Table of Contents
- [00:00:00] OpenAI is growing slower than anticipated
- [00:15:03] Not that it's nothing. I mean, it's going
- [00:30:10] using it
- [00:45:16] that Marcus is pointing out okay I'll I'll
OpenAI is growing slower than anticipated [00:00:00]
Alex Kantrowitz: OpenAI is growing slower than anticipated.What does that say about the broader AI story?Elon Musk and Sam Altman meet in court and Anthropix valuation is approaching $1 trillion.That and more is coming up on a Big Technology Podcast Friday edition [00:00:00 → 00:00:15]
right after this.Welcome to Big Technology Podcast Friday edition where we break down the news in our traditional cool-headed and nuanced format.We have a great show for you today.So much news to break down, [00:00:15 → 00:00:27]
including open ai's user and potentially revenue miss we'll talk about the internal numbers the company's response and what it means for the rest of the ai story we also have musk and sam altman [00:00:27 → 00:00:40]
in court anthropic is raising at a 900 billion dollar valuation and of course a big week for big tech earnings so we'll have so much to discuss in just a short amount of time joining us as [00:00:40 → 00:00:53]
always on friday is ranjan roy of margins ranjan great to see you welcome back good to see you alex a lot to cover this week a lot to cover and it's weeks like this where we see some data that comes [00:00:53 → 00:01:05]
in and in the data you can start to see some broader stories and really where the ai trend is moving i guess trend is selling it short but you get what i'm saying all right let's go to our [00:01:05 → 00:01:19]
for a story here OpenAI misses key revenue user targets in the high stakes sprint towards its ipo from the wall street journal opening i recently missed its own target for new users and revenue [00:01:19 → 00:01:29]
stumbles that have raised concern among some company leaders about whether it will be able to support its massive spending on data centers chief financial officer sarah fryer has told [00:01:29 → 00:01:40]
company leaders that she is worried the company might not be able to pay for future computing contracts if revenue doesn't grow fast enough board directors have also more closely examined [00:01:40 → 00:01:51]
the company's data center deals in recent months and questioned chief executive sam altman's efforts to secure even more computing power despite the business slowdown opening is of course [00:01:51 → 00:02:03]
pushing back on this story to me honestly like the revenue numbers is one thing like obviously this is a new category you are building you're gonna have revenue misses right that sort [00:02:03 → 00:02:16]
of comes with the territory but to me the bigger part of this story is that open ai had a goal to hit a billion chat gpt users by the end of the year in 2025 it missed it's still [00:02:16 → 00:02:28]
not even announced that number so the latest that we have is 900 million active users of chat gpt that came in February 2026, and the billion is yet to be found.Now, of course, it's still a big product, but we saw torrid growth last year and some big moments [00:02:28 → 00:02:45]
with the Studio Ghibli stuff.Voice, of course, was important.Now that consumer story is tailing off, and it makes me wonder about the future of consumer products and generative AI. [00:02:45 → 00:02:56]
Ranjan Roy: So what do you think?Well, are they an enterprise company or a consumer company? [00:02:56 → 00:03:02]
I think the new focus mantra, the new pivot to enterprise, I've been saying this for months now, that they have to have some kind of general focus and decision and strategic direction. [00:03:02 → 00:03:17]
Or is it codecs and actually the developer communities where they're going to see growth?But I think going from 900 to a billion, it is kind of amazing because GPT image two went mildly viral, certainly as much as the Studio Ghibli stuff.When was that? [00:03:17 → 00:03:37]
Six months ago, eight months ago, whatever it was.It's it all times a flat circle.An AI years, right?Yeah, exactly.Could have been last week for all I know.But this is exactly where, like, maybe you don't need to get to a billion users and that's okay. [00:03:37 → 00:03:55]
And it seems like the strategic direction they're going.I saw one thing that showed they went from 3 million users in Codex to 4 million.And that was impressive.And that is impressive. [00:03:55 → 00:04:06]
But trying to do everything all at once and actually pushing back when this reporting is coming out rather than Sarah Fryer saying, you know what? [00:04:06 → 00:04:18]
We're okay not hitting a billion users, and that's fine because the way we're building our business is not purely going to be on chat GPT consumer growth.But they're trying to have it every way. [00:04:19 → 00:04:31]
And I think that is potentially setting them up for issues as like more official numbers come out as they try to push to IPO.Okay, so I definitely let's put a pin in the enterprise side of things and opening eyes response to the story. [00:04:31 → 00:04:45]
Alex Kantrowitz: By the way, I have it on from a spokesperson.This is ridiculous.We are totally aligned on buying as much compute as we can and are working hard on it together every day. [00:04:45 → 00:04:53]
So OpenAI is vigorously disputing the fact that they are wondering about whether they should buy more compute.And you could even say that that's going to be their strategic advantage over Anthropic as this battle heats up. [00:04:53 → 00:05:07]
And we'll talk about Anthropic's forthcoming fundraising pretty soon.But I think that we'll get to enterprise.We've been talking a lot about enterprise.But I am curious to hear your perspective on the fact that this has sort of hit a wall with consumers. [00:05:07 → 00:05:21]
Let's take all the data points together.ChatGPT should have been at a billion.It's not.consumer sentiment or sentiment overall about AI, extremely negative. In fact, had somebody come into the comments on Spotify and be like, I heard an ad for your podcast, [00:05:21 → 00:05:38]
FU and FAI. Like that's how negative, I'm like, well, I'm not even the industry. I'm being critical here, you know, but that, but I would be just the very fact that I'm talking about AI got [00:05:39 → 00:05:49]
me a double, a double FU this morning. And then the last thing, and I think this is important, And this is new data that I got from Apptopia.So this is exclusive to the podcast here.Daily active user growth across all AI apps. [00:05:49 → 00:06:05]
So that includes Perplexity and Claude and the Geminis of the world.ChatGPT, growth is not just tailing off, it's down.So you can see that, you know, while like the space is growing overall, the growth is has completely flatlined. [00:06:05 → 00:06:26]
And it's been down, I think, from according to Autopia, four of the past five months.So this is like this is a real slowdown.So what's happening? [00:06:26 → 00:06:34]
Ranjan Roy: Well, does the Autopia data actually I mean, their name is Autopia include like app usage or is it mobile web and web usage? [00:06:35 → 00:06:45]
