Google & the Pentagon Discuss a Classified AI Deal, Siri Team at a Coding ‘Bootcamp,’ OpenAI’s New Pricing for ChatGPT Ads

0:13Aaron Tilley:Welcome everyone to the Informations TI TV. My name is Akash Pasricha. It is Thursday, April 16th. Anthropic has officially released its new Claude Opus 4.7 model. The company says it is better at long-running tasks and better at handling images. The information exclusively reported earlier this week that Anthropic was working on this model, and I spoke with Stephanie Palazzolo yesterday. She was the reporter who broke that news. You can check out more of her analysis on our YouTube page. We will link the video in our show notes. Today on the show, the information has exclusive reporting on Google's discussions with the Pentagon.

0:52Aaron Tilley:We'll bring on our Google reporter, Erin Wu, to share with us what she has learned plus our e-commerce reporter and guian published exclusive reporting about open ai's plans for new pricing for its chat gpt ads and our apple reporter wrote a really interesting story about members of the company siri division going to a coding boot camp ahead of a major product revamp and we will wrap the show today with a segment with our partner pwc we are talking about the implications AI is having on how companies play in the value chain. It's going to be a fun show, so let's get right on into it. The Information Publish exclusive reporting that Google is in talks with the government for a new AI deal.

1:34Aaron Tilley:Our Google reporter, Erin Wu, broke that story. She joins me now to discuss things. Erin, what do we know about these talks and this new deal?

1:42Erin Woo:Yeah, so Google is in the process of negotiating with the Pentagon a deal to put its Gemini AI models onto classified systems. The talks are still ongoing, but everything that I hear is that they are progressing towards a deal. People feel good about these talks.

2:00Aaron Tilley:Okay, so what is this deal supposed to be? How does it compare and contrast to the other deals that we heard Anthropic was negotiating and the deal that OpenAI inevitably did end up signing with the Pentagon?

2:13Erin Woo:So I want to emphasize that the situation is still fluid and this could change. But the deal that Google is negotiating with the Pentagon appears to be pretty similar to the deal that OpenAI ultimately got with the Pentagon. And so they're discussing an agreement to let the Pentagon use Google's AI models for all lawful use. Google has also proposed some language that would potentially prevent Google's AI models from being used for domestic mass surveillance, autonomous weapons, including Target, without what they call, quote, appropriate human oversight. And so they're also similar to the provisions that OpenAI ultimately got in their deal.

2:51Right.

2:52Aaron Tilley:And just to recap, these are the provisions that Anthropic wanted, couldn't get OpenAI, signed the deal, and then they kind of snuck in those provisions after the fact, right?

3:03Erin Woo:Yeah, so these are the provisions that Anthropik essentially refused to drop. So Anthropik refused to drop safeguards on using its AI for fully autonomous lethal weapons and for mass domestic surveillance. OpenAI's deal includes provisions about this, but lawyers I talked to at the time criticized this deal for saying that essentially these extra provisions that they put in didn't actually prevent these things because OpenAI had also agreed to all lawful use. Okay.

3:36Aaron Tilley:How big is this deal for Google?

3:41Erin Woo:So this is under, or this is a continuation of the$200 million deal that Google signed with the Pentagon last summer. So this is a sizable deal, but it's obviously very small for Google compared to its overall revenue.

3:55Aaron Tilley:And remind us here, Aaron, I mean, we've talked on the show about these deals and why they're important is the idea here that if Google can show that we can satisfy even the strictest requirements from customers around how our technology is used, or maybe, you know, just if we can satisfy the government, we should be able to satisfy any other commercial customer. Is that the logic here? Right.

4:20Erin Woo:I mean, I think it would certainly be a signal to commercial customers, but really this is the outgrowth of a years-long process with Google to rebuild its relationship with the Pentagon. So in 2018, Google famously declined to renew its Project Maven contract with the military after there was a huge employee outcry about it. And so that really became the primary example of liberal tech activism. And in the years since they've set up this whole Google public sector division, they've changed the AI principles that they published in the wake of that episode. And so they're really trying to break back into military work, and this deal would be a continuation of that.

4:58Aaron Tilley:Did you get any color on how employees are reacting to these discussions going on right now?

5:04Erin Woo:So honestly, the answer is I don't think employees know about them. So one of the big themes...

5:09Aaron Tilley:But now you reported it, so now they know.

5:12Erin Woo:Please reach out to me on Signal. I think it'll be really interesting to see how employees react to this deal. So there is a lot of employee unhappiness about the OpenAI deal, which seems to be quite similar to this deal. So it'll be interesting to see if there's protests about this deal, if people are unhappy, or if at this point people are like, okay, this is what we expected to be happening.

