Latest / AI Fire Daily / 🎙️ EP 345: Apple Alleges Trade Secret Theft & Google Unveils TimesFM-3

Transcript

🎙️ EP 345: Apple Alleges Trade Secret Theft & Google Unveils TimesFM-3 · AI Fire Daily

• 0:00
  Imagine walking out of a bank with a literal
• 0:03
  duffel bag full of cash. Right. You get caught
• 0:06
  by the police and you just look at them and tell
• 0:08
  them, you know, it is entirely the bank's fault
• 0:11
  for leaving the vault door unlocked. Yeah. Which
• 0:13
  is a wild defense. It's insane. But that is quite
• 0:17
  literally the defense OpenAI is using right now
• 0:20
  in this massive corporate espionage lawsuit against
• 0:23
  Apple. Welcome to this custom tailored deep dive,
• 0:27
  by the way. Our mission today is to help you
• 0:29
  navigate the rapidly escalating, really high
• 0:33
  stakes reality of the AI revolution. Because
• 0:36
  you need to know where the billions of dollars
• 0:37
  funding this are actually heading. Exactly. And
• 0:40
  to figure that out, our stack of sources today
• 0:42
  comes from industry dispatches, including AI
• 0:45
  Fire, Belay, and some very revealing technical
• 0:48
  release notes from Google. Yeah, we're really
• 0:50
  cutting through the noise today. We are. We're
• 0:52
  going to map out everything from that high stakes
• 0:54
  Apple lawsuit to a massive leap in time series
• 0:56
  forecasting, all the way down to why these tools
• 0:59
  might literally be what's slowing down your personal
• 1:02
  computer. Oh, yeah. The PC lag is real. It really
• 1:04
  is. But let's start on the high stakes battlefield.
• 1:08
  So Apple just came out swinging against OpenAI.
• 1:11
  They claim they have shocking evidence against
• 1:13
  this former Apple engineer, Chang Liu, who now
• 1:18
  works over at OpenAI. Right, the hardware guy.
• 1:20
  Yeah. According to the filings, while he was
• 1:22
  transitioning out of Apple, he allegedly took
• 1:24
  a confidential circuit schematic. Which, just
• 1:27
  to give some context, is an incredibly sensitive
• 1:29
  piece of IP. It's not just a sketch. No. I mean,
• 1:32
  for anyone outside of hardware engineering, a
• 1:34
  circuit schematic is the literal microscopic
• 1:37
  blueprint of how a processor routes power and
• 1:41
  data. Wow. It takes billions of dollars in R
• 1:44
  &D to figure out the exact design. It's the crown
• 1:46
  jewels. And Apple is claiming he didn't just
• 1:49
  like accidentally download it. They allege he
• 1:51
  used a tool that shares the exact name of an
• 1:54
  internal Apple app to access the data. Yeah,
• 1:56
  that implies a level of intent. Right. And they
• 1:59
  also claim OpenAI knew Lou still had access to
• 2:02
  Apple's network. But the kicker, the real ticker,
• 2:05
  is that Apple alleges once Lou realized he was
• 2:09
  being investigated, he actually asked an OpenAI
• 2:12
  colleague to help him destroy the digital evidence.
• 2:15
  Yeah, the destruction of evidence allegation,
• 2:17
  that brings a completely different level of severity
• 2:19
  to this whole thing. It really escalates it.
• 2:21
  It does. It shakes the entire conversation from,
• 2:23
  you know. a potential accidental data retention
• 2:26
  issue to a claim of malicious coordinated corporate
• 2:30
  theft. OK, let's unpack this, because OpenAI's
• 2:33
  legal defense brings us right back to that bank
• 2:35
  vault analogy. The unlocked door. Exactly. OpenAI
• 2:39
  is arguing that Lou only had this access because
• 2:41
  of Apple's poor off -boarding and security practices.
• 2:44
  They are essentially saying, hey, your IT department
• 2:47
  failed to revoke his credentials fast enough,
• 2:49
  so you can't blame us for what he did with that
• 2:52
  access. The audacity is staggering to me, especially
• 2:55
  when you consider, what, there are reportedly
• 2:57
  over 400 former Apple employees at OpenAI. right
• 3:00
  now. Yeah, over 400. And if that defense actually
• 3:03
  holds up in court, it sets a wild precedent for
• 3:06
  the tech industry. It really does. But what's
• 3:07
  fascinating here is that this case is no longer
• 3:10
  just about one rogue employee making a mistake
• 3:14
  or even about one stolen blueprint. What do you
• 3:16
  mean? You have to look at the specific legal
• 3:19
  remedy Apple is asking the judge for. They aren't
• 3:22
  just asking for monetary damages. They're asking
• 3:24
  the court to temporarily block OpenAI from developing
• 3:28
  hardware based on Apple technology while the
• 3:31
  whole case is litigated. Wait, really? So this
• 3:33
  is less about punishing one engineer and more
• 3:35
  about just freezing OpenAI's hardware assembly
• 3:39
  lines before they even get built? Precisely.
