Latest / Elon Musk Podcast / Elon Musk's Company replaces workers with AI

Transcript

Elon Musk's Company replaces workers with AI · Elon Musk Podcast

• 0:00
  Elon Musk's new project, Macro Hard, uses an artificial
• 0:04
  intelligence system to completely automate and replace
• 0:06
  the functions of entire software companies.
• 0:08
  Yeah. That initiative, which is also
• 0:11
  known as Digital Optimus, is basically a joint venture
• 0:15
  between Tesla and XAI, and they are blending physical automotive
• 0:19
  hardware with high level reasoning models to actually
• 0:22
  perform digital labor. So how does an electric vehicle
• 0:26
  manufacturer transform its infrastructure into a provider
• 0:30
  of autonomous white collar workers?
• 0:32
  Well, the architecture of this system, it mimics human
• 0:34
  cognition by using a dual process model.
• 0:37
  OK, dual process meaning two parts.
• 0:38
  Exactly. It relies on 2 complementary
• 0:41
  systems working together to emulate how a person approaches
• 0:44
  a complex task. So system 2 is XAI's Grok model.
• 0:48
  You can think of Grok as the thinking brain.
• 0:50
  It handles the strategic planning, the high level
• 0:52
  reasoning, and, you know, the overarching navigation of
• 0:55
  whatever complex workflow needs to be done.
• 0:57
  So if you assign the system a massive project like reconciling
• 1:01
  a year's worth of corporate expenses.
• 1:03
  Right. In that scenario, Grok is the
• 1:05
  manager sitting in the office figuring out the logical steps
• 1:08
  required to finish the job. Then you have system one, yeah,
• 1:11
  which is a Tesla developed AI agent acting as the instinctual
• 1:15
  executor. Yes, exactly.
• 1:17
  So if Grok is the manager deciding what needs to happen,
• 1:20
  the Tesla agent is the hands on worker actually doing it.
• 1:23
  It acts on the immediate sensory input to carry out the strategy,
• 1:27
  right? So Grok looks the blueprint, and
• 1:30
  the Tesla agent actually swings the hammer and lays the bricks.
• 1:33
  That is a perfect way to put it. And the way System One functions
• 1:36
  is incredibly specific. It is entirely visual.
• 1:39
  Visual. Yeah.
• 1:40
  The Tesla AI agent processes A continuous video feed of the
• 1:44
  computer screen covering the last five seconds of activity.
• 1:47
  OK, now. And based entirely on that
• 1:49
  visual information, it autonomously moves the mouse and
• 1:53
  types on the keyboard to execute the task.
• 1:55
  Wait back up. It is physically observing a
• 1:57
  video feed of the screen instead of plugging into the code.
• 2:00
  Exactly. Yeah, By bypassing traditional
• 2:04
  software integrations and interacting with the user
• 2:06
  interface exactly as a human would, the AI is no longer
• 2:10
  limited by compatibility. That makes sense.
• 2:12
  Because traditional automation requires a back end connection,
• 2:16
  you need an API or you have to pay a team of engineers to write
• 2:19
  custom code just to make two separate pieces of software talk
• 2:23
  to each other. Which is always a massive
• 2:25
  headache. It creates massive friction for
• 2:27
  cororations, but because Digital Optimist uses the exact same
• 2:31
  visual interface a person uses, it opens up the ability for the
• 2:35
  AI to operate legacy software, proprietary enterprise tools,
• 2:39
  and complex web interfaces seamlessly.
• 2:42
  So it changes the entire approach to enterprise
• 2:45
  automation because you eliminate the need for an IT department to
• 2:48
  build a bridge between applications.
• 2:50
  Exactly. That is like hiring an employee
• 2:52
  who only needs to look at a monitor rather than rewriting
• 2:55
  the company's entire back end infrastructure.
• 2:57
  I mean, think about the software you use at your job every day.
• 3:00
  If you bring a new human accountant into an office, you
• 3:03
  set them in front of a computer, show them the spreadsheet, and
• 3:06
  they start clicking and typing. Right, You do not have to
• 3:08
  rebuild the accounting software so their biological brain can
• 3:11
  plug directly into the corporate server.
• 3:13
  Exactly. By mimicking that exact human
• 3:17
  behavior, this system sidesteps all the friction of software
• 3:21
  integration. You just give the AI the exact
• 3:24
  same access you would give a human worker, and it just starts
• 3:26
  working. And the hardware running the
• 3:29
  system is where Tesla's specific manufacturing advantages really
• 3:33
  come into play. Right, the chips.
• 3:35
  Yeah, Macro Hard operates on Tesla's in house AI4 chip, which
• 3:39
  costs roughly $650 and that is paired with Xai's NVIDIA
• 3:43
  servers. And those AI 4 chips are already
• 3:45
  in the car. Exactly.
