Latest / Today’s AI News / OpenAI Raises $122B, Claude Code Source Code Leaks, Americans Are Using AI More But Trusting It Less

Transcript

OpenAI Raises $122B, Claude Code Source Code Leaks, Americans Are Using AI More But Trusting It Less · Today’s AI News

• 0:00
  Imagine a single whale so massive it changes the tide of
• 0:04
  the entire ocean just by turning around.
• 0:07
  That is quite the image, right? But also, that's exactly what
• 0:11
  Open AI's record shattering $122 billion funding round feels like
• 0:16
  today. They aren't just swimming in the
• 0:17
  market, they literally are the market.
• 0:20
  Oh, absolutely. So welcome back to today's AI
• 0:22
  News, brought to you by Nynex Productions.
• 0:25
  Our mission for this deep dive is to unpack this historic cash
• 0:29
  influx. We're going to look at open AIS
• 0:31
  tightening strategy and how that sharply contrasts with, well, a
• 0:34
  highly unusual code leak from their biggest rival, Anthropic.
• 0:38
  Which was incredibly messy, yeah.
• 0:39
  Yeah, super messy. We're also breaking down some
• 0:42
  really cool new methods developers are using to wrangle
• 0:44
  AI agents. And then we're digging into a
• 0:46
  startling new poll that reveals how you, the public, really feel
• 0:50
  about this whole AI boom. It's a landscape defined by
• 0:54
  extreme friction right now. I mean, the sheer velocity of
• 0:56
  the capital and the tech is it's entirely disconnected from how
• 1:01
  the broader public is processing these changes.
• 1:03
  It really is. So let's start with the money.
• 1:04
  Let's do it the undeniable center of gravity.
• 1:07
  Because if you've only been skimming the headlines lately,
• 1:10
  you probably thought rumors of executives leaving open AI or,
• 1:14
  you know, sudden pivots like pausing that sort of video tool.
• 1:17
  Right whispers of a company in crisis.
• 1:19
  Exactly. But their balance sheet tells a
• 1:21
  completely different story. They just closed $122 billion
• 1:26
  funding round. Which is why.
• 1:27
  At an 852 billion dollar valuation, I mean this is the
• 1:32
  largest single fundraise in venture history.
• 1:35
  By a mile. You really have to look at who
• 1:38
  is writing those checks to understand what's actually
• 1:40
  happening. You've got Amazon, NVIDIA and
• 1:43
  SoftBank anchoring 110 billion of this.
• 1:46
  And you know, NVIDIA is basically just investing in its
• 1:49
  own forward revenue. They're making sure Open AI has
• 1:51
  the cash to keep buying Nvidia's next Gen. chips.
• 1:54
  The ultimate circular economy, right?
• 1:56
  Exactly. And SoftBank is just building a
• 1:58
  financial Moat so deep that basically no competitor can
• 2:01
  cross it, right? But Amazon's deal is the one
• 2:04
  that really stands out. OK, let's unpack this, because
• 2:06
  this is the part that kind of broke my brain.
• 2:07
  Amazon's deal includes this specific AGI clause.
• 2:10
  Right, the artificial general intelligence clause.
• 2:12
  Yeah. So if Open AI achieves AGI,
• 2:15
  which is basically an AI that outsmarts humans at any economic
• 2:20
  task, this clause triggers a reset of their financial terms.
• 2:24
  It does. But wait, if an AGI clause
• 2:26
  triggers a reset, doesn't that mean investors are financial
• 2:29
  incentivize for open AI to delay achieving its ultimate stated
• 2:32
  goal? Like why would you punish them
• 2:34
  for crossing the finish line? Yeah.
• 2:36
  I mean, it sounds totally counterintuitive, right?
• 2:38
  It looks like a penalty, but it's actually not an incentive
• 2:41
  to delay. It's a structural hedge.
• 2:44
  What do you? Think about the economic paradox
• 2:46
  of AGI. If one private company
• 2:49
  successfully build a true God machine, traditional capitalism
• 2:53
  just fundamentally breaks down. Oh wow.
