Latest / Today’s AI News / Dorsey Says AI Replaces Managers, SpaceX $1.75T IPO, OpenAI Pays Workers to Train Their Replacements

Transcript

Dorsey Says AI Replaces Managers, SpaceX $1.75T IPO, OpenAI Pays Workers to Train Their Replacements · Today’s AI News

• 0:00
  Welcome back to today's AI News, brought to you by 9X Roductions.
• 0:03
  Imagine a traditional cororate ladder where every rung is a
• 0:07
  manager passing messages U and down.
• 0:09
  Now imagine ripping out all those middle rungs and replacing
• 0:12
  them with an allseeing highseed elevator.
• 0:15
  That is exactly what is happening in the corporate world
• 0:17
  right now, and it's shaking up the very foundation of how
• 0:20
  companies operate. Yeah, it really is a complete,
• 0:23
  you know, architectural shift. Like we are totally moving from
• 0:26
  those old hierarchical human networks straight into direct
• 0:30
  algorithmically routed information flows.
• 0:32
  Right. And well, the mission of our
• 0:34
  deep dive today is to basically bypass all the usual corporate
• 0:38
  jargon. We really want to understand how
• 0:40
  artificial intelligence is structurally altering
• 0:42
  businesses, global markets and believe it or not, even the way
• 0:46
  you manage casual arguments in your private group chats.
• 0:49
  Oh, absolutely. Which is critical because the
• 0:51
  pace of change has officially crossed over from theoretical
• 0:55
  white papers into the actual practical reality of how we
• 0:58
  work. Exactly.
• 1:00
  So let's jump right into the biggest shift in human capital
• 1:03
  we've seen this year. That starts with this massive
• 1:06
  high stakes bet over at Block. And the raw facts are just, I
• 1:11
  mean, they're pretty staggering. They really are.
• 1:13
  Back in February, Block cut over 4000 employees.
• 1:16
  That is more than 40 of their entire workforce, and Jack
• 1:20
  Dorsey, the CEO, just co-authored a post arguing that
• 1:23
  this was not a response to economic weakness.
• 1:27
  No, not at all. He basically framed it as a
• 1:29
  deliberate opening move, like he actually wants to replace his
• 1:32
  company's entire middle management layer with AI.
• 1:35
  OK, let's unpack this, because on the surface, it sounds like
• 1:39
  they're trying to turn their company's Slack history and
• 1:41
  digital paper trail into like an omniscient robot supervisor.
• 1:45
  Right. Yeah.
• 1:46
  But how does an AI actually replace the nuanced job of a
• 1:49
  human manager? I mean, a manager doesn't just
• 1:51
  forward emails, right? They read the room, they notice
• 1:53
  when an engineer is burning out. They navigate complex
• 1:56
  personality clashes. Can an algorithm really do that?
• 1:59
  Well, what's fascinating here is the specific mechanical argument
• 2:03
  Dorsey is making. And you know why it works
• 2:05
  specifically for a company like Block?
• 2:08
  He isn't actually arguing that an AI can put its arm around a
• 2:11
  struggling employee. Oh OK, So what is the argument
• 2:15
  then? He is arguing that in a
• 2:16
  traditional company, the primary structural function of a manager
• 2:20
  is simply to route information. You know, they take status
• 2:22
  updates from the bottom, summarize them, and pass them
• 2:25
  up. Then they take strategic
• 2:26
  directives from the executive team, break them down and pass
• 2:29
  them down. Wow.
• 2:30
  OK, so it's like treating human beings like routers in a
• 2:34
  corporate Wi-Fi network? Precisely now, because Block
• 2:37
  operates as a remote first company, almost every single
• 2:41
  decision, product, plan, design, iteration, and debate already
• 2:44
  exists as a digital record. Right, it's all captured in the
• 2:47
  cloud anyway. Exactly, and this gives an
• 2:49
  artificial intelligence the exact world model training data
• 2:52
  it needs. When the AI has a live, real
• 2:55
  time understanding of the entire business, it can instantly route
• 2:59
  that information up, down and laterally.
