Latest / Connor with Honor: AI, Real Estate and Real Life / The AI Industry Just Got NEW RULES: And One of Them Should Terrify Every Parent

Transcript

The AI Industry Just Got NEW RULES: And One of Them Should Terrify Every Parent · Connor with Honor: AI, Real Estate and Real Life

• 0:00
  SPEAKER_00: You ever walk into a room and realize everybody knows something that you don't?
• 0:04
  SPEAKER_00: That's what it felt like watching Gents and Wong take the stage yesterday at GTC in San Jose.
• 0:10
  SPEAKER_00: But while most people are sleeping, the entire AI industry just got a new set of rules.
• 0:14
  SPEAKER_00: A trillion dollars in chip orders.
• 0:16
  SPEAKER_00: Robotaxis coming to your city next year.
• 0:19
  SPEAKER_00: An open source model that just made half the competition irrelevant.
• 0:23
  SPEAKER_00: And a lawsuit against one of the biggest names in tech that should make every parent in America stop what they're doing and pay attention.
• 0:30
  SPEAKER_00: Welcome to AI with Honor, the daily download for Tuesday, March 17, 2026.
• 0:35
  SPEAKER_00: I'm coming to you from Santa Clarita, California, and today we definitely have a loaded show.
• 0:40
  SPEAKER_00: Let's get into it.
• 0:41
  SPEAKER_00: So yesterday, NVIDIA kicked off GTC 2026.
• 0:44
  SPEAKER_00: And if you don't know what GTC is, think of it as the Super Bowl for artificial intelligence.
• 0:49
  SPEAKER_00: 20,000 people packed into the SAP Center in San Jose, otherwise known as the SAP Center.
• 0:56
  SPEAKER_00: Sorry, couldn't resist.
• 0:58
  SPEAKER_00: 450 sponsors, a thousand sessions, 2,000 speakers, and the man in the lather jacket, Jensen Huang.
• 1:06
  SPEAKER_00: He walked out and said something that stopped the room cold.
• 1:09
  SPEAKER_00: He said NVIDIA inspects at least$1 trillion in revenue from its newest AI chips through 2027.
• 1:16
  SPEAKER_00: Let me say that again because the number deserves a second take.
• 1:20
  SPEAKER_00: $1 trillion.
• 1:23
  SPEAKER_00: That's$1,000 billion or$10,000 million.
• 1:27
  SPEAKER_00: Anyway, I'm not going to get into it.
• 1:29
  SPEAKER_00: Last year at the same event, the number was$500 billion.
• 1:32
  SPEAKER_00: It doubled in 12 months.
• 1:34
  SPEAKER_00: Not because of hype, not because of promises, because the purchase orders are already on the table.
• 1:39
  SPEAKER_00: The chips driving this and call are called Blackwell and VeraRubin.
• 1:44
  SPEAKER_00: And every major tech company on the planet is standing in line to get them.
• 1:48
  SPEAKER_00: Microsoft, Meta, Google, Amazon, all of them.
• 1:51
  SPEAKER_00: And NVIDIA just opened the 20th anniversary of CUDA, the software framework that powers everything they do.
• 1:58
  SPEAKER_00: Jensen made a point about that a lot.
• 2:01
  SPEAKER_00: He said the price of their older Ampere chips actually had gone up in the cloud.
• 2:06
  SPEAKER_00: Not down, but up.
• 2:08
  SPEAKER_00: Because demand for AI compute is so high that even the old stuff is getting more expensive.
• 2:12
  SPEAKER_00: That tells you something about where this market is headed.
• 2:15
  SPEAKER_00: Now, here's where it gets really interesting.
• 2:18
  SPEAKER_00: Jensen announced something that almost nobody saw coming.
• 2:21
  SPEAKER_00: A new chip called Grok 3, G-R-O-Q, Language Processing Unit.
• 2:27
  SPEAKER_00: NVIDIA acquired the startup Grok late last year for$20 billion.
• 2:32
  SPEAKER_00: Biggest deal they've ever done.
• 2:34
  SPEAKER_00: Grok was founded by the same people who created Google's Tensor Processing Unit.
• 2:38
  SPEAKER_00: So NVIDIA didn't just buy a company, they bought the team that built Google's original AI chip.
• 2:44
  SPEAKER_00: And now, taking that technology and they built a chip specifically designed to make AI faster at the point of response.
