The Man Leading Amazon’s AI Reboot

[00:00:00] [SPEAKER_02]: Zscaler extended its Zero Trust architecture with powerful AI engines trained by 500 trillion

[00:00:06] [SPEAKER_02]: daily signals to prevent ransomware and AI attacks that target business.

[00:00:11] [SPEAKER_02]: Zscaler Zero Trust plus AI.

[00:00:14] [SPEAKER_02]: Learn more at zscaler.com slash zero trust AI.

[00:00:22] [SPEAKER_03]: Welcome to Tech News Briefing.

[00:00:24] [SPEAKER_03]: It's Thursday, September 26th.

[00:00:26] [SPEAKER_03]: I'm Zoe Thomas for The Wall Street Journal.

[00:00:29] [SPEAKER_03]: If former President Donald Trump wins the election in November, Elon Musk could be given

[00:00:35] [SPEAKER_03]: a role in cutting government red tape.

[00:00:38] [SPEAKER_03]: We'll look at what Musk's past interactions with federal regulators could tell us about

[00:00:42] [SPEAKER_03]: how he might approach this job.

[00:00:45] [SPEAKER_03]: And then Amazon is working to improve its standing in the artificial intelligence race.

[00:00:51] [SPEAKER_03]: And it's turning to one of the chief architects of its voice assistant Alexa to do it.

[00:00:56] [SPEAKER_03]: We'll tell you about Rohit Prasad and his strategy.

[00:01:02] [SPEAKER_03]: But first, last week Elon Musk said he planned to sue the Federal Aviation Administration,

[00:01:09] [SPEAKER_03]: claiming regulatory overreach after the agency alleged that Musk's company SpaceX violated

[00:01:15] [SPEAKER_03]: rocket launch regulations.

[00:01:17] [SPEAKER_03]: The FAA said it's seeking about $633,000 for three different alleged violations during

[00:01:24] [SPEAKER_03]: two SpaceX launches last year.

[00:01:26] [SPEAKER_03]: This isn't the first time Musk has squared off with regulators, and his actions could

[00:01:31] [SPEAKER_03]: give us a taste of what's to come if he's put in charge of a government efficiency commission

[00:01:36] [SPEAKER_03]: in a second Trump administration.

[00:01:38] [SPEAKER_03]: Here to tell us more is our columnist Tim Higgins.

[00:01:41] [SPEAKER_03]: Tim, it wasn't that long ago that Musk seemed to be a fan of the FAA.

[00:01:45] [SPEAKER_03]: What changed?

[00:01:46] [SPEAKER_01]: Elon has been critical of late of the FAA, kind of arguing that the agency isn't moving

[00:01:52] [SPEAKER_01]: fast enough when it comes to space-related issues.

[00:01:57] [SPEAKER_01]: You have to remember SpaceX is in this rapid kind of growth period.

[00:02:01] [SPEAKER_01]: It's sending a lot of rockets into space for its satellite division called Starlink.

[00:02:07] [SPEAKER_01]: It's expanding that business.

[00:02:08] [SPEAKER_01]: It's also hauling things into outer space for clients.

[00:02:12] [SPEAKER_01]: And then also, it's working towards its very large rocket that is part of SpaceX's bigger

[00:02:19] [SPEAKER_01]: goal of reaching Mars and also getting to the moon.

[00:02:22] [SPEAKER_01]: And so delays are nerve-wracking for them.

[00:02:25] [SPEAKER_01]: Now in the past, Elon has said that delays were usually attributed to SpaceX, not to

[00:02:31] [SPEAKER_01]: the FAA.

[00:02:32] [SPEAKER_01]: But clearly there's some tension here, and it's part of a broader tension that you see

[00:02:36] [SPEAKER_01]: with Elon Musk and just the Biden administration as a whole.

[00:02:40] [SPEAKER_01]: You have to remember Elon has become much more political in the last few years.

[00:02:45] [SPEAKER_01]: He often talks about how he thinks some of this is politically motivated because of the

[00:02:49] [SPEAKER_01]: Democratic Party that is in power in the White House.

[00:02:51] [SPEAKER_01]: And he has sided with Republicans in particular, has strongly and vocally endorsed former

[00:02:58] [SPEAKER_01]: President Donald Trump.

[00:03:00] [SPEAKER_03]: The issue with the FAA isn't the first time that Musk has squared off against a regulator.

[00:03:04] [SPEAKER_03]: What's another prominent example of this?

[00:03:07] [SPEAKER_01]: Tesla, the electric car company, which is working to develop robot taxis, driverless

[00:03:13] [SPEAKER_01]: car technology.

