Looks like "during and after" part is quite explicit. The relevant parts in question an end-user must be allowed to replace are battery and display.

In theory, as long as the OEM has ANY battery and display provisioning procedure that's relevant to "full functionality", an end-user must be provided with means of unbricking.

https://www.da.vidbuchanan.co.uk/blog/android-c2pa.html

C2PA (the your-own-photo DRM that doxes you and requires locked bootloader to work) finally got the well deserved treatment.

Turns out you're actually covered. So the remaining problem with Minimal is mostly that you may want more DR or a bit more filmic rendering, but at least it provides reasonable noise patterns and minimal/zero LTM.

As iPhone 18 Pro (Max?) has a variable aperture, I wonder if they'll automatically shrink it to allow the clueless to shoot document photos while being way too close... I've heard too many times that "camera got worse after updates" in this aspect, even though it's just physics. Yes, you should use telephoto for the clearest looking document photos otherwise, just mind the motion.

https://fixupx.com/johnternus/status/2097672038781763614

Major warning: iPhone Duo is basically a foldable version of (currently not existing) base iPhone 18, it's not considered a Pro even for the pure software features that have no business being locked and lack a form factor justification. Contrary to speculations, I'd say it's unlikely that it's the sensor from 17e, even though "Hybrid Focus Pixels" are an odd sight. Unless 17e was artificially gatekept from 4k120 (rather than due to inability to offer a staggered/DCG/DAG sensor mode at that frame rate and skipping it because of it), the sensor on Duo is twice faster.

Regardless, it's an odd sight to see a 48~50MP sensor that isn't 4PD on a product so high-end. Not because I'm a fan of using 12 million microlenses on 48 million photosites, but because it seems very desirable to device vendors, thanks to sole amount of phase data they provide for PDAF. This at least does suggest that it's a smaller format than 18 Pro. EXIF will later tell us the crop factor and therefore the sensor format.

Apple Watch 12 offers a currently opt-in feature that basically uploads the conversations it captures into the cloud.

I know many people are concerned about completely offline hardware trying to sense you 24/7 for something convenient that you may want. For example, Mi 17 is capable of waking up the screen when you specifically move your eyes at it, likely thanks to a low power front camera sensor mode and Qualcomm's support for that. I liked it, but it probably eats battery.

That's one thing. But *uploading* constantly sensed data from people who never gave you consent for it, especially without homomorphic encryption (that was used with Enhanced Visual Search and still managed to break people's trust when it got enabled by default in update), is simply way too much. I've seen an article from TechRadar that completely omitted the important part: it is an online feature based on servers attesting their legitimacy. For something meant to constantly listen.

FAQ:
- Yes, it's currently less likely that this data gets anywhere near advertisers and government agencies compared to conversations with Google bots.
- Yes, it's much more invasive, as you don't have to wake it up every time and the whole premise is about listening not just to you, but also to people who are nearby enough.
- Yes, an attested cloud is not an extension of your device. It's still moving your own and others' data out of your offline control, justified only by claim that the cloud holders implemented technical roadblocks for extracting this data. And so they did. But would they sign a contract that bankrupts the company in case of a sufficiently advanced or powerfully sanctioned breach? No.


Tldr: don't assume you're talking in private if you see an Apple Watch 12 on someone's wrist.

I know that most of people are either logged in or swear by not using AI in any form, but to whom it may concern, the login-less ChatGPT got upgraded to GPT-5.6 Luna and it's now capable of spending over a minute on your tasks. If you're somehow still using Google search, this is a much better way to find any information that wouldn't be your first result.

Of course, there are means of getting access to better models anonymously, but this is quite a leap regardless. Two years ago it'd be a jaw-dropping model you'd need to pay solid money for. I think it's additionally interesting that the industry started talking about intentionally slowing down, right when I'd expect it to do that naturally (unless they move on from the LLM path). That makes free, login-less access comparatively much more valuable.

It is faster in the end tho