Google goes wild, again... 11 things you missed at I/O

Fireship www.youtube.com

Kurze Zusammenfassung

Das Video fasst die wichtigsten Ankündigungen von Google I/O 2025 zusammen, darunter die Vorstellung von Gemini 2.5, das andere KI-Modelle übertrifft, und Gemini Flash. Es werden neue KI-gestützte Tools wie Stitch zur automatischen UI-Generierung, Jules für die KI-gestützte Codeerstellung und Flow zur Erstellung filmischer Inhalte vorgestellt. Weitere Highlights sind Project Astra für die Echtzeit-Umgebungserkennung, Project Beam für die 3D-Umwandlung von Videos und Verbesserungen bei Google's Video- und Bildgeneratoren. Abschließend wird die Einführung neuer CSS-Primitive für Karussells hervorgehoben.

  • Gemini 2.5 übertrifft andere KI-Modelle.
  • Neue KI-Tools für UI-Generierung, Codeerstellung und Filmerstellung.
  • Project Astra und Beam für verbesserte Realitätserkennung und 3D-Umwandlung.
  • CSS-Primitive für Karussells vereinfachen die Webentwicklung.

Googles neue KI-Modelle und Abonnements

Google hat Gemini 2.5 Deepthank vorgestellt, das alle anderen Modelle übertrifft, sowie das günstigere Gemini Flash, das ebenfalls beeindruckende Leistungen zeigt. Um die volle Leistung von Gemini nutzen zu können, ist das neue Google AI Ultra-Abonnement erforderlich, das zunächst 124,99 $ pro Monat für 3 Monate kostet und danach auf 249 $ pro Monat steigt. Im Gegenzug erhält man Zugriff auf die volle Leistung von Gemini und 30 TB Speicherplatz. Die offenen Gemma-Modelle ermöglichen es Entwicklern, eigene Produkte zu entwickeln und damit Geld zu verdienen, wobei die Leistung von Gemma 3N an Claude Sonnet 3.7 heranreicht.

KI-Agenten im Browser und neue Tools

Gemini wird bald im Browser verfügbar sein und mit dem neuen Agent-Modus Aktionen auf Websites ausführen können, wie z. B. das Klicken auf Schaltflächen und das Ausfüllen von Formularen. Tools wie Stitch generieren automatisch UIs, während Jules KI-gestützte Agenten codiert. Google unterstützt das Model Context Protocol, was die Integration dieser Technologien für Entwickler vereinfacht.

Project Astra und Android XR Brillen

Project Astra nutzt einen Live-Video-Feed und KI mit niedriger Latenz, um die Umgebung zu beschreiben. Android XR-Brillen sind real und werden diese Technologie integrieren, um mit Produkten wie den Meta Smart Ray-Ban-Brillen zu konkurrieren. Project Beam wandelt 2D-Videos mithilfe von KI in 3D-Erlebnisse um.

Verbesserungen bei Video- und Bildgeneratoren

Google's Video- und Bildgeneratoren VO und Imagen haben bedeutende Verbesserungen erfahren. Es ist nun möglich, Videos mit Ton zu generieren. Diese Technologien werden in dem neuen Tool Flow zusammengeführt, das sich an Hollywood-Studios richtet und möglicherweise in Zukunft für die Erstellung von Inhalten auf Plattformen wie Netflix verwendet wird.

Neue CSS-Primitive für Karussells und Hostinger als Sponsor

CSS verfügt über neue Primitive zum Erstellen von Karussells, wodurch der Codeaufwand im Vergleich zu jQuery deutlich reduziert wird. Hostinger, der Sponsor des Videos, bietet Hosting-Lösungen und virtuelle private Server (VPS) an, auf denen man eigene Server mit Tools wie Koolifi konfigurieren kann, um beispielsweise Next.js zu hosten.

Fasse jedes YouTube-Video in Sekunden zusammen

Hol dir Zusammenfassungen wie diese für jedes Video, in deiner Sprache, mit der kostenlosen Summ-App.

Laden im App Store
Jetzt bei Google Play

Transkript

Vollständiges Transkript von „Google goes wild, again... 11 things you missed at I/O“. Klicke auf einen Zeitstempel, um das Video an dieser Stelle zu öffnen. Du brauchst ein anderes Video? Nutze den YouTube-Transkript-Generator.

