Latest / Today’s AI News / Biohub's $500M AI Push, Cancer Caught 3 Years Early, and AI That Thinks Like a Chef

Transcript

Biohub's $500M AI Push, Cancer Caught 3 Years Early, and AI That Thinks Like a Chef · Today’s AI News

• 0:00
  Welcome back to today's AI News, brought to you by 9X
• 0:03
  Productions. It is great to be here.
• 0:05
  So imagine if human biology wasn't a mystery, but a massive
• 0:08
  software program, and we finally decided to hit compile on the
• 0:11
  source code. Today, we're looking at what
• 0:14
  happens when tech giants pour hundreds of millions into
• 0:18
  treating our cells like a data set just waiting to be decoded.
• 0:21
  Yeah, we have a really fascinating deep dive ahead for
• 0:24
  you today. And.
• 0:24
  We absolutely do. Our mission for this deep dive
• 0:27
  is to really explore how AI is jumping out of those little text
• 0:32
  boxes and directly analyzing the physical world.
• 0:35
  Right, the real world. Exactly, from predicting
• 0:37
  cellular biology and finding invisible diseases to literally
• 0:41
  reverse engineering how human beings taste food.
• 0:44
  It's wild stuff. It really is.
• 0:45
  So let's just jump right into our first major development.
• 0:49
  Mark Zuckerberg and Priscilla Chan's Biohub just announced a
• 0:53
  massive, I mean massive $500 million AI biology swing.
• 0:57
  Yeah, That is just a staggering amount of capital to deploy into
• 1:00
  a single scientific vertical. I mean, they're calling this the
• 1:03
  Virtual Biology Initiative, right?
• 1:05
  And the underlying premise here is to push AI to the point where
• 1:08
  it can run highly accurate predictive biology simulations.
• 1:13
  And the breakdown of the funding, you know, it tells a
• 1:15
  very specific story about where the bottleneck actually is.
• 1:19
  One of them for. Sure, like 400 million of that
• 1:22
  half billion is going straight to data generation in imaging
• 1:26
  tech. Wow.
• 1:28
  Yeah, and the remaining 100 million goes to external
• 1:31
  research labs. They're partnering with some
• 1:33
  heavy hitters, too. NVIDIA, The Allen Institute ARC.
• 1:37
  Right, trying to build an ecosystem.
• 1:39
  Exactly, building open data sets and models to predict how human
• 1:42
  cells behave. But I kind of have to push back
• 1:45
  on the core philosophy. OK, lay it on me.
• 1:48
  So we look at the early days of large language models, right?
• 1:51
  The solution was basically just scrape the Internet and throw
• 1:54
  massive amounts of text at the machine.
• 1:56
  Just feed it everything right? But are they just trying to
• 1:58
  brute force human biology the same way by throwing massive
• 2:01
  amounts of data at it? I mean, language has rules that
• 2:04
  we invented. That is a very good point.
• 2:06
  Biology is so messy and chaotic, is it really that reducible?
• 2:11
  Well, honestly, that tension you just described, that is the
• 2:13
  defining scientific question of this decade.
• 2:16
  Yeah, Yeah. I mean, you have people like
• 2:18
  Google Deepmind's Demis Hasabis right?
• 2:21
  Who recently predicted that AI could eventually just end
• 2:24
  disease? Which sounds like science
• 2:25
  fiction. It does, but Biohub is putting
• 2:28
  serious real world money behind that exact same thesis and it
• 2:32
  all centers on scaling laws. OK, unpack that for us.
• 2:36
  So we know that scaling up data and compute cracked human
• 2:39
  language. We also know it cracked protein
• 2:41
  structure, which was, you know, a massive biological milestone.
• 2:44
  Right, the whole alpha fold breakthrough.
• 2:46
  Exactly. But the tension is that a living
• 2:48
  cell isn't a static protein. It's incredibly dynamic.
• 2:52
  It's constantly reacting to viruses, temperature changes,
• 2:57
  genetic mutations. It's alive.