Alex Kantrowitz: it includes app usage so that's interesting because we also and i'm going to get to this in a moment but maybe it's worth bringing up now if you are a user of this of these apps [00:06:45 → 00:06:58]
your usage is actually up but the gross addition of users is slowing down okay gross i mean you know sort of the number not like as an this is a nasty yeah this is we're a gross in that come on [00:06:58 → 00:07:11]
Ranjan Roy: our listeners know gross our listeners know gross i sure do well well hold on to clarify the it's actually a declining aggregate gross number of users in these apps is what the data [00:07:11 → 00:07:26]
is showing or it's the additional slow down yeah yeah i mean at the this kind of base so that doesn't surprise me the there i do believe that there is like everyone who is interested has [00:07:26 → 00:07:42]
downloaded a chat gpt a gemini cloud whatever else has started to use it i think even out of my personal experiences friends family everything like everyone already has it on their phone 900 [00:07:42 → 00:07:56]
million i mean i think like in the us it's probably reached relative saturation so to me like the actual growth side of it is not as much of a concern i do think like how do they find those [00:07:56 → 00:08:13]
next hundred million users is it like that you don't hear a lot of talk around international growth and strategy from these companies and this whole market like i don't know like in india [00:08:13 → 00:08:26]
obviously china is going to be its own very very specific market like in africa like where is the next kind of vector of growth because when you're at 900 million you've tapped out the us and pretty [00:08:26 → 00:08:38]
much i'm guessing as much as you're going to and then as long as the average person who is using it is using it more it's still still you know moving in the right direction well let me let me push [00:08:38 → 00:08:53]
on this a little bit further in enterprise we're seeing all these different use cases right [00:08:53 → 00:08:59]
Alex Kantrowitz: we're seeing of course the agentic use cases that we talk about all the time but we're also seeing purpose-built apps for finance for legal for medicine right all over the place any industry [00:08:59 → 00:09:13]
you look there's a purpose-built gpt app that's actually proving valuable taking off building users and having real like significant valuations there's a new one i hear about every [00:09:13 → 00:09:24]
week consumer it hasn't happened that way you would think that with the technology this powerful there would be a breakout of consumer apps and we're going to get into big tech earnings in a [00:09:24 → 00:09:33]
bit meta is case in point right they have had this technology they're trying to build a consumer app with it yes they're trying to develop the foundational models but they're also working [00:09:33 → 00:09:41]
the applications it's just not taking off with consumers is my is my point well you think i'm [00:09:41 → 00:09:46]
Ranjan Roy: wrong about this yes completely and this is my going to be my rant for the week or one of many potentially but it's interesting like the entire meta ecosystem experience is now powered by ai [00:09:46 → 00:10:02]
like the way everyone talks about ai does not have to be like yes meta ai the chat experience i don't know anyone that's using it i know they put out some crazy numbers and i'm sure people [00:10:02 → 00:10:14]
get kind of looped into interacting with the chat experience but every time you scroll your instagram feed the recommendation engine that's powering the ad that is being served to you this [00:10:14 → 00:10:25]
was meta's like greatest i mean they broke out of the apple ios 14.5 prison and kind of showed that they can why everyone is more addicted to instagram than ever every ad that's being created probably [00:10:25 → 00:10:38]
has ai component to it like i think actually facebook is just one big ai slop fest if you've logged in recently so like it is it i think the end user having a chat bot experience like chat [00:10:38 → 00:10:54]
gpt is where everyone's head goes into but in reality so much of consumerization spotify the number of ai generated songs for better or for worse that are showing up on the platform and [00:10:54 → 00:11:06]
getting plays is increasing so i think the big kind of like disconnect here is everyone is thinking consumer generative ai or consumer ai overall is are people downloading and asking [00:11:06 → 00:11:20]
questions to a chatbot meanwhile every existing consumer experience restaurants on doordash are creating much more engaging images using like it's happening everywhere so i think to me that is the [00:11:20 → 00:11:34]
real consumer ai application not how many people are using chat gpt and apparently amazon amazon even has like these little ai podcasts about their product they're about their products and [00:11:34 → 00:11:46]
Alex Kantrowitz: katie netopolis from business insider was like playing one of the podcasts about i think eczema cream and no diaper rash cream yeah you can write your own questions and the host will address [00:11:47 → 00:11:58]
Ranjan Roy: it and she just writes like my butt hurts and they're like that's a great question katie okay so wait hold on okay go ahead no i'm not going to stop you no amazon like the growth in rufus from what i've been hearing is actually spectacular i've been [00:11:58 → 00:12:17]
using rufus more myself now what's rufus rufus is amazon's ai actually it is a chat experience for the most part but it's basically so it's like you can ask questions you can either ask questions [00:12:17 → 00:12:31]
directly in an Amazon product page. Now my Amazon, and probably because I've been using it more, the entire left rail when I log in is actually Rufus. So they are pushing people more towards [00:12:31 → 00:12:43]
it. Again, you ask questions. It not only gives you recommendations, you can ask questions about a product. Does this have USB-C charging when I was getting something recently? But also they're [00:12:43 → 00:12:55]
actually injecting their entire amazon ads business directly within rufus as well so like when we've been talking about well chat gpt have ads they're already building out this entire [00:12:55 → 00:13:06]
ai advertising ecosystem directly so i think but it's embedded in the product it's not someone going to chat gpt and chat gpt shopping has not taken off in the way everyone was expecting six [00:13:06 → 00:13:19]
to eight months ago meanwhile amazon is figuring it out so i think there's so many pockets and And I know I work in the AI industry and I want to be biased, but you know I can be very skeptical about this.But this one, I have to push back on. [00:13:19 → 00:13:34]
Consumers are engaging with AI more than ever.Okay, let me push back on this one more time, then we can move on to our other stories. [00:13:34 → 00:13:42]
