing biological age. Let's
• 0:59
  kick things off with this massive math claim.
• 1:02
  OpenAI dropped a staggering official report earlier
• 1:06
  today. They are using an unreleased, highly advanced
• 1:09
  model right now. Right, this model... is significantly
• 1:13
  more capable than GPT -6 Astra. But the model
• 1:18
  itself isn't the craziest part here. It's how
• 1:20
  they actually deployed the technology. They scaled
• 1:23
  it up to 10 ,000 concurrent AI agents. Whoa.
• 1:27
  I mean, imagine scaling to 10 ,000 agents talking
• 1:30
  for 88 hours. It is genuinely hard to wrap your
• 1:32
  head around. They essentially created a massive
• 1:35
  digital research facility overnight. These agents
• 1:38
  worked continuously for almost four straight
• 1:40
  days. And they weren't just computing simple
• 1:42
  numbers in isolation either. They were actively
• 1:45
  collaborating and communicating the entire time.
• 1:47
  They acted like a giant specialized digital construction
• 1:50
  crew. Exactly. You had different agents acting
• 1:52
  as architects and builders. They were passing
• 1:54
  complex mathematical blueprints back and forth
• 1:57
  continuously. It's like stacking Lego blocks
• 1:59
  of data, but the Legos are PhD level math. That
• 2:02
  is a great way to visualize the mechanism. They
• 2:05
  sent millions of distinct messages during this
• 2:07
  run, and it cost millions of dollars in sheer
• 2:10
  computing power. They focused all this energy
• 2:13
  on the Niger -Stokes equations. Let's unpack
• 2:16
  why these specific equations matter so much.
• 2:19
  They're incredibly fundamental to understanding
• 2:21
  our physical world. They describe exactly how
• 2:24
  different fluids flow and move. Yeah, they explain
• 2:27
  how water currents move through the ocean. They
• 2:29
  explain how air moves over a jet engine wing.
• 2:32
  But predicting this fluid movement mathematically
• 2:34
  is notoriously difficult. The math eventually
• 2:37
  breaks down and becomes completely chaotic. The
• 2:39
  Clay Mathematics Institute named it a Millennium
• 2:42
  Prize problem. That means there is a $1 million
• 2:44
  bounty for solving it. The proofs OpenAI generated
• 2:47
  focuses on a very specific phenomenon. It shows
• 2:50
  something called finite time blow -up in fluid
• 2:53
  dynamics that uses smooth external force to prove
• 2:56
  this outcome. Let's translate that complex idea
• 2:59
  into plain English. It means showing a fluid's
• 3:01
  behavior becomes mathematically impossible to
• 3:04
  calculate. The equations literally break down
• 3:06
  at a certain point. Right, and proving that breakdown
• 3:08
  satisfies one of Clay's formulations. But here's
• 3:12
  where the story gets really interesting. OpenAI
• 3:15
  didn't just output a standard academic research
• 3:18
  paper. They backed this entire claim up with
• 3:21
  a lean proof. A lean proof is a math proof verified
• 3:25
  by a computer program. This is a huge deal for
• 3:28
  mathematical verification. It completely removes
• 3:31
  human error from the initial checking process.
• 3:33
  The computer physically guarantees that the logical
• 3:36
  steps are sound. It gives mathematicians a much
• 3:38
  stronger foundation for verification. But there's
• 3:41
  a massive controversy brewing in the math world
• 3:43
  right now. The community is deeply divided over
• 3:46
  this news today. Oh, yeah. Human mathematicians
• 3:49
  are feeling incredibly sidelined right now. Two
• 3:51
  prominent researchers were already working on
• 3:53
  closely related results. Kristen Buckmaster and
• 3:56
  Levent Alpuj were making significant progress.
• 3:58
  They were doing the hard, slow work of human
• 4:00
  mathematics. But they had used open AI tools
• 4:03
  during their daily research. OpenAI strictly
• 4:06
  denies accessing their specific user data directly.
• 4:10
  They say they didn't just steal the researchers
• 4:12
  private work, but they admitted something that
• 4:14
  raises some really big questions. They cannot
• 4:16
  rule out indirect influence from de -identified
• 4:18
  usage data. This means anonymized research patterns
• 4:22
  might have trained the swarm. It makes you wonder
• 4:25
  how these massive swarms actually learn. Did
• 4:29
  they absorb the human intuition without realizing
• 4:31
  it? Because the Clay Institute still lists Navier
• 4:34
  -Stokes as totally unsolved today. So it remains
• 4:37
  a claimed solution, not an accepted one. It'll
• 4:40
  take human experts many months to verify everything.
• 4:43
  So does a computer proof count if human mathematicians
• 4:47
  can't read it? Well, if the software perfectly
• 4:49
  verifies the underlying logic of it all, human
• 4:52
  comprehension might just be an optional bonus.
• 4:54
  So if the machine checker says yes, human reading
• 4:57
  is optional. It is a radical shift in how we
• 4:59
  define human knowledge. That ability for AI to
• 5:02
  check its own work isn't just theoretical math.