• 2:56
  Right Intellectual property, compounding returns, normal
• 2:59
  vendor contracts, all that becomes completely obsolete
• 3:01
  overnight. So Amazon is basically putting a
• 3:04
  legal safety valve in place. So they're saying if you change
• 3:06
  the world, all bets are off. Exactly, they are acknowledging
• 3:10
  that if AGI is turned on, we enter a post scarcity reality
• 3:15
  and they need the right to immediately renegotiate their
• 3:18
  relationship with the creator. Which honestly just raises the
• 3:21
  question of who actually gets to decide if AGI has been achieved
• 3:24
  to trigger that calls in the 1st place I.
• 3:26
  Know it's a total legal paradox, but it shows how seriously these
• 3:29
  tech giants are treating the timeline.
• 3:32
  For sure. But if we look at the present,
• 3:34
  Open II is aggressively making money right now like $2 billion
• 3:39
  a month in revenue. That pace of scaling is just
• 3:42
  unprecedented. Yeah, it's growing four times
• 3:44
  faster than Alphabet and Meta did at this exact same stage.
• 3:47
  And what's fascinating here is where that money is coming from.
• 3:51
  Over 40% of their revenue is now enterprise, strictly business to
• 3:55
  business, OK. That is a huge shift.
• 3:57
  Right, and it's their fastest growing segment.
• 3:59
  That one metric completely recontextualizes all those
• 4:02
  recent operational pivots, like shutting down Sora and
• 4:05
  abandoning those consumer side quests.
• 4:07
  So they aren't just an open-ended research lab having
• 4:09
  fun anymore. Not at all.
• 4:11
  They are ruthlessly skating to where the enterprise money is.
• 4:14
  They want to build a unified super app.
• 4:17
  Merging ChatGPT codecs for coding and autonomous agents all
• 4:21
  into 1B to B ecosystem. Exactly.
• 4:24
  They're prepping for a massive IPO.
• 4:26
  You don't get to a trillion dollar valuation with flashy
• 4:29
  video experiments. You get there by becoming the
• 4:32
  foundational operating system for corporate America.
• 4:35
  That makes total sense. And you know, seeing open AI
• 4:38
  tighten up their corporate strategy like that makes the
• 4:40
  absolute mess over at Anthropics stand out so much.
• 4:43
  Oh, the timing was incredible. Right, because Empropic's whole
• 4:47
  brand identity, their entire market differentiator is safety
• 4:51
  and security. Responsible AI company.
• 4:53
  Exactly. And yet they just experienced
• 4:56
  this highly uncharacteristic super public code leak.
• 5:00
  Yeah, they accidentally pushed the source code for their
• 5:01
  unreleased AI coding tool. Clawed code right out onto a
• 5:05
  public registry. Just sitting there for anyone to
• 5:07
  see. Over 1900 files, half a million
• 5:10
  lines of code. And developers obviously pounced
• 5:12
  on it immediately. Oh, within hours the GitHub
• 5:16
  mirror had something like 4000 stars and over 7000 forks.
• 5:20
  The dev community ripped it apart.
• 5:22
  And they found 44 hidden feature flags and three entirely
• 5:27
  unreleased projects. Including some serious
• 5:29
  architectural stuff like cross session memory.
• 5:32
  So the AI remembers what you were coding yesterday.
• 5:34
  Exactly, and deep planning systems for multi step agent
• 5:37
  autonomy. They even found internal code
• 5:40
  names like Capybara, which looks like a clawed 4.6 variant
• 5:43
  already on version 8 internally. OK, but here's where it gets
• 5:46
  really interesting. Because alongside all that
• 5:49
  serious enterprise tech, they found code tracking every time a
• 5:52
  user swears A clawed Yep. It's a swear.
• 5:54
  Tracker and an unreleased virtual AI terminal pet code
• 5:57
  named Buddy with 18 different species, rarity, tears and
• 6:01
  stats. Like there is a stat literally
• 6:03
  called chaos and another called snark.
