Humane’s HP Exit & the Foundation Model Slugfest

[00:00:00] Hi, I'm Eric Newcomer. And I am Madeline Rynbarger. And this is the Newcomer Podcast. Each week, Eric and I discuss the VC deals and the drama that went down. Let's do it! Here we go! Welcome back to the podcast. Lots to cover in AI this week. Big wins, big raises, some dramatic, not so wonderful exits.

[00:00:27] A lot of chest thumping, you know, I feel like everyone wants to make sure they're getting headlines this week. But yeah, one, I don't know, happy failure or failure that exceeds expectations. We're of course talking about Humane. HP agreed to acquire Humane for $116 million. It's much less than I think the total they raised was a little over $230 million over their lifespan.

[00:00:51] But it's not nothing for a product that had such a disastrous rollout as Humane did when it first launched. Expectations are everything here. You know, they had the advantage of having this sort of Apple sheen. You know, their former Apple executives, key to some of Apple's product design. They're launching their first AI pin. So there's a lot of, oh my God, this could be huge. And then it, you know, clearly didn't meet the hype.

[00:01:21] It was a brutal, brutal launch for them. I do feel like there's some points I've heard some chatter, you know, that while it's definitely not the worst exit that could have happened for something like this, there's some kind of dark irony of, you know, some Apple engineers that went off to start their own thing that will be now leading a new division with an HP to integrate AI into PCs rather than, you know, building your own thing.

[00:02:21] And also, you know, AI native hardware executives. I think there's such a premium on AI talent still. This could just be viewed as a very savvy acqui-hire and we'll see what they end up building with this. I don't think it was, you know, I think it was just maybe a case of something too early. There's lots of potential around this. And lots of times the first ones are always the first to fail. But we'll see if, you know, AI devices without a screen really take off. This one clearly did not. But I see the vision.

[00:02:48] This fits into the sort of, I don't know, the Jet, Quibi, Humane. I'd say the three companies, you know, Jet was the big e-commerce play. Mark Lurie, Quibi was Katzenberg and Meg Whitman going big on short video. And Humane, you know, this AI device that you can sort of wear and talk to.

[00:03:10] I mean, I think all of them represent sort of super hype-y, lots of funding, where immediately it was sort of like, do you really need to go big for this to work? You know, and Jet, you know, the argument was where we're going against the Amazons of the world. So to have this scale, we need to like pour a bunch of money into it. But I mean, it worked out for them. They sold to Walmart. So that was sort of the test case.

[00:03:38] I mean, some people think it was an extremely, what, a $3 billion acquirer for Mark Lurie. But, and then Quibi was a disaster. Quibi, it did. I still think like, man, they could have done it small. Like, it's not clear for me on that one. Why? It was just such a huge launch all at once. Right, right. Like, right. Just like, do, I mean, it's just like, I don't know. We're in the content business. And like, we, I wrote a, we wrote, you know, I quit my job, but you know, I wrote some stories, saw if they could work.

[00:04:08] And then when people liked the stories, we expanded out. I don't, I don't quite, I understand that they wanted high production value content, but I, I don't know. I think they could have done it a little leaner than they did. I think Humane, you know, it's like, it's a hardware device. So on the one hand, you needed to spend money to, to get, and clearly they didn't even have enough money to do everything they want.

[00:04:29] It felt like there was too much polish on like the device, like the actual physical, you know, in this sort of apply sense that they had this like beautiful, like little device. And it felt like, honestly, a lot of the challenges, the sort of voice communication, it also felt like they wanted this whole sort of interface. There's a reason that people like software companies as investments more than hardware companies. Software companies are leaner and you can iterate and you think AI is making that more possible.

[00:04:56] But for some reason, Humane decided we need to build a custom device just for AI. And I mean, maybe they think they just didn't get enough money to prove it, but yeah, they didn't prove it. I mean, it doesn't seem like a lot of these AI hardware devices have much utility beyond what you could already do on a smartphone or a different device that you already have. Which then you should just be a magical. Exactly. But like the rabbit R pin that was, you know, a humane competitor.