• 3:41
  OpenAI has been heavily signaling to investors
• 3:44
  that they want to build custom chips, maybe consumer
• 3:46
  devices. Right. The hardware pivot. Exactly.
• 3:49
  Yeah. Apple is taking this specific case and
• 3:51
  using it as a lever to legally stall a major
• 3:54
  competitor's entire hardware division. It's a
• 3:57
  strategic maneuver to just hamstring arrival
• 3:59
  and discovery and injunctions for years. Wow.
• 4:03
  But it's not just. the hardware blueprints being
• 4:06
  fiercely guarded. The actual data being pumped
• 4:09
  into these models is causing like an even bigger
• 4:12
  legal earthquake on the software side. Oh, absolutely.
• 4:15
  The music industry. Yeah, the music industry.
• 4:17
  So we're seeing Sony Music and Warner Music suing
• 4:20
  Anthropic, basically accusing them of massive
• 4:23
  ongoing IP theft regarding the lyrics and the
• 4:26
  data used to train their models. Right. But looking
• 4:29
  through the source material, there's a highly
• 4:31
  specific detail that jumps out. The labels aren't
• 4:34
  just suing the corporate entity. The lawsuit
• 4:36
  directly names Mpropix CEO Dario Amodei. Yeah,
• 4:40
  they went right for the top. Why name the CEO
• 4:42
  directly instead of just going after the corporate
• 4:44
  treasury? What's the strategy there? Well, that
• 4:46
  is a very deliberate legal tactic known as piercing
• 4:49
  the corporate veil. OK. Normally, a corporation
• 4:53
  acts as a shield, right? If the company gets
• 4:56
  sued. the executives' personal bank accounts
• 4:59
  and their reputations are somewhat protected.
• 5:01
  Right. They're insulated. Exactly. By naming
• 5:04
  Dario Omode directly, the music labels are applying
• 5:07
  intense personal pressure on AI executives. They
• 5:11
  want the leadership to feel the heat for these
• 5:14
  data scraping practices personally. So it's a
• 5:16
  warning shot. It sends a chilling message to
• 5:18
  the entire industry. If you scrape our copyrighted
• 5:22
  data to train your billion -dollar models, we
• 5:24
  will literally come after your personal assets.
• 5:26
  Man. And the reason these companies are willing
• 5:29
  to fight tooth and nail in court and risk these
• 5:32
  massive liabilities is because the financial
• 5:33
  rewards are scaling at an almost terrifying speed.
• 5:37
  Wee money is just unreal right now. It is. Just
• 5:39
  looking at the sources, chat GPT ads recently
• 5:42
  crossed a $1 billion annualized run rate. And
• 5:46
  they hit that in less than 200 days after launching
• 5:48
  the platform. Which is a fundamentally disruptive
• 5:50
  metric. I mean, a billion dollars in under 200
• 5:53
  days. Tens of thousands of advertisers, right,
• 5:55
  across like 40 countries. Yeah. Advertisers are
• 5:57
  realizing that conversational interfaces capture
• 5:59
  consumer intent so much better than a traditional
• 6:02
  search engine page filled with, you know, blue
• 6:05
  links. Right. The budget is shifting at a phenomenal
• 6:08
  rate. And it's not just software revenue either.
• 6:11
  It's the physical infrastructure. NVIDIA just
• 6:14
  threw down a $3 .5 billion investment in MediaTek.
• 6:18
  Yep. So as big tech, you know, Google, Amazon,
• 6:22
  Microsoft, as they try to build their own custom
• 6:24
  AI chips to rely less on NVIDIA, MediaTek is
• 6:27
  stepping up to help manufacture those custom
• 6:30
  chips. Which normally you'd think is a huge threat
• 6:32
  to NVIDIA. Right. But here's where it gets really
• 6:34
  interesting. NVIDIA isn't just content selling
• 6:37
  the picks and shovels for the gold rush anymore.