Alex Kantrowitz: First of all, I would say, and we've had this debate before, I think, you really have to take the recommendation engines, the AI-based recommendation engines, and put them in one category.and then the generative experiences in another category. [00:13:43 → 00:13:56]
We've had AI-based recommendation for a long time, like feed sorting and ad serving.But what I'm talking about specifically is how does generative AI translate into real consumer experiences?And yes, you can chat with Amazon [00:13:56 → 00:14:11]
and you can listen to a podcast about diaper cream.You know, that's all exciting.But what I'm saying is, where are like the wave of consumer applications that you know we might have expected you know there's no a you know remember character.ai [00:14:11 → 00:14:29]
like there's no like ai character or ai friend app that's that's taking off there's no like explore history app that's taking off there's no like you know ai stylist app that's taking off [00:14:29 → 00:14:42]
there's no ai prominent ai dietician that's taking off etc there are definitely you know categories of consumer products that just do not have a consumer, a generative AI application taking off in a way that you thought it would. [00:14:42 → 00:14:59]
And then again, like you're seeing this slowdown in chat GPT growth. [00:14:59 → 00:15:03]
Not that it's nothing. I mean, it's going [00:15:03]
Alex Kantrowitz: Not that it's nothing. I mean, it's going to hit a billion users.The question is when, but like even opening, I and they said they were stretch goals, but even open AI anticipated that it would hit a billion and it just hasn't.So what's your response there? [00:15:03 → 00:15:18]
Ranjan Roy: So like, actually, this is actually a perfect example. Are you, I'm guessing this is as far away from your everyday habits as possible, but have you ever used a dress up app? [00:15:18 → 00:15:31]
No, this is not something I've used. But that was a very good prediction ahead of time.Well, no, this is another, like working very closely in the retail and consumer world. [00:15:32 → 00:15:41]
This is something we'd started experimenting in my previous experience at Adore Me, like, Like virtual dress-up apps and try-on apps actually have been exploding in popularity.Then you have Google. [00:15:42 → 00:15:55]
Actually, within Google Shopping, virtual try-on is actually gaining a lot of ground where you can actually find a model exactly your size.You can even upload your own picture, and then you can actually try on items within the Google Shopping experience. [00:15:55 → 00:16:09]
Those are all generative experiences.Those are all not going to show up in an Apptopia like ChatGPT experience.But I do think, again, it's being integrated into the things people are doing every day. [00:16:09 → 00:16:25]
And also, LLMs are feeding into an Instagram like their recommendation engines.It's no longer just machine learning anymore.So it's still embedded in there as well.Okay. [00:16:25 → 00:16:36]
Alex Kantrowitz: Okay. Look, I think the reason why I'm bringing this up and the reason why I wanted to start the show this way is because, well, we have, of course, this concrete data point from OpenAI. [00:16:36 → 00:16:47]
But obviously, everybody is, every company is making this pivot into some form of agentic type of experience like the Codex and the Claude Codes of the world and the enterprise move. [00:16:47 → 00:17:01]
And so my question really is, are they making this move from a position of strength where like they have, you know, you would like to have massive growth of chat GPT, but to see that there's potential in these this enterprise and agentic application and say, OK, we're just going to place our bets there. [00:17:01 → 00:17:19]
Or are they moving out of a position of weakness where like, oh, it's not growing as much anymore.And now we have to make our move.So that's where I can turn and get skeptical again. [00:17:19 → 00:17:28]
Ranjan Roy: I think they're moving from, it is a strategic mistake.I think rather than, and my kind of like hot take on this is when you have like a company that's a developer first culture, everyone is going to get more excited about Codex. [00:17:28 → 00:17:48]
And why is everything like moving to the command line?Most average people are never going to do anything from a command line interface.Yet so many of these projects, so many of these products are moving in that direction.People get very excited. [00:17:48 → 00:18:02]
I even see all this stuff around how everyday users are going to be actually in the command line using Codex.No, they're not.I think it's a bias within these organizations because they're developer-first cultures.I think it's a mistake. [00:18:02 → 00:18:19]
I think there's a lot of opportunity from everything I was saying.And actually, again, Amazon, I think, gets it.You don't see Amazon.They know this is our product.This is our business.This is our customer. [00:18:20 → 00:18:32]
So we are going to embed generative experiences or AI-first experiences throughout.And we're going to move things in that direction.And that's where I think everyone is rushing there.This is what I work in.And everyone, and again, you're seeing like, [00:18:33 → 00:18:49]
like Anthropic had this historic run and suddenly 4.7, you just see all this negative sentiment come out around costs and people instantly start stepping back little.And then Codex comes in and 5.5.And like, it's, I don't think [00:18:49 → 00:19:05]
when everyone is rushing towards the same thing that for a company like OpenAI that has such a foothold in consumer, it's the right decision.Your advice, Topin, I would really be like, stick with consumer, don't give up on the [00:19:05 → 00:19:19]
Sora type stuff and try to own the consumer side of generative AI as opposed to shifting to Codex.Yeah. Unless they're almost accepting Google will beat them at it.Interesting.Which is not unreasonable. When you are Google and you're already on the... [00:19:21 → 00:19:40]
the, I don't know, did you see this study around how like Google, I mean, in an evil way, like giving Chromebooks to every student in America and now the actual YouTube utilization, [00:19:40 → 00:19:51]
like YouTube usage during school hours is up like exponentially, but I mean, for better for worse.Oh, hopefully they're watching big technology podcasts there.Well, as long as the first graders of America are just, actually my son who was in first [00:19:51 → 00:20:08]
grade he if i ever play our podcast in the car when we're driving he gets so mad and he's like this is the most boring thing ever so i'm sorry i don't think the first graders that that demographic [00:20:08 → 00:20:23]
is our biggest fan these are the people that were angering first graders and anti-ai listeners hate mail from both these are the things he's gonna he's leaving two star reviews without me [00:20:23 → 00:20:33]
even knowing on my phone but ron john i mean i okay so this is a thing that my other side of it [00:20:33 → 00:20:39]
Alex Kantrowitz: is even though these let's just take this stuff to be true even if it were true revenue miss user miss but deeper engagement i would say open ai is heading in the right direction with codex i mean [00:20:39 → 00:20:49]
