Latest / Today’s AI News / OpenAI vs Anthropic on Cyber AI Access, Nvidia Bets on Quantum, Anthropic Redesigns Claude Code

Transcript

OpenAI vs Anthropic on Cyber AI Access, Nvidia Bets on Quantum, Anthropic Redesigns Claude Code · Today’s AI News

• 0:00
  Welcome back to today's AI News, brought to you by 9X
• 0:02
  Productions. Glad to be here.
• 0:04
  Imagine a neighborhood dealing with a sudden string of
• 0:07
  burglaries. Do you only give high tech
• 0:10
  security systems to the 40 richest houses, or do you hand
• 0:14
  out master keys to the entire neighborhood watch?
• 0:17
  That's exactly the debate over cybersecurity access tearing the
• 0:21
  AI world apart today. It really is.
• 0:23
  I mean, it's this massive philosophical divide over the
• 0:27
  very infrastructure of how we defend the digital world, right?
• 0:30
  And we're seeing 2 completely opposing strategies deploy in
• 0:35
  real time. So in this deep dive, we are
• 0:37
  unpacking this rapid evolution in AI access and automation.
• 0:41
  We're looking at who actually gets the most powerful tools
• 0:44
  being built right now. Yeah, starting from elite
• 0:46
  cybersecurity defenders. Exactly.
• 0:48
  And then moving into quantum physics labs and going all the
• 0:51
  way down to how you might soon be automating your everyday web
• 0:54
  browser. I've got a lot of ground to
• 0:55
  cover. We really do, including some
• 0:57
  quick kits in a pretty amazing community workflow at the end.
• 1:00
  So let's just jump straight into that Neighborhood Watch debate.
• 1:03
  Yeah, the cyber arms race between Open AI and Anthropic is
• 1:07
  really the defining story here. Absolutely.
• 1:09
  So Open AI just rolled out GPT 5.4 Cyber which is a model
• 1:13
  specifically built and tune for defensive security work, right?
• 1:18
  And the way they are deploying this is through something they
• 1:20
  call the Trusted Access for Cyber Initiative.
• 1:23
  OK, so how does that work? Basically, if you are a
• 1:26
  cybersecurity professional and you pass their ID verification
• 1:29
  checks, you get access. They are opening this up to
• 1:32
  thousands of verified defenders worldwide.
• 1:35
  Thousands. Wow.
• 1:36
  Yeah, and this is a direct counter move to what Anthropic
• 1:39
  did last week with their Mythos rollout, where they restricted
• 1:42
  access to this highly vetted whitelist of like 40 major tech
• 1:46
  giants and government contractors.
• 1:48
  Right, so Anthropic is taking the walled garden approach.
• 1:51
  Exactly. Their philosophy is that when
• 1:53
  you create a tool with potentially dangerous dual use
• 1:56
  capabilities you know you keep restricted, you only give it to
• 1:59
  organizations that have massive internal compliance and security
• 2:03
  teams. Which makes sense on paper.
• 2:05
  It does, but Open AI is making the exact opposite bet.
• 2:09
  Their lead researcher on this food Matan.
• 2:12
  He argued that cyber defense is a team sport.
• 2:15
  A team sport, right? Yeah, and that no single company
• 2:18
  should be in the business of picking winners and losers in
• 2:21
  the security landscape. OK, let's unpack this for a
• 2:23
  second. If you give thousands of people
• 2:25
  the ability to reverse engineer compiled software, which is, you
• 2:29
  know, the marquee feature of GPT 5.4 Cyber I.
• 2:33
  Think that's the big selling point?
• 2:34
  Aren't you essentially handing out a treasure map to bad actors
• 2:38
  who managed to slip through those ID checks?
• 2:41
  I mean, how does this team sport philosophy actually work in
• 2:44
  practice? Well, to understand Open AI's
• 2:46
  logic, we really have to look at the fundamental asymmetry of
• 2:49
  cyber defense. OK, historically, the attacker
• 2:52
  always has the advantage, always the only need to find a single
• 2:56
  vulnerability to breach a system.
• 2:58
  Just one crack in the armor. Exactly.
• 3:00
  But the defender, they have to secure every single endpoint,
• 3:03
  every line of code perfectly 24 hours a day.
• 3:06
  Which is impossible for human teams to do perfectly.