5:39Aaron Tilley:Now, on the flip side, I wonder if you got any color on how people in the Pentagon feel towards Google? Because I know that, look, I mean, we've talked about Emil Michael, the former Uber executive that is now sort of running tech for the Pentagon in many ways. They have all these connections to Silicon Valley. They have allegiances. They have rivalries. How do people in the Pentagon feel towards Google specifically? Yeah.

6:10Erin Woo:So I think that the relationship has warmed a lot in the past few years. Like, Obviously, the Pentagon is a big place. Not everyone feels the same way. But people I talked to who are fairly close to this said that it seems like the Pentagon has really gotten over their distrust of Google after Project Maiden. And like if anything, compared to Anthropic, Google is now viewed as more of the adult in the room. So they have a closer relationship with the Pentagon than the other frontier labs, OpenAI and Anthropic, because of these years that they've put into building the relationship and hiring like ex-government workers, ex-government contractors, veterans.

6:45Erin Woo:At the same time, though, they are still the third place cloud provider behind Amazon and Microsoft.

6:51Aaron Tilley:And last question for you, Aaron, I mean, if you just take a step back and think about why this story is important, why the idea of tech companies doing work with the government is important. I mean, I think it's very clear we should be paying attention to this, but I wonder if there is a broader trend here that you are observing, which is, you know, is it a national security story? Is it a story about AI in the government? it? Like, I'm just trying to sort of put this into context.

7:22Erin Woo:Yeah, I think it's, I mean, really all of those things. Like, I would say that over the past year and a half, one of the most interesting stories has been tech company and AI companies in particular relationship with the Trump administration, how companies in many ways have tried to cozy up to the Trump administration in a way that we certainly did not see during Trump one. So I think this is interesting in light of that, but it is also specifically an AI story. of like AI is becoming much more powerful and what AI is able to do, the kinds of deals that these companies are able to make. I mean, this is stuff that's going to start to have, if not already is having real world impact.

8:00Aaron Tilley:Well, and you know, also we know that Anthropic has this new model out, Mythos I'm talking, not the model that they released today, which is separate. But I mean, And once again, it could just be a matter of time, you know, until Google or OpenAI has their own version of that model and has to decide, you know, well, are we giving it to the government? What does that relationship look like? And that obviously starts with these discussions, which is let's just start with the plain vanilla models and figure out to what extent the government is going to use those. So a really interesting set of discussions.

8:37Aaron Tilley:Aaron, I want to thank you for coming on and sharing with us what you know. That is Erin Wu, our Google reporter here on TI TV. OpenAI is making a big change to its ads product. That is according to an exclusive report from Anne Guillen, the information's e-commerce reporter. I want to bring her on to share with us more what she knows. And what is this change that OpenAI is pursuing? Hey, Akash. So it's not quite a change. OpenAI is really, really moving to rapidly expand their advertising business, which they just launched at the beginning of February. So in my story this week, I am writing about how OpenAI has told some advertisers that they'll be launching some new kinds of ad campaigns and some new pricing models in some cases in as soon as, you know, the coming days.

9:29So it really is kind of a super quick ramp up on OpenAI's part. And they're really kind of rapidly moving to try to scale up this ad platform and, you know, make inroads with marketers. So in the coming days, some advertisers will be able to set up ad campaigns in ChatGPT that charge advertisers based on how many clicks the ads get. not just how many views the ads get, which is the original pricing model that they launched with back in February. And then they've also indicated that at some point in the near future, they will introduce what is known in the ad industry as conversion-oriented campaigns.

10:12So those are ads and campaigns that are aimed at getting people to take a specific action, whether it's making a purchase or downloading an app or creating an account somewhere. um so open ai has indicated that those kinds of campaigns are also in the works although it is a little bit less clear what the timeline is for that so uh like i said it's really kind of less of a change and more just uh open ai kind of laying the groundwork to scale this up as quickly

10:42Aaron Tilley:as they can so give us a little bit of advertising 101 here because i mean in in the uh the the customer funnel, like a marketing funnel, whatever they call it, right? I mean, impression is the first thing you get, and then you got to click, and then do you actually buy? That's the funnel of conversions. So they were prioritizing impressions as a way of pricing their ads. So is the idea here that they will, I mean, they're able to charge more for clicks and conversions? Is it a matter of how you measure it? Or I mean, what does this mean for that? Not necessarily. There certainly are different pricing models for these different kinds of campaigns.

11:23So I think that is something that OpenAI is clearly working on. We've reported, you know, the initial commitments for the impressions-based ads, the views-based ads, those were quite high. And in talking to several advertisers and several advertising agencies, that was a big kind of piece of feedback from some of them was that, you know, some of these these prices for these initial commitments that OpenAI wanted, you know, they were very high. And I think especially because a lot of advertisers still view this as super experimental, it's not something that's becoming a core part of their marketing strategy or their budget yet.