if you think about anthropic right last july i was in anthropic speaking with dario he was happy that they were making 4 billion arr now they're at 35 potentially there is a tremendous market [00:20:49 → 00:21:03]
opportunity to go after with this agent style use case in the enterprise and so to me like if open ai thinks that they can pass anthropic because they're going to have more capacity and [00:21:03 → 00:21:16]
Ranjan Roy: potentially on par or better models go there no i mean i work in that at writer like that's i mean i see it firsthand it's very attractive and it's like when it's working it works very fast and But it's competitive. [00:21:16 → 00:21:31]
It's also like, for a company of OpenAI's size, again, at Rider, we've been enterprise only for our entire life.So that's the game.OpenAI, it hasn't been the game.And they have this asset of 900 million users.They can be integrated directly within everyone. [00:21:32 → 00:21:52]
And the important thing here is you can grow revenue fast. [00:21:52 → 00:21:58]
And I do think this is all ahead of the big IPO race and battle here because you can grow revenue a lot faster by getting a bunch of developers using your tool, them not paying attention and token maxing and just blowing out tokens. [00:21:58 → 00:22:14]
and you'll increase consumption, you'll increase revenue very quickly.But that's a short-lived phenomenon versus you have every person in the US, you own the verb to search with AI is to chat GPT something.Like that is a tremendous asset. [00:22:14 → 00:22:31]
And I think they're kind of seeding it to Google right now.Okay. [00:22:31 → 00:22:36]
Alex Kantrowitz: Well, I think we'll just have to watch this play out there.I don't think there's any real answer here.But it'll be a very interesting battle as it continues to play out.As OpenAI does this, of course, it has the thorn in its side of Elon Musk. [00:22:37 → 00:22:59]
And I'm curious how, if you've been watching the trial between OpenAI and Musk this week, and if you have any thoughts on whether this trial will lead to anything of consequences. [00:22:59 → 00:23:13]
Of course, Musk is suing OpenAI for taking his money, going from a charity to a for-profit, unjustly enriching themselves and betraying the charitable trust value that is taking place this week.What's your read on it? [00:23:13 → 00:23:29]
Ranjan Roy: It's rare that listeners will hear me agreeing with Elon Musk.but I think this is one case like it feels like at a very simple logical level this is they were a non-profit and that was the entire founding story for a long time I mean they are a non-profit [00:23:30 → 00:23:52]
hold on what if the current status so much happens that I can't even remember have they converted or yes they've converted but they still have the nonprofit arm that is a certain yes that owns [00:23:53 → 00:24:07]
a certain amount of yeah like we've joked for a long time around how opaque the structure is i think it put ceylon musk in a pretty good just from a very human logical like if you're trying [00:24:07 → 00:24:21]
to convince a jury i think it's a pretty good argument i think there's been zero accountability for any large technology company for so many years that the idea that anything would ever happen that [00:24:21 → 00:24:38]
would actually derail the business because there's just so much vested interest in it like I don't know I don't I the cynic in me just assumes nothing will actually happen maybe there's a fine [00:24:38 → 00:24:49]
there musk and Sam put on a good show but do you think there will actually be any consequence [00:24:49 → 00:24:57]
Alex Kantrowitz: coming out of the trial no i don't think so i mean maybe there should maybe there will be a fine to open ai because they'll have to end up paying that money to the non-profit but i agree with [00:24:57 → 00:25:09]
you i think that elon has a leg to stand on here i mean he gave 30 plus million dollars to found this thing and he currently has like no share in it at all i don't see how that's fair and of [00:25:09 → 00:25:21]
course the open a argument is like well elon gave this as a donation to a charity he can't look at it as an investment and i'm like well of course he gave it to as a donation to a charity you were [00:25:21 → 00:25:32]
a charity you set up that structure with him in the beginning if you began as a for-profit he would have looked at as looked at it as an investment now i know musk is trying to get [00:25:32 → 00:25:42]
musk and sam altman remute sorry he's trying to get sam altman and greg brockman removed from the top of opening i don't think that's going to happen but i wouldn't be stunned if the [00:25:42 → 00:25:53]
jury ended up siding with musk here and of course it's advisory so we'll see what the judge does i don't think the judge is going to blow up open ai but there could be some consequences [00:25:53 → 00:26:03]
Ranjan Roy: like but what though a couple yeah billions billions going you think billions from the for-profits of the non-profit i wouldn't be stunned i mean i guess it's like a significant [00:26:03 → 00:26:15]
Alex Kantrowitz: amount of billions and by the way that could hamper the whole you know build out the can imagine you're an investor and you put all this money in for them to you know have database [00:26:15 → 00:26:25]
capacity to compete against anthropic and then you have to and has to go elsewhere i don't know [00:26:25 → 00:26:29]
Ranjan Roy: well okay so the interesting part here is one the fact that grok is a direct competitor xai like it just makes the whole thing even just richer i think in terms of how they're approaching [00:26:30 → 00:26:49]
this did you did you see that elon was like promoting the ronan farrow sam walton article across twitter x yeah so talk about what happened there yeah so users were reporting it was actually [00:26:49 → 00:27:03]
like a new ui experience almost of like having an article pop up both in the standard ad format elon retweeting it but also even like just popping up at the bottom of your screen the ronan farrow [00:27:03 → 00:27:20]
new york article about some sam altman having many faces and which did you read it was for if you've been following sam altman and open ai for a long time there wasn't anything groundbreaking [00:27:20 → 00:27:32]
in it but surprises there yeah yeah but it painted a pretty strong picture especially if you're not following closely but it's still funny to me that like this bastion of free speech and non-manipulated [00:27:32 → 00:27:45]
speech supposedly of x literally the owner going to trial is able to kind of just manipulate and [00:27:45 → 00:27:53]
Alex Kantrowitz: control what people are seeing do you think that open ai kind of has a zuck winklevi argument to make here which is like if you were if you were so smart you would create a facebook but you [00:27:53 → 00:28:07]
didn't like they could point to the fact that like most of the value has been created by them and elon has sunk billions into building xai which has had mixed results oh that would be [00:28:07 → 00:28:21]