• 3:08
  Right. So open a eyes thesis is that by
• 3:11
  distributing an incredibly powerful defensive tool to
• 3:14
  thousands of analysts, you flip that asymmetry.
• 3:17
  Oh, I see. You create this massive
• 3:19
  distributed immune system where patches and defenses are
• 3:23
  generated faster than attackers can exploit the vulnerabilities.
• 3:26
  That's fascinating, but let's dig into that reverse
• 3:29
  engineering capability because that is what makes this model so
• 3:32
  highly sought after. It really is.
• 3:34
  When we talk about reverse engineering compiled software,
• 3:37
  we are talking about taking the raw machine code, like the ones
• 3:41
  and zeros or the assembly language that a computer
• 3:45
  actually reads. The absolute lowest level.
• 3:47
  Right. And trying to translate it back
• 3:49
  into human readable programming logic like C++ or Python.
• 3:55
  Which is an incredibly tedious, grueling process for human
• 3:58
  analysts. I can imagine.
• 3:59
  Yeah, because when software is compiled, all the helpful things
• 4:02
  a developer writes, you know, the variable names, the
• 4:05
  structural comments, the logical, all of it is completely
• 4:08
  stripped away. You are left with raw binary
• 4:10
  operations. So if an analyst wants to figure
• 4:13
  out if a piece of compiled software contains hidden malware
• 4:16
  or a back door, they traditionally have to spend
• 4:19
  weeks, literally weeks, mapping out memory registries and CPU
• 4:24
  instructions just to understand what the program is trying to
• 4:26
  do. And GPT 5.4 Cyber just bypasses
• 4:29
  that manual mapping entirely. Exactly.
• 4:31
  It uses pattern recognition to look at the binary structures
• 4:34
  and instantly infer the high level logic.
• 4:37
  It maps those assembly patterns to known source code structures
• 4:41
  and flags malicious intent without ever needing the
• 4:44
  original code. It's wild.
• 4:46
  It drastically reduces the time it takes to dissect the thread
• 4:49
  from weeks to seconds. Seconds.
• 4:51
  That is a massive leap. It is, but your underlying
• 4:54
  concern about the raw can abilities of these models
• 4:57
  falling into the wrong hands, that's clearly shared at the
• 4:59
  highest levels. I mean, we saw Treasury
• 5:01
  Secretary Besson summon Wall Street leaders for an emergency
• 5:04
  briefing last week specifically regarding the hacking
• 5:07
  capabilities of Anthropics Mythos model.
• 5:10
  The capability jump we are seeing in this generation of
• 5:12
  models is profound. We're watching the foundation of
• 5:16
  next generation digital security being laid while the architects
• 5:19
  are still, you know, debating the blueprints.
• 5:22
  Debating the blueprints, that's a great way to put it.
• 5:25
  And, you know, the asymmetry in cybersecurity is all about
• 5:28
  protecting our current classical computing infrastructure, but
• 5:32
  the hardware underpinning that infrastructure is about to be
• 5:35
  completely overshadowed by what Nvidia's doing in the quantum
• 5:38
  space. Oh man, Nvidia's Izing release
• 5:41
  is genuinely pushing the boundaries of what is
• 5:44
  computationally possible right now.
• 5:46
  It really is. So NVIDIA released Ising, which
• 5:48
  is the first family of open source II models designed
• 5:51
  specifically for quantum computers.
• 5:53
  The goal here is to solve the massive technical roadblocks
• 5:56
  that have kept quantum technology stuck in research
• 5:58
  labs for years, specifically the issues of calibration and error
• 6:02
  decoding. Yeah, those have been huge
• 6:04
  bottlenecks. Exactly.
• 6:05
  So their first model automates machine tuning.
• 6:08
  It turns what used to be a days long manual job for physicists
• 6:11
  into something that takes like a few hours.
• 6:14
  Which is a game changer for research.
• 6:16
  Base and the second model fixes errors as they happen, hitting 2
• 6:19
  1/2 times the speed and three times the accuracy of the best
• 6:22
  open source alternative out there.
• 6:24
  Jensen Wong is actually calling this the operating system of
• 6:27
  quantum machines. And honestly, the adoption rate
• 6:30
  speaks volumes. So over 20 major institutions
• 6:33
  deployed it at launch. Really like who?