• 2:58
  Right. It's a messy, complex reality
• 3:01
  and currently AI biology data sets they Max out at around 1
• 3:05
  billion cells. Which sounds huge.
• 3:08
  It sounds huge, yeah, but the human body has something like 30
• 3:11
  trillion cells. Oh wow.
• 3:12
  OK, so 1 billion is basically nothing?
• 3:14
  Pretty much, which is why Biohub's Alex Rives actually
• 3:18
  said they need an order of magnitude more data just to see
• 3:21
  if this approach even works. So that explains the $400
• 3:24
  million going straight to data generation?
• 3:26
  Exactly. The big bet is whether those
• 3:28
  established scaling laws will hold true for living tissues if
• 3:33
  they can generate enough data. The theory is the AI will learn
• 3:36
  the hidden grammar of cellular biology.
• 3:39
  It's a hidden grammar. I like that.
• 3:40
  Yeah, and if it learns that grammar, it can predict how a
• 3:44
  disease will spread or how a cell will react to a drug
• 3:48
  entirely in a simulation. So we're essentially talking
• 3:51
  about treating the human body like a system that can be
• 3:53
  debugged. That is exactly what they're
• 3:55
  aiming for, but. You know, the crazy thing is we
• 3:57
  don't actually have to wait five years for Biohub to finish
• 4:00
  building this massive foundation.
• 4:02
  No, we don't. AI is already decoding our
• 4:04
  bodies to save lives today. Yeah.
• 4:06
  So shifting gears slightly, the Mayo Clinic just published new
• 4:09
  data on an AI model called Red Made.
• 4:12
  Yes, and it completely redefines early detection.
• 4:15
  It really does. This is honestly one of the most
• 4:18
  robust real world applications of computer vision and
• 4:22
  healthcare that we've seen. For sure.
• 4:23
  So red Made do is an AI that reads standard CT scans, and the
• 4:28
  headline here is that it is spotting pancreatic cancer up to
• 4:32
  three years before doctors typically find it.
• 4:35
  Three years, That is huge, right?
• 4:38
  But the methodology is what really caught my attention.
• 4:41
  They didn't just give the AI scans of people who were already
• 4:44
  visibly sick. Right, they ran a retroactive
• 4:46
  study. Exactly.
• 4:47
  They fed the AI nearly 2000 routine CT scans that have been
• 4:51
  cleared. Like human specialist looked at
• 4:53
  them years ago and said yeah, these are completely normal.
• 4:56
  Normal scans. Yeah, but they belong to
• 4:58
  patients who eventually, unfortunately developed
• 5:00
  pancreatic cancer. And the AI looked at those
• 5:03
  normal scans and flagged 73% of the cases early.
• 5:06
  Wow. And at the two year mark before
• 5:09
  a clinical diagnosis, the AI was spotting roughly 3 times as many
• 5:13
  early cancers as highly trained radiologists 3.
• 5:15
  Times as many. Yeah, you really have to ask
• 5:17
  like what is the machine seeing that a 20 year veteran
• 5:21
  oncologist is missing? Well, the model is analyzing
• 5:24
  hundreds of quantitative imaging features, right?
• 5:27
  It's looking at microscopic textures and structural
• 5:30
  gradients in the tissue because you know a human doctor is
• 5:33
  looking for a physical mass, a shadow, a lump.
• 5:37
  Something obvious. Exactly.
• 5:38
  But the AI isn't looking for a lump.
• 5:40
  It's looking at the statistical relationships between the pixels
• 5:44
  to find these invisible ghosts of a disease that won't fully
• 5:48
  manifest for years. It's literally seeing a spectrum
• 5:52
  of reality that human eyes cannot perceive.
• 5:54
  And the context here, the biological context, is so
• 5:57
  critical for why this matters. Tell us about that.
• 6:00
  Well, pancreatic cancer is notorious, right?
• 6:03
  It hides deep in the abdomen and it rarely shows any symptoms
• 6:06
  until it has already spread. It's a really tough one.