Ranjan Roy: has that been said yet because if you're listening sam that's the argument like i feel this this whole thing is for show i mean i think like it they've both recognized and elon's trying to kind [00:28:23 → 00:28:36]
of like cut them at the knees ahead of their ipo boost xai like obviously there's a strong show element and that would be the greatest it's like how's xai going bro like you already paid [00:28:36 → 00:28:50]
your 44 billion for x and for twitter and you're jamming that into everyone as much as possible but we built something people love we basically like invented this entire industry right now [00:28:50 → 00:29:03]
Alex Kantrowitz: How are you doing?I will say, though, there are Grok users out there.I was flying back from Vegas to New York and sat next to a guy that drives the subway.And I was like, you're talking about AI.And he goes, yeah, I use Grok. [00:29:03 → 00:29:17]
I don't have to badger it to give me an answer I want.So there is appeal out there, but clearly it's not.It's not as far as like the big businesses go.It's not holding a candle to open AI or Anthropic right now. [00:29:17 → 00:29:29]
Ranjan Roy: Well, what is the, what do you think the grok strategy is in this?Do you think they're going to go pivot to enterprise away from consumer? [00:29:29 → 00:29:40]
Alex Kantrowitz: No, they should, speaking of the opening and consumer, maybe they should lead into bad room. That other, AI girlfriend that Musk made that could be the potential, you know, growth area there. Well, just from a business standpoint, [00:29:40 → 00:29:51]
Ranjan Roy: And maybe if OpenAI is truly kind of moving away from consumer, it does open it up.But I guess, why has Meta AI...I mean, it's not a good product, but the actual chat bot experience from every time I've tried using it. [00:29:51 → 00:30:11]
using it [00:30:10]
Ranjan Roy: using it.But to me, it still feels like if you already have the consumer's undivided attention, they don't have to open up another app and experience like someone should be killing it on this whether [00:30:10 → 00:30:24]
it's meta whether it's elon and x like but it hasn't happened yet this is the point i was making at the beginning of the show all right thank you for seeing the light i guess that chat but i [00:30:24 → 00:30:39]
guess google google has shown google for real i don't think so yeah no because do you think google is this great hit consumer ai chatbot i mean gemini like funneling users from your core [00:30:39 → 00:30:54]
experience to a standalone app google has shown they are able to do that far more successfully than meta i mean i think like based on gemini's numbers in the consumer market they've shown you [00:30:54 → 00:31:08]
Alex Kantrowitz: can do that okay before we go to break because we have a lot more to cover today you highlighted a a section of dialogue in this court case you want to share a little bit about why that's important [00:31:08 → 00:31:21]
Ranjan Roy: and what it is there's a few interesting really interesting parts that came out so far in the trial including elon playing like logical jiu jitsu about like it's a yes or no question [00:31:21 → 00:31:35]
that's like asking me do you beat your wife which i don't know like doing that in a courtroom is just so ridiculous to me as though it's like you're i did high school debate and like that [00:31:35 → 00:31:45]
felt like the kind of thing you would do when you're a freshman but more important relative to the industry musk was asked do you know what distillation is by open ai's lawyer william sabbitt [00:31:45 → 00:31:58]
he's it means to use one ai model to train another model and he was asked has xai done that with open AI? Musk replied, generally, all the companies do that. So that's a yes, partly. Musk continued, [00:31:58 → 00:32:13]
distillation is a technique where a smaller AI model is trained to mimic the behavior of a larger, more capable model, making it cheaper and faster to run while preserving much of its performance. [00:32:13 → 00:32:23]
So it's actually, and he continued, the Sabbath, has open AI technology been used in any way to develop XAI. Musk, it is standard practice to use other AIs to validate your AI. I think this is [00:32:24 → 00:32:37]
significant because I think the distillation conversation when it comes to Chinese models and deep seek has been a pretty loaded one. And if the fact that he's just admitting this openly [00:32:37 → 00:32:52]
and saying it confidently, still from a commercial perspective, what does that mean is kind of crazy to me like you would think and maybe there i guess there's probably not a lot of law and regulation [00:32:52 → 00:33:05]
around not doing this but it's still i don't know again from a purely commercial perspective i was shocked that he was saying this were you yeah definitely no it's it's stunning and clearly it's [00:33:05 → 00:33:16]
Alex Kantrowitz: happening everywhere and it goes to sort of a question i asked greg brockman last week which is that like is it going to be economically viable to train these models if you just get distilled [00:33:16 → 00:33:28]
and i don't know there's coming there there may come a point where you know right now we're seeing real leaps in every every new model to a degree and it might come a point where it sort of levels [00:33:28 → 00:33:41]
out and once that does you know how far is the distillation going to be behind the proprietary stuff probably not that far and so that sort of gets to the question of well do we end up seeing sort of intelligence at a certain point, [00:33:41 → 00:33:55]
commoditize and compute at a certain point, commoditize.And we end up in a price war because everything is basically delivering the same.And so then you compete on price.I mean, that's sort of, that was the logic behind this kind of memorable quote [00:33:55 → 00:34:12]
that Mark Cuban gave me in the episode we did on Wednesday, where he said, OpenAI is shitting money away at scale.Because that's his belief is effectively, [00:34:12 → 00:34:21]
Ranjan Roy: you kind of get to that place what do you think around john well are you saying that the models will be commoditized and it will be about product and price yeah that that could that could be the [00:34:21 → 00:34:32]
case okay okay just checking just checking i'm advancing this theory i'm not i'm not you know sort of throwing it out i think it's well actually on that i don't know if you saw like one of the [00:34:32 → 00:34:46]
more on the topic of both distillation and price there's a lot of hype around deep seek v4 is supposed to be again like top level frontier model at a fraction of the cost that almost certainly [00:34:46 → 00:34:59]
like proudly is distillation at its core and then like i don't know have you seen like brian chesky who's i think been on the show a few times he was talking about yeah okay so they're [00:34:59 → 00:35:12]
talking about using quen from alibaba from a cost perspective that basically and i do think moving to a world where let's say you use anthropic and open ai to actually build [00:35:12 → 00:35:23]
