Google & Claude: The AI hype trains are taking different tracks!

00:00:00: I'm recording this at the end of a week that was packed

00:00:03: with AI-related launches. And you're right,

00:00:08: every week is packed with AI-related launches.

00:00:11: It can be rather annoying, but this week we had Google I.O.,

00:00:15: and therefore Google gave us tons of AI stuff, and we'll dive into

00:00:19: that, and we had the Anthropic event

00:00:23: where they launched their Claude 4 models, and that was also

00:00:27: pretty interesting. Also because we can clearly see that

00:00:31: these companies are following totally different

00:00:35: strategies, and we can see that the models developed by

00:00:39: these companies focus on totally different things.

00:00:42: And in general, we might therefore see that we maybe

00:00:46: don't get AGI, general intelligence, through

00:00:50: those LLMs anytime soon. Instead, we might get a lot of

00:00:54: specialized products and models, and that's a

00:00:58: rather interesting future, and in my opinion, the more

00:01:02: likely future than having that, uh, all-knowing

00:01:06: super AI. But we'll see. And I actually wanna start

00:01:10: with Anthropic and Claude 4 because when I'm recording this, it

00:01:13: was just yesterday that they launched this all.

00:01:17: And they launched two new models, two new versions

00:01:21: of the Claude 4 model, so to say, and if you watched their

00:01:24: livestream or if you read their announcement blog post, which of course,

00:01:28: like many other sources, is linked below this episode, then you'll quickly

00:01:32: notice that they're only talking about coding.

00:01:36: Their model is only about coding, or their

00:01:39: models are only about coding, and that seems to be

00:01:43: the niche. It's not really a niche, I guess, it's one of the

00:01:47: things these, uh, large language models can do best, but

00:01:51: it's their thing they're specializing on clearly.

00:01:55: And it was kind of obvious, I guess, because with Claude

00:01:58: 3.5 and 3.7, though that model

00:02:02: had some problems, but with those older models that were released a

00:02:06: couple of weeks or months ago, we could already see that they were primarily

00:02:10: strong in generating code. And in general, these large

00:02:14: language models are decent at generating

00:02:18: code, at least up to a certain point, and I'll also get back

00:02:22: to that. So it's no surprise that Claude 4 is all about code and that

00:02:26: they only mention (laughs) that in their announcement essentially, and their

00:02:30: Claude 4 seems to do really well. Now, I

00:02:33: personally haven't used it a lot. I played around with it a bit.

00:02:36: It's available through Cursor and VS Code, for

00:02:40: example, but not so much in many other tools,

00:02:44: and we'll see how well it does because I do remember that

00:02:48: when Claude 3.7 was announced, it also was super amazing and

00:02:52: all these AI influencers on X hyped it up like they do with every model, of

00:02:56: course, and then we all found out, well, it's not that great.

00:02:59: It kind of does random stuff, it does stuff you don't want it

00:03:02: to do, it changes random parts of your code, and we'll have to

00:03:06: see if Claude 4 is any better there. But what's interesting is that

00:03:10: Anthropic is clearly becoming that code large

00:03:14: language model company. They want to win the code

00:03:18: generation game, which is a difficult game to win because it's not

00:03:22: like they're the only model creator that wants to win there, but

00:03:26: it seems to be their thing. And of course, the general

00:03:29: question is what's the end game for large language

00:03:33: models and code generation? You could of course

00:03:36: argue that in a year or so, no

00:03:40: code or almost no code will be written by human developers, and there

00:03:44: have been some CEOs that followed that

00:03:48: reasoning or that shared, uh, that idea, that, uh, way

00:03:52: of thinking. I don't agree at all, and I think it's total

00:03:56: nonsense to assume that this will be the future

00:03:59: anytime soon. What's clear, of course, is that these

00:04:03: models can be great assistants, and I've said that for, for months now.

00:04:07: I've been using these tools, like probably all of you, for months now

00:04:11: as assistants, no matter if that's GitHub Copilot and the GitHub

00:04:15: Copilot Agent mode in VS Code or if it's Cursor or

00:04:18: Windsurf or any other extension or IDE.