[00:03:14] [SPEAKER_01]: These kind of areas are in some ways operating in gray zones.

[00:03:18] [SPEAKER_01]: Where the regulations are still trying to catch up with the technology.

[00:03:22] [SPEAKER_01]: And you've seen him over time tend to use a tactic where he is ultra-aggressive in how

[00:03:27] [SPEAKER_01]: he handles these regulators.

[00:03:29] [SPEAKER_01]: In part, it seems to be helpful in pushing through his agenda.

[00:03:33] [SPEAKER_01]: The governments don't want to be seen as standing in the way of technology or progress.

[00:03:38] [SPEAKER_01]: And he kind of knows how to use his sway.

[00:03:41] [SPEAKER_01]: That said, he tends to also be ultra-aggressive when it comes to more earthbound or more traditional

[00:03:48] [SPEAKER_01]: regulations as well.

[00:03:50] [SPEAKER_01]: Whether it's with the SEC a number of years ago, which took issue with his infamous tweet

[00:03:56] [SPEAKER_01]: saying that he had funding secured to take Tesla private.

[00:03:59] [SPEAKER_01]: It turned out maybe the deal wasn't so done and they charged him with misleading investors.

[00:04:05] [SPEAKER_01]: Elon and Tesla ultimately settled with the government, paid some fines.

[00:04:09] [SPEAKER_01]: This is why he's under this kind of order that he's not supposed to tweet immaterial

[00:04:14] [SPEAKER_01]: information about Tesla without running it by the company.

[00:04:17] [SPEAKER_01]: But the point being is that he's had this kind of defiant relationship with regulators

[00:04:21] [SPEAKER_01]: for a while.

[00:04:22] [SPEAKER_03]: Why does his relationship with regulators matter now?

[00:04:26] [SPEAKER_01]: Elon's defiance with regulators takes on new meaning now that former President Trump has

[00:04:32] [SPEAKER_01]: said he plans to name the tech billionaire as the head of a special commission to look

[00:04:39] [SPEAKER_01]: for ways to reduce government, whether it's spending or regulations.

[00:04:43] [SPEAKER_01]: Really kind of allowing Elon to get into the weeds and start attacking the rules and regulations

[00:04:48] [SPEAKER_01]: that he thinks are in the way of business.

[00:04:51] [SPEAKER_01]: The idea is here's this very public person who would be potentially making recommendations

[00:04:57] [SPEAKER_01]: on places to cut, whether it's spending or rules.

[00:05:00] [SPEAKER_01]: That's an interesting position for him to be in.

[00:05:03] [SPEAKER_03]: All right, that's our columnist Tim Higgins.

[00:05:06] [SPEAKER_03]: Coming up, insiders at Amazon say the company has struggled to catch up with OpenAI, Google

[00:05:12] [SPEAKER_03]: and other companies in the AI race.

[00:05:14] [SPEAKER_03]: We'll tell you about the man tasked with rebooting Amazon's AI approach after the break.

[00:05:25] [SPEAKER_02]: AI may be the most important new computer technology ever, but AI needs a lot of processing

[00:05:31] [SPEAKER_02]: speed and that gets expensive fast.

[00:05:33] [SPEAKER_02]: Upgrade to the next generation of the cloud, Oracle Cloud Infrastructure or OCI.

[00:05:38] [SPEAKER_02]: OCI is the single platform for your infrastructure, database, application development and AI needs.

[00:05:44] [SPEAKER_02]: Do more and spend less like Uber, 8x8 and Databricks Mosaic.

[00:05:49] [SPEAKER_02]: Take a free test drive of OCI at oracle.com slash wallstreet, oracle.com slash wallstreet.

[00:06:01] [SPEAKER_03]: Rohit Prasad is one of the chief architects of Amazon's Alexa voice assistant.

[00:06:06] [SPEAKER_03]: Now Amazon has put him in charge of a new team with thousands of people to develop AI products

[00:06:12] [SPEAKER_03]: for an upgraded Alexa and other businesses.

[00:06:15] [SPEAKER_03]: Because while Amazon's cloud computing division is said to benefit massively from the AI boom,

[00:06:21] [SPEAKER_03]: the company also doesn't want to risk losing its position as a frontrunner in tech innovation.

[00:06:27] [SPEAKER_03]: Here to tell us more is our reporter Sebastian Herrera.

[00:06:30] [SPEAKER_03]: So Amazon's AI voice assistant Alexa is integrated into over 500 million devices around the world.

[00:06:37] [SPEAKER_03]: So Sebastian, how is it that Amazon is playing catch up in the AI race?