[00:05:24] And again, it fell into the same problem of what's the use case for this if I have a phone? Like, right. It seems to be that this whole category, while exciting, and I appreciate the need to, I appreciate the desire to be removed to some extent from screen. And be able to move more freely throughout the world and interact with technology. But like with any wearable that we've seen outside of, you know, things that enhance your screens, like AirPods or Apple Watch, you know, it doesn't seem like... Well, those are the two devices. Exactly.

[00:05:52] Feels like Apple Watch or AirPods. These are devices we already have that in some ways are meant to get us off our screens. And I mean, the problem is they're locked down and sort of controlled by Apple. But you would think somebody would come up with a strategy that, yeah, on the watch or on our AirPods that gets us away from screens and allows us to interact with AI.

[00:06:16] And I, unfortunately, what, Apple is still in the world of Siri, which is what, the worst voice platform around. I mean, we haven't even talked... Sorry, I just... Well, we haven't even talked about, you know, Amazon, which dumped tons of money into like getting Alexa's into everybody's house. I have some in a box somewhere. And I mean, I'm curious what you think about this. I'm not... And, you know, we're friends of...

[00:06:41] Most of my best friends run a voice company and they're big voice bulls. Godspeed. But I don't know. I feel like I'm the class... I write professionally. So my bias is clear. You know, I guess I'm on a podcast, but I consume mostly writing. I think writing is a very efficient way to get information. And I don't really see myself wanting to go from typing to a computer to talking to one necessarily. I don't know. Do you feel differently?

[00:07:10] Well, I do see there being some utility in voice, maybe more as a listening vehicle to get off of your phone and interact with more directly in that sense for either an assistant. If it... Again, the key is always if it works seamlessly, then yeah, I could find that to some degree useful if I'm, you know, trying to complete some task that normally would require a screen, but I could just do it with my voice in my home in sort of the fantasy of what an Alexa or Siri could do as an agent. That would be awesome.

[00:07:38] I would absolutely appreciate that because I could, you know, focus... When I'm on my screen, I could focus on work. And when I think of something, I could instantly complete that task. And that would definitely be helpful. However, it's not there yet. Wait, I'm sorry. Give us... What type of task are you imagining here? Or... Like ordering groceries if I'm looking in the fridge and I could just say, you know... You're just like walking around. Walking around. Seamless, hands-free. Oh, I need to write that down. Hey, put this in my notes app. But all of this is things that are, again, really...

[00:08:05] Apple has the advantage here if they were to release something like this that would work. Because it would just integrate into my existing devices that I already manage. So I think that's something that's tough for a startup to break into as well in this space. I guess I agree with you that voice in the sort of like ambient moments wandering around, if it could just capture the like five seconds of inspiration so you don't have to open up your phone, find the right app, you know, that's sort of where the opportunity... And I guess to bring it back to humane is sort of the argument for the device that's

[00:08:35] just sort of on you, whereas it feels like your iPhone is still in the world of you'd have to pull up the specific app to record it. And it's not... And Apple in particular seems to have set up this privacy situation where they're not just going to say, oh, all of a sudden, Apple is the phone that... iPhone, we ambiently record you all the time so that you can just shout out into the world, you know. But maybe Android or somebody will, you know, get in that game. But even then, it doesn't feel like you need a secondary device for that.

[00:09:03] It feels like that's something that could just be captured with your existing technology. So hardware, it's a tough game. There's a lot of incumbents. They do things well. Well, just on voice, I mean, you know, one of the buzziest Silicon Valley companies right now, do you know what I'm going to say? Granola. Yes. Which has become somewhat controversial within the newcomer company. I'm sort of a granola scold. Or I just hate the idea that it's recording me without anybody having to tell me.

