(Preview) Amazon Introduces an AI-Powered Alexa, Is OpenAI the New BlackBerry?, The Social Costs of Removing Friction

**Andrew Sharp** (0:00)
Hello and welcome to a free preview of Sharp Tech.

**Ben Thompson** (0:09)
You made a brand promise, and you've built in capabilities to do these sorts of things that you have to rebuild from scratch. If Brian uses his Alexa to turn his lights on and off, guess what the LLM-enabled version of Alexa better be able to do?

**Andrew Sharp** (0:25)
Right.

**Ben Thompson** (0:26)
And better be able to turn his lights on and off, right?

**Andrew Sharp** (0:29)
That's the thing, probabilistic household chore maintenance doesn't sound all that attractive. It's like deterministic solutions make sense.

**Ben Thompson** (0:38)
Right. Are they going to turn on or off, right? It's like shorting his lights, right? It's like you're not going to know until you try. And so that's number one. They have all this capabilities that were built on the old way of doing things that they have to re-implement, and they have to re-implement to the same level of reliability they had before when that previous level of reliability entailed offloading a huge amount of the complexity to the user, and not just in a technical perspective, but from a mental mindset perspective, where I blame myself when it doesn't work. And again, given what these things were capable of, that was a good place to have the product, where users had to do the work. But now if they want to do this, they have to re-do that. And that is even before then you get to part two, which is just doing this stuff at scale. Like how does the cost structure work? Like right now you have a free version of Chat GPT, but if you actually want the decent functionality, you have to pay the subscription, is there going to be Alexa subscription? Or like, how are you going to pay for the inference? How are you going to do this at scale, at cost? All the like obvious bits about doing an LLM. But I think it's that first part that's the most important. Is Alexa Plus going to turn Brian's lights off consistently? And it has to be able to do that, which is a significant increase in what are expectations of LLMs today. Part of the whole thing with people's LLMs, they know they're going to hallucinate. They know they're going to screw up. That's very different from our expectations of these digital canisters with disembodied voices inside of them.

**Andrew Sharp** (2:13)
Yes. No, exactly. I mean, it's a real technical challenge. And The Verge adds that nobody in the media got to try all the new capabilities associated with Alexa Plus, and they won't be able to try the new capabilities until late March at the earliest. A footnote on the Alexa Plus page indicates it won't be coming all at once either. Some features represented are in development and will be released in future updates, was part of the demo, which I think speaks to the complexity that you're describing there.

**Ben Thompson** (2:48)
It's exactly the same as Siri, right?

**Andrew Sharp** (2:49)
And how tough a nut it is to crack. Yeah, no, exactly. And Apple Intelligence, I mean, we're months in to the Apple Intelligence era, and we're still waiting for some of the features that we saw in the Apple Intelligence commercials six months ago. So all of this stuff is difficult. Andy Jassy did say that new beautiful hardware would be coming this fall.

**Ben Thompson** (3:12)
What Amazon is very well known for, beautiful, beautiful hardware.

**Andrew Sharp** (3:16)
Exactly. Just channeling the spirit of jobs. So Bob in Michigan asks, is Alexa Plus too late? Is this like the Windows phone where network effects have taken off and Amazon's success with dumb Alexa makes it hard for them to make the jump? What do you think about that?

**Ben Thompson** (3:36)
It's pretty interesting.
As a side note, let's talk about Siri. I think in the fullness of time, Apple probably wishes they had never launched Siri, and then today they could launch Siri undergirded by an LLM, and it's this incredible experience, and they could slowly build up the Siri, like all the capabilities that are in Siri now, which doesn't work as well as Alexa and Google Voice, but the things like turning on your lights and turning those sorts of things. If you could start from scratch and from an Apple perspective, that would be totally fine because the iPhone has not been selling because of Siri capability. It's been selling because it's an iPhone. And so they could start rolling out a voice assistant in 2025 and be fine and start from scratch with a completely new sort of capability. And honestly, maybe this is what they should have done with Siri, is have an option that's like, if you want Siri as is, that's one. We have a new voice called Joey. I don't know whatever you want. And you could choose that options. Joey will not be able to do anything Siri does. It can have a good conversation. It can't turn your lights on and off. We are going to add that over time, instead of this sort of attempting to retrofit Siri. Maybe that would have been a better approach, but they could do that because they, the iPhone has sold itself for years, and people buy iPhone for other reasons.