• 6:40
  They are buying the railroads to ensure everyone
• 6:43
  still has to pay their toll to move the gold.
• 6:46
  That's the perfect analogy. Because they're demanding
• 6:48
  something specific in this deal, right? Yes,
• 6:50
  NVLink. To understand why NVLink is the ultimate
• 6:54
  railroad, you have to realize how AI models actually
• 6:57
  work inside a data center. Okay, lay it on me.
• 6:59
  A model like ChatGPT is way too massive to fit
• 7:03
  on a single chip. It requires thousands of chips
• 7:05
  acting together as one giant brain. So the bottleneck
• 7:08
  isn't usually how fast a single chip can think.
• 7:11
  It's how fast they can talk to each other. Exactly.
• 7:14
  NVLink is NVIDIA's ultra high -speed connection
• 7:16
  cable for that communication. So by investing
• 7:19
  $3 .5 billion into MediaTek and demanding that
• 7:22
  NVLink integration, NVIDIA is ensuring that even
• 7:26
  if a tech giant decides to use a competitor's
• 7:29
  chip, they still have to use NVIDIA's gauge of
• 7:31
  rail to connect them. Wow. They still have to
• 7:34
  plug it into the NVIDIA -controlled ecosystem.
• 7:36
  It protects their data center dominance, even
• 7:38
  as the industry desperately tries to pivot away
• 7:40
  from them. That is wild. And speaking of locking
• 7:43
  in dominance, the government sector is officially
• 7:45
  making its move. Oh, the Pentagon deployment.
• 7:48
  Yeah, the Pentagon just adopted specific, highly
• 7:50
  secure versions of ChatGPT and Grok built entirely
• 7:53
  for their civilian staff. And according to the
• 7:55
  sources, this secure AI portal already has over
• 7:58
  1 .7 million users deployed. But, and this is
• 8:02
  crucial, if you look closely at that deployment
• 8:04
  list from the sources, there is a glaring omission.
• 8:08
  Anthropic. Right. You have ChatGPT, you have
• 8:10
  Grok. But Claude, which is Anthropit's model,
• 8:13
  the same company currently fighting the music
• 8:15
  industry, is entirely missing from this secure
• 8:17
  portal. Missing out on a 1 .7 million user Pentagon
• 8:21
  contract feels like a pretty massive blow. It's
• 8:24
  incredibly damaging. Government contracts are
• 8:26
  uniquely sticky. Yeah, once you're in, you're
• 8:30
  in. Exactly. Once the Department of Defense...
• 8:33
  integrates a specific architecture, trains almost
• 8:35
  2 million people on it, and builds all their
• 8:38
  compliance protocols around that specific API,
• 8:41
  they almost never switch vendors. So Anthropic
• 8:44
  is just left out in the cold on one of the biggest,
• 8:47
  most recession -proof deployments out there.
• 8:50
  Exactly. Okay, so the infrastructure is locking
• 8:52
  in, the government contracts are locking in,
• 8:54
  billions of dollars are flowing. But while all
• 8:57
  that business maneuvering is happening, the actual
• 8:59
  capabilities of the AI are making quantum leaps.
• 9:02
  Right, moving way beyond just chatbots. Exactly.
• 9:05
  We're talking about predicting the future of
• 9:07
  complex enterprise systems. The sources highlight
• 9:09
  this massive breakthrough from Google research
• 9:11
  called Times FM3. This is a next -gen zero -shot
• 9:16
  AI for complex time series forecasting. From
• 9:20
  generating text to predicting time series data
• 9:22
  is profound. Reading the technical release notes,
• 9:25
  the specs are just wild. It features 330 million
• 9:28
  parameters and was pre -trained on over one trillion
• 9:31
  time points. A trillion data points. A trillion.
• 9:34
  But for those of us who aren't data scientists,
• 9:37
  when they call this a zero -shot AI, what does
• 9:41
  that actually mean in plain English? Zero -shot.