but then start to cost optimize towards cheaper models and maybe it's within their ecosystems maybe it's just an open free-for-all in terms of any model i do think that's where things will go [00:35:24 → 00:35:35]
know but did you see there's apparently like a house of representatives recommendation around like banning the use of Chinese models and like actually calling out Airbnb specifically really [00:35:35 → 00:35:53]
Alex Kantrowitz: no I haven't seen that I mean I have seen I mean if you look at like apps like perplexity for instance like they'll allow you to use like the open AI or anthropic models or Kimmy K2 [00:35:53 → 00:36:05]
which they have, of course, like they've downloaded the weights, they've post-trained on their own.They've sort of given their own version of that model.But I just don't see the Chinese models going away because ultimately, if you ban the Chinese models, [00:36:05 → 00:36:18]
aren't you effectively saying like you're banning open source?I mean, there are the NVIDIA NEMO models, which are open source.But outside of that, it's mostly a China thing.Well, actually, so here, so two Republican-led House committees, [00:36:18 → 00:36:30]
Ranjan Roy: They're probing specifically Airbnb and AnySphere, which is the owner of Cursor, over their use of Chinese models.So I found this really interesting specifically because after, like, we didn't talk about Meta and Manus last week. [00:36:30 → 00:36:47]
I mean, to me, like, first of all, we could definitely get into what's going to potentially happen there.But China, that's like quite the salvo. [00:36:47 → 00:36:56]
You know, like you cannot acquire our technology even after that technology has moved to Singapore and trying to get out of the CCP oversight to actually say we're blocking that transaction. [00:36:56 → 00:37:11]
To me, actually, like the U.S.-China tech Cold War like heated up significantly when that happened. [00:37:12 → 00:37:19]
And then when I saw this, that the Republican House committees are actually throwing out this idea that you cannot use Quen or other Chinese models, I think it's going to get – I mean, that whole Jensen-Dworkesh exchange is going to become a far more significant or bigger story this year. [00:37:19 → 00:37:40]
Alex Kantrowitz: Okay.This week, I'll just say one thing, then we really need to go to break.This week, I heard probably the best explanation of what Jensen's position is. [00:37:41 → 00:37:49]
Which is effectively, if you don't sell the American or NVIDIA tech stack into China, you will force the Chinese model makers to build, to optimize basically algorithmically on Chinese chips, like chips from Huawei. [00:37:49 → 00:38:07]
In the event that they are able to make those optimizations and in some ways outpace the American models or become an appealing alternative, they could potentially build those on Huawei chips alone and not make it compatible on the NVIDIA stack. [00:38:07 → 00:38:28]
and then do their own form of export controls on the U.S. or to the rest of the world and basically have control over AI.So let's say they make state-of-the-art models built on Huawei chips.They could hold the U.S. back from actually using those [00:38:28 → 00:38:43]
and effectively restrict our ability to have cutting-edge AI.By putting that constraint on them, you sort of put yourself under the barrel in that way [00:38:43 → 00:38:54]
Ranjan Roy: where you could potentially not have access to the AI that you want.I think that's a circular but reasonable argument.But question, should large tech companies in the U.S. be allowed to use Chinese models? [00:38:54 → 00:39:10]
Yes. I mean, you should be able to download the weights, do the work on your own, and then run them. I think so.okay but only the open source side of it not directly connecting to the Alibaba quite [00:39:11 → 00:39:25]
infrastructure the same way you would to an anthropic yes or no it depends what you're doing yes or no question God yeah you know Mr Senator I'm gonna say I'm gonna say yes I'll [00:39:25 → 00:39:39]
say yes I don't have a problem with it for now until we see those things I would say it's not lead to i don't think it will lead to like a clear catastrophe right away like as air is the is the [00:39:39 → 00:39:51]
Alex Kantrowitz: fact that you can't say so i mean i don't know this is kind of a weird thing to go a weird rabbit hole to go down but is the fact that you like can't get straight answers about tiananmen square [00:39:51 → 00:39:59]
Ranjan Roy: going to impact which hotel or apartment room you book on airbnb that would be weird well maybe feng shui i say this with the taiwanese mother-in-law could start injecting itself into [00:39:59 → 00:40:13]
there be successfully maybe that would be a much clearer understanding exactly so maybe this is we're both arguing okay that form of soft power i'm i'm i'm for all right let's go to break we'll [00:40:13 → 00:40:24]
Alex Kantrowitz: go to break and come back and talk a little bit about big tech earnings and prediction markets right after this and we're back here on big technology podcast friday edition just to continue going on with my conversation or my point here about [00:40:24 → 00:40:38]
AI consumer, if you look at the earnings that came in this week, if you were a cloud company, you were very happy. If you were building AI consumer apps or you were building for consumers, [00:40:38 → 00:40:50]
you were either not happy or you were thrilled that you didn't invest a lot into AI. So let's just break it down. This is from CNBC. You look at Google Cloud. Google Cloud grew 63%, [00:40:50 → 00:41:04]
percent 23 billion 20 billion dollars this is by far the strongest growth growth rate for any period since google started breaking out cloud results in 2020 that's massive aws by the way [00:41:04 → 00:41:16]
stuck in the 17 18 growth rate range for the past few years grew 28 microsoft grew 40 azure [00:41:16 → 00:41:25]
if you are providing the ai infrastructure for you know this enterprise build out you are doing [00:41:27 → 00:41:34]
Ranjan Roy: really well what do you think about this ron john i mean the numbers are insane 63 at that scale i mean and i guess it reflects this is like a public company earning breakout that kind of [00:41:34 → 00:41:48]
tells the anthropic story as well that we keep hearing about through fuzzy arr numbers here we have a clear 63 growth to 20 billion dollars in a quarter for google crowd cloud is nuts like i [00:41:48 → 00:42:02]
think yeah it's i feel the will there be demand or are we over building capacity it seems like [00:42:02 → 00:42:12]
Alex Kantrowitz: that question has been answered do you see any holes in that story yeah so here's a tweet from gary marcus sheer insanity amazon google microsoft and meta collectively are spending more money than [00:42:12 → 00:42:22]
the manhattan project every single month more than 20x the manhattan 12x the manhattan project every year and what do they have to show for it none are making major profits on ai none has a [00:42:22 → 00:42:32]