• 3:01
  That shifts the perspective entirely so AI has the context
• 3:06
  of the whole company simultaneously.
• 3:08
  Right. Whereas a traditional middle
• 3:09
  manager usually only has the context of their specific, you
• 3:13
  know, siloed department. You no longer need a human layer
• 3:17
  to summarize a project or schedule a Tuesday check in
• 3:20
  meeting. That is wild.
• 3:21
  And because that routing function is automated, Dorsey
• 3:24
  says, the human beings left at block now fall into just three
• 3:28
  distinct categories. OK.
• 3:30
  What are the new roles? First, you have the builders.
• 3:32
  Those are the engineers, designers, the creatives
• 3:34
  actually making the product. Second, you have problem owners,
• 3:37
  individuals responsible for specific outcomes and
• 3:40
  bottlenecks, but who do not manage people.
• 3:43
  Got it. And the third.
• 3:45
  And you have player coaches, people who still contribute
• 3:47
  directly to the product but also focus on developing the talent
• 3:50
  and well-being of the builders. Wow, that completely eliminates
• 3:54
  the traditional supervisor who just walks around asking for
• 3:57
  status updates. Yeah, the status update is dead
• 4:00
  and I actually want you, the listener, to stop and consider
• 4:03
  your own workplace for a second if all of your emails, your
• 4:07
  daily stand up meetings, your shared documents were fed into
• 4:10
  an AI that instantly knew the real time status of everything.
• 4:14
  Would your current managerial layer still be necessary?
• 4:17
  Exactly. Or are they just acting as that
• 4:19
  information router? I mean, it's a slightly
• 4:21
  terrifying thought if your entire career is built on
• 4:24
  project management, but it also perfectly bridges us into the
• 4:28
  current market dynamics because, well, if AI can legitimately
• 4:32
  streamline a massive complex operation like block and strip
• 4:36
  out that much overhead. The massive valuations we are
• 4:39
  seeing suddenly start to make mathematical sense, right?
• 4:42
  Which brings us to SpaceX. They just filed for June debut
• 4:46
  on the public markets and it is poised to be the largest initial
• 4:49
  public offering in the history of the world.
• 4:51
  The scale of this offering is almost difficult to comprehend.
• 4:54
  It really is. They are targeting A valuation
• 4:57
  north of $1.75 trillion. They're looking to raise
• 5:01
  somewhere between 50 and $75 billion in cash.
• 5:05
  To put that in perspective, that would more than double the
• 5:08
  previous global record, which was Saudi Aramco's $29 billion
• 5:11
  offering back in 2019. Surpassing a state owned oil
• 5:15
  monopoly with rockets and satellites.
• 5:17
  It's just wild. But here's where it gets really
• 5:19
  interesting. I'm looking at the structure of
• 5:20
  this and Elon Musk actually absorbed his artificial
• 5:23
  intelligence company XAI into SpaceX just before making this
• 5:28
  IPO filing, right? He.
• 5:29
  Folded them together. And frankly, I'm confused by the
• 5:32
  financial logic here. The AI side of this new mega
• 5:35
  company reportedly pulls in under $1 billion in revenue.
• 5:39
  Yeah, that's right. Meanwhile, the rocket and
• 5:40
  satellite business is generating roughly 20 billion.
• 5:43
  Yeah. So I mean, is this just a
• 5:45
  strategic stunt, like slapping the hottest tech buzzword onto a
• 5:49
  rocket company to juice the stock price before going public?
• 5:52
  Well, if we connect this to the bigger picture, it is far more
• 5:55
  strategic than a simple evaluation stunt.
• 5:57
  This is really about creating an impenetrable hardware, software
• 6:01
  and data monopoly loop. OK.
• 6:03
  So it's not just about the immediate revenue of the AI
• 6:05
  side. Exactly.
