OpenAI Is Getting Into the Business of War

[00:00:03] Exchanges, the Goldman Sachs podcast featuring exchanges on rates, inflation, and U.S. recession risk.

[00:00:12] Exchanges on the market impact of AI.

[00:00:15] For the sharpest analysis on forces driving the markets and the economy,

[00:00:20] count on exchanges between the leading minds at Goldman Sachs.

[00:00:23] New episodes every week. Listen now.

[00:00:33] Welcome to Tech News Briefing.

[00:00:35] It's Thursday, December 5th.

[00:00:37] I'm Danny Lewis for The Wall Street Journal.

[00:00:40] Have you ever had a weather app swear it wasn't raining in the middle of a downpour?

[00:00:45] We'll find out why you can't always trust your weather app and ways to get better predictions.

[00:00:50] And then, for years, many U.S. tech companies kept the defense industry at arm's length.

[00:00:56] Now, one of the biggest names in artificial intelligence development, OpenAI, is getting

[00:01:01] into the business of war.

[00:01:03] The chat GPT maker has struck a deal with defense startup Anduril.

[00:01:08] Our reporter Deepa Sitharaman explains how this is the latest example of Silicon Valley's

[00:01:12] efforts to forge deeper ties with the Pentagon.

[00:01:18] But first, weather apps are a convenient way to get a glimpse at the day's forecast so you

[00:01:23] know what to wear.

[00:01:24] But they can be fickle and aren't always accurate.

[00:01:28] Our personal tech columnist, Nicole Nguyen, dug into why this happens and how to get better

[00:01:32] weather predictions.

[00:01:33] She joins us now.

[00:01:35] Nicole, why aren't weather apps always accurate?

[00:01:38] Weather apps use data from these publicly available models.

[00:01:43] And these models spit out a bunch of numbers.

[00:01:48] And it's up to our weather apps to simplify those numbers and turn it into an icon, say,

[00:01:54] partly cloudy, rainy, etc.

[00:01:57] It doesn't have much nuance.

[00:01:58] So it's making its best guess.

[00:02:00] And when that guess is wrong, you are stuck in the rain, unfortunately.

[00:02:05] How has weather forecasting tech changed in recent years?

[00:02:08] The underlying models are these very complicated mathematical models that require supercomputers

[00:02:15] all around the world to crunch the numbers.

[00:02:17] And it's so intensive that some only crunch the numbers every six hours.

[00:02:22] And those computers themselves have become faster, more competent.

[00:02:26] And so they can run more types of mathematical equations for more grids across the planet faster.

[00:02:35] And that means that the accuracy has gone from a good forecast looking ahead three days to a good

[00:02:43] forecast looking ahead over the next seven days.

[00:02:46] I talked to a meteorologist who worked in the 70s, and he said the accuracy has gone up by a day per

[00:02:53] decade.

[00:02:54] Okay, so what are some ways then that people can get more accurate forecasts from their apps?

[00:02:58] The trick is to look at many different apps, which is probably the answer people don't want,

[00:03:04] is to have to do more work when it comes to the weather.

[00:03:08] There is a tool called Forecast Advisor, which looks at forecast accuracy for your zip code

[00:03:15] or city or state.

[00:03:16] So if you go to Forecast Advisor, boop in your zip code or city, look at your top three sources.

[00:03:22] This site, I will say, does not rank Apple weather or Google weather for technical reasons,

[00:03:27] but it's a good indication of which sites, which was surprising to me, are best for your location.

[00:03:34] One for San Francisco, for example, is called Forca, which I was not familiar with.

[00:03:39] And it is mostly, mostly spot on.

[00:03:43] Another tip is to look at short-term data or real-term data.

[00:03:47] So instead of a seven-day forecast, just look at what's happening tomorrow or within the next three

[00:03:51] days.

[00:03:52] Real-time data is also very helpful.

[00:03:55] You can look at radar apps, which seem overwhelming at first because you're like,

[00:04:00] am I cosplaying a meteorologist?

[00:04:02] But once you dig into these radar apps, radar maps, it becomes self-evident.

[00:04:08] You know, where that cloud is moving is where rain will be.

[00:04:11] And you can see where the models think a storm is moving.

[00:04:15] And sometimes that's a little bit more information than just the icon in your weather app.

[00:04:20] That was personal tech columnist Nicole Nguyen.

[00:04:24] Coming up.

[00:04:25] Until recently, OpenAI banned its technology from being used for military purposes.

[00:04:30] Now it's signed a deal with defense startup Anduril.

[00:04:33] What changed?

[00:04:35] That's after the break.

[00:04:43] Kuntenservice contactieren?

[00:04:44] Für viele Menschen ist das der beste Weg, einen schönen Tag zu ruinieren.

[00:04:48] Aber bei Zendesk sorgen wir für eine bessere Customer Experience.

[00:04:51] Besser für die Großmutter.

[00:04:53] Besser für die Floristin.

[00:04:54] Besser für den jungen Mann in Haus Nummer 3a.

[00:04:57] Besser für sie.

[00:04:58] Besser für alle.

[00:04:59] Denn während einige behaupten, dass der Kunde immer recht hat, sagen wir, dass KundInnen

[00:05:03] immer Menschen sind.

[00:05:04] Und da wir auch Menschen sind, wollen wir etwas Gutes für uns alle tun.

[00:05:08] Zendesk.

[00:05:09] Customer Experience mit KI für Menschen gemacht.

[00:05:16] OpenAI is getting into the business of war.