technical moat a massive price war is inevitable a few of their customers are seeing major returns on investment greatest capital misallocation in history i mean here's the question is what these [00:42:32 → 00:42:46]
cloud services divisions are seeing this big massive bump in revenue just downstream of the major amounts of money that the anthropics and the open ai are raising and sort of okay not [00:42:46 → 00:43:01]
you know not quite sustainable without those big fundraising and by the way big fundraising moments [00:43:01 → 00:43:06]
Ranjan Roy: and by the way a lot of that fundraising is coming from them what do you think okay i like your circular funding and again and actually a lot of that funding is in the form of cloud credits [00:43:06 → 00:43:18]
that is recognized as revenue i'm not saying that's 100 sure what's happening but maybe yeah yeah yeah so on one side i feel like again if you listen regularly you know i can be very [00:43:19 → 00:43:31]
skeptical and i will open ai or anthropic have a successful ipo i'm not sure i feel like gary marcus and ed citron and them like i wish they just said okay something positive or impressive [00:43:31 → 00:43:47]
has happened like not everything gary has to a degree he did say that claude code is a combination of neurosymbolic systems and machine learning which is fair which is fair like [00:43:47 → 00:44:01]
llms on their own without a harness without a product without like all of this okay all right at least gary's recognizing it i do think the investment in the infrastructure side [00:44:01 → 00:44:14]
it's it's interesting because maybe okay maybe the one argument against this is obvious that the demand is there and they got to keep building is maybe if i take what he's saying and [00:44:14 → 00:44:25]
extrapolate a bit is the idea that the economics of how they're investing are flawed that like the building out assuming constant price at today's growth and today's like revenue the fact that it will scale linearly or exponentially like that. [00:44:25 → 00:44:42]
Maybe it's true that as costs come down, the amount they've invested, if DeepSeek v4 and Quen and others and people are using open source and the actual cost goes down dramatically, [00:44:43 → 00:44:57]
Alex Kantrowitz: then it could be pretty bad capital allocation.Right.I mean, I think we can't, even though the use cases are there, which they are, right?And even though this won't go to zero, [00:44:57 → 00:45:09]
Ranjan Roy: we cannot discount the fact that there could be a collapse here because of the very factors that Marcus is pointing out okay I'll I'll say and it's true no one understands the economics [00:45:09 → 00:45:23]
that Marcus is pointing out okay I'll I'll [00:45:16]
Ranjan Roy: that Marcus is pointing out okay I'll I'll say and it's true no one understands the economics of any of these businesses right now like the what the act what is a true margin will again [00:45:16 → 00:45:30]
we've seen it with Anthropic that just that insane spectacular growth the pushback on price and understand it after four seven came out and recognizing that a lot of it is subsidized [00:45:30 → 00:45:43]
anyways so what are the expenses to anthropic will eventually have a more clear picture of like and then how all that relates to the infrastructure side i guess it's fair no one [00:45:43 → 00:45:56]
what is an average margin for an ai business no one knows yet exactly so that is something that But I don't know.I think we need to keep coming back to on this show. [00:45:56 → 00:46:09]
Alex Kantrowitz: At first, it was like, is this technology going to work?The technology is working.And the question is, these business decisions that are being made, there's no other way to really describe them than YOLO decisions.Nobody knows what's going to happen here. [00:46:10 → 00:46:25]
The demand is coming in.But it's a brand new category.There's bumps in the road.And we could end up seeing a price collapse. [00:46:25 → 00:46:34]
Ranjan Roy: I also actually, when you say YOLO, it kind of makes me think like the executives, the CEOs of these companies are all in the same circle, which makes this interesting too. [00:46:34 → 00:46:47]
So like when everyone around you that you have known, respected, hated, just like that is your basically social circle or like professional circle, your closest LinkedIn connections is saying the same thing.It's going to exacerbate how you think. [00:46:47 → 00:47:06]
Like, yeah, it is interesting to me that, and it's a very, the Musk Altman trial reminds us, this is a very, very small group of people that have known each other, competed against each other. [00:47:06 → 00:47:19]
I mean, you know, had spats with each other, like, remember when Zuckerberg was in Musk, the cage match, like all types of interactions.And they're all speaking, they're all thinking the same exact thing. [00:47:20 → 00:47:35]
maybe that's another reason everyone could be wrong well that's sort of what makes what Apple [00:47:35 → 00:47:41]
Alex Kantrowitz: has done even though Apple did try to make this happen which has made what Apple's done quite impressive that they decide hey we don't want to spend on foundational models I'm kind [00:47:41 → 00:47:51]
of going 180 on Apple honestly they let me just say they they had iPhone sales grow 21.7 percent they don't have AI on the iPhone Siri sucks this is just the counterpoint to what we've [00:47:51 → 00:48:04]
saying they had quarterly sales of 100 111 billion i think i foreshadowed it earlier by saying you know in consumer you're probably unhappy if you spent a lot or you're happy if you didn't spend [00:48:04 → 00:48:18]
Ranjan Roy: anything and when i said that second part i was referencing apple i mean if their ineptitude and incompetence and god do i hate siri but if that ends up helping them in the long run because [00:48:18 → 00:48:37]
by sheer virtue of incompetence they did not go all in on building their own models and investing in ai infrastructure and that ends up being the right decision god bless john turnus and his reign [00:48:37 → 00:48:50]
Alex Kantrowitz: because i mean could happen i mean conventional wisdom now is like oh apple you did a good thing and now you're selling your mac minis by the way in the earnings call they talked about how [00:48:50 → 00:49:01]
mac mini has become an important part of the ai agent infrastructure and they've also talked about how the new siri is coming this year so you might end up getting the best of both worlds i believe [00:49:01 → 00:49:14]
you i'll believe it when i see it i'll believe honestly if they do this i will take back of many of the negative things i've said about tim cook actually i'm going to say something positive [00:49:14 → 00:49:24]
Ranjan Roy: about siri today do you know alexa plus cannot translate into chinese my wife was asking and we actually have an alexa plus and siri both kind of like next to each other and then she [00:49:24 → 00:49:37]