• 6:06
  Yes, Xei's standalone revenue is a fraction of the rocket
• 6:09
  business. Yeah, but by absorbing it,
• 6:11
  SpaceX becomes the very first USAI era mega listing to hit the
• 6:16
  public markets. They are completely beating
• 6:18
  heavyweights like Open AI and Anthropic to Wall Street.
• 6:21
  Oh wow, so the market has been starved for a pureplay AI giant
• 6:26
  to invest in and Musk is serving them this hybrid behemoth
• 6:29
  instead. But how do rockets and AI
• 6:32
  actually loop together mechanically?
• 6:34
  Think about the data through Starlink.
• 6:36
  SpaceX controls a massive global data transmission network.
• 6:40
  They are gathering unprecedented amounts of telemetry and real
• 6:43
  world data. Right.
• 6:45
  So they own the pipeline. Exactly.
• 6:46
  You feed that unique data into XI's algorithms to make them
• 6:49
  smarter. Then you deploy those smarter
• 6:51
  algorithms into Tesla's Optimus robotics or into the autonomous
• 6:55
  navigation of the Rockets themselves.
• 6:57
  Oh I see. It is a closed loop of hardware
• 6:59
  collecting data, software learning from IT, and robotics
• 7:03
  executing it. In the physical world, no rival
• 7:06
  can match that infrastructure. That makes perfect sense.
• 7:10
  It's not just a rocket company anymore.
• 7:11
  It's an infrastructure company for autonomous intelligence.
• 7:14
  But you know, I also noticed that structurally this IPO is
• 7:18
  highly unusual. About 30% of the shares will be
• 7:22
  open to everyday retail investors.
• 7:24
  Which is incredibly high for an offering of the scale.
• 7:27
  Usually, institutional investors and massive hedge funds swallow
• 7:31
  up all the early action. Right, But there's a massive
• 7:34
  catch. It comes with a special 2 tier
• 7:36
  voting structure. Even though the public is
• 7:39
  financing this massive $50 billion expansion, Musk retains
• 7:43
  absolute, unbreakable control over the company's direction
• 7:46
  after goes public. Yeah, he is democratizing the
• 7:48
  financial investment, but he is is certainly not democratizing
• 7:51
  the governance. It's a very calculated move.
• 7:54
  But you know, to build these trillion dollar closed loop
• 7:58
  ecosystems and to build AI models capable of replacing
• 8:02
  those corporate managers we talked about earlier, you need
• 8:05
  models that actually understand how to do the work.
• 8:08
  They can't just scrape public Reddit forms anymore.
• 8:11
  They need highly specific, nuanced human expertise.
• 8:15
  Which perfectly transitions us to how they are actually getting
• 8:19
  that data, essentially by paying seasoned professionals to dig
• 8:23
  their own graves. Yeah, it's a big grim when you
• 8:25
  phrase it like that. Well, according to a deep dive
• 8:27
  from Business Insider, Open AI is running a massive internal
• 8:30
  initiative called Project Stagecraft through a platform
• 8:34
  called Handshake AI. They are paying up to 4000
• 8:37
  freelance professionals roughly $50.00 an hour to build hyper
• 8:40
  specific occupation based training data.
• 8:43
  Right. And this raises an important
• 8:45
  question about the definition of valuable data at this current
• 8:48
  stage of the AI race. What do you mean?
• 8:50
  Well, for years, a data labeling was manual labor.
• 8:53
  It was asking people to click on the stoplights in a photo or
• 8:56
  verify if a sentence made grammatical sense.
• 8:58
  Right, the classic keep tcha stuff.
• 9:00
  Exactly, but that era is over. The goal now is to
• 9:04
  mathematically map complex knowledge work, not manual
• 9:08
  labor. And the professions they're
• 9:10
  recruiting are intense. I mean, we are talking about
• 9:13
  commercial aviation experts, clinical pharmacists, plant
• 9:16
  scientists and senior human resources specialists.