• 6:08
  It is the five year survival rate sits below 15% because by
• 6:12
  the time human eyes can clearly see the tumor on a scan, the
• 6:16
  window for life saving surgery has usually just closed, right?
• 6:19
  So spotting those invisible precursors 3 years early, I mean
• 6:22
  that alters the entire trajectory of the disease.
• 6:25
  But here's my question. Historically implementing some
• 6:29
  cutting edge diagnostic tool, it requires hospitals to buy
• 6:33
  millions of dollars in new specialized hardware.
• 6:35
  Right. Or it forces patients into, you
• 6:37
  know, really invasive procedures.
• 6:39
  Exactly. So how is the Mayo Clinic
• 6:41
  actually getting this to patients without causing massive
• 6:45
  bottlenecks in the system? That is the genius of the Red
• 6:48
  Momo deployment. It relies entirely on the
• 6:51
  routine CT scans patients are already getting.
• 6:54
  Oh, interesting. Yeah.
• 6:55
  So you might go in for a scan because of, say, a suspected
• 7:00
  kidney stone or just unexplained abdominal pain.
• 7:03
  And while your doctor is looking at your kidneys, this AI is just
• 7:06
  running silently in the background on the hospital
• 7:09
  servers. Just double checking.
• 7:10
  Exactly. Double checking the pancreas for
• 7:12
  those invisible mathematical textures.
• 7:14
  It introduces 0 friction to the medical system.
• 7:17
  No new hardware, no extra appointments.
• 7:19
  It's just this invisible safety net.
• 7:21
  That is incredible. And you know that ability to
• 7:23
  spot invisible patterns? It isn't just locked away in
• 7:25
  $1,000,000 hospital servers. And well, no it is not.
• 7:28
  The exact same underlying architecture finding the hidden
• 7:31
  rules of a data set, is now running on personal laptops.
• 7:35
  Right on our desks. Yeah, to decode something much
• 7:37
  more personal, which is our own creativity, Individuals are
• 7:41
  building bespoke AI workflows for their daily creative lives.
• 7:45
  We are seeing a rapid democratization of these pattern
• 7:48
  recognition systems right now. You really are.
• 7:50
  The tools to build highly personalized AI agents are just
• 7:53
  accessible to anyone. You can totally bypass the need
• 7:56
  for a computer science degree. Case in point, today we have a
• 7:59
  guide on how to build a custom blog writing agent with
• 8:03
  absolutely 0 code. Zero code love.
• 8:06
  That using a tool called Lang Flow.
• 8:09
  Basically you can build a sub agent that perfectly mimics your
• 8:12
  specific writing style and it runs locally on your machine for
• 8:15
  free, minus the fractions of a cent you pay for the language
• 8:17
  model API calls. And setting an architecture up
• 8:20
  locally you know that offers massive advantages for data
• 8:23
  privacy. You own the tool exactly.
• 8:25
  But for a non-technical user tuning in, how do they actually
• 8:29
  construct this? So instead of writing lines of
• 8:31
  code, Langflow basically gives you a visual canvas.
• 8:34
  You just snap building blocks or nodes together.
• 8:39
  Exactly. You download it, you select the
• 8:41
  blog writer template and that gives you all the basic
• 8:44
  plumbing. Then you add a simple text input
• 8:46
  block called Topic where you type what you want the blog to
• 8:49
  be about. OK, pretty straightforward.
• 8:51
  But the critical step, the part that makes it actually sound
• 8:55
  like you, is the reference block.
• 8:57
  You take your absolute best existing blog post, paste it
• 9:01
  into a block, and wire it directly into the AI's
• 9:04
  generation node. So you are giving the model a
• 9:07
  stylistic mirror. A stylistic mirror.
• 9:09
  That's a great way to put it. Instead of trying to write a
• 9:11
  really complex prompt explaining your tone, you just give it the
• 9:14
  raw data. Let the AI extract your pacing,
• 9:18
  your vocabulary, your exact sentence structure.
• 9:20
  Right then you just plug in your API key for Open AI or Anthropic
• 9:24
  and test in the playground. It pulls your topic, runs it
• 9:27
  through the mirror of your reference text, and outputs a
• 9:29
  draft. That is so powerful.