• 2:52
  SPEAKER_00: This is not about training anymore.
• 2:54
  SPEAKER_00: This is about inference.
• 2:56
  SPEAKER_00: Inference is the moment where AI actually talks back to you.
• 3:00
  SPEAKER_00: When you ask Claude a question and it starts typing.
• 3:02
  SPEAKER_00: When a voice agent picks up the phone and it starts speaking, that's inference.
• 3:07
  SPEAKER_00: And they want that moment to be instantaneous.
• 3:10
  SPEAKER_00: The GROC 3 LPU is expected to ship in quarter three of this year.
• 3:14
  SPEAKER_00: A full rack, they're calling it.
• 3:16
  SPEAKER_00: The LPX will hold 256 of these processors.
• 3:20
  SPEAKER_00: And it's designed to sit right next to the Vera Rubin system.
• 3:24
  SPEAKER_00: Think of it like this.
• 3:26
  SPEAKER_00: You got the brain, now you got the mouth.
• 3:28
  SPEAKER_00: The brain does the deep thinking, the mouth delivers the answer.
• 3:32
  SPEAKER_00: Like so many people we know, they lack the deep thinking part.
• 3:35
  SPEAKER_00: And NVIDIA just made both of them dramatically faster.
• 3:38
  SPEAKER_00: Jensen also indicated the updated versions of the LPX already on the roadmap for future years.
• 3:44
  SPEAKER_00: This isn't a one-off product, this is a new category they're creating.
• 3:48
  SPEAKER_00: But Jinsen still wasn't done.
• 3:50
  SPEAKER_00: He introduced something called NemoClaw.
• 3:52
  SPEAKER_00: It's a developer toolkit that lets you build and run AI agents with a single command.
• 3:57
  SPEAKER_00: You tell it what you want, it finds the right open source model, it downloads it, builds you an AI agent, one command.
• 4:04
  SPEAKER_00: The kind of thing that used to take a development team two weeks of work a solo operator can do in five minutes.
• 4:11
  SPEAKER_00: If you're building anything with AI right now, that should grab you by the collar and pull you in because the barrier to entry just dropped to almost zero.
• 4:20
  SPEAKER_00: And the people who figure out how to use these tools first are the ones who are going to eat.
• 4:25
  SPEAKER_00: Now let's talk about what might be the biggest physical AI announcement all year.
• 4:29
  SPEAKER_00: Uber and NVIDIA announced a partnership to launch fully autonomous robotaxis in 28 cities across four continents by 2028, starting right here.
• 4:39
  SPEAKER_00: Los Angeles and San Francisco.
• 4:41
  SPEAKER_00: The first half of 2027, that's next year.
• 4:46
  SPEAKER_00: Not someday, not eventually next year.
• 4:49
  SPEAKER_00: The technology running these vehicles is called Alpamayo.
• 4:56
  SPEAKER_00: It's a reasoning-based AI model built specifically for autonomous driving.
• 5:01
  SPEAKER_00: And it doesn't just see the road, it thinks about the road, it ponders the road, it uses chain of thought logic to reason through unpredictable situations.
• 5:11
  SPEAKER_00: Construction zones, a kid chasing a ball into the street, a motor cop pulling somebody over, a delivery driver making, I added the motor cop, a delivery making, a delivery driver making a U-turn where they shouldn't be.
• 5:24
  SPEAKER_00: Al Pomeo processes all of that in real time and makes a decision.
• 5:28
  SPEAKER_00: Not a programmed response, a reasoned decision.
• 5:31
  SPEAKER_00: It's like the difference between a rookie who freezes and a veteran who reads the scene.
• 5:36
  SPEAKER_00: That analogy isn't random.
• 5:38
  SPEAKER_00: If you've been on the street, you know exactly what I mean.
• 5:40
  SPEAKER_00: Jensen said the chat GPT moment for physical AI has arrived.
• 5:46
  SPEAKER_00: Here we go.
• 5:47
  SPEAKER_00: And when you when you've got Uber's global network plus Nvidia's full software stack, plus BYD, Nissan, Gili, Izuzu, and Hyundai all building level four vehicles on the same platform, it's not a concept card and an auto show.
• 6:05
  SPEAKER_00: That's a fleet.
• 6:06
  SPEAKER_00: Each city will get a phased rollout.