[00:09:33] That's true. As reporters, it's very important on our jobs to let our sources know if we're recording them or not. Exactly. So I would prefer to not be automatically assuming that everyone's recording. Right. On the other hand, you know, I sort of want to use it. So it's one of these like norms where you're resistant, but you're like, I don't know if the floodgates are going to open. Maybe I'll just sort of, I'll get on board. You know, you can't, you can't stop that level of societal change. I mean, they claim, you know, my understanding is, you know, it's not record.

[00:10:02] But this, it's very semantic, but you know, they're, they process it, you know, what's like, is Zoom recording you if you're not actually saving it, but it, the computer hears it, you know, it's gone. So if Zoom is not recording you, then similarly, this is listening and then it turns it into the text and then it's gone. That's sort of, I think, the defense. But it is like, it's, I think it's like a word for word transcript.

[00:10:27] So if you want some deniability on what's said in your conversations, granola would seem to undermine that. I spent like an hour looking through like state laws on how they talk about voice recording. And they're not, you know, they're just varied, you know, it's, it's like, and there's all these sort of stray words that I imagine if you were fighting about granola, you know, you'd be like, well, sure, maybe you can say it's not recording. But they're, you know, they'll say like transmit.

[00:10:53] With granola, to me, in some ways, it is argument for text. So you're having these terrible voice meetings and they're like, oh, let's make sure you have the text transcript. And so some of this AI stuff, we're constantly teasing out, you know, is it the voice is inefficient and we need to get a succinct text record or, you know, but I don't know. What do you think about that?

[00:11:17] I think, I mean, it feels like there's a utility for both, but it does seem like everything is still sort of landing on text, which then would imply the need for a big screen to manage all of this text for you. And I think maybe with LLMs, there's a good voice to text interface is pretty slick at this point, as we know from granola. That seems to be a medium where this is working rather well. And there's lots of product innovation in this space. And why mess with something that's, you know, working well?

[00:11:44] I feel like outside of Silicon Valley, there's still a lot of people resisting AI. Like, or I find my own family members are very reluctant to use AI. I'm curious, are you seeing that? When you're out about among the non-Silicon Valley inclined, what's your read on their AI embrace? Among the non-technical masses? The normie masses. The normie masses. Among the normies.

[00:12:10] When I return to the normies, I, well, I have, I feel like I'm in a, I'm in a bubble a bit around a lot of my network is still very in the tech world outside of this job. But I do think that there is some resistance. When I was, I was home last weekend with my family for, in Austin, Texas, which is a tech heavy city, to be fair, for the President's Day weekend. And it seems like there's certainly some skepticism, especially when just talking with, you know, family friends coming through for the holiday.

[00:12:39] People were much more, I guess, cautious, if not outright skeptical or a little suspicious of a lot of this in a way that I think we forget about in this beat.

[00:13:20] And I think that there's a lot of the things that we're going to do, you know, that we're going to be able to experiment with ChatGPT. That Google can meet people sort of where they are. We've also talked about this theme that the, the blank text box is itself intimidating and that a lot of these AI products are going to have to be more narrow. Anyway, let's talk about the foundation models though. Who had the biggest, many billions on the belief that they will get the world to play with their text box. Who do you have the biggest news this week? It was a range.

[00:13:46] Obviously, Mira Maradi, former OpenAI CTO, her startup that has been long rumored is officially launched. Thinking machines. What changed? Didn't we all, she just came on the record and gave some friendly interviews and said, yeah, that thing you'd heard, it's, it is happening. She didn't even say how much money they raised or anything, right? There has been, there's been no public discussion from thinking machines about how much they did raise. Mira, come on the podcast. You have so many questions to answer.

[00:14:13] I, I just want, I don't think I've seen this out there, but you know, she seemingly was somewhat of a Sam Altman critic, right? The board oust Sam, puts her in charge of OpenAI. And then she rallies to Sam. Sam comes back and then she leaves the company. I don't know. Like what's, I just feel like that story doesn't make sense.

[00:14:37] Like if the board believes in you, many of many employees, some of whom are now defecting to your company, believe in you. Why not just like be the leader of OpenAI? Like I, if you have a desire to run a company, yeah. So you could have run it all. Like, and it really does feel like if she had stuck with the board, she could have won. The biggest theory would just be like that she read the employee base and felt like Sam was going to win.