turned around and asked siri and siri was able to translate something into chinese so siri's got something i guess it's alexa plus not the other leading ones but say i siri won one battle okay [00:49:37 → 00:49:50]
Alex Kantrowitz: well that is probably more than it's won in any time in recent history so we got to give one to siri man apple again don't down apple i think that's something i'm learning all right let's [00:49:50 → 00:50:02]
end today talking a little we have some prediction market news news this is a recurring theme that comes up on the show about the prediction markets and we have a story ron you can take us away about [00:50:02 → 00:50:14]
Ranjan Roy: senators banning themselves from prediction market trading yeah the u.s senate unanimously how rarely do we see something along bipartisan lines barring senators from trading on prediction markets [00:50:14 → 00:50:27]
and obviously cal sheet and polymarket i mean we apparently on it was a few weeks ago cal she said it suspended one u.s senate candidate and two candidates for the house of representatives [00:50:28 → 00:50:40]
for political insider trading on their own campaigns there was this crazy story where a u.s army special forces master sergeant actually was charged with using classified information [00:50:40 → 00:50:53]
around the maduro capture that he was part of that mission to bet on which is just like insane still to me like the most dystopian thing imaginable but it's it's nice to see the u.s senate actually [00:50:53 → 00:51:07]
Alex Kantrowitz: restricting themselves from doing something absurd yeah no i think that there's a growing recognition that some of the prediction market activity is can be very cancerous to a society [00:51:07 → 00:51:21]
can be unfair to voters to sorry to gamblers which is like i guess they should know better yeah voters no one cares about but yeah but well i mean on the other hand you could say well they're [00:51:21 → 00:51:33]
Ranjan Roy: actually like more accurate now so what do you think about that where do you stand on that so i've seen that argument and kind of like the companies themselves almost use that argument that [00:51:33 → 00:51:46]
if a small number of people are kind of driving the market in the actual accurate direction using insider information that makes the market more accurate which is true but it doesn't make me [00:51:47 → 00:51:59]
Alex Kantrowitz: like this any better and it also rigs it against everybody else and i think it is i think if you look at it on a whole there is a serious you know this stuff has only recently been legalized and [00:51:59 → 00:52:11]
it's kind of taken as normal today and i say this is someone who likes to put like a couple dollars on the game when I'm watching anything and put it on like the FanDuel odds. But there [00:52:11 → 00:52:25]
is without a doubt a lot of healthy activity here, but also a lot of extremely cancerous activity here. And it's almost like you're seeing a society that can't help itself. So let me tell you one [00:52:25 → 00:52:36]
story before we leave. There's this quarterback in college football at Texas Tech. His name is Brendan Sorsby. He just entered a gambling addiction program for sports betting that [00:52:36 → 00:52:48]
could end his college career. This is according to Matt Schick from ESPN. And then Schick posts a article from CBS Sports about the fact that he could miss the season. This is the second paragraph [00:52:48 → 00:53:01]
of that article. And this really annoys me. Texas Tech was an overwhelming favorite to repeat as Big 12 champions after acquiring Sorsby this offseason, but now has moved to an even money [00:53:01 → 00:53:15]
at plus 100 via FanDuel Sportsbook after Monday's news. The Red Raiders' projected win total has also decreased, going from 11.5 at opening to 10.5 victories, and Sorsby is no longer on FanDuel's [00:53:15 → 00:53:28]
Heisman odds list after opening at plus 2,500 just outside of the top 10. CBS, allow me to address you for a moment you are writing an article about a quarterback with a serious [00:53:28 → 00:53:41]
gambling addiction problem that may cost him a season in the ncaa and potentially send him right to the pros where his life you know may be destroyed because his draft standing will not [00:53:41 → 00:53:52]
be anywhere close to where it was before maybe destroyed is too strong but it won't be what it was before you have no less than three mentions of the odds movement from said person's life [00:53:52 → 00:54:04]
Ranjan Roy: destroying activity with hyperlinks directly out to those exact bets to those bets now i don't say [00:54:04 → 00:54:14]
Alex Kantrowitz: this likely get a grip cbs sports don't do this this is just a you know it is a it propels people into the situations that sorsby finds himself and i don't understand how we have a [00:54:14 → 00:54:29]
society who is looking at this and saying we have no problem here i this is disgusting this is crazy [00:54:29 → 00:54:37]
Ranjan Roy: like actually the this is a good call out for this is the most kind of like weird example of like obviously like how much sports sites have been incorporating odds into even just like tv [00:54:37 → 00:54:53]
broadcasting into every like their websites apps everything but yeah that is quite do you think someone even do you think this is just ai generated and the logic around all these like [00:54:53 → 00:55:07]
incorporating bets is already built into the cms and like or do you think they someone actually sat down and was like i'm going to do this or do you think someone had to do it and actually felt sick [00:55:07 → 00:55:20]
to their stomach which of those three oh god i mean i don't know if it which one would be better to be honest someone's got to get on the phone with barry weiss and say you know don't do this [00:55:20 → 00:55:31]
please i mean out of all their problems this is a pretty bad one though i'm gonna do it i'm writing a letter to the editor i'm gonna do it i'm doing it dear barry first time caller long time listener [00:55:31 → 00:55:44]
listen we got to talk about cbs sports hey you got all that go in and check alex has a screenshot in our prep doc here i want to confirm do those click out directly to the bet because that is that's the [00:55:44 → 00:55:58]
single most horrifying thing i can imagine yes i'm gonna find out for you right now oh i don't have i [00:55:58 → 00:56:04]
Alex Kantrowitz: don't have it well i guess we will end on that uplifting note ron john i mean lord almighty I didn't think we could get more depressing than OpenAI's missed billion user number, but I think we found it here.So we'll end on the doom and gloom. [00:56:04 → 00:56:20]
I think generative AI is showing up in consumer experiences.There we go.Now, excuse me while I put a polymarket bet on when OpenAI will announce that number.Yeah.Just kidding.I won't do that.All right, everybody.Thank you for listening to Rajan. [00:56:20 → 00:56:34]
Thanks for being here again.Great to see you guys always.Have a good week.See you next week.All right, everybody.See you next week, and we will be back next time on Big Technology Podcast. [00:56:34 → 00:56:44]