• 9:31
  But here's a pro tip for you listening to take this to the
• 9:33
  next level. You can share this local tool
• 9:36
  using an MCP server, specifically Claude MCP.
• 9:41
  Let's define that real quick, because MCT is becoming a really
• 9:43
  foundational concept. Go for it.
• 9:46
  It stands for Model Context Protocol.
• 9:48
  Right. So think of Claude as this
• 9:50
  brilliant but isolated project manager sitting in the cloud.
• 9:54
  By setting up an MCP connection, you are essentially giving that
• 9:57
  project manager a secure phone line to call the highly
• 10:01
  specialized ghostwriter living on your local laptop, right?
• 10:05
  So Claude can now summon your personalized style match blog
• 10:08
  tool automatically whenever you ask it to draft something.
• 10:12
  You imagine having a digital ghostwriter living right on your
• 10:15
  machine. The connective tissue between
• 10:16
  the Mayo Clinic story and the slang flow guide is just
• 10:19
  fascinating to me. Well, whether an AI is analyzing
• 10:22
  the microscopic texture of ACT scan or analyzing the cadence
• 10:26
  and rhythm of your writing style, it is proving
• 10:28
  exceptionally good at extracting the hidden rules of a system
• 10:31
  from just raw data. Oh, for sure.
• 10:34
  Which actually brings us to an incredible story about
• 10:36
  extracting rules from a system that a computer literally cannot
• 10:41
  experience. OK, I know.
• 10:42
  Where are you going with? This the human sense of taste.
• 10:44
  Yes, a food robotic startup called Kaikaku AI just published
• 10:49
  a paper called Epicure and they are claiming this is a ChatGPT
• 10:53
  moment for food. I mean that is a very bold
• 10:56
  framing. A ChatGPT moment it is.
• 10:59
  How are they achieving a language model breakthrough for
• 11:02
  our taste buds? So the researchers started with
• 11:04
  a massive, messy data set of over 6000 real world recipe
• 11:09
  ingredients. They clean them up into about
• 11:11
  1000 usable food items. Then they fed the AI the
• 11:15
  ingredient list of countless recipes.
• 11:17
  They just showed the model how humans group ingredients
• 11:20
  together. Just the combinations.
• 11:21
  Right, just companies. And here is what is wild.
• 11:24
  The AI never saw any chemical breakdowns.
• 11:26
  Yeah, really. Really.
• 11:28
  It was never trained on human taste labels like sweet, salty,
• 11:31
  or sour. Yeah.
• 11:32
  Yet, purely by analyzing the combinations, the model
• 11:36
  successfully identified all 5 basic human tastes, it correctly
• 11:40
  ranked different types of Peppers by their exact spiciness
• 11:43
  level, and it accurately groued ingredients by global regional
• 11:47
  cuisines. That is just a master class in
• 11:49
  latent space mapping. Break that down for us latent
• 11:52
  space so. To visualize a latent space,
• 11:55
  imagine the AI creating this massive 3 dimensional map inside
• 11:59
  its brain. It starts placing ingredients in
• 12:02
  this space based entirely on their relationships in the data.
• 12:05
  Even though it doesn't know what taste actually is, it notices
• 12:09
  that ingredients used in similar context clustered tightly
• 12:12
  together in this 3D map. But how does it understand heat?
• 12:16
  Like how does it understand spiciness without a chemical
• 12:18
  profile? Well, think about the
• 12:20
  mathematics of human preference. The AI doesn't know a jalapeno
• 12:24
  burns your tongue, right? It has no tongue.
• 12:26
  Right, obviously. It just observes that
• 12:28
  malematically, whenever a jalapeno is introduced to a
• 12:30
  recipe, human beings almost always introduce a dairy product
• 12:34
  like sour cream, cheese or yogurt to the equation.
• 12:37
  Oh, to cool it down. Exactly.
• 12:39
  Every recipe modification passed down through generations is just
• 12:43
  a data point. Every swap or pairing is a
• 12:46
  signal about what humans think works.