[00:15:06] I mean, the dark theory would be that Sam has some sort of like leverage point over her or is a very intimidating figure. But it doesn't make any sense to me that she rallied all the team to defend Sam against the board and then leaves and is undermining her old company by building a competitor. And she did take around 20 engineers from OpenAI to come to Thinking Machines Lab. Reminiscent of Ilya's company, you know, Safe Superintelligence also poached a ton of engineers.

[00:15:34] So it seems like all the co-founders have gone off to fund their own thing and, you know, pull away from what could have just been, you know, forwarding towards the mission of AGI with the company that has the most momentum behind it.

[00:15:50] In a previous episode, I think I went on a rant about this very problem, which is there's an insane collective action problem where you would think all the top talent should stay at the same company and ensure that one of them is super valuable. But instead, every individual actor says, well, maybe I should have more spiritual control and more equity over the smaller company that can surely raise a lot of money.

[00:16:15] And so then they all sort of like shoot at each other and maybe it's going to undermine their collective valuation. I wonder if it really does come down to just the equity conversation, not to sound, you know, like, oh, it's about the funding and not the mission, because all of these companies are very mission driven. A lot of them don't even have products yet for these new offshoots. At one point, you have to consider that like a good benefit of leaving is to own your own thing and create your own thing. We're fans of that, you know, at an independent media company.

[00:16:44] So, I mean, you are driven to the mission to your point. It does seem counterintuitive. I mean, maybe the mission ultimately would just involve I'm the one who can do it, you know, at a certain point when you lose that faith in what's happening. You say I'm the one that can do it better and that you get the added benefit of the equity when you do it. One theme I think we wanted to talk about is AI is making it easy to build small companies, right? I think the New York Times, Aaron Griffith, just had a piece highlighting that fact.

[00:17:12] The new dick measuring in Silicon Valley is no longer how much money you raise and how many employees you have. It's how much revenue you can generate with as few as employees as possible. I mean, I think people mid journeys in the list. Who else? Oh, Cursor for sure is on the top of the list. And also just the speed in which they've generated revenue is such an important metric, too. And you can do it with these tiny teams in the age of AI. Isn't this what the media class has wanted? Focus on profitability and unit economics.

[00:17:41] I mean, now, honestly, I don't know how ominous was Aaron's story about this. I mean, there's the implicit like, oh, maybe AI is really coming for jobs if Silicon Valley is now proudly doing things with as few humans as possible. To your point there, the first jobs that were seemingly automated away by AI were software engineers with coding assistants. This is a use case for this technology that's coming very well. And anyone can start a company with a much smaller team because of these tools.

[00:18:09] So ironically, a lot of what is the bread and butter of Silicon Valley is something that is one of the first to go in the sort of automation fears around how much productivity AI can unlock. Which gives Silicon Valley moral credibility and high ground. It's like, if we're willing to kill our own jobs, we can kill yours, too. You know? I mean, I am not like a Luddite. I just absolutely do not think we should in any way slow down technology because of our perceived impact on jobs.

[00:18:39] Like, we should try to build the best stuff we can as efficiently as possible. And I think we need to fix that on the sort of policy side and build a better safety net and say, oh, maybe someday everybody doesn't have to work. Obviously, that is not the administration we have right now, which is just like, oh, let's kill every government job in the world. Like, it doesn't make any sense. But that's another episode. Let's talk about Grok. I can't believe we've gotten this far in AI.

[00:19:05] Grok is honestly, there's an argument that everybody else is trying to make news to overshadow what their new model release. It's been performing quite well on a lot of benchmarks. I don't have the latest on the leaderboard. Well, I saw some debate, you know, it's so early, but, you know, they have these chat bot arenas, I think. But then you don't know if it's apples to apples.