• 6:10
  SPEAKER_00: First, data collection vehicles to learn the roads, then supervised operations, then fully driverless.
• 6:16
  SPEAKER_00: Waymo still leads the pack right now with services in Phoenix, San Francisco, and Los Angeles.
• 6:22
  SPEAKER_00: Tesla is pushing its camera-based approach.
• 6:25
  SPEAKER_00: But this Uber NVIDIA play just changed the entire competitive landscape because NVIDIA isn't building cars.
• 6:31
  SPEAKER_00: They're building the operating system that every car will run.
• 6:35
  SPEAKER_00: Now let's shift to something that the open source community is absolutely buzzing about.
• 6:39
  SPEAKER_00: Mistral.
• 6:40
  SPEAKER_00: The French AI company dropped a brand new model yesterday.
• 6:44
  SPEAKER_00: Mistral Small4.
• 6:46
  SPEAKER_00: Don't let the word small fool you.
• 6:48
  SPEAKER_00: This thing's a beast.
• 6:50
  SPEAKER_00: 119 billion total parameters.
• 6:52
  SPEAKER_00: But here's the trick.
• 6:54
  SPEAKER_00: It only activates 6 billions of billion of them at any given moment.
• 6:58
  SPEAKER_00: It uses something called mixture of experts architecture.
• 7:02
  SPEAKER_00: Sounds like a DD game.
• 7:04
  SPEAKER_00: 128 experts built into the model, but only four of them fire on any given token depending on what you're asking.
• 7:11
  SPEAKER_00: It's like having 128 specialists in a hospital.
• 7:14
  SPEAKER_00: But the triage nurse only sends you to the four you actually need.
• 7:19
  SPEAKER_00: Nobody else gets involved, nobody else bills you.
• 7:21
  SPEAKER_00: That's why it's fast.
• 7:22
  SPEAKER_00: 40% faster than the previous version.
• 7:25
  SPEAKER_00: Triple the throughput.
• 7:27
  SPEAKER_00: And it supports a 256,000 token context window.
• 7:31
  SPEAKER_00: That means you can feed an entire code base, a complete manuscript, and it holds all of the memory all at once while reasoning through your question.
• 7:40
  SPEAKER_00: But here's the real game changer: configurable reasoning.
• 7:44
  SPEAKER_00: You can set a parameter called reasoning effort.
• 7:47
  SPEAKER_00: Set it to none and you get fast, direct, snapped answers.
• 7:51
  SPEAKER_00: Set it to high and the model slows down.
• 7:54
  SPEAKER_00: Thinks step by step and gives you deep analysis like a research assistant sitting across the table.
• 7:59
  SPEAKER_00: One model, one deployment, one API call, one toggle between speed and depth.
• 8:04
  SPEAKER_00: You used to treat a fast model and a smart model running side by side.
• 8:09
  SPEAKER_00: That's what you used to need.
• 8:11
  SPEAKER_00: Route some questions here, route others there.
• 8:14
  SPEAKER_00: That's over.
• 8:16
  SPEAKER_00: One model now is going to do both, and it's open source under Apache 2.0.
• 8:22
  SPEAKER_00: That means anyone, any company, any developer can use it, modify it, deploy it commercially for free.
• 8:28
  SPEAKER_00: On Live Codebench, it beat GPT-OSS 120B while producing 20% less output.
• 8:36
  SPEAKER_00: Less output at equal quality means lower inference cost for everyone.
• 8:41
  SPEAKER_00: They have discussed this as AGI or ASI only being a few good decisions away, and this is probably how they're going to get to that bar.
• 8:51
  SPEAKER_00: The implications for small businesses are enormous.
• 8:54
  SPEAKER_00: If you're running a voice agent, if you're running a chat bot, if you're paying per token for APA calls to a closed model, this could slash your costs overnight.
• 9:03
  SPEAKER_00: Any Mistral and Mistral didn't stop there.
• 9:07
  SPEAKER_00: They announced a strategic partnership with NVIDIA.
• 9:10
  SPEAKER_00: They're co-funding something called the Neomotron Coralition, a group of eight AI labs building frontier open source models together.
• 9:19
  SPEAKER_00: Mistral brings the architecture and training techniques.
• 9:22
  SPEAKER_00: Leave it to the French.
