To make this actually testable in the future, here’s my bar for affordable Level 5: • ≤$30k to buy (cheap new car), or <$0.65/occupied mile as a robotaxi • No geofence, remote operator, or special infrastructure • No dependence on GPS or centimeter-accurate HD maps — a normal map + onboard perception should be enough • Drives anywhere on Earth that a competent human can reasonably drive • Handles weather, poor markings, construction, closures, crowded streets, events, pedestrians, cyclists, hand signals, police directing traffic, and informal negotiation with humans • Handles novel situations and bad/outdated maps • Completes the whole trip: pickup/dropoff, rerouting, pulling over, parking, and recovering when things go wrong • Never needs a human to take over That’s the bar I think is still 5-10 years away.

To make this actually testable in the future, here’s my bar for affordable Level 5: • ≤$30k to buy (cheap new car), or <$0.65/occupied mile as a robotaxi • No geofence, remote operator, or special infrastructure • No dependence on GPS or centimeter-accurate HD maps — a normal map + onboard perception should be enough • Drives anywhere on Earth that a competent human can reasonably drive • Handles weather, poor markings, construction, closures, crowded streets, events, pedestrians, cyclists, hand signals, police directing traffic, and informal negotiation with humans • Handles novel situations and bad/outdated maps • Completes the whole trip: pickup/dropoff, rerouting, pulling over, parking, and recovering when things go wrong • Never needs a human to take over That’s the bar I think is still 5-10 years away.

Here are some things which each use as much water as a data centre: A medium-sized office building A golf course A hotel A leisure centre A swimming pool A large restaurant A secondary school A school A small hospital A shopping centre A supermarket A commercial laundry A large bakery A brewery A dairy farm A conference centre A spa A football stadium An ice rink A large theatre A museum A prison A garden centre A busy car wash

Affordable Level 5 self-driving requires AGI and is still at least 5-10 years away. First you need multimodal multiplexed AGI, then you need real-time inference, then on an energy-efficient, low-cost edge chip. Many steps left.

Agree. Future of transportation is personalized. People may romanticize busses, but will pick a private autonomous driver every time if price is right. Still need to build more tunnels in cities though!

after a century of uninterrupted growth, for 15 years, between 2005 and 2020, America did not grow electrical generation. we completely forgot what it meant to grow industrial capacity. politicians constantly talked about "reindustrializing" and "bringing back high-paying blue collar jobs". then AI actually did it. for the first time in 20 years we started reindustrializing America's heartland. creating a massive boom of skilled, high-paying blue collar labor. an enormous construction boom. a new industry emerged to finance the expansion of our aging grid. a grid that needed to grow anyway, if we were going to become a modern, highly electrified economy. and all of a sudden the tech companies and hyperscalers showed up with wads of cash, completely willing to pay for a massive grid expansion. and all the politicians that talked so tough about "bringing back industry and blue collar jobs" panicked when the rubber hit the road. because the NIMBYists had become used to stagnation and freaked out when it came time to build a couple transmission lines. all that tough talk down the drain, because some boomers have AI anxiety and think a warehouse full of computers is unsightly next to a mattress warehouse and a vape shop. probably the most pathetic political episode I've ever witnessed. the only question that really matters is: is it still possible to build anything at all in this country? or have we completely lost the political will?

In my ~14 months at @OpenAI, one of the most surprising and genuinely wonderful things about it, has been the full leadership support for full-duplex models (gpt-live series), even when it seemed deeply improbable that they would work. And yet, if they did work, it was obvious how magical they could be. I am quite sure this is not an exception but a general rule for research projects at @OpenAI. There were so many moments when the problem felt impossibly hard. But one thing kept being true: it was never clear why it shouldn’t work. And almost every time we understood the problem a little better, it became a little easier to solve. I’m pretty sure there are very few environments in the world where a bet like this could have been made and sustained. So grateful to be part of this wonderful place.

Codex is my personal trainer, programming my @tonal workouts. Asked Codex to install the Android emulator, install the Tonal app, and create a new weekly program with custom lifts.