• 12:49
  That makes total sense. The AI is reverse engineering
• 12:52
  human pain and pleasure purely by mapping how we historically
• 12:55
  combine items to balance those sensations.
• 12:58
  It's essentially treating human recipes like prompts for our
• 13:01
  biology. That's a great way to look at
• 13:02
  it. And Kaikaku?
• 13:04
  They aren't stopping at academic papers either.
• 13:06
  They're taking this learned mathematical understanding of
• 13:09
  flavor and pairing it with their robotics arm.
• 13:12
  To do what? To create autonomous food
• 13:15
  infrastructure for commercial kitchens.
• 13:17
  Wow. Yeah, so the AI is designing
• 13:19
  menus suggesting precise ingredient substitutions that a
• 13:22
  human chef might never even consider, and creating physical
• 13:26
  products based on its latent space map of human taste and
• 13:29
  texture. So it's not just generating text
• 13:31
  anymore. It is chef intuition translated
• 13:34
  into algorithmic robotics. Precisely.
• 13:37
  And, you know, to power all of these massive leaps from
• 13:40
  decoding cellular biology to reverse engineering flavor
• 13:43
  profiles, the underlying tech ecosystem has to move at
• 13:46
  breakneck speed. It really does.
• 13:48
  The foundational models, the developer tools, they're
• 13:51
  evolving daily at this point. Yeah, so let's do some rapid
• 13:54
  fire updates, the quick hits and everything else in AI today.
• 13:57
  Let's do it. What is leading the pack?
• 13:59
  So open AI just released GPT 5.5.
• 14:03
  Oh, huge. Yeah, this is their new Frontier
• 14:05
  model. Yeah, but the major focus isn't
• 14:07
  just better chat capabilities, it is heavily geared toward
• 14:10
  agentic coding and actual computer use.
• 14:14
  And we should say the distinction between chatting and
• 14:16
  computer use is vital, right? Previous models could write code
• 14:20
  and just hand it to you to execute, but computer use means
• 14:24
  the AI can actually take control of a virtual environment.
• 14:27
  It can drive the car. Exactly.
• 14:28
  It can open the terminal, run the code, see the error, and fix
• 14:32
  it autonomously. It moves the AI from just being
• 14:35
  an advisor to an active participant.
• 14:37
  And Speaking of active participants, Cursor just
• 14:40
  released their SDK. Another big one.
• 14:42
  Yeah, cursor has obviously been the dominant AI coding
• 14:44
  environment, but this totally changes their ecosystem.
• 14:47
  Right, because an SDK or software development kit allows
• 14:51
  third party developers to take Cursor's underlying agent
• 14:54
  technology and build it directly into their own products.
• 14:58
  So it's everywhere now. Exactly.
• 15:00
  It breaks Cursor's powerful agents out of their specific
• 15:03
  application and makes them portable across the broader
• 15:06
  software landscape. Mistral AI also made a major
• 15:09
  move. They launched Vibe Remote Agents
• 15:12
  and these are powered by their new Open Weights Medium 3.5
• 15:16
  model. These agents run cloud sessions
• 15:19
  that allow for parallel coding tasks.
• 15:21
  So let's clarify open weights first for everyone.
• 15:23
  It basically means the core mathematical parameters of the
• 15:26
  model are available for developers to download and
• 15:29
  modify. Right, unlike a closed system
• 15:31
  like open AI's. Exactly.
• 15:33
  But the parallelization, that is the real breakthrough here.
• 15:36
  How so? Normally if you ask an AI to
• 15:38
  build a complex feature, it has to finish task A before it
• 15:41
  starts task B. But Vibe breaks that bottleneck.
• 15:44
  It allows the AI to clone itself in the cloud, working on
• 15:47
  multiple disparate parts of a code base simultaneously.
• 15:50
  That's going to speed things up so much.
• 15:52
  Unbelievably so. Google is also removing some
• 15:55
  bottlenecks, specifically in workflow friction.
• 15:58
  They just added native file creation capabilities to Gemini.