00:04:23: These models can be really good at that.

00:04:25: By the way, not just when you're using them integrated into an IDE

00:04:29: or code editor, but also as a partner for having

00:04:32: discussions and for evaluating different approaches or

00:04:36: architectures when you're planning a project, when you're setting up a more

00:04:40: complex project. That's what I'm doing a lot, uh, I'm using Google

00:04:43: Gemini 2.5 Pro all the time for discussing

00:04:47: different strategies, different ways of implementing certain

00:04:51: features in a Go application I'm currently building which I'll hopefully launch

00:04:55: soon. So I'm doing that all the time, and

00:04:59: that's where they're really, really valuable, and of course, that will

00:05:03: stay that way in the future and they will probably become more valuable there.

00:05:07: But when it comes to generating all the code and

00:05:11: building more complex applications and not getting lost along

00:05:15: the way, when it comes to these things, sorry, that's, that's just

00:05:19: not going to happen anytime soon, that's totally

00:05:23: nonsense in, in my opinion. But of course, for building individual

00:05:26: features with help of AI, they're amazing and that's going to stay that way and you should use

00:05:30: them for that, in my opinion, uh, you should definitely at least give them a try.

00:05:34: And I've mentioned it before, I got a course on how to

00:05:38: efficiently use, uh, AI tools as a developer if you wanna learn

00:05:41: more about that, I'll link it below the video.

00:05:43: I also got some other AI courses which you'll also find there, but no matter if you take the course or

00:05:47: not, give those AI models a shot.... for using them, uh,

00:05:51: as, as assistants when writing code.

00:05:53: And that seems to be the Claude thing, though of course they

00:05:57: clearly, uh, have a different goal. They wanna replace

00:06:01: developers in the end, I guess, because they definitely

00:06:05: wanna, uh, keep on pushing and build models that are

00:06:08: capable of building entire projects on their own or with

00:06:12: almost no human input, I guess. And we'll see how that

00:06:15: goes, uh, and we'll have to see how good Claude 4 is.

00:06:19: But yeah, here it is and Anthropic is clearly

00:06:23: the coding large language model company for now and we'll see

00:06:26: if that changes again in the future, but that seems to be their thing.

00:06:31: Google, on the other hand, is totally different.

00:06:33: They also had their keynote and lots of other sessions

00:06:37: at Google I/O this weekend. It was all about AI, as

00:06:41: you might imagine, and they announced lots of

00:06:45: initiatives, research projects, a new super

00:06:49: expensive paid plan that gives you early

00:06:52: access or exclusive access to certain models and features,

00:06:57: uh, if you're in the US. Uh, us peasants from

00:07:01: other countries are not included for now, but if you're in the US and

00:07:05: you're willing to pay 250 bucks per month, after

00:07:08: the first three months where it's a bit cheaper, uh, then you get all that

00:07:12: extra access. And clearly Google is

00:07:16: trying to achieve a more general form of

00:07:20: intelligence. They're going all in on agents.

00:07:23: They announced a couple of research projects, specifically

00:07:26: Astra and Marina, that are all about giving you,

00:07:30: uh, agentic features no matter if it's integrated

00:07:34: into Gemini or if it's on Google devices, like their

00:07:38: Google smartphones, the Pixel devices, uh, which should be capable

00:07:42: of helping you on all kinds of tasks without a lot of input from

00:07:46: you. So they're definitely going all in there.

00:07:49: They announced View 3, uh, a video generation

00:07:52: model which looked pretty amazing. Imagen-4, their image

00:07:56: generation, uh, model which, uh, should be pretty good,

00:08:00: though (laughs) I just tested it before recording this and it failed to

00:08:04: create an illustration, but what can you do?

00:08:06: Uh, maybe it was also the old model. I'm not even sure which model I'm using there, to be fair.

00:08:11: But yeah, they announced a lot and I will say the Google Gemini

00:08:15: models, the 2.5 models, are pretty amazing.

