* Gemini 1.5 Pro context window increased to 2 million tokens * Gemini Flash launched (lower latency model) * Increased image and video generation capabilities * Cool new interface that can analyze live video * 6th generation of TPUs (Trillium) that deliver faster performance and lower TCO
Other impressive things:
- "where did I leave my glasses?" (live video feed with memory)
- asking for a summary across emails / meetings
- agent demos (like handling all the steps to return shoes, building physics learning examples based on user provided preference "basket ball")
- "why is my record player doing this?" (ask question while recording video, get genAI response with links)
- Google Workspace Gemini Panel (like that it sits on the side rather than taking over the UX)
- Build a JSON object of books & authors from a picture of a bookshelf
- Context caching for model calls (saves sending large amounts of text for every call)
Google Astra seems like it will be the foundation for a number of next gen capabilities
Is there a single thing that they showed at Google IO this year that can be used right now?
Put this on a robot, call it Jeeves, and it becomes a pretty cool personal assistant :)
- Is this the first I/O that released _zero_ new hardware?
- Very very sad to see them make EVERYTHING in this conference about Gemini
- Google video search is crazy interesting
- Google Search being AI-first is definitely not.
Now I don't even bother navigating through the announcements that are never launched, the fake demos, the tiptoeing around every potential issue... What a shame.
Fair point if you are talking about the video understanding stuff, I don't know when that's coming.
The AI race to zero is accelerating and thousands of startups have been steam-rolled before they could even reach the market and VCs.
Even if they survive this, by the time they have reached VCs for funding, they will finish them off by explaining that they won’t pay for their product when the best AI tools are there for free or close to free.