• 6:05
  I love that so much. It's like finding out the
• 6:07
  strictest, safest kid in class has a secret diary tracking your
• 6:10
  bad words and, you know, a virtual pet with ACA Austat.
• 6:14
  It really shatters that ultra serious corporate veneer.
• 6:18
  But honestly, building a snarky virtual pet into an enterprise
• 6:21
  coding tool makes a lot of sense if you think about it.
• 6:25
  Really. How so?
• 6:26
  Well, software engineers spend 10 to 12 hours a day staring at
• 6:29
  highly abstract, super sterile terminal environments.
• 6:34
  The cognitive load is brutal. That is very true.
• 6:36
  O by gamifying the interface, adding an emotional anchor like
• 6:40
  BBDY, you create Doamine loop. Yeah, it reduces burnout and
• 6:44
  turns the AI into a collaborative presence rather
• 6:47
  than just a blank prompt. That actually is really smart,
• 6:50
  but Anthropic is saying this was just human error.
• 6:53
  No customer data was exposed. Does that defense hold up given
• 6:57
  how bad this looks? Technically, yes, it holds up OK
• 7:00
  because of what was actually leaked.
• 7:02
  They only exposed the command line interface, the CLI layer.
• 7:05
  So what does that mean in plain English?
• 7:07
  Think of the underlying model weights, the billions of
• 7:09
  parameters that actually make Claude Smart as the brilliant
• 7:12
  chef in a restaurant, and the secret recipes OK.
• 7:15
  The brain. Right, the CLI layer is it's
• 7:17
  just the waiter. It takes your order, walks to
• 7:19
  the kitchen and brings the food back.
• 7:21
  And Tropic basically leaked the blueprints for their waiter, not
• 7:24
  the chef. Got it.
• 7:25
  So open source rivals already have waiters anyway.
• 7:28
  Exactly. They do that routing logic for
• 7:29
  free. So from a competitive
• 7:31
  standpoint, there's no massive technical loss, but the
• 7:35
  reputational damage that is very real to major leaks in a week
• 7:41
  for a company whose entire identity is we are the safest,
• 7:45
  it just gives the Internet and their competitors so much to
• 7:48
  talk about it. Definitely does.
• 7:50
  Now, Speaking of those deep planning features Anthropic
• 7:52
  accidentally showed off, we're seeing these models being pushed
• 7:55
  to act like massive super apps. They are taking on incredibly
• 7:58
  complex multi step workflows. Right, but developers on the
• 8:01
  ground are hitting a massive wall trying to keep these AI's
• 8:04
  focused. Context rot.
• 8:06
  Yes, context rot. What exactly is that?
• 8:08
  It's. A fascinating limitation.
• 8:10
  So transformer models use these attention mechanisms.
• 8:13
  They have to weigh the relevance of every single word in your
• 8:16
  prompt against every other word. But when you start dumping
• 8:20
  hundreds of pages of API docs, passcode and long conversational
• 8:25
  histories into the context window.
• 8:27
  It just gets confused. Basically, the attention heads
• 8:29
  get completely overwhelmed, the signal to noise ratio degrades,
• 8:34
  and the AI just loses the thread of what it's supposed to be
• 8:37
  doing. And that's when it starts
• 8:38
  hallucinating. Exactly.
• 8:40
  Inventing code that doesn't exist, using outdated protocols,
• 8:43
  all of that. While the sources highlighted a
• 8:46
  really clever and completely free workflow to fight this,
• 8:50
  It's a Chrome extension called Mark Snip.
• 8:53
  Mark Snip is brilliant. It really is.
• 8:54
  So the workflow is you install mark snip and you use it to
• 8:58
  download a clean markdown file of the exact documentation you
• 9:01
  need. It strips out all the garbage.
• 9:03
  Right, no web format saying no ad scripts, no navigation bars,
• 9:06
  just pure text data. Then you save that file in a
• 9:09
  specific folder in your working repo like agent context.
• 9:12
  And that folder becomes a physical boundary for the.
