Latest / In The Loop / EP15 - Google, OpenAI, Meta, Anthropic & The Three Battles To Own All AI

Transcript

EP15 - Google, OpenAI, Meta, Anthropic & The Three Battles To Own All AI · In The Loop

• 0:00
  This week's discussion will follow three big AI
• 0:02
  battlegrounds. Google and Open AI are now
• 0:04
  taking aim at anthropic ownership of the AI coding
• 0:07
  arena. Then there's Meta trying to take
• 0:08
  Open AI dominance in the consumer AI market.
• 0:11
  And then finally, we have Open AI slow but steady assault on
• 0:14
  Google search market. What we're witnessing here is a
• 0:17
  high stakes battle between trillion dollar companies on
• 0:20
  shaping the next computer paradigm because these companies
• 0:23
  want to own everything. They want to be the place that
• 0:26
  you and me visit every single day.
• 0:28
  In today's episode, I'm going to tell this story.
• 0:31
  This is in the loop with Jack Horton.
• 0:33
  I hope you enjoy the show. So for the past year, the AI
• 0:46
  coding space has had a very established hierarchy because
• 0:50
  for some time now, Claude's Anthropic has been just a go to
• 0:53
  model for most programmers. They've been well and truly out
• 0:56
  in front as a leader. They've even raised, you know,
• 0:59
  over $3 billion pretty much based on this advantage alone.
• 1:02
  But last week on May 6th, it looks like some things are going
• 1:06
  to start changing because Google just released Gemini 2.5, which
• 1:10
  is just two weeks before their annual developer conference
• 1:13
  where they launch most of their exciting big products.
• 1:15
  So you can tell they're probably gearing up for something quite
• 1:18
  exciting. Gemini 2.5 immediately took over
• 1:21
  Claude in the Web dev arena leaderboard, and they even broke
• 1:25
  the 1400 ELO rating barrier, which is the first model to ever
• 1:29
  break that threshold. And that's the gold standard for
• 1:32
  coding. And there's been a lot of
• 1:33
  validation from developers, AI companies too.
• 1:37
  There's been quote shared by Google from, for example, Celia
• 1:39
  Salberti at Cognition Labs, who were the creators of the Devon
• 1:43
  engineering agent. If anyone remembers that they
• 1:45
  were quite famous Once Upon a time.
• 1:47
  But he says that it feels more like a senior developer because
• 1:51
  of the one shot prompt that can do massive refactors and make
• 1:55
  really big judgement calls. However, what I would say here
• 1:58
  is that I'd take everything with a pinch of salt because this
• 2:01
  benchmark story gets quite nuanced as you start to dig into
• 2:04
  things. Because while Gemini dominates
• 2:06
  that LM Arena test, it actually falls behind on other
• 2:09
  benchmarks. For example, there's a benchmark
• 2:11
  called the live bench contamination free evaluation.
• 2:14
  This type of evaluation is called contamination free
• 2:17
  because it basically means that the AI models are tested on
• 2:21
  questions and assessments, which has none of the data that it
• 2:26
  would have been trained on. So it can't memorize or
• 2:29
  regurgitate something it might have seen its training data.
• 2:31
  So it's supposed to be quite balanced and neutral.
• 2:34
  And on this test, Open AI O3 mini model actually outperforms
• 2:38
  it when it comes to coding accuracy.
• 2:39
  So as you can tell, the benchmark space gets quite
• 2:42
  confusing and nuanced. My personal preference for this
• 2:44
  is one shot questions and things that you could do really well on
• 2:49
  one model, try and do it on another, and then judge it
• 2:51
  yourself. But the main take away here is
• 2:54
  that Google just topped one of the most important assessments
• 2:57
  and benchmarks that will influence developers.
• 3:01
  And ultimately, Google's strategic play here isn't about,
• 3:03
  you know, just model performance.
• 3:05
  It's about pure integration because they're weaving Gemini
• 3:08
  directly into Android Studio, Google, Firebase, Cloud Code.
• 3:13
  So any product where they own developer attention, they have
• 3:17
  more developers using them than most providers right now.
• 3:20
  And they've really started to steal the show.
• 3:22
  But while this is happening, Open AI aren't just, you know,
• 3:25
  sleeping on this and doing nothing.