0:00 Yesterday, Google unleashed the most intense barrage of AI hype bait, Trust Me Bro, benchmarks, and paywalled vaporware the world has ever seen, and it was awesome. Gemini 2.5 Deepthank crushed all other reasoning models like 03. And even the cheap Gemini Flash model was able to beat off 03 from the top spot. In the Chrome browser, Gemini will soon have the ability to navigate and use websites for you. Meanwhile, this new tool called Stitch will automatically generate and design UIs, which means you're really screwed if you're a web designer. AI is going to be designing websites for AI agents to use in the AI powered browser. Meanwhile, Jules is an async agentic coding agent that will code your AI agents, making coders obsolete yet again. I'm afraid of no man. But this new AI tool called Flow scares me. This thing could actually make loweffort faceless edge lords on YouTube obsolete by generating entire

0:47 cinematic adventures with AI. But that's just the tip of the sloppberg. In today's video, we'll break down 11 gamechanging things you need to know from Google IO 2025. It is May 21st, 2025, and you're watching the code report. When I went to watch the keynote, I saw that they had turned off comments, and it's obvious this cowardly move is because they already knew what the top comment would be. And that's why we're going to talk about the worst thing first, Google's new AI Ultra payment plan. I'm already on my second second mortgage to afford all my AI subscriptions. But now to get access to the full girth of Gemini's Power, you need the new Google AI Ultra subscription, which is available at the low low price of $124.99 per month for 3 months, then $249 a month after that. And if you buy right now, they'll also throw in 1 month free and an absurd 30 TB of storage for all your generated

1:32 slob. But that might be worth it because Gemini is now the best across all LM Arena categories. Not surprisingly, it completely crushed its own Trust Me Bro benchmarks and the vibes are strong with this one. And Gemini also managed to be Pokemon on its own. But what might be more exciting is the release of Gemma 3N. Gemma models are open. So as a developer, you can use them to make money for your own products. And the performance of this one is approaching Claude Sonnet 3.7. Now, as I mentioned before, Gemini is also going to be coming in your browser soon. And its new agent mode, formerly Project Mariner, will allow it to take actual actions on websites like click buttons, fill out forms, and automatically generate your opinion to post a comment to a Fire Ship video. At the end of 2024, I told you that 2025 was the year of agents. OpenAI just launched the codeex agent. GitHub copilot launched agent mode and now

2:18 Gemini has its own creatively named agent mode. And agents will be used in tools like Stitch and Jules to automatically generate UIs, code them, and interface with tools like Figma. The best news though is that Google supports the model context protocol, which makes it much easier for developers to integrate this tech. That's surprising because Google would normally go off and invent its own architecture that nobody asked for and then kill it a few years later. But one of the most impressive things at IO is Project Astra. It takes a live video feed and uses low latency AI to describe the world around you. I've been using it to identify which mushrooms are edible in the wild, and it's actually pretty useful. We also found out that Android XR glasses are real, and eventually this technology will be integrated into them to compete with things like the Meta Smart Ray-B bands. But an even more futuristic announcement was Project Beam, a tool

3:05 that takes 2D video and uses AI to transform it into a 3D experience. That'll make your daily stand-up meetings much more realistic, and I think it could be a gamecher for one other industry that I can't mention in this video without being demonetized. In addition, VO and Imageen, Google's video and image generators, also got huge upgrades. And what's crazy is that you can now generate video that also has sound. And she commands your awe with every breaking light. And this is all being put together in a new tool called Flow, which is targeted at Hollywood Studios and someday might be used to generate all the content that you binge on Netflix. And that brings us to the final thing you need to know, which is absolutely insane and crazy, and I can't even believe I'm saying this, but CSS has new primitives for building carousels. Instead of 10,000 lines of jQuery, you can now build something like this with just a few hundred lines of

3:51 CSS. And once you've done that, you'll need a server to deploy it to, like Hostinger, the sponsor of today's video. Not only do they provide fully managed hosting solutions, but also virtual private servers where you can deploy anything. For under 10 bucks per month, you could be running your own server with predictable pricing and a respectable two CPUs and 8 GB of RAM. In fact, when you create a server, you can automatically configure it with tools like Koolifi, which makes it possible to host Nex.js on your own VPS with minimal pain. If you're looking for freedom and an awesome developer experience, check out Hostinger with the link below. This has been the code report. Thanks for watching and I will see you in the next

Share

Weitere Video-Zusammenfassungen