Gemini Nano. It's in Pixel Recorder and Gboard but multimodal capabilities are coming soon. It's very limited to a couple phones and not many companies outside of Google have access, but it has been used in production since last year.
Apple does not have on device AI in production today. Microsoft's "Copilot+PCs" launched less than 2 weeks ago, but are significantly more powerful machines.
This feature seems entirely pointless, especially for this device (which should just be called the Pixel Fold 2).
Since the device is basically square (resolution 2,076 x 2,152) why would you not just rotate 90 degrees to change the side by side to top bottom like we can today?
Because people don't understand the difference between using ChatGPT running in a datacenter somewhere via API calls and having a tiny model actually running on device.
Even quantized down to 4bit the best 7B models barely run locally on the best mobile hardware that exists today. GPT4 is ~250x larger than that 7B model lol
Yeah, I like the camera bump on my Pixel Fold since it shouldn't go all the way across the back and is still square-ish. These are similar but worse. Pixel 6 was peak design IMO.
Host computer is an old i7 desktop running proxmox. I don't think that's the problem.
I just think a native dashboard in the apps would be better for a bunch of reasons and the Home Assistant project is big enough now to support/warrant it.
Are you are implying that somehow going native gets you faster network transmission speeds?
No. If the components are native they can be loaded immediately without network transmission. The network calls for the current status of entities can happen in parallel and would complete much faster than one large webpage load.
And your standard cards can not be modded with CSS unless you install something 3rd party that allows that.
Pixel Fold. It takes ~2 seconds on gigabit WiFi. Slow 4G networks take a lot longer than 2 seconds. Could be pretty much instant regardless if it was native.
I work in mobile and respectfully disagree with it having no impact on time to show. You might not notice much difference if you are inside your house using gigabit WiFi, but go load your dashboard on a slow 3G/4G network lol
Sure, there is a higher development cost but you get what you pay for in this case. The project already supports native apps for Android, iOS and Mac.
Drag n drop is great but honestly what I want is a native UI on each platform.
I realize that they aren't reinventing the wheel with the UI updates this year so it's not really comparable/fair to say but I just want something faster and more responsive (especially on mobile). It seems crazy to me that in 2024 the fastest hardware available still takes a couple seconds to show my dashboard on a cold boot of the app.
I don't but I have a couple of the presence sensors from the company and like them a lot. They sort of require some tuning but have been quite reliable since.
Damn, I guess I'm the only one with TasmoAdmin installed lol.
Edit: Is the Meteorologisk institutt (weather) integration really used by 77%? I read a couple HA weather threads and didn't see a single person saying they use it.
If you have a zigbee network the aqara leak detector is a good option. It doesn't require a hub, cloud or any setup. And if you are willing to wait you can get them on AliExpress for a lot cheaper than Amazon.
Gemini Nano. It's in Pixel Recorder and Gboard but multimodal capabilities are coming soon. It's very limited to a couple phones and not many companies outside of Google have access, but it has been used in production since last year.
Apple does not have on device AI in production today. Microsoft's "Copilot+PCs" launched less than 2 weeks ago, but are significantly more powerful machines.