• 3:27
  This week they've taken a huge bet and a bet that's trying to
• 3:31
  take on Anthropic and Google here.
• 3:33
  They just made a $3 billion acquisition of a company called
• 3:36
  Windsurfer, which is basically an AI coding tool.
• 3:40
  And this says a huge amount about how important this space
• 3:44
  really is because that's 1% of their total valuation.
• 3:48
  So not a small amount of money because of all those different
• 3:51
  use cases. I did an entire episode on some
• 3:53
  of the most popular use cases last week.
• 3:56
  Coding is still really high up and coding is expensive.
• 3:59
  Programmers are expensive. They take a long time to
• 4:02
  recruit. There's a lot of competition.
• 4:04
  So there's a real obvious business case to reduce the
• 4:07
  number of developers required in a process.
• 4:10
  And if that enables our developers to produce more
• 4:12
  stuff, to sell more stuff at a lower cost for them, again, the
• 4:15
  business case only gets bigger. OK, so moving to battleground #2
• 4:29
  which is the consumer AI space. So as you know, when it comes to
• 4:33
  consumer AI, so just people getting either using it
• 4:36
  day-to-day for random tasks, paying either nothing or a low
• 4:41
  or the lowest membership tier for an individual license
• 4:43
  ChatGPT is well out in front. They've got 400 million weekly
• 4:48
  active users. But now Meta, so Zuckerberg is
• 4:51
  going for that market share because they've just launched
• 4:54
  the Meta AI app powered by their latest language model called
• 4:58
  Llama four. And the numbers Meta shared are
• 5:00
  quite interesting because they reported that 500 million people
• 5:03
  have tried it's AI features. And Mark Zuckerberg in its Q1
• 5:07
  earning calls said that Meta is on track to reach 1 billion
• 5:11
  monthly active iterations across all platforms.
• 5:13
  And he said that the speed of adoption has even beaten their
• 5:16
  internal expectations. Meta's advantage in this space
• 5:19
  is obvious huge leverage. They have got a ridiculous
• 5:23
  network of people across Facebook, Instagram, WhatsApp,
• 5:27
  so an insanely big distribution channel to reach people.
• 5:31
  And I'm sure that they can find ways to integrate personal data
• 5:34
  within their language models or their their applications they
• 5:37
  launch. And usually when people get a
• 5:39
  more personalized experienced, they prefer it.
• 5:42
  Now when it comes to 1 to one conversations with a chat bot,
• 5:45
  I'm unsure if that'll play out in the same way because there's
• 5:48
  a degree of trust that you need to have to share challenges,
• 5:53
  needs, wants with a chat bot. So obviously the damage done a
• 5:58
  few years ago with everyone wearing their tinfoil hat with
• 6:01
  Facebook selling data has definitely not left the public's
• 6:05
  perception of that brand. Now obviously this seems quite
• 6:07
  exciting for Meta, but I will say that his language also says
• 6:10
  a lot. He says has tried on
• 6:12
  interactions. It's not the same as saying
• 6:14
  weekly or monthly active users. So when he tells us those
• 6:17
  numbers, I will then truly gauge how successful it's been so far.
• 6:21
  Until then, I see it as more marketing than actually any
• 6:24
  proof it's working. I can't help but notice the
• 6:27
  interesting approach that Mark Zuckerberg and Meta are taking
• 6:31
  here because obviously they've invested a huge amount of money
• 6:33
  into Rayman glasses into the metaverse.
• 6:36
  None of those bets are paid off yet and they may never pay off
• 6:39
  and they may be just years too early, but the things they're
• 6:42
  investing in could be the next consumer platform instead of an
• 6:45
  iPhone. I mean, interestingly, Apple
• 6:47
  came out recently and said that they don't think the iPhone or a
• 6:51
  smartphone will be that consumer device forever.
• 6:54
  They think it's in the in on their on the road at the moment
• 6:57
  to replacing that. They think that it will be
• 7:00
  replaced in the coming years. So could the Ray Ban glasses
• 7:04
  represent that next consumer device or platform?
• 7:07
  Maybe You can tell I'm hedging here because I'm not really
• 7:10
  certain that I back Meta to produce both good hardware or
• 7:13
  good software. I think they buy products of
• 7:16
  other people really well and integrate them even better.