• 9:44
  Basically means the model can accurately predict
• 9:47
  outcomes for a specific industry or a specific
• 9:50
  data set that it was never explicitly trained
• 9:52
  on. Oh, wow. Yeah. It has seen so much data that
• 9:56
  it just intuitively understands the underlying
• 9:58
  math of how things fluctuate over time, whether
• 10:00
  that's grid energy usage or retail supply chains
• 10:03
  or server traffic. Okay. The sources also detail
• 10:06
  that it supports both univariate and multivariate
• 10:09
  forecasting. To make this concrete, let's use
• 10:12
  an analogy. Yeah. Because I want to make sure
• 10:13
  I'm getting this. Sure, go for it. Is it like,
• 10:14
  instead of trying to predict how much flour a
• 10:16
  bakery needs and then separately predicting how
• 10:18
  many eggs they need. This model looks at the
• 10:21
  upcoming weather, a holiday weekend, past sales,
• 10:24
  and predicts the entire recipe and the business
• 10:26
  outcome in one single breath. That is exactly
• 10:28
  it. And that one single breath part is actually
• 10:32
  the most important technical lead here. Really?
• 10:34
  Yeah. The technical release notes that Times
• 10:36
  FM 3 generates the entire forecast horizon in
• 10:39
  one forward pass. Older models operate step by
• 10:43
  step. They predict Tuesday, then use Tuesday
• 10:46
  to guess Wednesday, and so on. So if Tuesday
• 10:48
  is slightly wrong, Friday is a complete disaster.
• 10:51
  Exactly. The errors compound. By doing this in
• 10:54
  one forward pass, it looks at the entire context
• 10:57
  and outputs the whole future timeline simultaneously.
• 11:00
  The computational efficiency is a massive leap
• 11:03
  forward. And it doesn't just give you one blind
• 11:05
  guess, right? It outputs what they call nine
• 11:07
  probability quantiles. Right. Which gives risk
• 11:10
  managers a statistical range of likely outcomes.
• 11:13
  And it beats models like Kronos 2, Toto 2 .0.
• 11:16
  Yeah. It's available on GitHub right now. But,
• 11:19
  and this is a big... But there is a key limitation
• 11:21
  in the fine print. The license. Yeah. The pre
• 11:24
  -trained weights currently use a non -commercial
• 11:26
  license, which means businesses can't actually
• 11:29
  use it for commercial production workloads yet.
• 11:31
  Right. Google is basically treating the open
• 11:33
  source community as an unpaid R &D lab right
• 11:35
  now. Hey, find all the bugs for us. Exactly.
• 11:39
  They want researchers to stress test it and build
• 11:41
  integrations like the BigQuery integration mentioned
• 11:44
  in the sources. But they are keeping the commercialization
• 11:47
  leash very tight. They're showing us the future,
• 11:49
  but... They aren't quite letting us sell it yet.
• 11:51
  Not yet. But, you know, while Google is focused
• 11:54
  on these massive predictive models for the enterprise,
• 11:57
  the day -to -day reality for you and me is shifting
• 12:01
  toward autonomous agents doing the busy work.
• 12:03
  Right, the agentic workforce. Yeah. Let's fire
• 12:06
  through some of these tools from the sources
• 12:07
  because the rollout is so rapid. We've got Hermes
• 12:10
  agent version 0 .21 .0, which has bots mode and
• 12:14
  sub -agent steering. Which essentially means
• 12:16
  the AI can create... mini versions of itself
• 12:19
  to go do different tasks. Right. Then there's
• 12:22
  interactive sessions, which lets AI drive the
• 12:25
  full software lifecycle, coding, testing, deployment,
• 12:28
  all of it. Wild. Fodor Video Agent is turning
• 12:30
  scripts into editable motion graphics. Brandjet
• 12:33
  is taking live buying signals from like Reddit
• 12:36
  and X and turning them into qualified leads with
• 12:39
  automated outreach. It's a total workflow takeover.
• 12:42
  And even Gemini Enterprise is getting rooms,
• 12:44
  which is this shared workspace where teams define
• 12:47
  goals. and the AI keeps the work focused. But...
• 12:51
  Out of all of these, the one that really stopped
• 12:53
  me in my tracks was GrokBot. Oh, the purchasing
• 12:56
  feature? Yes. GrokBot now lets you build custom
• 12:59
  agent workflows, save them, share them. And crucially,
• 13:02
  it can make purchases after getting your approval.
• 13:05
  Yeah. I got to push back on this on behalf of
• 13:08
  everyone listening. I want to save time. I really
• 13:09
  do. But giving a bot access to make purchases
• 13:13
  even with approval feels like a massive leap
• 13:16
  of faith. Are we actually ready to give AI the
• 13:19
  company credit card? If we connect this to the
• 13:22
  bigger picture, you are touching on the massive
• 13:25
  psychological shift we are going through. It
• 13:27
  feels risky. It does. Because we are shifting
• 13:29
  from AI as a chat interface to AI as a digital
• 13:32
  employee. When it's just text, you still have
• 13:35
  the agency. When the bot has the credit card,
• 13:37
  the psychological barrier is huge. So how are
• 13:39
  they getting people to adopt this? Well, the
• 13:41
  shift is actually fueling an entire cottage industry
• 13:44
  of upskilling. Just look at the sources. AI Fire
• 13:49
  and Belay are heavily promoting this passive
• 13:52
  income AI course. Oh, right. Mia's full system.