• 16:02
  Oh, that's incredibly useful. Yeah, you no longer have to copy
• 16:04
  and paste text outputs, Jim and I can now autonomously export
• 16:09
  its work directly into Google Docs, Sheets, Slides, Microsoft
• 16:13
  Word, Excel and Markdown files. I mean, it sounds like a minor
• 16:17
  quality of life update, but in enterprise environments
• 16:20
  eliminating that friction of data transfer between the AI and
• 16:24
  the final presentation format that saves countless hours at
• 16:27
  scale. Oh for sure.
• 16:29
  Shifting over to the creative economy, 11 Labs launched 11
• 16:32
  Music, right? This is a full streaming
• 16:34
  platform featuring AI song generation, remixing and AI
• 16:38
  assisted track creation. They already have over 4000
• 16:41
  artists on board, but the crucial detail here is that they
• 16:43
  are offering creator payouts. Which is huge because that
• 16:46
  addresses the biggest point of friction in generative media
• 16:49
  right now. Right, getting paid.
• 16:50
  Exactly. By building an ecosystem where
• 16:53
  artists can actually monetize AI assisted creations, 11 Labs is
• 16:57
  really attempting to align the economic incentives of the
• 17:00
  creators with the advancement of the AI tools.
• 17:04
  We also saw a major international release today.
• 17:06
  Yeah, Xiaomi dropped MIMO V 2.5 Pro.
• 17:11
  OK, It's a very powerful new open source model.
• 17:14
  And this really highlights the intense global competition
• 17:17
  happening right now. Yeah, it really does.
• 17:19
  Chinese tech giants are consistently releasing open
• 17:22
  source models that rival or occasionally even benchmark
• 17:25
  higher than Western counterparts, right?
• 17:28
  It forces the global developer community to constantly evaluate
• 17:31
  the geopolitical origins of their underlying architecture.
• 17:34
  And that ties directly into the security landscape.
• 17:36
  Open AI released a comprehensive cybersecurity action plan today.
• 17:40
  Interesting timing. Yeah, they're working directly
• 17:42
  with the US government and industry leaders on threat
• 17:44
  coordination to sort of democratize cyberdefense.
• 17:48
  Because cybersecurity in the AI era is just an asymmetric
• 17:51
  battle. Bad actors are using language
• 17:54
  models to write polymorphic malware, execute highly
• 17:57
  personalized phishing campaigns, and at an unprecedented scale.
• 18:02
  Yeah, so Open AI's plan to democratize defense is really an
• 18:05
  attempt to put equally powerful AI agents in the hands of the
• 18:09
  defenders to automate threat detection before a breach even
• 18:13
  occurs. Exactly.
• 18:14
  And finally, on the regulatory front, two US House committees
• 18:17
  have officially opened probes into two major tech companies
• 18:20
  over their use of Chinese AI model.
• 18:22
  This is a big development. It is.
• 18:24
  They are investigating any Shere, the makers of Cursor,
• 18:27
  whose Composer 2 tool is built on the Kimi model, and they're
• 18:30
  also probing Airbnb over their use of an agent built on Quinn.
• 18:34
  And viewing this strictly through the lens of supply chain
• 18:37
  mechanics, it just reveals how deeply intertwined global AI
• 18:40
  development has become. Oh, absolutely.
• 18:42
  You have American consumer and enterprise tools relying on
• 18:44
  foundational models developed overseas.
• 18:47
  Lawmakers are signaling that the origin of the foundational layer
• 18:50
  matters just as much as the application layer.
• 18:52
  Every major tech company is going to have to audit their
• 18:55
  tech stack to navigate this evolving regulatory landscape.
• 18:59
  It's going to be a massive undertaking, definitely.
• 19:01
  Well, from massive geopolitical supply chains down to the
• 19:04
  biology of our cells, the scale of these stories today is
• 19:08
  immense. It really is.
• 19:09
  But I want to bring this deep dive back to the human level.