• 9:20
  But how do you actually map someone's expertise?
• 9:23
  They aren't just typing out a textbook.
• 9:24
  Right, not at all. The work they're doing for that
• 9:27
  hourly rate is exhaustively detailed.
• 9:29
  These experts are instructed to create incredibly specific
• 9:33
  personas and simulate entire multi day professional
• 9:36
  workflows. Wait, so they have to act it
• 9:38
  out? Essentially, yes.
• 9:39
  They aren't just giving the AI the final answer to a problem.
• 9:43
  They have to provide the starting context, the
• 9:45
  intermediate goals, the blind alleys they went down, the
• 9:47
  reference materials they consulted and the final
• 9:49
  deliverables. Wow, so if you are a pharmacist
• 9:52
  in this program you aren't just listing drug interactions?
• 9:54
  Exactly. You are simulating how your
• 9:56
  brain actually works when a patient presents with a rare
• 9:59
  multi drug interaction compounded by specific
• 10:01
  underlying health conditions. Oh, so you have to document
• 10:05
  everything? Yeah, which medical database you
• 10:07
  check first, how you weigh conflicting risk factors, and
• 10:10
  how you formulate your final recommendation.
• 10:12
  You are literally charting the invisible cognitive architecture
• 10:15
  of your own expertise. Capturing the thought process
• 10:19
  itself, not just the output, but, you know, reading through
• 10:22
  the report, there was a quote from one of the contractors that
• 10:24
  really stopped me in my tracks, they told the reporter, Quote,
• 10:28
  we all were aware that we were basically training AI to replace
• 10:32
  us. Yeah.
• 10:32
  That is heavy. It is.
• 10:34
  I have to ask, if Open AI is successfully mapping the
• 10:37
  cognitive workflows of aviation experts and pharmacists right
• 10:41
  now, what does this mean for the timeline of artificial General
• 10:44
  intelligence? Are we looking at a much shorter
• 10:46
  runway to AGI? All the indicators suggest the
• 10:49
  timeline is compressing rapidly. I mean AGI requires exactly this
• 10:53
  kind of granular cross disciplinary reasoning.
• 10:56
  The fact that Open AI is spending this much capital to
• 10:58
  catalog specialized task by task knowledge shows they are
• 11:01
  preparing the foundation for a massive leap in autonomous
• 11:04
  capabilities. That makes a lot of sense.
• 11:06
  In fact, internal leaks show Open AI is already drafting
• 11:10
  policy papers focused on Severe economic disruption.
• 11:13
  They are openly discussing how society will need to rethink the
• 11:17
  entire social contract. Right when the engineers
• 11:20
  building the machine start proactively writing policy
• 11:22
  papers about how to handle the economic fallout of the machine,
• 11:25
  we should probably pay close attention.
• 11:27
  But the massive data collection from Stagecraft is actually
• 11:30
  what's allowing researchers to distill all that complex
• 11:33
  reasoning down into much smaller, Hyerefficient models
• 11:37
  and entirely new modalities. Absolutely.
• 11:39
  Let's look at the RAID fire breakthroughs in our quick hits
• 11:41
  today because we are shifting from these massive Cororate
• 11:44
  stories to some really wild technical leaps.
• 11:48
  First, UZAI just rolled out GLM 5V Turbo.
• 11:52
  This is being called a vision coding model.
• 11:54
  It's a huge leap for visual processing.
• 11:56
  Yeah, it literally looks at screenshots, design drafts, or
• 11:59
  even a rough sketch on a napkin and directly generates runnable
• 12:03
  computer code from what it sees. But I'm struggling to understand
• 12:06
  the mechanism here. How does a machine look at a
• 12:08
  flat arrangement of pixels and understand how to code a working
• 12:11
  app? Well, it requires a leap from
• 12:13
  basic optical character recognition to deep semantic
• 12:16
  spatial understanding so that the model doesn't just, you
• 12:20
  know, see the text on the screen.