• 3:46
  The AI4 chip is already installed in all current
• 3:49
  generation Tesla vehicles. It is completely compatible with
• 3:52
  all AI4 equipped vehicles. Meaning a parked car can process
• 3:56
  office work while sitting in a driveway.
• 3:58
  Imagine pulling your car into the garage after work, putting
• 4:01
  it in park, and while you sleep, the car's computer starts
• 4:05
  working a night shift as a corporate accountant.
• 4:08
  It is wild to think about. A vehicle sitting idle in a
• 4:11
  garage is essentially a high-powered computer doing
• 4:14
  nothing for the vast majority of its existence.
• 4:17
  And Tesla plans to deploy millions of dedicated Digital
• 4:21
  Optimist units at their Supercharger stations to tap
• 4:24
  into approximately 7 gigawatts of available power.
• 4:28
  Right. They are taking advantage of the
• 4:29
  physical infrastructure they have already built across the
• 4:32
  entire country, and this directly connects to the massive
• 4:36
  compute bottleneck facing the technology industry right now.
• 4:39
  Because power is the big issue. Yeah, traditional AI data
• 4:42
  centers face severe power grid delays.
• 4:45
  Building a new server facility requires securing massive tracts
• 4:48
  of land, obtaining complex local permits, and waiting for utility
• 4:52
  companies to connect heavy duty power lines.
• 4:54
  Which takes forever. Getting the physical
• 4:56
  Transformers installed often takes years of bureaucratic
• 4:58
  fighting, but utilizing the Supercharger network completely
• 5:02
  bypasses this limitation. Tesla already owns the real
• 5:05
  estate and they already secured the high capacity power
• 5:08
  connections to charge vehicles. Transforming charging stations
• 5:11
  into a decentralized industrial scale data center.
• 5:14
  Exactly. So the charting station down the
• 5:17
  street double S as a server farm.
• 5:19
  It creates a massive distributed computing network without
• 5:22
  waiting for local governments to approve grid upgrades.
• 5:26
  You take 7 gigawatts of power, which represents a massive
• 5:30
  portion of the total energy consumed by all data centers in
• 5:32
  the United States, and you instantly monetize that idle
• 5:36
  capacity. Instead of waiting three years
• 5:38
  to build a centralized server warehouse, the compute power is
• 5:41
  distributed across thousands of parking lots and rest stops.
• 5:44
  If you've got a decent mic and laptop and some free time,
• 5:47
  Babble Audio is paying people to record speech data and annotate
• 5:51
  audio for AI training. No minimums, no fixed hours.
• 5:55
  You work when you want and get paid weekly via PayPal, Venmo,
• 5:58
  or bank transfer. They pay per recorded or
• 6:01
  annotated hour, plus bonus challenges for hitting weekly
• 6:04
  goals. And if you sign up through our
• 6:06
  link, you get priority processing and a $15 bonus.
• 6:09
  Links in the show notes. So getting back to it, the
• 6:12
  economic target of this project is explicitly stated in its
• 6:15
  name. Macro Heart.
• 6:16
  Right Macro Heart is a direct reference to Microsoft,
• 6:19
  reflecting the goal of automating clerical work,
• 6:21
  accounting, human resources and coding.
• 6:24
  Because Microsoft Score business relies on selling Productivity
• 6:28
  Tools for human workers to use. Exactly, and Macro Heart aims to
• 6:31
  replace the workers entirely. The cost structures really
• 6:34
  explain why this specifically targets traditional enterprise
• 6:36
  software. A Digital Optimist unit could
• 6:39
  theoretically be hired for a fraction of the cost of a
• 6:41
  traditional employee. Oh, absolutely.
• 6:43
  We are talking around 500 to $1000 a month compared to a
• 6:46
  massive white collar salary and benefits package.
• 6:49
  Right. When a corporation hires you for
• 6:51
  administration, they are paying for your salary, your health
• 6:54
  insurance, your payroll taxes, and the physical office space
• 6:57
  you occupy. Paying a flat monthly fee for an
• 7:00
  autonomous agent scatters that entire financial model.
• 7:04
  It severely limits the need for large human administrative teams
• 7:09
  and per seat software licensing models, and that opens up
• 7:13
  unprecedented profit margins for early adopters.
• 7:16
  Because software companies usually charge per user.
• 7:18
  Exactly. Think about a standard corporate
• 7:20
  setup. If a corporation replaces 500
• 7:22
  human resource employees with five autonomous AI agents, they
• 7:26
  no longer need to pay for 500 individual software licenses.
• 7:30
  Wow. The company adopting the AI sees
• 7:32
  massive cost savings on their balance sheet, while the
• 7:34
  traditional software vendor permanently loses its recurring
• 7:37
  revenue. But relying entirely on an AI
• 7:40
  for security critical or regulatory software introduces
• 7:44
  massive risks. Yes, it really does.