• 6:36
  We're talking Harvard, Cornell, Fermalab, Sandia National Labs,
• 6:41
  Ion Q, UC Santa Barbara. The heavyweights.
• 6:45
  Yeah, the heavyweights of physics.
• 6:47
  To understand why this software is so critical, we have to look
• 6:50
  at the physical limitations of quantum bits or quibbits.
• 6:53
  Right, because quibbits are notoriously fragile.
• 6:56
  Incredibly fragile. A tiny fluctuation in
• 6:58
  temperature from a stray cosmic ray or even just ambient
• 7:01
  electromagnetic noise can cause them to lose their quantum
• 7:04
  state. Which is called decoherence,
• 7:06
  right? Exactly decoherence.
• 7:08
  And when that happens, it creates an error in the
• 7:10
  calculation. Historically, quantum error
• 7:13
  correction has required immense computational overhead.
• 7:16
  Traditional algorithms essentially use multiple
• 7:19
  physical qubits just to monitor a single logical qubit.
• 7:22
  That seems really inefficient. It is.
• 7:25
  The problem is that classical algorithms are often too slow.
• 7:28
  By the time the algorithm registers the error syndrome and
• 7:31
  calculates the fix, the Quibid has already further degraded.
• 7:34
  Oh wow. Yeah, the system spends more
• 7:36
  time trying to decode its own errors than it does running the
• 7:40
  actual quantum algorithm. OK, so Izing steps in to bypass
• 7:44
  that traditional algorithmic bottleneck using predictive
• 7:47
  pattern recognition. Precisely, the neural network
• 7:50
  anticipates and decodes the error patterns in real time,
• 7:54
  vastly outperforming those classical heuristic methods.
• 7:57
  That's incredible. So what does this all mean?
• 7:59
  It it sounds like NVIDIA isn't trying to mine the gold
• 8:02
  themselves, They are mass producing the absolute best
• 8:06
  pickaxes and shovels for the quantum gold rush, right?
• 8:08
  That is exactly what they're doing.
• 8:10
  The brilliance of Nvidia's move isn't just the technical
• 8:12
  solution, it's the distribution strategy.
• 8:14
  Think of it like building the electrical grid for a new city.
• 8:17
  NVIDIA is laying the power lines and standardizing the voltage by
• 8:21
  making Izing open source. You don't hear what specific
• 8:24
  quantum hardware you build or what scientific applications you
• 8:27
  plug into the wall as long as you are running on their grid.
• 8:31
  It's the exact same playbook they use everywhere else.
• 8:34
  It's that ecosystem lock in. They're doing it with self
• 8:38
  driving cars with their Alpa Mayo platform and they're doing
• 8:41
  it for robotics with Isaac GR00T.
• 8:44
  Exactly the same. They release the foundational AI
• 8:47
  layer for free. It becomes the industry standard
• 8:49
  because of its performance, and suddenly NVIDIA owns the
• 8:52
  infrastructure beneath an entirely new paradigm of
• 8:55
  computing. And they do it before the market
• 8:58
  even fully materializes. Yeah, it positions them to
• 9:00
  dominate a projected $11 billion quantum market just by being the
• 9:05
  default language that all quantum hardware speaks.
• 9:08
  It's a brilliant strategy, and you know, we've just looked at
• 9:11
  these massive abstract physics problems in quantum labs.
• 9:15
  But AI is also rapidly automating the absolute most
• 9:18
  common piece of software on Earth, which is your web
• 9:21
  browser. Yes, the Gemini integration in
• 9:24
  Chrome. This represents a major shift
• 9:26
  from passive web browsing to active task delegation.
• 9:30
  It's wild. So Google rolled out Deep Gemini
• 9:34
  integration directly into the Chrome settings under a new tab
• 9:37
  called AI Innovations. And if you are using the
• 9:41
  standard version, you can enable a feature to share your current
• 9:44
  tabs with Gemini. So say you open three different
• 9:47
  tabs for a product you want to buy.
• 9:49
  You just type the AT symbol in the prompt box to attach those
• 9:52
  specific tabs and ask it to compare features and pricing.
• 9:55
  And it synthesizes the live data across those tabs instantly.
• 9:59
  Works even without the AI proop here.
• 10:01
  Exactly. It completely transforms the
• 10:03
  browser from a simple window where you view documents into
• 10:06
  like a localized synthesis engine.