[00:06:42] [SPEAKER_00]: The reason that OpenAI has been so successful and Google as well,

[00:06:46] [SPEAKER_00]: is that they started to work on large language models, these large AI models

[00:06:51] [SPEAKER_00]: that power generative AI, that power technology like chat GPT.

[00:06:56] [SPEAKER_00]: They started working on this stuff years ago.

[00:06:58] [SPEAKER_00]: And Amazon was not as focused on this technology as its competitors were.

[00:07:04] [SPEAKER_00]: And it certainly was not as focused on it when it comes to its Alexa assistant.

[00:07:08] [SPEAKER_00]: So Amazon started focusing on these large AI models later than its competitors did.

[00:07:15] [SPEAKER_00]: And by the time that it started focusing on it,

[00:07:18] [SPEAKER_00]: the competitors like Google, like OpenAI, like Microsoft were already a lot ahead.

[00:07:23] [SPEAKER_03]: So let's talk about Rohit Prasad, the man tasked with leading Amazon's efforts.

[00:07:28] [SPEAKER_03]: What roles has he held inside the company?

[00:07:31] [SPEAKER_00]: So he was brought on to help launch Alexa, which was launched around 2014-2015.

[00:07:36] [SPEAKER_00]: And he really rose up the ranks in Alexa, and he eventually became their head scientist of Alexa.

[00:07:44] [SPEAKER_00]: And in the past year, as Amazon's AI goals have become more broad,

[00:07:48] [SPEAKER_00]: his role has evolved to essentially leading its most ambitious AI team.

[00:07:53] [SPEAKER_00]: That's where he's evolved to today.

[00:07:55] [SPEAKER_03]: You recently wrote a profile of Rohit Prasad.

[00:07:59] [SPEAKER_03]: Did you speak to him for that?

[00:08:00] [SPEAKER_00]: I did not speak to him.

[00:08:02] [SPEAKER_00]: Amazon declined to make Rohit Prasad available for the piece.

[00:08:05] [SPEAKER_00]: But what I do know about Rohit and speaking to friends of his and colleagues of his

[00:08:10] [SPEAKER_00]: is that he is someone that's really passionate about bringing complicated tech to the masses.

[00:08:17] [SPEAKER_00]: He's spoken a lot about just the challenges of getting generative AI

[00:08:21] [SPEAKER_00]: out there and getting it to be reliable.

[00:08:23] [SPEAKER_00]: And something that he's explained a lot is that getting AI to reason as a human does

[00:08:30] [SPEAKER_00]: is one of the most challenging parts of it.

[00:08:32] [SPEAKER_00]: That's why we see even OpenAI and Google with their AI models, and with ChachiPT.

[00:08:38] [SPEAKER_00]: Sometimes they have imperfections, right?

[00:08:40] [SPEAKER_00]: Sometimes they hallucinate.

[00:08:41] [SPEAKER_00]: And this idea of getting a machine to reason like a human is a very complex part of this

[00:08:47] [SPEAKER_00]: technology, and it's something that Rohit Prasad has spoken about a lot.

[00:08:52] [SPEAKER_03]: So how has Prasad tackled the challenges of building up Amazon's AI efforts?

[00:08:57] [SPEAKER_00]: Well, Amazon's AI efforts are very much in flux right now.

[00:09:00] [SPEAKER_00]: And he is trying to wrangle just their goals and charge ahead.

[00:09:06] [SPEAKER_00]: What Amazon's been focusing on is building in-house AI models that could compete with

[00:09:11] [SPEAKER_00]: the likes of OpenAI and Google and others.

[00:09:14] [SPEAKER_00]: Their hope has been that these AI models can not only be beneficial for them with products

[00:09:20] [SPEAKER_00]: like Alexa and making Alexa smarter, but also be able to be attractive enough for other businesses

[00:09:25] [SPEAKER_00]: to use.

[00:09:26] [SPEAKER_03]: Sebastian, you mentioned the release of ChachiPT in 2022.

[00:09:30] [SPEAKER_03]: How did that change the approach Prasad and his team are taking?

[00:09:34] [SPEAKER_00]: ChachiPT was really a moment not just for the industry, but for Amazon as well.

[00:09:40] [SPEAKER_00]: When ChachiPT was released, some of Prasad's team, they were really surprised with how

[00:09:45] [SPEAKER_00]: powerful it was and how easy it was to use.

[00:09:49] [SPEAKER_00]: One interesting anecdote I got is that ChachiPT organized code at times better than Amazon's

[00:09:55] [SPEAKER_00]: own internal systems for Alexa.