• 12:21
  It understands the functional relationship of the visual
• 12:23
  components. OK, like how?
• 12:25
  It recognizes that a specific rectangular shadow represents A
• 12:28
  clickable button, or that a small triangle next to text
• 12:33
  implies a drop down menu. It maps that visual intent
• 12:36
  directly to the syntax of a programming language.
• 12:39
  Wow, so it's like having a senior developer looking over
• 12:42
  your shoulder, instantly translating your napkin sketch
• 12:46
  into a working app. Exactly.
• 12:47
  It removes an incredible amount of friction.
• 12:50
  And Speaking of LEAPS, you really have to look at this
• 12:52
  industry shift toward efficiency with RCAI and liquid AI.
• 12:56
  Oh right, the efficiency pivot. Grouping those two together
• 12:59
  makes total sense. Yeah, RCAI just introduced their
• 13:02
  Trinity large thinking model. This is an open weight reasoning
• 13:05
  model that rivals top tier proprietary models like Opus 4.6
• 13:09
  on industry benchmarks, but at 120th the cost.
• 13:12
  Slashing compute by 95% is a game changer and liquid AI.
• 13:16
  Liquid just released the LFM 2.5350 M It's a remarkably tiny
• 13:22
  model built specifically to excel at using external tools
• 13:26
  like calculators or web search. It's outperforming models twice
• 13:29
  its size. We are moving toward highly
• 13:31
  capable, cheap on device reasoning.
• 13:34
  That makes perfect sense, but it's not just coding and
• 13:37
  reasoning getting these upgrades right.
• 13:39
  Alibaba just launched 12.7 Image.
• 13:42
  It's a unified model that handles image generation and
• 13:45
  precise editing. Right, with text rendering
• 13:47
  across 12 languages, yeah. But the standout feature is that
• 13:50
  it can maintain character consistency across up to 12
• 13:53
  different images from a single prompt.
• 13:55
  I know that has historically been the Holy Grail for AI video
• 13:59
  and comic generation. How do they actually maintain
• 14:01
  that consistency mathematically? It all comes down to how the
• 14:04
  model manages its latent space. It is the multi dimensional
• 14:07
  mathematical map where it stores concepts. 12.7 image anchors the
• 14:11
  core mathematical representation of the character's features in
• 14:14
  that latent space. Oh, so it locks the identity?
• 14:17
  Exactly. It only alters the surrounding
• 14:19
  noise, like the lighting or the background, while keeping the
• 14:22
  core identity strictly locked. And as these visual and creative
• 14:26
  tools get more advanced, evaluating them becomes
• 14:28
  incredibly difficult, which is why Contra Labs emerging from
• 14:31
  stealth mode today is so fascinating.
• 14:34
  Right. And just to clarify for those
• 14:35
  unfamiliar, operating in stealth mode means a startup has been
• 14:38
  building their product in total secrecy and now that they have
• 14:41
  launched their entire business model is fascinating, they
• 14:44
  aren't building AI generation tools.
• 14:46
  No. They are building evaluation
• 14:48
  platforms specifically designed to grade AI creative tools.
• 14:52
  They are building data sets and leaderboards focused entirely on
• 14:55
  benchmarking human creative taste.
• 14:57
  Quantifying good taste so we can teach it to an algorithm.
• 15:00
  Incredible. Now, that push for cheap,
• 15:02
  accessible models is exactly what puts this immense power
• 15:06
  into the hands of everyday users, which brings this entire
• 15:09
  deep dive back down to the ground.
• 15:11
  Right, the community workflows. Exactly, we've covered trillion
• 15:15
  dollar valuations and mapping professional expertise, but how
• 15:19
  are regular people using these tools right now?
• 15:22
  Well, today we have a brilliant workflow from Thomas LA User out
• 15:26
  in France I. Love this one.
