Oran Barak

Today I spoke with two frinds from Google AI and learned that the KPI for engineers at Google and many large tech companies has fundamentally changed. A key metric now is how many tokens you are able to use. The more tokens you use, the greater your potential contribution—one person can now accomplish the work that once required ten or even a hundred people. #HerPowerAI #Google #WorkModel #KPI

1

To all the engineers in my network who might or might not know what test-driven development is: The California government’s approach to innovation is like test-driven development. In some cases it actually *IS* test-driven development. I just found out that California’s failed SB 1047 would have “required developers to perform risk assessments of models prior to release, with guidance from the Government Operations Agency.” (wikipedia) There’s one place where the analogy to test-driven development falls short, though. Whereas in TDD you know what you’re testing for, SB 1047 doesn’t even know what it’s trying to regulate, as its goal is to “mitigate the risk of catastrophic harms from AI models so advanced that they are not yet known to exist.” Unfortunately, the state senator who introduced SB 1047 is back at it and still doesn’t realize he is trying to put the cart before the horse. Source: Politico

2

On-device AI becomes powerful only when it works reliably on real hardware. In this Melange Walkthrough, Yeonseok Kim and Sun-Q Kim break down how the full on-device pipeline works, from model selection and quantization to real-device benchmarking and SDK deployment. Select. Benchmark. Deploy. This is how production-ready on-device AI should operate. Watch the full walkthrough below. #OnDeviceAI #EdgeAI #MobileAI #AIInfrastructure #ZETIC #Melange

26

Amazon's Trainium chips are getting all the headlines, but here's what they don't tell you about actually running production workloads on them. I've deployed AI infrastructure on every major chip platform over the past two years. While AWS tours focus on theoretical performance, we've learned the hard truths about Trainium in production. The chip delivers on promise for specific workloads - training large language models sees 40-60% cost savings versus A100s. But the ecosystem is still maturing. Driver stability issues hit us on 3 different client deployments. Memory bandwidth becomes the bottleneck faster than expected on mixed workloads. Most teams jump to Trainium because of cost projections without testing their specific models first. Run a pilot with your actual training pipeline before committing infrastructure spend. What's been your experience with custom AI chips versus traditional GPU setups? Worth the complexity trade-offs? #AIInfrastructure #AWSTrainium #MLOps #CloudComputing #DevOps

1

🚀 Can Waymo become a $1T company by 2035? Waymo’s recent $16 billion investment round at a $126 billion valuation suggests the market thinks so. But to reach a $1T valuation at a 6–8× revenue multiple (Apple/Google territory), Waymo would need ~$150B in annual revenue. I stress-tested this in my Autonomy Scale ROI model. Here’s what the math says: 1️⃣ The “Uber Plus” Trap 🛑 Uber reported ~211M trips/week in 2023. For Waymo to hit $150B, they need to facilitate 500M trips/week. The Insight: You can’t get there just by stealing ride-hailing market share. Waymo must convert a significant % of the 10B weekly trips currently taken in private vehicles. This requires a shift from "luxury service" to "utility infrastructure." 2️⃣ The Operational "Lights-Out" Lever 🛠️ Supporting 500M trips/week requires ~1.3M vehicles (~400× today). At that scale, the binding constraint isn’t hardware—it’s labor per vehicle-hour. Today, Waymo operates at roughly 42 vehicles per remote operator. To reach $1T territory, we need a “lights-out” depot model where total labor cost (remote assist + depot ops) drops toward ~$0.50 per vehicle operating hour. Why this matters: At 50% utilization, that’s ~$0.15 labor cost per productive mile. Once labor is decoupled from mileage, density effects compound faster than incremental hardware savings. 3️⃣ Destination Commerce: Beyond the Fare 💸 This is the biggest unlock. If Waymo only charges for the miles, it’s a commodity. The Opportunity: The real $1T play is becoming the platform that links transportation to commerce. Think: enterprise mobility contracts, bundled subscriptions, or "sponsored trips" where the destination (a retailer or restaurant) subsidizes the fare. The trip economics must extend beyond the fare itself. Airports monetize retail. Airlines monetize loyalty. Amazon monetizes Prime beyond shipping. Mobility will likely follow the same pattern. The Verdict: $150B revenue is mathematically possible, but only if Waymo stops being a "car company" and starts being a "spatial commerce platform." Try the simulation yourself: I’ve updated the Autonomy Scale ROI tool to allow for these "Hyper-Scale" scenarios, including inputs for long-term labor costs and platform-based revenue. 👉 https://lnkd.in/gYQDWprB Is $1/mile margin sustainable at million-vehicle scale? Or does competition compress it to pennies? Curious to hear operator views. #AutonomousVehicles #Robotaxi #UnitEconomics #Waymo #Tesla #Uber #Wayve #PhysicalAI #Mobility #Robotaxi #ProductManagement #ScalingAI

27

10 Comments

I'm very excited to share the news that today, Linaro, supported by Arm, is launching CoreCollective marking the next phase of all-inclusive collaboration in the Arm ecosystem. Check out https://lnkd.in/enz4rFZn for more details.

20

"Intel is shuttering its automotive architecture business and laying off most of its staff as part of a broader restructuring at the chipmaker." https://lnkd.in/erw7EaEK

7

1 Comment