Elon’s Flying Car + Harley Finkelstein on Raising Entrepreneurial Kids

Amber Mac (0:00)

Pin verification is a straightforward Uber safety feature that helps make sure you get into the right vehicle. It's easy to make a mistake and get into the wrong car, especially when it's late at night or you're in a crowded spot and all the vehicles look similar. Pin verification adds an extra step to help confirm you're getting into the vehicle matched to your trip. It's easy to enable and customize pin verification through safety preferences. In the Uber app, go to account, safety, safety preferences. From there, you can select pin verification and select verify rides with pin, so all your rides require pin verification. If you prefer to use it only at night, you can toggle to night rides only. Once you've been matched with a ride, you'll see a pin in the app. When you arrive, your driver will ask for your pin. The driver enters your pin into the driver app. The app will only allow the trip to begin if the pin matches. For more information about pin verification, you can visit Uber online. And thank you so much to Uber for supporting The Amber Mac Show.

SPEAKER_2 (0:59)

The Amber Mac Show on Canada Talks.

Amber Mac (1:02)

This week on The Amber Mac Show, what one tech company thinks about the future of robots and the work humans do, and what happens when AI agents break the law.

Jeff MacArthur (1:10)

And our guest for this episode is Shopify's Harley Finkelstein, who joins us to talk about inspiring the next generation of entrepreneurs. Plus, I'm going to do a deep dive into Elon Musk's flying car on The Other Mac. But first, let's dive into the latest tech headlines.

AI company Anthropic recently released a study entitled What Work Can Robots Do? in which they outline how AI models could replace people for about half of all the work done in the USA today. And when you add robots combined with AI, that number is about 81%. But an important caveat to that, one that actual people may take some comfort in, is that robots are less expensive than just employing real humans for only about 0.3%. So I think that's good news, at least.

Amber Mac (1:59)

That is definitely good news. What's also really interesting in this report is how they break down the different environments in which robots could effectively do physical tasks. This starts at E0, where robots would be unable to do the job at all. Some examples of this would be work by hairdressers dying someone's hair or carpenters erecting scaffolding to E1, which involves purpose-built robot environments like a factory floor or a food assembly line to E2, structured human work facilities like a hospital pharmacy or a container port, and finally E3, or unstructured work environments like those encountered by a Waymo robot taxi or John Deere autonomous tractor. So really fascinating to see the breakdown of this, Jeff, in terms of understanding the different environments where robots may make sense and maybe where they don't.

Jeff MacArthur (2:51)

Absolutely, and neat for hairdressers to stick out there as being something difficult to do and when we wouldn't necessarily think of that. But yeah, it's kind of cool. And Anthropic then combines this and other data to calculate the occupations that are most exposed to being taken over by robots on a scale of zero, meaning the job just can't be done by current robots to three.

So the occupations at the top of this scale include things like taxi and other drivers that basically clock in between 2 and 2.2 on the scale, moving down through things like refuse and recyclables, collecting to excavating and surface mining activities. So it's really worth taking a read through. Anthropic publishes lots of interesting stuff, and you can find this and anything else at anthropic.com/research, if you want to take a look in a little more depth there.

Amber Mac (3:43)

I do want to say too, just to push on that side a little bit, I've been using that research site for a lot of my keynotes. And like you said, just a fabulous resource in terms of information, love how transparent they are with a lot of this information. So that is a good thing. Now, sticking with big AI companies, OpenAI has revealed yet more instances of its AI hacking into websites and the like, demonstrating that the company doesn't seem to be getting much better at keeping its agents from engaging in potentially illegal behavior.

In one case, OpenAI's agent found exposed credentials and made cryptocurrency withdrawals from an otherwise locked wallet. As OpenAI reported, and I'm quoting here, the model questioned whether it had permission, then reasoned that a public recovery phase and valueless test network tokens made the funds acceptable to use. The deployment succeeded and charged transaction fees to the account. This case shows that a model can substitute its own judgment about likely harm for the account holder's permission, then act on that judgment without authorization, end quote.