• 13:55
  Exactly. They promised to teach users this system
• 13:58
  with no tech skills required. They had to package
• 14:02
  these autonomous agents into like zero to profit
• 14:05
  certification courses just to make the idea of
• 14:08
  a bot with a credit card feel manageable to the
• 14:11
  average person. They have to make it sound like
• 14:12
  a get rich quick tool so you ignore the risk.
• 14:15
  Basically, yeah. Well, as we deploy these digital
• 14:17
  employees everywhere, we are starting to feel
• 14:19
  the friction, like literally feeling the bloat
• 14:22
  in our daily digital lives. Of the true cost
• 14:24
  of convenience. Yes. Here's a nugget from the
• 14:26
  sources that is so relatable. If your PC feels
• 14:29
  slower lately, AI features might literally be
• 14:32
  the problem. Those background tasks. Yeah. the
• 14:34
  background tasks, the neural processing units.
• 14:37
  The sources note that a few of these settings
• 14:39
  are actually easy to disable. And if you're trying
• 14:41
  to play a video game, your computer will thank
• 14:43
  you. We're dragging down our hardware for features
• 14:45
  we didn't even ask for. And it's not just our
• 14:47
  hardware getting bogged down. It's our writing.
• 14:50
  The sources highlight this new tool called the
• 14:52
  LLM Cliche Highlighter. I love this tool. It
• 14:55
  is so funny. It's designed to spot up to 38 common
• 14:59
  AI writing cliches just to clean up. obvious
• 15:03
  elemisms like, you know, delve or in today's
• 15:06
  fast paced digital landscape. It's exhausting
• 15:09
  reading that stuff. It is. So what does this
• 15:11
  all mean? Are we just entering a bizarre loop
• 15:13
  where we build AI tools to spot the lazy work
• 15:17
  of other AI tools, all while our personal computers
• 15:20
  slow to a crawl? This raises an important question,
• 15:22
  really. I mean, if our daily text is full of
• 15:24
  38 predictable cliches and our hardware is lagging
• 15:27
  out, the AI revolution is currently coming with
• 15:30
  a very noticeable tax. on authenticity and computing
• 15:33
  performance. It's a huge tax. Well, let's quickly
• 15:36
  recap this journey we've been on today. We started
• 15:38
  with corporate espionage Apple suing OpenAI over
• 15:41
  microscopic blueprints and the music industry
• 15:44
  going directly after Anthropic's CEO. The stakes
• 15:46
  are incredibly high. Then we looked at the billion
• 15:49
  -dollar moats, ChatGPT's massive ad revenue,
• 15:52
  NVIDIA forcing NVLink on MediaTek, and the Pentagon
• 15:56
  locking in their preferred AI vendors. We broke
• 15:59
  down Google's Times F - predicting complex futures
• 16:03
  in one forward pass. And finally, we looked at
• 16:05
  autonomous agents taking our credit cards and
• 16:08
  the cliches slowing down our PCs. It's a lot
• 16:10
  of friction for one ecosystem. It is. And I want
• 16:13
  to leave you with one final provocative thought
• 16:15
  to mull over. Earlier, we talked about OpenAI's
• 16:18
  defense against Apple. They blamed Apple's poor
• 16:20
  off -boarding security for the data breach. Right,
• 16:23
  the unlocked vault. But now we are giving autonomous
• 16:26
  AI bots like Grok the ability to make purchases
• 16:28
  and using agents to drive entire software deployments.
• 16:32
  If the very creators of this technology point
• 16:34
  the finger at poor security practices when things
• 16:37
  go wrong internally, what happens when an autonomous
• 16:40
  agent with your credit card experiences a security
• 16:42
  failure in your personal workflow? That's a scary
• 16:45
  thought. Who takes the blame then? The bot. the
• 16:48
  developer, or you for leaving the vault door
• 16:51
  unlocked. Thank you so much for joining us on
• 16:53
  this deep dive. Keep exploring, keep questioning
• 16:55
  the tools you use, and we'll catch you next time.