• 9:24
  SPEAKER_00: Nvidia brings the DGX cloud compute that would otherwise cost tens of millions.
• 9:29
  SPEAKER_00: Their first joint project is a base model that will underpin the upcoming Nemotron 4 family.
• 9:35
  SPEAKER_00: That is a direct shot at OpenAI and at Google and at every closed model provider that charges you a premium for access.
• 9:43
  SPEAKER_00: And then today, Mistral even went further.
• 9:46
  SPEAKER_00: They announced Mistral Forge, a new platform that lets enterprises build custom AI models trained on their own data.
• 9:54
  SPEAKER_00: Their CEO, Arthur Mensch, says the company is on track to surpass$1 billion in annual recurring revenue this year from a French startup that didn't exist three years ago.
• 10:05
  SPEAKER_00: That's what happens when you combine real engineering talent with the right business model.
• 10:10
  SPEAKER_00: That's the power of open source done right.
• 10:12
  SPEAKER_00: Now I need to talk about something that matters to every parent, every grandparent, every coach, every teacher that watches my show.
• 10:19
  SPEAKER_00: Three teenagers in Tennessee just got filled a class, filed a class action lawsuit against Elon Musk's XAI, the company that makes Grok G R O K.
• 10:30
  SPEAKER_00: These are real girls.
• 10:31
  SPEAKER_00: Two of them are still minors.
• 10:33
  SPEAKER_00: They're alleging that XAI's image generation tools were used to create sexually explicit images and videos of them using their real photos.
• 10:42
  SPEAKER_00: School pictures, family photos, yearbook images turned into something no child should ever have to see themselves depicted within.
• 10:51
  SPEAKER_00: The images were discovered on Discord and Telegram.
• 10:54
  SPEAKER_00: They were being shared and traded like currency among predators.
• 10:58
  SPEAKER_00: One of the girls found out she got an anonymous Instagram message linking to altered photos and videos of herself.
• 11:04
  SPEAKER_00: Her high school yearbook picture manipulated into explicit content.
• 11:09
  SPEAKER_00: Let that set for a second.
• 11:11
  SPEAKER_00: Now the perpetrator didn't use Grok directly.
• 11:15
  SPEAKER_00: They used a third-party app that ran on Grok's AI algorithm.
• 11:19
  SPEAKER_00: But the lawsuit argues that XAI knowingly licensed technology to these app makers, many of them overseas, specifically to outsource the liability.
• 11:29
  SPEAKER_00: The Center for Countering Digital Hate estimated that in just 11 days, Grok generated approximately 3 million sexualized images.
• 11:37
  SPEAKER_00: 23,000 of those appeared to depict children.
• 11:40
  SPEAKER_00: That's roughly 190 images per minute.
• 11:43
  SPEAKER_00: Google and OpenAI watermark their AI-generated images so you can trace them to where they to whence where they came from.
• 11:51
  SPEAKER_00: XAI has not adopted that standard.
• 11:53
  SPEAKER_00: California's attorney general has already opened an investigation.
• 11:57
  SPEAKER_00: Multiple lawsuits are piling up, and the plaintiffs in this case aren't just asking for money.
• 12:02
  SPEAKER_00: They want to make their business decision to allow this kind, they want to make the business decision to allow this kind of contact so financially devastating that no company would ever consider it again.
• 12:14
  SPEAKER_00: And I'll say this, I spent over 20 years as a cop.
• 12:18
  SPEAKER_00: I started 18 years old working inside a county jail in Roosevelt County, New Mexico.
• 12:24
  SPEAKER_00: Not patrol, not a desk, inside the cages.
• 12:27
  SPEAKER_00: And at 18, you learn a lot by people that are much older and much more dangerous.
• 12:32
  SPEAKER_00: Where you learn very fast what happens when power is given to people without accountability.
• 12:38
  SPEAKER_00: And what I see happening with these tools where there are no guardrails, where there's no watermark, and where there's no audit trail, it looks familiar.
• 12:46
  SPEAKER_00: It's the same pattern.
• 12:47
  SPEAKER_00: Pattern without restraint always ends the same way.
• 12:51
  SPEAKER_00: The badge doesn't make you look good.
• 12:53
  SPEAKER_00: The restraint does.
• 12:54
  SPEAKER_00: The same thing applies to AI.
• 12:56
  SPEAKER_00: The capability doesn't make you valuable.