• 9:15
  AI yes, you prompt the agent and say only read the files in the
• 9:19
  agent context folder and use that as your absolute source of
• 9:22
  truth. It's an incredible way to
• 9:23
  isolate context. It's literally like giving me AI
• 9:27
  an open book test. Instead of letting it
• 9:29
  hallucinate answers from its hazy memory of the entire
• 9:32
  Internet, you are physically handing it the exact textbook
• 9:35
  chapter it needs to pass the test.
• 9:37
  And restricting it from looking anywhere else.
• 9:39
  Exactly, and the hallucination rate just plummets.
• 9:43
  Plus pro tip here, you can use this for anything.
• 9:46
  Grabbing the meat of long articles, Twitter threads, dense
• 9:50
  government sites. Yeah, anyone can use it to
• 9:52
  control their models data, diet. Which is so empowering for
• 9:55
  users. But, you know, while developers
• 9:57
  are finding all these clever ways to optimize AI tools, we
• 10:00
  really need to look at how the rest of the world feels about
• 10:03
  this breakneck pace. The public sentiment is
• 10:05
  definitely shifting. A new Quinnipiac University poll
• 10:08
  just came out and it shows a massive disconnect.
• 10:11
  AI usage has jumped 14%. People are definitely using it.
• 10:15
  For sure, top uses are research at 51%, writing at 28 and data
• 10:19
  analysis at 27, so the adoption is there.
• 10:21
  But the trust is sinking. Cratering.
• 10:23
  Honestly, jobbing society spiked 14 points. 70% of people now
• 10:28
  expect AI to shrink their job opportunity.
• 10:31
  That is a staggering #70%. And there is this sharp income
• 10:35
  divide that really jumped out to me. 52% of people earning over
• 10:39
  $200,000 say AI does more good than harm.
• 10:43
  But if you look at those making under 50,000. 60% of them say it
• 10:47
  does more harm. Also, only 5% of the public
• 10:50
  believes AI developers represent their interests. 5% and 74
• 10:55
  percent want more government regulation.
• 10:57
  Yeah, the data is incredibly polarized.
• 11:00
  So what does this all mean? It's basically an adoption boom
• 11:02
  wrapped in a panic attack. People are using it more than
• 11:05
  ever to do their jobs while simultaneously terrified it's
• 11:09
  going to steal that exact same job.
• 11:11
  It's a really complex dynamic. If we connect this to the bigger
• 11:14
  picture, you have to look at the mechanics of Labor versus
• 11:18
  capital. OK, break that down.
• 11:19
  High earners, right? They're generally in strategic
• 11:22
  roles. For them, AI is a lever.
• 11:25
  It automates administrative friction, amplifies their output
• 11:28
  and makes them look better without threatening their core
• 11:31
  job. They are using the AI.
• 11:32
  Right, but lower wage earners are often doing exactly the kind
• 11:36
  of repetitive, process oriented tasks that these models are
• 11:41
  explicitly built to automate. So they are facing direct labor
• 11:44
  substitution. Exactly.
• 11:46
  For them, AI isn't the lever, it's a replacement.
• 11:49
  And historically, whenever a transformative technology scales
• 11:52
  faster than the social infrastructure can absorb it,
• 11:55
  you see this exact kind of public besimism.
• 11:58
  Which usually leads to what? Regulatory backlash.
• 12:01
  The optimism inside the Silicon Valley bubble is clashing so
• 12:04
  hard with Main Street fear right now.
• 12:06
  And Despite that fear, the tech itself is just plowing forward.
• 12:10
  Let's run through today's quickest ecosystem updates to
• 12:12
  show just how fast things are moving.
• 12:14
  Let's do these quick hits. We are seeing models rapidly
• 12:17
  expanding and shrinking at the same time.
• 12:19
  On the expanding side, Google just launched VO 3.1 Lite.
• 12:22
  A very budget friendly video generation model.
• 12:25
  Yep, up to 8 seconds of video at half the cost of their fast
• 12:28
  model. Then there's Fauna, which is a
• 12:30
  built in creative agent for model routing and prompting.