[00:19:25] So from the Twitter arguments among the AI crew that I saw, it seems still an open question of whether Grok was better on the same playing field or if it was giving itself some sort of edge. Part of this is what Grok had some of the most, if not the most, GPU access to train this, which then you don't know if that's a positive in judging what they've been able to produce or not. Right. It's like, oh, it's scary because they had so many GPUs.

[00:19:52] But now that it's out, it's like, oh, it wasn't so much better. In the era of DeepSeek, is that really the flex that it once was? You can train a model that is incrementally better on a much lower stack. Although, of course, you know, there's some debate around how much of that cost of training was actually really for the actual training itself. However, it still seems that it was marginally less, but still with Grok, it is impressive that they have this massive GPU cluster.

[00:20:18] Maybe less impressive that the model made a good amount of progress, but more incremental progress with this amount of power. But it's still notable that they have this cluster. It gives them a lot of leverage in this in the space. And I would not count out XAI and Grok on the talent front, even if they're not, you know, in the lead with OpenAI and Anthropic at this point in terms of model performance. And they have access to Twitter and the disinformation miasma that it is for good and bad.

[00:20:47] You know, they're going to, I don't know. I'm, yeah. Elon seems to be going off the deep end. Just before we got on this, he was on CPAC. I'm actually in D.C. right now. I think they're around here. I'm here for a venture in the capital, a wonky D.C., D.C. conference that speaking with Kara Swisher. Anyway, but Elon was on stage. He seemed like under the influence and like was losing a stream of consciousness. And at the same time, Grimes, his, was she his wife?

[00:21:17] No, just his, the mother of some, the mother of some of his children was tweeting at him that they, one of their children is having a medical emergency and she's been able to contact him. Like, I don't know. And obviously this week, the Ukrainian, the abandonment of Ukraine tonally has been despicable. So I don't know this. Obviously, I am no Trump and no Elon Trump fan.

[00:21:42] But I just feel like this could have been the week that things jumped the shark and that they will hopefully lose the country. But yeah, we'll, we'll see. With that wonderful, dark, calamitous note about the state of our country, let's jump into a really major funding round that feels more relevant than ever. Eat your carrots. No politics. We're talking about business again. What's the funding round? Oh, the funding round is Saronic, which is based out of Austin, Texas.

[00:22:10] And yes, it is kind of on theme, autonomous boats. So drone boats for the military. For warfare. Okay, good. So Elon and Trump will be well armed. Raked in a good 600 million in Series C funding in a round led by Elad Gill. Some other participants were General Catalyst had not backed this company yet, but they did back it for the first time in this round.

[00:22:33] And then some existing investors, including A16Z, Caffeinated Capital, and 8VC participated in this round, which really feels like at this point, kind of a who's who of who backs, you know, American dynamism style defense tech from the mega funds. We have Elad, who was an early investor in Anderil. We have A16Z investor in Anderil, 8VC investor across all of these things, Joe Lonsdale, and then General Catalyst with their global resilience practice and focus led by Paul Kwan.

[00:22:59] It's an A-list ring of investors for a massive funding round for autonomous boats. We support defense tech investing. Requires some... Sorry. I'm in a dark mood about the American government at the moment, but previous... Before today, I think I've supported defense tech investing, but if we're... I don't know if the boats are going to go to war on behalf of Russia, that will be deeply troubling.

[00:23:24] Another honorable mention funding round and General Catalyst deal, Together AI, which helps startups fine tune and train models through rented GPU space, raised $305 million at a $3.3 billion valuation. It was co-led by General Catalyst and Prosperity 7 Ventures, which I believe is out of Saudi Arabia. But then Salesforce Ventures, NVIDIA, Kleiner Perkins, Emergence, and Lux also participated in the round.

[00:23:52] So big up round in a short period of time. And Coding started... Startup Codium is apparently raising... Anyway, it's boom time. It's still... I mean, AI... We're not getting the same, you know, foundation model investments that we once did, but applications and infrastructure clearly are going to go strong this year. And a lot of the tier one VCs have a lot of appetite for it, have not slowed down. Cool. Well, I think that's our show. We'll see you next week. See you next week.