• 5:04
  It's bleeding into how these models execute real
• 5:07
  world tasks. This perfectly connects to our second
• 5:10
  major topic today. The relentless pace of AI
• 5:13
  progress is just staggering right now. Individual
• 5:16
  models are completely dominating standard human
• 5:19
  benchmarks. GPT -6 Astra just achieved something
• 5:23
  totally wild in education. It actually aced South
• 5:26
  Korea's full CSAT college entrance exam. Yeah,
• 5:30
  it scored a perfect 450 points. That is one of
• 5:33
  the hardest standardized tests on Earth. And
• 5:36
  it became the very first AI to achieve this milestone.
• 5:39
  Right. And it did it with incredible computational
• 5:41
  efficiency. It'd be rival models like Claude
• 5:44
  and Gemini quite handily. But the progress isn't
• 5:47
  just happening in standardized testing environments.
• 5:49
  Meta is flexing some serious autonomous research
• 5:52
  muscle today, too. Their autonomous research
• 5:54
  system just competed fiercely on Kaggle. For
• 5:56
  those unaware, Kaggle is a massive data science.
• 5:59
  platform. Companies post real world problems
• 6:01
  and offer cash prizes for solutions. It is a
• 6:04
  brutal proving ground for data scientists. Meta's
• 6:06
  system beat almost 4 ,000 human competitors on
• 6:09
  real tasks. It actually finished in the top 10
• 6:12
  overall. People don't realize why Kaggle is so
• 6:14
  incredibly difficult. Real world data is remarkably
• 6:18
  messy and full of missing values. It requires
• 6:21
  serious human intuition to clean and organize
• 6:23
  it. The fact an agent navigated that autonomously
• 6:27
  is a massive leap. Absolutely. The money following
• 6:30
  this autonomous technology is also totally staggering.
• 6:33
  A software company called Cognition just raised
• 6:36
  $2 billion. Their total valuation jumped to $26
• 6:39
  billion in four months. Their annualized revenue
• 6:42
  jumped from $492 million to $900 million. They
• 6:46
  build autonomous AI software engineers for major
• 6:49
  tech companies. It proves that AI coding is a
• 6:51
  highly lucrative space. We are seeing incredible
• 6:54
  adoption rates across the entire industry. But
• 6:57
  things are moving just as fast on the consumer
• 6:58
  side. ChatGPT Images 2 .5 is officially rolling
• 7:02
  out to users today. It promises much faster generations
• 7:05
  and far better image fidelity. ByteDance is also
• 7:08
  preparing a massive new project right now. They're
• 7:10
  focusing heavily on real -time spatial video
• 7:12
  generation. They're building an entirely new
• 7:15
  world model from scratch. A world model is an
• 7:17
  AI that simulates realistic physical environments.
• 7:20
  Zhang Yiming is personally leading this new development
• 7:23
  project. A new sedans -based spatial model could
• 7:27
  launch next month. Independent creators are already
• 7:29
  finding clever ways to use this. One creator
• 7:32
  broke the whole video editing process down beautifully.
• 7:35
  They shared 10 simple prompts to automate social
• 7:38
  media video editing. I still wrestle with prompt
• 7:41
  drift myself. Oh, yeah. Prompt shrift is when
• 7:44
  an AI slowly ignores your initial instructions.
• 7:47
  Exactly. So having exactly 10 reliable video
• 7:50
  editing prompts sounds amazing. It takes you
• 7:52
  from raw footage to finished content instantly.
• 7:55
  But there are definitely dark clouds forming
• 7:58
  in this rapid development. Security is becoming
• 8:00
  a massive issue for regular users today. Hackers
• 8:03
  are actively stealing tokens from unsuspecting
• 8:05
  people online. Tokens are the digital currency
• 8:08
  used to pay for AI processing power. Yeah, hackers
• 8:10
  are completely draining cloud users' money. You
• 8:13
  really have to watch out for the early warning
• 8:18
  signs. And that brings up another massive issue
• 8:20
  regarding consumer privacy. Meta's new Muse agent
• 8:24
  is raising some very serious questions. It is
• 8:27
  now live inside your email and calendar apps.
• 8:30
  It reads everything to help organize your daily
• 8:33
  digital life. It even connects to your personal
• 8:35
  payments and health apps. It is totally free
• 8:37
  to start using right now. But you must hand over
• 8:40
  massive amounts of personal access. It needs
• 8:43
  to read your entire life to actually assist you.
• 8:46
  It analyzes your doctor's appointments and your
• 8:48
  grocery receipts simultaneously. So are we trading
• 8:51
  our ultimate privacy for Muse's convenience?
• 8:54
  Well, you basically surrender your digital secrets
• 8:56
  to get the perfect automated assistant. Right.
• 8:59
  You pay for the ultimate digital assistant with
• 9:01
  your personal data. It is a profound daily tradeoff
• 9:04
  that everyone has to navigate now. Sponsor. Yeah.