12:04I think OpenAI needs to kind of introduce more of this flexibility, more, you know, range of pricing models in order to be able to get more advertisers up and running on the platform. So I think it's not necessarily like higher or lower prices. I think OpenAI just knows if they want to become a serious ad platform that is really, you know, indispensable for marketers, they need to kind of have all these different ad campaigns in order to be taken seriously. And I kind of view it in a way as it trying to build out the function that, for example, Amazon already has, right?

12:47Aaron Tilley:We've had our colleague Catherine Perloff on the show earlier talking about how Amazon already has a huge ad tech business. They have a lot of advertising experience and they're offering more to the advertisers in the way of data and stuff like that. So this seems to be OpenAI saying, oh, boy, we better get our act together. Right. OpenAI knows they're in big leagues now. And so they've got to very quickly catch up with everyone else. And I mean, they have big ambitions. We've reported that OpenAI has told some of their investors recently they expect more than$2 billion in ad revenue this year, which is, you know, crazy.

13:29They're just getting this off the ground and already are projecting$2 billion in revenue. And then$11 billion in ad revenue next year, which would make it bigger than Snap, Pinterest, a lot of ad platforms that have been around for, you know, many, many years. So clearly they have big plans.

13:46Aaron Tilley:I was struck by this line in your story. You spoke with somebody who was talking about some of the challenges with advertising in chatbots, period. And this is not focused on OpenAI. This is just the new game that we're playing. And the quote here that stood out to me is they said, it's not a true keyword targeting strategy. Every prompt looks different. Can you explain to me what they might have meant by that, what they were indicating? Right. Right. So it's not it's and I think we're seeing this across the ad industry as people try to wrap their heads around what does advertising in chatbots and in LLMs even look like?

14:26It's not, you know, it's not a one-to-one copy of, you know, maybe the previous era of advertising that was very SEO, very search optimization oriented, where you're putting in lists of thousands and thousands of keywords that you're trying to target your ads around. And because you can't predict what an LLM will say in its response, it's hard to promise advertisers, you know, every time this word is mentioned or this topic is mentioned, your brand will also show up with an ad. So I think that's something that a lot of marketers are really, and the platforms as well, too, are trying to really kind of wrap their heads around, like, what does this model even look like?

15:12And I think also that kind of unpredictability is making it difficult for, again, both the platforms and the marketers to think about how do we really measure this? Because I think that's something else that's come up in a lot of my reporting is that marketers and advertisers are very used to a lot of detail about how their ads are performing and very granular data and measurement about how their ads are performing. And so I think that's an area where OpenAI will really need to, again, kind of quickly scale up what they can offer advertisers because those measurement and tracking capabilities and that access to data, that's definitely, I would say, you know, kind of the number one thing on their list that they're looking for.

15:59Great.

15:59Aaron Tilley:Well, Anne, I want to thank you for coming on. That is Anne Guillen, our e-commerce reporter here at The Information. show. The information published exclusive reporting about Apple Siri staffers ahead of a big product revamp. Our Apple reporter Aaron Tilley and co-executive editor Amir Afradi wrote that story. Aaron joins me now to share with us what he knows. Aaron, what's going on with Apple Siri staffers? Yeah, so they did a, you know, fairly sizable reorg again, and they really seem to be focusing and shrinking down the team in a lot of ways. So there's now going to be 60 people working on the core Siri product.

16:43There's going to be 60 people working on evaluations of the product. And then a big chunk of the team is being moved off of it and onto this multi-week bootcamp for AI coding. So is this vibe coding?

17:02Aaron Tilley:Like what's the idea here of Coding Bootcamp? Yeah, so this is the modern way of coding. It's using AI tools to increase your productivity. And it's obviously sweeping the rest of the tech industry and including Apple. Apple internally is really starting to adopt these tools quite quickly, mostly anthropic cloud coding tools. So this is just like, this is just going like wildfire inside the company. And, you know, Siri, the Siri team is quite, has been around for a while. They're often seen as a bit bloated inside the company. So, you know, it's uncertain to me right now if this is a way of trying to get this team up to snuff, getting them improving their skill sets, or it's a matter of trying to just slim down the team because it's not working well.

18:06Aaron Tilley:okay so how does this then relate to the timing of siri because you pointed this out in your story i mean we're like what two months away from the big siri revamp at least that's what we're expecting but these people not busy working on that i mean i you know they're very busy right now they're getting their their the whole team is working really hard to get this ready for wdc the worldwide developers conference in June, or they always release new software. And yeah, that's really the timing of this reorg. It is quite interesting given just how busy they are working towards this goal for this launch.

18:51And I think it shows just like how difficult it is and how maybe be like the team isn't structured right for for for getting this ready i think i think they're um yeah under tremendous pressure and there's a big team that that isn't maybe ready for what's to come next i don't know if their skill set's ready so and i just want to make one thing