• 10:08
  But the Google AI Pro tier pushes this into autonomous
• 10:11
  territory. Oh yeah, with AI Pro you can
• 10:14
  enable a feature called Let Gemini Browse for you.
• 10:16
  That sounds intense. It is.
• 10:18
  You give it a high level command like find me a cheaper product
• 10:22
  with these same features, and Gemini actually creates an
• 10:25
  execution plan. It uses Chrome to navigate the
• 10:28
  web, find the product and present you the options.
• 10:31
  Right. And here is the pro tip that is
• 10:33
  kind of wild. You can even authorize Gemini to
• 10:36
  use your saved Chrome passwords to sign into your accounts and
• 10:40
  execute those plans on your behalf.
• 10:42
  Which is a huge deal. Yeah, and here's where it gets
• 10:45
  really interesting, but also slightly terrifying for me.
• 10:49
  Letting an AI use my saved passwords to log into my
• 10:52
  accounts and execute plans. That feels like handing my
• 10:55
  physical wallet to a personal shopper I just met on the
• 10:58
  street. That's a very fairpoint.
• 11:00
  I mean, if you're feeling a bit uneasy at the thought of handing
• 11:02
  over your saved passwords to an AI to go shopping for you,
• 11:05
  you're definitely not alone. Yeah, it requires an incredible
• 11:07
  level of trust in Google security boundaries, but what we
• 11:11
  are witnessing is the fundamental evolution from
• 11:13
  declarative user interfaces to intent driven interfaces.
• 11:18
  Moving from clicking buttons to simply stating a goal.
• 11:21
  Exactly when you browse, traditionally you are navigating
• 11:25
  the document object model or Dom of a website.
• 11:29
  You visually scan the page, you find the Add Cart button, you
• 11:32
  click it, you navigate to checkout, type in your details
• 11:36
  and hit submit. The agentic model of Gemini is
• 11:40
  doing all that Dom navigation for you in the background.
• 11:43
  It reads the code of the website, identifies the
• 11:45
  necessary input fields, and mimics human interaction to
• 11:49
  achieve the goal you stated. It's moving from AI as a passive
• 11:53
  conversational advisor to an active autonomous agent.
• 11:56
  It fundamentally changes our relationship with the software.
• 11:59
  You are no longer the driver navigating the roads, you're
• 12:02
  just sitting in the back seat giving the destination to the
• 12:04
  chauffeur. That's it.
• 12:05
  And if everyday consumers are starting to hand over a browser
• 12:08
  control to agents, you have to realize the developers who
• 12:11
  actually build these tools are undergoing an even more radical
• 12:14
  workflow shift. Which brings us perfectly to
• 12:17
  Anthropics complete makeover of the Claude Code desktop app.
• 12:21
  Oh, this is a big one. Yeah, they have redesigned this
• 12:24
  tool around the reality that developers are no longer just
• 12:27
  chatting with a single AI assistant, they are
• 12:30
  orchestrating multiple parallel AI sessions simultaneously.
• 12:34
  It's essentially a command Center for an any genetic
• 12:36
  workflow now, right? The new interface includes A
• 12:38
  sidebar to manage all these live and recent sessions, and you can
• 12:42
  filter them by project status. It even features auto archiving
• 12:46
  when a pull request on GitHub is merged or closed.
• 12:49
  That's super clean. Very clean and they have a drag
• 12:52
  and drop layout for customizable workspaces, plus an integrated
• 12:56
  editor and terminal. So you don't even have to switch
• 12:58
  tools. Exactly.
• 12:59
  You can run tests, edit files, review changes, preview HTML
• 13:03
  without leaving the app, but the most significant addition is a
• 13:06
  research preview called Routines.
• 13:08
  Right routine. This allows developers to run AI
• 13:10
  tasks on a schedule, trigger them via an API, or have them
• 13:14
  execute automatically based on specific GitHub events.
• 13:17
  Wow, Anthropic stated vision here is a future where
• 13:20
  developers spend less time writing code and more time
• 13:23
  directing 1/2 human half AI workforce.
• 13:25
  Which is a massive paradigm shift.