• 9:07
  So we have these agents. operating in entirely
• 9:09
  digital spaces. But what happens when that autonomous
• 9:12
  logic tackles physical biology? This brings us
• 9:15
  to our final and most fascinating segment. This
• 9:18
  is where AI starts directly impacting the human
• 9:21
  body. In silico medicine published some incredible
• 9:24
  new clinical trial data today. They were testing
• 9:27
  a specific drug called Rentocertib. Their AI
• 9:31
  system completely designed this unique medical
• 9:33
  drug. It was originally meant to treat idiopathic
• 9:37
  pulmonary fibrosis. That is a terrible disease
• 9:39
  where scarring slowly chokes the lungs. The patient's
• 9:42
  lung tissue becomes incredibly thick and stiff
• 9:45
  over time. The AI specifically picked the biological
• 9:48
  target protein for this disease. Then it literally
• 9:51
  drew the molecular structure of the drug itself.
• 9:54
  They ran a human clinical trial over the past
• 9:57
  year. It included 42 human patients suffering
• 10:00
  from the severe condition. They took deep blood
• 10:02
  samples from all of these treated patients. Then
• 10:05
  they compared these samples against the massive
• 10:07
  UK biobank database. They used six different
• 10:11
  aging clocks to analyze these specific samples.
• 10:13
  Aging clocks are AI models that read age markers
• 10:16
  from your blood. Yeah, they analyze tiny biological
• 10:18
  changes to determine your true biological age.
• 10:21
  They look at things like DNA methylation and
• 10:23
  protein levels. All six clocks rated the treated
• 10:26
  patients as biologically younger. One group's
• 10:28
  age estimates fell by 2 .7 to 3 .5 years. Two
• 10:32
  sec silence. That is a massive biological reversal
• 10:35
  from a simple lung drug. It is a completely unexpected
• 10:38
  systemic benefit for these patients. It's like
• 10:41
  finding out your cough medicine accidentally
• 10:43
  fixes your car's engine. You just don't expect
• 10:45
  a lung treatment to reverse cellular aging. Totally.
• 10:48
  But underneath it all, human biology is highly
• 10:51
  interconnected. The AI targeted a specific protein
• 10:54
  responsible for lung scarring. That same protein
• 10:57
  might regulate cellular aging across the entire
• 10:59
  body. Here is a really crucial detail from the
• 11:02
  study data. The dosing that showed the clearest
• 11:05
  aging signal was unique. It was completely different
• 11:08
  from the dosing for lung function gains. This
• 11:11
  potentially shows deep biological effects beyond
• 11:14
  just basic disease relief. It suggests the drug
• 11:17
  is interacting with fundamental aging mechanisms
• 11:19
  directly. Dario Amode, the CEO of Anthropic,
• 11:23
  had a great quote recently. He said, real medical
• 11:26
  breakthroughs will turn the tide on AI sentiment.
• 11:29
  He believes actual biological cures will work
• 11:31
  better than marketing. When people see their
• 11:33
  relatives cured, their fear of AI vanishes. Stories
• 11:36
  like this show the public what medicine may soon
• 11:39
  become. We are only at the starting line of this
• 11:42
  incredible race. AI -designed drugs are just
• 11:45
  now trying to reach the consumer market. We are
• 11:47
  going to see a flood of new molecules very soon.
• 11:50
  Could aging just be another engineering problem
• 11:52
  AI is solving? It seems biology is basically
• 11:55
  a vast software program that AI is debugging.
• 11:58
  Exactly. Biology might just be complex code that
• 12:01
  AI is deciphering. It completely changes how
• 12:04
  we view the human body's limits. We stop seeing
• 12:06
  aging as inevitable and start seeing it as fixable.
• 12:09
  Let's step back and look at the big picture today.
• 12:11
  These sources all point to one massive, undeniable
• 12:15
  shift in technology. AI is no longer just mimicking
• 12:18
  human text anymore. No, it is actively generating
• 12:21
  novel truths about our physical universe. We
• 12:23
  see a 10 ,000 agent swarm cracking millennium
• 12:26
  math problems. We see AI agents beating top human
• 12:29
  researchers on messy Kaggle datasets. And we
• 12:32
  see AI -drawn molecules reversing biological
• 12:34
  clocks in humans. These systems are uncovering
• 12:37
  things we never knew before. They're pushing
• 12:39
  the absolute boundaries of what is scientifically
• 12:41
  possible. It is a profound transition from artificial
• 12:44
  intelligence to artificial discovery. This leaves
• 12:47
  us with a really deep question to ponder today.
• 12:50
  If AI swarms can solve our hardest mathematical
• 12:52
  proofs so quickly. And if they can engineer age
• 12:56
  -reversing drugs in a matter of hours. What exactly
• 12:58
  is the role of the human scientist a decade from
• 13:01
  now? Are we becoming just the managers of these
• 13:04
  AI geniuses? Think about that as you go about
• 13:07
  your day. Check your own daily workflows right
• 13:09
  now. What is one complex problem in your life
• 13:12
  today? Could you unleash an AI agent on it right
• 13:15
  now? You might be surprised by what it can discover
• 13:17
  for you. Thank you for joining us for this deep
• 13:19
  dive. We truly appreciate you spending your time
• 13:21
  with us today. Out to your own music.