[00:09:58] [SPEAKER_00]: And so at that point, they realized that their AI models were not up to speed.

[00:10:03] [SPEAKER_00]: And if they wanted to create a better future for Alexa and one that could compete, they

[00:10:08] [SPEAKER_00]: really needed to get their models up to speed.

[00:10:11] [SPEAKER_00]: Part of the challenge for Amazon is that they're such a big company.

[00:10:15] [SPEAKER_00]: And these teams that have been working on these AI models are really big.

[00:10:20] [SPEAKER_00]: There's thousands of people working on this.

[00:10:22] [SPEAKER_00]: And sources that I've talked to said that sometimes that's been a negative for Amazon's

[00:10:27] [SPEAKER_00]: AI goals, that teams have been too bloated, that the models are taking forever to fine

[00:10:32] [SPEAKER_00]: tune, that some of these deadlines have been rushed because Amazon is trying to catch up

[00:10:37] [SPEAKER_00]: to speed with open AI and others.

[00:10:39] [SPEAKER_00]: Interestingly enough, Andy Jassy, the CEO, recently said that he wants to make Amazon

[00:10:44] [SPEAKER_00]: in general more nimble.

[00:10:46] [SPEAKER_00]: And so there's a focus right now on Amazon and making their processes less bureaucratic

[00:10:51] [SPEAKER_00]: that could benefit some of the hardships that these AI teams have had.

[00:10:55] [SPEAKER_03]: I'm curious how the infusion of large language models into Alexa affected the voice assistants

[00:11:01] [SPEAKER_03]: reliability.

[00:11:03] [SPEAKER_00]: Well, interestingly enough, as Amazon infused large language models into its Alexa assistant,

[00:11:10] [SPEAKER_00]: it actually became worse at doing simpler tasks.

[00:11:14] [SPEAKER_00]: So in some tests, for example, they lost the reliability to turn on the lights at 90% plus

[00:11:20] [SPEAKER_00]: accuracy.

[00:11:21] [SPEAKER_00]: And that was a really big issue for Amazon because one of the great things that Alexa

[00:11:25] [SPEAKER_00]: does is that it does simple tasks like turning on the lights at a high accuracy rate.

[00:11:30] [SPEAKER_00]: And Rohit Prasad is someone who's spoken about those sort of challenges that when you

[00:11:34] [SPEAKER_00]: put more information into the machine, it basically has information overload and then

[00:11:39] [SPEAKER_00]: it can't do the things that it did great before.

[00:11:42] [SPEAKER_00]: And for Amazon, they've been hesitant to release a product that maybe answers a question in

[00:11:48] [SPEAKER_00]: a more human-like way, but all of a sudden doesn't tell you what the weather is like

[00:11:52] [SPEAKER_00]: in your local town with accuracy.

[00:11:55] [SPEAKER_00]: Amazon declined to talk about any sort of delays or troubleshooting issues that it's

[00:11:59] [SPEAKER_00]: had with its AI models and with Alexa.

[00:12:03] [SPEAKER_00]: So they did not speak about some of these issues and certain reliabilities with Alexa

[00:12:08] [SPEAKER_00]: going down as they put more AI technology into it.

[00:12:12] [SPEAKER_03]: That was our reporter Sebastian Herrera.

[00:12:15] [SPEAKER_03]: And that's it for Tech News Briefing.

[00:12:16] [SPEAKER_03]: Today's show was produced by Julie Chang with supervising producer Catherine Millsop.

[00:12:21] [SPEAKER_03]: We had additional support from Trina Menino.

[00:12:24] [SPEAKER_03]: I'm Zoe Thomas for The Wall Street Journal.

[00:12:26] [SPEAKER_03]: We'll be back this afternoon with TNB Tech Minute.

[00:12:29] [SPEAKER_03]: Thanks for listening.

[00:12:35] [SPEAKER_02]: AI may be the most important new computer technology ever, but AI needs a lot of processing

[00:12:40] [SPEAKER_02]: speed and that gets expensive fast.

[00:12:43] [SPEAKER_02]: Upgrade to the next generation of the cloud, Oracle Cloud Infrastructure or OCI.

[00:12:47] [SPEAKER_02]: OCI is the single platform for your infrastructure, database, application development and AI needs.

[00:12:54] [SPEAKER_02]: Do more and spend less like Uber, 8x8 and Databricks Mosaic.

[00:12:59] [SPEAKER_02]: Take a free test drive of OCI at oracle.com slash wallstreet oracle.com slash wallstreet.