• 12:59
  SPEAKER_00: The discipline to use it responsibly does.
• 13:02
  SPEAKER_00: If you have kids or if you have grandkids, if you care about anybody under 18, this is the AI story you should be paying the most attention to today.
• 13:10
  SPEAKER_00: Not the chips, not the models, not the trillions of dollars.
• 13:13
  SPEAKER_00: This one.
• 13:14
  SPEAKER_00: Let me connect a couple more dots before I pull it all together.
• 13:17
  SPEAKER_00: IBM just completed its acquisition of Confluent.
• 13:21
  SPEAKER_00: If you don't know Confluent, that's the company behind real-time data streaming for enterprises.
• 13:27
  SPEAKER_00: What this means is that IBM is making real-time data the engine of its entire enterprise AI and agent strategy.
• 13:34
  SPEAKER_00: When you hear companies talk about AI agents being the future, they need data flowing in real-time to work, not batched, not delayed, but live.
• 13:43
  SPEAKER_00: That's what acquisition is about.
• 13:45
  SPEAKER_00: IBM is betting that the companies who win the next decade will be the ones whose AI agents can react to information as it happens, not after the fact.
• 13:54
  SPEAKER_00: Also worth noting, Nebus Group, a cloud infrastructure provider focused on AI, announced a$3.775 billion capital raise through convertible notes and loans.
• 14:08
  SPEAKER_00: Their stock dropped 10% on the news because of dilution fears.
• 14:13
  SPEAKER_00: But here's the context.
• 14:14
  SPEAKER_00: The day before, they announced a massive new AI infrastructure deal with Meta.
• 14:19
  SPEAKER_00: Demand for cloud computing capacity is so strong that companies like Nebias are raising billions just to build fast enough.
• 14:26
  SPEAKER_00: The infrastructure layer of AI is becoming its own gold rush.
• 14:30
  SPEAKER_00: The model makers get the headlines.
• 14:32
  SPEAKER_00: The infrastructure builders get the revenue.
• 14:34
  SPEAKER_00: Now let me pull everything together and tell you what I think this all means for us.
• 14:38
  SPEAKER_00: We're at an inflection point.
• 14:40
  SPEAKER_00: Jensen Wong sees a trillion dollar infrastructure build out.
• 14:43
  SPEAKER_00: Uber sees autonomous vehicles in 28 cities.
• 14:46
  SPEAKER_00: Mistral open source models that match or beat the biggest closed systems on Earth.
• 14:51
  SPEAKER_00: IBM sees real-time data agents' enterprise future.
• 14:55
  SPEAKER_00: And all of it is happening right now.
• 14:57
  SPEAKER_00: Not in five years, not in a decade, this year and next, 2026, 2027.
• 15:03
  SPEAKER_00: The question isn't whether AI is going to change things.
• 15:06
  SPEAKER_00: The ship sailed a long time ago.
• 15:08
  SPEAKER_00: The question is whether you're going to be someone who uses it or someone it happens to.
• 15:12
  SPEAKER_00: If you're a small business owner still answering every phone call yourself, there's a voice agent that can handle it 24 hours a day, seven days a week for less than a cost of one part-time employee.
• 15:22
  SPEAKER_00: You take that energy that you use doing that task that you can pass along and you build something else and something better.
• 15:28
  SPEAKER_00: If you're a real estate agent still hoping your next client finds you on Zillow, there's a system that you can put at the top of Google Maps in your city every single day that will do it for you so you can go build something better and get more clients.
• 15:41
  SPEAKER_00: If you're a content creator still spending four hours editing 10 minutes of video, there are tools available right now that cut your pauses, lock your eye contact off the lens, and make you look like you got a full production team behind you.
• 15:53
  SPEAKER_00: None of this requires a computer science degree.
• 15:55
  SPEAKER_00: It requires a decision, a decision to start.
• 15:58
  SPEAKER_00: And that's what this show is about.
• 16:00
  SPEAKER_00: Every single day, democratizing this technology, making it accessible, making it make sense for real people with real businesses and real lives.
• 16:09
  SPEAKER_00: Because the gap between those who use AI and those who don't isn't going to shrink.
• 16:14
  SPEAKER_00: It's going to become a canyon.
• 16:16
  SPEAKER_00: And I would rather see you standing on this side of it.