• 7:19
  Look at WhatsApp, look at Instagram and how well have
• 7:22
  integrated them into their network.
• 7:23
  But I'm unsure about whether they'll be able to pull off a
• 7:26
  real consumer device. That's hardware and software
• 7:28
  where I combined. I can't help but think that
• 7:30
  surely Apple will come in and do a good job here, even if they
• 7:33
  are fumbling the bag at the moment.
• 7:35
  But Meta's narrative here that they say, for example, 50% of
• 7:38
  their active users have been engaging with the Voice
• 7:40
  Assistant via the glasses. So that's quite an interesting
• 7:44
  play because it's literally putting AI on your face, which
• 7:47
  is one step away from it being in your ear all the time.
• 7:49
  Now, whether Meta can really take over open AI's consumer
• 7:53
  market share, I don't know. And whether Apple can actually
• 7:57
  step in and produce good hardware for new devices and new
• 7:59
  products, I also don't know anymore.
• 8:01
  But I personally don't associate, let's say, a social
• 8:04
  media app with a chat bot to get things done with.
• 8:07
  Maybe they can win within the therapy and companionship space.
• 8:10
  If you never listened to last week's episode, go back and
• 8:12
  listen to that. You'll understand exactly what
• 8:14
  I'm talking about here. But in short, it's a very big
• 8:17
  use case right now, which is people asking for advice on
• 8:19
  their life and purpose. But me personally, I use AI
• 8:23
  typically 70 to 80% of the time on work related tasks.
• 8:27
  So it's a lot of my weekly active usage.
• 8:30
  And I'm unconvinced whether people really can be persuaded
• 8:32
  to turn to a Meta AI app to do those type of tasks.
• 8:35
  But there are a lot of personal use case tasks that I think Meta
• 8:38
  could really own. Again, personal relationships,
• 8:41
  companionship, shopping, all those type of use case that are
• 8:44
  deeply personal and very related to social media.
• 8:47
  They could do quite well on think, but open AI are not going
• 8:50
  to make this easy. I've seen lots of rumours about
• 8:52
  them trying to partner with hardware creators and their
• 8:56
  consumer playbook in 2025 right now is quite clear.
• 8:59
  It's increased the power of each model.
• 9:02
  So using that with the new reasoning models, GPT 4 O, for
• 9:04
  example, Step 2 is increase the ubiquity of those models.
• 9:08
  So they've given those best models to people paying £0 or
• 9:12
  they've produced even more powerful models for people
• 9:14
  paying £200 a month. And then they integrate new
• 9:17
  technology vertically to own more market share.
• 9:21
  So they've integrated web browsing, now, shopping, plug
• 9:24
  insurance because they want ChatGPT to be that one app to
• 9:28
  rule them all in the AI assistant space.
• 9:30
  I'm just going to pause here for a second because 55% of you
• 9:33
  listening right now aren't actually subscribed.
• 9:35
  And it honestly helps so much if you subscribe, review, and share
• 9:39
  an episode into a Slack channel or a Microsoft Teams channel
• 9:42
  with your colleagues or your friends.
• 9:44
  It would literally exponentially grow the podcast and continue
• 9:48
  it's really fast growth. Anyway, thank you and let's
• 9:50
  return to the episode. OK, so the final battleground
• 10:03
  here is the search and shopping space.
• 10:06
  Perhaps the most audacious move right now is how well open AI
• 10:10
  seem to be slowly eating into Google search market.
• 10:13
  There's other companies like Perplexity worth mentioning, of
• 10:15
  course, they're really, really big.
• 10:18
  But compared to open AI, actual distribution and total active
• 10:22
  users, they're tiny. And so for those that don't
• 10:25
  know, Open AI rolled out a shopping feature in Traction BT
• 10:28
  this month, so you can now search for top five best laptops
• 10:33
  for XYZ factors and it will provide no ads and a breakdown
• 10:37
  of the exact laptops it recommends.
• 10:39
  It's reasoning. You can ask it in the
• 10:41
  conversation for then reviews by different people to help you
• 10:44
  decide and then go and buy it or from a single conversation.