[00:05:19] The developer behind the AI chatbot ChatGPT has announced a new deal with Anduril, a defense

[00:05:25] company that specializes in drones and anti-drone defense systems.

[00:05:29] It's a big, and potentially lucrative, change for the AI developer.

[00:05:34] And it could signal deeper ties between the tech industry and the U.S. Department of Defense.

[00:05:39] Our reporter Deepa Sitharaman has been covering this story.

[00:05:42] She joins us now.

[00:05:43] And we should note before we get into it, News Corp, owner of the Wall Street Journal,

[00:05:48] has a content licensing partnership with OpenAI.

[00:05:51] Deepa, what do we know about the deal between OpenAI and Anduril?

[00:05:54] At a high level, what we know is that OpenAI is allowing Anduril to use its technology in the

[00:06:03] company's defensive tools.

[00:06:07] So, like counter drones and things like that.

[00:06:10] And very explicitly, the companies will not allow the OpenAI technology to be used in combat weapons.

[00:06:19] But that's kind of it.

[00:06:21] We don't know how much money is changing hands.

[00:06:25] We don't know where this technology might be deployed.

[00:06:29] We don't know all the ins and outs and in what manner it might be deployed.

[00:06:35] These are things that we've asked the company, but we're told that it was too sensitive to answer.

[00:06:43] When you say using OpenAI's tech and Anduril's defensive tools, what exactly do you mean?

[00:06:49] So, our understanding now is that OpenAI's AI technology, so these are models like the one underlying chat GPT.

[00:06:59] They will start to incorporate that technology into the software that Anduril has built to maintain its counter drone systems.

[00:07:11] These are systems that detect and assess and track drones.

[00:07:16] And if they spot a threatening drone, then what happens is the military or whoever the Anduril customers are can use electronic jamming, drones, and other means to take it down.

[00:07:32] And the idea here and the hope here is that these AI systems can improve the accuracy and speed at which Anduril systems can detect and respond to drones,

[00:07:44] and the idea being that that minimizes casualties among U.S. troops.

[00:07:50] So this is kind of a big shift because previously OpenAI explicitly barred its technology from being used for military and warfare purposes.

[00:07:59] Why are they changing their stance now?

[00:08:01] Yeah.

[00:08:02] So this has been an ongoing process for the last year.

[00:08:06] They had initially banned all kinds of military applications, including weapons.

[00:08:12] But then in January, there was a change in the company's usage policies.

[00:08:19] That the line that previously said that military and warfare usage was not allowed, that that line had been removed.

[00:08:31] And the company subsequently confirmed that it was working with the Defense Department on a slew of different projects.

[00:08:38] They're exploring a lot of other partnerships with the Defense Department that really are on the defensive side or analytical side of things.

[00:08:48] Where they've drawn a red line for now is we don't want these tools to be used to explicitly kill people.

[00:08:57] If it's defense or if it's just information gathering, they're increasingly comfortable with that.

[00:09:05] And then you've really seen since the election in November a big uptick in these types of announcements and deals across Silicon Valley.

[00:09:18] So it's not just OpenAI that's made a shift towards the Pentagon.

[00:09:24] It's the broader tech industry that's done the same.

[00:09:28] And along those lines, now that OpenAI and Anduril have this agreement, could we see other companies follow?

[00:09:35] Yeah. And in fact, you're seeing other companies do these kinds of deals this year.

[00:09:40] One of the things that's really picked up over the last year is this discussion broadly in the AI industry about AI companies serving as a strategic asset for the United States and bolstering national security at a time when there is rising competition in AI between the U.S. and countries like China.

[00:10:03] So the companies are trying to portray themselves as essential to the security backbone of the country.

[00:10:10] Besides OpenAI, we've seen its big partner, Microsoft, doing a lot of deals with the military.

[00:10:18] And then also Meta is now allowing the U.S. military to use its AI models.

[00:10:25] And Anthropic, which is another big AI startup that was founded by a number of OpenAI co-founders and senior executives, that company is also embracing the military and allowing its AI tools to be used through a partnership with Palantir.

[00:10:42] All these deals are such a striking contrast from 2018, which was not that long ago when there were internal protests within Google against Project Maven, which was a project that Google was embarking on with the U.S. military.

[00:11:00] That kind of internal employee protest actually led Google to back away from those deals.

[00:11:08] But we've seen a re-engagement there, too, with Google.

[00:11:12] And so a few years ago, this is very difficult to imagine.

[00:11:16] And over the last few years, there's been a real transition because it allows these companies to survive and thrive potentially.

[00:11:28] So if working on military technology becomes more common, what does this mean for the tech industry at large?

[00:11:34] The tech companies have often been seen as global players.

[00:11:38] And the CEOs of these tech companies have often been positioned as global leaders.

[00:11:44] What is happening now is that these companies are increasingly portraying themselves as partners to the American government.

[00:11:55] Like, they're no longer just global enterprises.

[00:11:58] They are American enterprises.

[00:12:00] And we've seen signs of that change over the last couple of years.

[00:12:04] And now it's just accelerating in this past year.

[00:12:08] That was our reporter Deepa Sita Raman.

[00:12:11] And that's it for Tech News Briefing.

[00:12:13] Today's show was produced by Julie Chang with supervising producer Catherine Millsop.

[00:12:17] I'm Danny Lewis for The Wall Street Journal.

[00:12:19] We'll be back this afternoon with TNB Tech Minute.

[00:12:22] Thanks for listening.