• 19:12
  Always the best place to end. We started out talking about
• 19:15
  massive tech companies trying to cure disease, but the most
• 19:19
  impactful story today comes from how one individual is using AI
• 19:24
  to fight their own medical battle right now.
• 19:26
  Oh wow. Yeah, I want to share a
• 19:28
  community workflow submitted by a listener, Patrick B, based in
• 19:32
  Boonesboro, Maryland. Hi, Patrick.
• 19:34
  A couple of months ago, Patrick received a cancer diagnosis.
• 19:37
  Man, I am so sorry to hear that. Yeah, and as you can imagine, he
• 19:41
  felt totally overwhelmed. The appointments, the complex
• 19:45
  medical concepts, insurance claims, treatment regimens.
• 19:48
  The cognitive overload that accompanies a severe medical
• 19:52
  diagnosis is just paralyzing. It really is.
• 19:54
  You're handed a fundamentally life altering reality and
• 19:58
  simultaneously asked to become an expert in medical
• 20:01
  administration. Exactly.
• 20:03
  Yeah, you have to navigate complex oncology jargon,
• 20:06
  coordinate a labyrinth of specialists, fight with
• 20:08
  insurance. It is a grueling full time job.
• 20:11
  Absolutely. So to help manage this, Patrick
• 20:14
  built a personalized, comprehensive dashboard inside
• 20:17
  Claude. Brilliant.
• 20:18
  He uploads all of his scans, appointment notes, and insurance
• 20:21
  documents. The AI synthesizes everything.
• 20:24
  It tracks his progress and generates simplified summaries
• 20:28
  that he can easily share with his family.
• 20:29
  Which spares him the emotional exhaustion of repeatedly
• 20:32
  explaining traumatic medical updates to everyone, right?
• 20:36
  But the most impactful element of this workflow is how Patrick
• 20:40
  described the AI's role in the clinic.
• 20:42
  He called it his patient advocate.
• 20:44
  Patient advocate. I love that.
• 20:45
  Because he has fed his entire medical history into this closed
• 20:48
  system, the AI has a holistic view of his exact case.
• 20:53
  It knows him like his doctors do, so it helps him generate
• 20:56
  highly targeted, informed questions to ask his oncology
• 21:00
  team before he ever walks into the room, and it ensures he
• 21:04
  stays strictly on top of his doctor's complex instructions.
• 21:07
  You know the medical system is incredibly fragmented and when
• 21:11
  you are fighting an illness you are at your most vulnerable.
• 21:13
  You don't have the baseline energy required to be the
• 21:16
  project manager for your own body.
• 21:18
  You really don't. By utilizing this technology as
• 21:21
  a cognitive exoskeleton, Patrick isn't just organizing PDFs, he
• 21:26
  is centralizing the administrative burden and
• 21:28
  reclaiming agency over his treatment plan.
• 21:31
  Reclaiming agency It. Is just a profound demonstration
• 21:34
  of how AI can step in as a tireless advocate during our
• 21:37
  hardest moments. I think that is the absolute
• 21:40
  best way to frame the promise of this technology reclaiming
• 21:43
  agency. A massive thank you to Patrick
• 21:46
  for sharing his workflow with the community.
• 21:47
  We're all rooting for you. Absolutely rooting for you,
• 21:49
  Patrick. Well, you know, that brings us
• 21:51
  to the end of our journey today. If you want to stay updated on
• 21:54
  how this technology is reshaping our world, make sure you hit
• 21:56
  that subscribe button right now. Please do.
• 21:59
  We'd also love it if you could take a second to rate the show.
• 22:01
  Five stars. It really helps us keep bringing
• 22:03
  you these deep dives. We will see you back for
• 22:05
  tomorrow's episode. And before you go, think about
• 22:08
  the through line. Today, AI is spotting invisible
• 22:10
  cancer patterns and routine CT scans, and it's mathematically
• 22:14
  reverse engineering the human experience of pace without ever
• 22:17
  having a tongue. If AI can uncover these latent
• 22:20
  invisible structures in our bodies and our food, what other
• 22:24
  hidden patterns in your daily human life is it silently
• 22:27
  decoding right now?