• 10:48
  And if the data is anything to go by, it's had a massive
• 10:51
  impact. Open AI disclosed that they've
• 10:53
  had 1 billion web searches made through ChatGPT in a single week
• 10:57
  for the first ever time. Now, it's really important to
• 11:00
  know this represents only 1% of Google's total search in a given
• 11:04
  week. But that's a huge leap for Open
• 11:07
  AI. And they've been telling
• 11:09
  investors in the market that the shift to conversational search
• 11:12
  to buy product is happening faster than anyone anticipated.
• 11:16
  I personally found it extremely interesting.
• 11:19
  I've tried it. I'm not someone that typically
• 11:21
  buys lots of things online. To be honest, I'm not that type
• 11:24
  of e-commerce lover. But I know a lot of people love
• 11:27
  shopping online. And what I did enjoy was being
• 11:29
  able to get reviews on products within the same conversation
• 11:32
  because that's how I typically make decisions.
• 11:33
  I find different products I'm interested in and then search
• 11:37
  for YouTube videos and other reviews to help me make that
• 11:39
  decision. And if all that can happen in
• 11:41
  one conversation really easily and quickly, that's extremely
• 11:44
  powerful. Now, Google, of course, won't
• 11:47
  just roll over. And they're doing a huge amount
• 11:48
  of investment innovation right now.
• 11:51
  They've launched their AI overviews, which have been used
• 11:54
  by over a billion people so far. They've got their new AI mode.
• 11:57
  So if you go to Google on their app, you can go to this AI mode,
• 12:01
  which I personally don't like. It feels like 2 separate
• 12:03
  products and it feels extremely lazy because they've not figured
• 12:06
  out how to integrate them both yet.
• 12:07
  And I think this really speaks to why Google are going to have
• 12:09
  an uphill struggle here because most of their revenue comes from
• 12:13
  people clicking those blue links in traditional search.
• 12:16
  And it shows ACNBC survey found in April that 52% of people
• 12:22
  trusted chat GB TS recommendations more than
• 12:24
  Google's. With many people signing
• 12:26
  Google's as heavy approach as a real issue and driver of that
• 12:31
  mistrust because they know that Google is trying to sell you
• 12:34
  something every time you ask anything.
• 12:36
  And it's pretty obvious that Open AI and Perplexity and other
• 12:39
  companies know this is the case because they're not monetizing
• 12:42
  through ads. And Sam Altman said in an
• 12:44
  interview in May 2025 to last a couple weeks ago, he thinks
• 12:48
  there's a better way to connect users to products than those
• 12:51
  traditional search methods. And using advertising, which I
• 12:55
  personally love, you might be able to tell throughout
• 12:57
  different podcasts that I'm not a massive fan of advertising.
• 13:00
  So I love that this choice puts a huge amount of pressure on
• 13:03
  Google's very ad heavy model. And I suspect when Perplexity
• 13:07
  launch their new web browser, they're going to do exactly the
• 13:09
  same. And the number of these
• 13:11
  companies popping up open AI Perplexity is starting to
• 13:15
  actually take away search market share.
• 13:17
  For example, in May 2025, Apple's senior vice president of
• 13:20
  services, so he's called LEQ Incorp, made a massive
• 13:25
  admission. He said that Apple witnessed its
• 13:26
  first ever drop in Safari web searches compared to the
• 13:30
  previous month, which uses Google and Google Pay a heck of
• 13:33
  a lot of money for that. And Eddy Q explicitly attributed
• 13:36
  this decline to the likes of Chachi, BT and Perplexity,
• 13:40
  adding that Apple is even considering adding these to the
• 13:44
  Apple devices and the search results from Safari.
• 13:47
  And if this partnership between Apple and Google starts to
• 13:49
  falter, it's very expensive and impactful for the ecosystem.
• 13:53
  Google pays Apple $20 billion per year to remain the default
• 13:56
  search engine in Safari and on devices, and that represents 18%
• 14:00
  of Apple's earnings and over 50% of Google search traffic comes
• 14:04
  from Apple devices. So these are huge numbers.
• 14:08
  So stepping back for a second, as I said, I think these real
• 14:11
  battles right now between these trillion dollar companies isn't
• 14:14
  about producing the best coding assistant or chat bot.
• 14:17
  It's really about who owns all the attention and really shapes
• 14:21
  how we interact with technology and information moving forward.
• 14:24
  Anyway, I hope you found today's episode helpful.
• 14:26
  I'm going to follow these stories really closely, so I'll
• 14:28
  see you next time.