• 7:46
  Especially regarding hallucinations and emergent bugs
• 7:49
  that human testers usually catch, human friction is
• 7:51
  actually a hidden safety feature in corporate workflows.
• 7:53
  That is a really good point. If a human accountant looks at a
• 7:57
  tax form and sees a number that makes no logical sense, they
• 8:01
  pause. They ask a clarifying question.
• 8:03
  Right. They do not just push it
• 8:05
  through. Exactly.
• 8:06
  An AI system might confidently hallucinate a tax rule and
• 8:11
  autonomously file thousands of erroneous documents before
• 8:14
  anyone notices. And the speed and scale of the
• 8:17
  system mean that emergent bugs cause catastrophic damage
• 8:20
  instantly. Yeah, you also see intense
• 8:23
  friction surrounding the project on a corporate level, too.
• 8:26
  Oh, the lawsuits. Yeah, Tesla shareholders have
• 8:28
  filed A lawsuit alleging a breach of fiduciary duty,
• 8:31
  claiming talent and hardware were diverted from Tesla to XAI.
• 8:35
  The lawsuit argues that resources belonging to a public
• 8:38
  company were funneled into a private entity controlled by the
• 8:41
  exact same CEO, and there is a direct contradiction at the
• 8:45
  center of this arrangement. Musk previously claimed Tesla
• 8:48
  had no need to license anything from XAI.
• 8:51
  He did say that. He argued that Tesla's real
• 8:53
  world artificial intelligence was completely separate from the
• 8:56
  language models being developed at XAI.
• 8:58
  But now Tesla has invested $2 billion into XAI, SpaceX has
• 9:03
  acquired XAI, and Digital Optimus explicitly relies on XA
• 9:08
  is Grok to function. It's totally integrated.
• 9:11
  The publicly traded hardware company is now completely
• 9:14
  reliant on the private software company to run its flagship AI
• 9:19
  project. And this entanglement opens up
• 9:21
  heavy regulatory scrutiny. From the FTC.
• 9:24
  Yeah, the US Federal Trade Commission is heavily focused on
• 9:27
  market concentration and the use of sensitive of personal data.
• 9:31
  Because macro hard records computer screens in real time to
• 9:34
  function. The privacy and compliance
• 9:35
  hurdles are massive. Because it sees everything.
• 9:38
  Everything. If an AI agent is processing
• 9:40
  payroll or handling medical records, it is constantly
• 9:44
  recording highly sensitive, protected information.
• 9:47
  The visual input mechanism that makes the AI so adaptable is the
• 9:50
  exact same feature that triggers intense regulatory alarm.
• 9:54
  Honestly, I think the internal shareholder lawsuit and the
• 9:56
  corporate entanglement pose the primary threat to the project
• 9:59
  survival. You think so?
• 10:00
  Yeah, you have massive institutional investors actively
• 10:03
  trying to force the return of intellectual property and
• 10:06
  resources. If a judge rules that the
• 10:09
  financial arrangement violates basic fiduciary duties, the
• 10:13
  entire collaboration between Tesla and XAI could be untangled
• 10:17
  or completely halted by court order.
• 10:19
  I see what you mean. The structural foundation of the
• 10:21
  project is actively under legal attack from the very people
• 10:25
  funding the hardware. Well, I actually think European
• 10:28
  and US regulators blocking the screen recording data collection
• 10:31
  is a much harder wall to climb. Really harder than the
• 10:34
  investors. Yeah, because shareholder
• 10:36
  lawsuits usually end with financial payouts or some form
• 10:39
  of corporate restructuring. You can appease angry investors
• 10:42
  with a settlement, that is, but regulatory agencies like the FTC
• 10:46
  enforce strict privacy laws that cannot be simply bought off,
• 10:49
  right? If the core function of System 1
• 10:51
  requires continuous video recording of user interfaces,
• 10:55
  and regulators classify that as an illegal collection of
• 10:58
  sensitive personal data, the technology fundamentally cannot
• 11:02
  operate. You cannot negotiate with a
• 11:04
  privacy law that outright bans your primary mechanism for data
• 11:08
  ingestion. Exactly.
• 11:09
  If the government says an AI cannot legally point a digital
• 11:12
  camera at a atient's medical file, the entire architecture of
• 11:16
  the visual agent stops functioning entirely.
• 11:19
  Macrohard represents a shift from assistive digital tools to
• 11:23
  autonomous cororate workers, powered by an unprecedented
• 11:27
  combination of automotive hardware and advanced reasoning
• 11:30
  models. And if a fully autonomous
• 11:32
  digital worker makes a catastrophic error in a
• 11:35
  company's tax filing or code base, who exactly is legally
• 11:38
  responsible for the fallout? If you're not subscribed yet,
• 11:41
  take a second and hit follow on whatever app you're using.
• 11:43
  It helps us keep making this. We appreciate you being here.