• 15:28
  Yeah. Thomas runs a private Discord
• 15:30
  server where he and his friends hang out, play video games, and
• 15:33
  inevitably debate politics. And as anyone who has ever been
• 15:37
  on the Internet knows, a group of friends arguing politics in a
• 15:40
  chat room can get incredibly heated.
• 15:43
  A tale as old as broadband. Exactly.
• 15:46
  So what does this all mean? Well, Thomas built this custom
• 15:50
  Taoist inspired bot powered by Claude.
• 15:52
  When two friends are getting into a vicious argument, someone
• 15:54
  can tag this bot. The bot instantly reads the
• 15:57
  room. It pulls all the recent
• 15:58
  messages. Right.
• 15:59
  Yeah. And it factors in the historical
• 16:00
  context of the specific participants, their shared
• 16:03
  inside jokes, their vocabulary. Then it sends all of that
• 16:06
  context via an API to Anthropics Haiku model and brings the
• 16:10
  response back into the Discord channel.
• 16:12
  But the brilliance is in the prompt design.
• 16:14
  The goal of the bot isn't to play judge and jury and declare
• 16:17
  who is right or wrong. No, it's entire programmed
• 16:20
  purpose is to bring perspective, clarity, and a little bit of
• 16:24
  self aware humor to the argument.
• 16:26
  Heavily inspired by Taoist philosophy, it actively de
• 16:30
  escalates the tension. I really love this workflow
• 16:32
  because it serves. This is the absolute perfect
• 16:34
  contrast to the block restructuring story we opened
• 16:37
  with. How so?
• 16:38
  Well, you have massive corporations using AI as a
• 16:41
  surgical tool to cut jobs, remove middle management, and
• 16:45
  ruthlessly maximize operational efficiency by treating humans
• 16:49
  like data routers. But here you have regular people
• 16:53
  using the exact same underlying technology to mediate human
• 16:56
  connection. They're using AI to de escalate
• 16:59
  conflict and inject philosophy into their group chats.
• 17:03
  It is the ultimate paradox of this technology.
• 17:05
  The exact same neural network can be a corporate grim Reaper
• 17:09
  or a digital peacekeeper, depending entirely on who was
• 17:11
  writing the prompt. It really is.
• 17:12
  And that contrast leads us to a really profound question.
• 17:15
  Oh, what's that? If artificial intelligence is
• 17:18
  now fully capable of acting as the omniscient, hyper efficient
• 17:22
  middle manager at our workplaces, routing our daily
• 17:25
  tasks, while simultaneously acting as the empathetic Taoist
• 17:29
  mediator in our private heated arguments with our friends, what
• 17:33
  eventually happens to human soft skills?
• 17:35
  Oh wow, that is a fascinating angle.
• 17:38
  Think about it. If we rely on an algorithm to
• 17:40
  tell us how to work together efficiently, and we rely on an
• 17:43
  algorithm to help us forgive each other, are we slowly,
• 17:46
  quietly outsourcing our patience, our empathy, and our
• 17:50
  interpersonal organization entirely to algorithms like a
• 17:53
  muscle that doesn't get used? Do our own capacities for
• 17:57
  conflict resolution begin to atrophy?
• 17:59
  That is exactly the kind of question we need to be asking is
• 18:01
  that this technology integrates into every single corner of our
• 18:04
  lives. It's not just about what the AI
• 18:06
  can do, it's about what we stopped doing once the AI does
• 18:09
  it for us. We'll leave you with that
• 18:10
  thought to chew on today. It's definitely something to
• 18:13
  ponder. If you want to stay ahead of
• 18:14
  this incredibly fast moving tech world and keep navigating these
• 18:18
  structural shifts with us, make sure you hit subscribe on the
• 18:21
  show right now. If you got some value out of
• 18:23
  this Dee Dock today, please take a second to rate the show 5
• 18:26
  stars. We'll see you back for
• 18:28
  tomorrow's episode.