• 13:27
  It is. It sounds like we're moving away
• 13:30
  from being the writer of the code and becoming the editor in
• 13:33
  chief of an automated newsroom. But do developers risk losing
• 13:37
  their fundamental coding muscles if they are just managing agents
• 13:41
  all day? That's a debate that comes up a
• 13:43
  lot. If you stop writing the syntax
• 13:45
  yourself, do you eventually forget how to code?
• 13:49
  But that concern really stems from misunderstanding what the
• 13:52
  actual bottleneck in software engineering is today.
• 13:55
  And the bottleneck is rarely typing speed.
• 13:57
  Right, it's architectural logic. The heavy lifting isn't writing
• 14:01
  a for loop or formatting a database query.
• 14:04
  The heavy lifting is orchestrating complex logic
• 14:07
  across highly integrated systems.
• 14:08
  That makes sense. When you have 5 autonomous
• 14:10
  agents working on different microservices within an
• 14:13
  application, your job elevates. You literally become that editor
• 14:16
  in chief. You have to understand the code
• 14:18
  base at a macro level to ensure that an agent optimizing the
• 14:21
  database doesn't break the logic of the agent building the user
• 14:25
  interface. Right code review, system
• 14:27
  design, and rigorous testing become the primary muscles a
• 14:30
  developer exercises. Precisely.
• 14:32
  You might not be writing every single word of the articles, but
• 14:35
  you had better know exactly what makes a cohesive newspaper, or
• 14:39
  the entire publication just falls apart.
• 14:41
  Absolutely. Well, these agentic workflows
• 14:43
  are trickling down to every single sector, which brings us
• 14:46
  perfectly to the rapid fire updates across the rest of the
• 14:49
  industry today. Let's do some quick hits.
• 14:52
  OK, so UKAI safety evaluators just announced that Anthropics
• 14:57
  Claude Mythos preview, which is that restricted model we
• 15:00
  discussed earlier, is the first AI to ever complete their 32
• 15:04
  step corporate hack simulation. And we really need to connect
• 15:08
  this back to what we were just talking about in segment one.
• 15:10
  When we talk about a 32 step hack, we are not talking about
• 15:13
  an AI simply guessing a password.
• 15:16
  We are talking about a sophisticated autonomous kill
• 15:19
  chain. The model performs
• 15:21
  reconnaissance on a network, identifies vulnerabilities,
• 15:24
  escalates its privileges, moves laterally across different
• 15:27
  servers, and exfiltrates data, all without human intervention.
• 15:31
  That is terrifying. It completely contextualizes
• 15:34
  Treasury Secretary Besson's emergency Wall Street briefing.
• 15:38
  When financial regulators see a model capable of autonomous
• 15:41
  network penetration at that scale, the threat landscape
• 15:44
  shifts overnight. It really does make open AIS
• 15:47
  push to get defensive tools like GPT 5.4 Cyber into the hands of
• 15:52
  thousands of analysts feel incredibly urgent.
• 15:54
  Absolutely. OK, moving from capabilities to
• 15:57
  sheer scale open AIS. Greg Brockman marked the
• 16:00
  company's 10th anniversary by framing AI as a shift to a
• 16:03
  compute powered economy. And the jaw-dropping metric here
• 16:07
  is that they are claiming roughly 1 billion weekly users
• 16:10
  across ChatGPT and codecs. 1 billion weekly users.
• 16:14
  I mean, I'm just trying to wrap my head around that to
• 16:16
  contextualize 1 billion weekly users for a technology that
• 16:19
  barely existed in the mainstream a few years ago.
• 16:21
  That is mind. Blowing It's staggering.
• 16:23
  That is roughly one out of every eight people on the planet
• 16:26
  querying their data centers every single week.
• 16:29
  The physical infrastructure, the power consumption, and the
• 16:32
  cooling requirements to support that level of inference.
• 16:35
  It's reshaping global energy markets.
• 16:38
  Compute really is becoming the foundational utility of the
• 16:40
  modern economy. It really is.
• 16:43
  We also seeing a massive consolidation of talent right
• 16:45
  now. The AI personal finance startup
• 16:48
  Hero is officially winding down operations, and its specialized
• 16:52
  staff is joining Open AI. Yeah, this is clearly a
• 16:55
  strategic talent acquisition. Open AI is absorbing teams that
• 16:59
  have deep domain specific expertise in financial
• 17:03
  compliance and data structuring, bringing those capabilities in
• 17:06
  house. To strengthen their enterprise
• 17:08
  offerings. I see.