• 12:33
  And Alibaba just dropped Quinn 3.5 Omni, which is a massive
• 12:38
  system for native text, image, audio and video understanding.
• 12:41
  Don't forget H Company, they launched Hollow Three, a brand
• 12:45
  new open weight computer use agent.
• 12:47
  And Salesforce overhauled their Slack Bot agent with 30 new
• 12:51
  capabilities, including reusable skills.
• 12:54
  It's just massive integration everywhere.
• 12:56
  But then on the shrinking side, Prism ML launched Bonsai.
• 13:01
  I love. This one It's a tiny open source
• 13:03
  model capable of running on standard consumer hardware, like
• 13:06
  right on your laptop. Proving you don't always need a
• 13:08
  billion dollar server farm for capable reasoning, right?
• 13:12
  But Speaking of server farms, Oracle just announced the
• 13:15
  largest layoff in their entire corporate history.
• 13:17
  Thousands of jobs gone. And they are openly crediting
• 13:20
  this major restructuring to a pivot toward AI infrastructure.
• 13:24
  You have Prism ML making AI tiny enough for your laptop, and then
• 13:27
  you have Oracle completely gutting its human workforce just
• 13:31
  to make room for massive AI server racks.
• 13:33
  Which brings that 70% job anxiety from the pull right into
• 13:37
  reality. The displacement isn't
• 13:39
  theoretical anymore. No, it's happening right now.
• 13:42
  It is a very stark macroeconomic reality, but I want to end on a
• 13:46
  slightly different note because our Community AI workflow
• 13:49
  segment today shows the complete flip side of that Oracle news.
• 13:54
  It's a great microeconomic counterpoint.
• 13:56
  Exactly. Oracle represents the macro job
• 13:59
  loss, but this story shows how AI can empower someone to
• 14:03
  completely reinvent themselves after losing a job.
• 14:06
  We're talking about Claire B. Yes, Claire is a 58 year old
• 14:09
  from New Zealand who describes herself as a total tech phobe.
• 14:13
  I think a lot of people can relate to that.
• 14:14
  Oh for sure. So she was made redundant last
• 14:17
  year, but instead of giving up she used clawed to build a brand
• 14:20
  new business from scratch, making tallow and kawakawa based
• 14:24
  bombs. And she?
• 14:25
  Didn't just use it to write emails or whatever.
• 14:27
  No, she used Clawed as a literal business partner.
• 14:30
  It helped her write her business plan, crunch her financials,
• 14:32
  perfect the chemical balance of her recipes, gave her ingredient
• 14:36
  advice. It even held her hand through
• 14:37
  setting up Shopify and SEO, right?
• 14:40
  Exactly. A self-proclaimed tech pho doing
• 14:43
  her own SEO because the AI walked her through it step by
• 14:46
  step, and she noted that Claude kept her encouraged when she
• 14:49
  felt overwhelmed by the technical friction.
• 14:51
  That is so important. While 70% of people are
• 14:54
  terrified of losing their jobs, Claire's story proves that AI is
• 14:57
  simultaneously acting as a free Co founder.
• 15:00
  It dramatically lowers the barrier to entry for
• 15:02
  entrepreneurship at literally any age.
• 15:04
  It absolutely does, and I think the duality is the perfect thing
• 15:08
  to think about today. I completely agree.
• 15:11
  So looking at Oracle's massive layoffs versus Claire's new
• 15:14
  business, ask yourself what unique human skills do you
• 15:18
  possess that an AI super app just can't replicate?
• 15:22
  And how can you start treating AI as your Co founder rather
• 15:25
  than your competitor? That is the ultimate question we
• 15:28
  all have to figure out right now, and that is all the time we
• 15:30
  have for a deep dive today. Be sure to hit that subscribe
• 15:32
  button to stay updated on all this crazy momentum.
• 15:35
  And if you enjoyed the show, please rate US five stars.
• 15:37
  We will see you back for tomorrow's episode.