00:08:18: I'm using them all the time, as I mentioned before, for discussing

00:08:21: how to implement certain features, how to tackle certain problems.

00:08:24: And of course also outside of coding, because unlike Anthropic,

00:08:29: Google clearly does not have a goal of only winning the coding

00:08:33: game, though like all AI companies, it is a lot

00:08:36: about coding and code assistance and coding agents, because

00:08:40: that's like the quick win, right? Large language models ultimately are

00:08:44: about generating text. Code is text, so that's there for a

00:08:48: natural fit and where a lot of work flows into.

00:08:52: But it's not just about that. As you can tell by the

00:08:56: benchmark numbers Google publishes, it's not all about

00:08:59: code. Um, and about... In general, their messaging.

00:09:03: It's about building those agents and about helping you with all kinds of tasks.

00:09:06: Of course, Google being Google, they're also trying to

00:09:10: integrate AI even deeper into their search.

00:09:14: Ultimately, search may just turn into a

00:09:18: chat in the end, like an AI chat where you can still search for

00:09:22: things, where you can still ask your questions, but then you'll get

00:09:26: a, a result, an answer, which of course will include links to

00:09:30: sources, but you will not primarily get those links as

00:09:34: Google worked in the past and still does to some extent today.

00:09:38: Instead you will get these summaries.

00:09:39: And that of course has interesting implications for

00:09:43: people creating content, writing news articles, blog posts,

00:09:47: you name it. Because if you're just feeding

00:09:52: the Google AI or other AIs for

00:09:55: their summaries and people never visit your website,

00:10:00: what's the point of writing blog posts and news articles in the first

00:10:04: place, right? So that will be an interesting future.

00:10:06: Um, ultimately, we may have AI summaries of content that was

00:10:10: also generated by AI and I'm not sure if I'm liking that future.

00:10:14: But anyways, that's where we're going, and that is what Google also

00:10:18: announced. They're adding more AI integration into

00:10:21: Google Search. They are adding this AI mode there where you

00:10:25: essentially have this chat mode. Um, so an, an

00:10:29: open, uh, AI chat without the need for a sign-up and

00:10:33: anything like that. And it of course makes sense because they can't just let,

00:10:37: uh, OpenAI and other providers eat their lunch with their models, but

00:10:41: of course it's an interesting challenge for Google, and

00:10:45: ultimately as mentioned, also for content creators.

00:10:49: Now, of course we could dive into all the details of

00:10:53: the different capabilities and projects and models that were

00:10:57: announced. You probably already heard a lot about them, and it'll

00:11:00: be interesting to see which of these projects

00:11:04: become reality, anyways, because announcing a

00:11:08: lot of stuff is one thing. Having good products in the end

00:11:12: is a totally different thing, and that's been the problem with

00:11:16: AI for the last two years. There's been a lot of hype.

00:11:19: There have been a lot of announcements.

00:11:21: And don't get me wrong, there have been a lot of great

00:11:25: products and services and capabilities and we're really

00:11:29: living in an amazing or a scary world, depending on how you look at it,

00:11:33: when it comes to AI, but there also have been a lot of

00:11:37: broken promises. I've heard lots of CEOs tell me

00:11:41: that, uh, developers will be replaced in six months, 12 months,

00:11:45: and that was two years ago. So lots of things

00:11:49: being said. You should not take it all that

00:11:53: seriously, but of course it'll be interesting to see, uh, where we'll be in a

00:11:57: couple of months from now. And we definitely got

00:12:00: interesting models, great coding assistants as mentioned, great

00:12:04: models for doing research and all that fun stuff.

00:12:07: And yeah, that's essentially what happened, I guess not just

00:12:11: this week but what also kept on evolving this

00:12:14: weekend. Therefore, please definitely let me also know what your

00:12:18: thoughts are about this entire development, about this week specifically,

00:12:21: AI in general, and what you think where we'll be

00:12:26: in, let's say, six months from now. I'll of course keep you updated

00:12:29: with my thoughts and so on, and um, wish you a great time. Bye.