• 17:09
  Exactly. Meanwhile, AWS rolled out Amazon
• 17:11
  Bio Discovery. This is a drug design platform
• 17:14
  that utilizes biological foundation models, but what
• 17:17
  makes it unique is that it has a built in physical lab network
• 17:20
  for testing and synthesizing the designs.
• 17:21
  Which is brilliant because it bridges the gap between boots
• 17:24
  and atoms. Instead of an AI just predicting
• 17:27
  how a protein might fold or how a molecule might bind in a
• 17:30
  digital simulation, the platform connects those digital
• 17:33
  predictions directly to automated physical labs to test
• 17:37
  the actual efficacy. It drastically shortens the
• 17:40
  feedback loop in pharmaceutical research.
• 17:42
  That's incredible. And lastly, for our quick hits,
• 17:45
  Baidu released Ernie Image. This is an 8 billion parameter
• 17:50
  open weight text to image model and despite its relatively small
• 17:53
  size, it is competing really closely with top rivals.
• 17:56
  Yeah, it just highlights that the open weight ecosystem
• 17:58
  outside the United States remains highly competitive.
• 18:02
  They're focusing heavily on parameter efficiency and making
• 18:05
  these models accessible. Well, we have covered billion
• 18:07
  dollar hardware infrastructure and billion user milestones
• 18:10
  today, but I always love remembering that the most
• 18:12
  profound application of this technology often happens at a
• 18:15
  highly individual scale. It's time for the community AI
• 18:18
  workflow. Yes it is and I absolutely love
• 18:21
  this one. Today we are looking at Will G
• 18:23
  from Nashville, TN. Will is almost 70 years old and
• 18:26
  he decided to use ChatGPT at work to build a comprehensive
• 18:30
  product item tracking system. He used Microsoft Forms, Lists
• 18:34
  and Power Automate. Well, that's a great stack,
• 18:36
  yeah. He created this QR based system
• 18:39
  so dock workers and delivery drivers can scan items on their
• 18:42
  phones for real time tracking, and by doing this he completely
• 18:47
  bypassed the need for his company to purchase expensive
• 18:50
  proprietary hardware scanners. It's amazing.
• 18:53
  And get this, because of this initiative, at almost 70 years
• 18:57
  old, he is now known around his workplace as the AI Guy.
• 19:02
  I love that so much, me. Too, It completely shatters the
• 19:05
  stereotype that you need to be a 20 something Silicon Valley tech
• 19:08
  bro to build highly effective AI infrastructure.
• 19:11
  It really does, and Will's workflow perfectly illustrates
• 19:14
  the exact shift we discussed earlier regarding developers.
• 19:17
  Oh. Right.
• 19:18
  Will didn't need to learn a program language.
• 19:20
  He didn't need to write the syntax for API calls or
• 19:22
  webhooks. He just used Chat GT as a
• 19:25
  translation layer. He understood the business logic
• 19:27
  right. He knew what data the dock
• 19:29
  workers needed to input, and he knew where that data needed to
• 19:32
  go. The AI simply provided the
• 19:35
  connector tissue between those accessible Microsoft tools to
• 19:39
  execute his logic. The core value is knowledge
• 19:42
  application. He became the architect of his
• 19:44
  own automated system, proving you don't need a computer
• 19:46
  science degree to solve real world problems with this stuff
• 19:49
  Which. Leaves all of us with a highly
• 19:50
  relevant question to consider. Let's hear it.
• 19:52
  If a 70 year old can become the AI guy at his company just by
• 19:56
  asking ChatGPT how to connect a few everyday Microsoft tools,
• 20:01
  what clunky legacy system in your own life or work?
• 20:03
  Place is just waiting to be automated by a single well
• 20:07
  crafted prompt. That is the perfect thought to
• 20:09
  leave you with. The master keys to these
• 20:11
  incredibly powerful tools are being handed out right now.
• 20:14
  The only question is what doors you're going to open with yours.
• 20:18
  Make sure you hit that subscribe button to stay updated on all
• 20:21
  these fast moving developments. And if you enjoyed today's deep
• 20:25
  dive, we would really love it if you could take a second to rate
• 20:28
  the show 5 stars. It really helps us out.
• 20:30
  Thank you so much for listening, and we'll see you back for
• 20:33
  tomorrow's episode.