12 hours with iOS27

1. Liquid Glass
   1. Compared with the earliest glass, it has basically turned straight into a big piece of plastic. Readability has improved, but it is no longer glass.
   2. The optional transparent/non-transparent sliders do not feel very different in use. I can only say they broke it.
2. Siri AI (note: according to the official wording, the current Siri AI is not the full-powered version)
   1. For things it cannot do, it simply does not answer, instead of saying “I don’t know”.
   2. It still cannot call its own search engine, and instead just pops up a search.
   3. Voice interaction feels unnatural in silence detection and interjection/interruption handling. Siri repeatedly says things it has already said before.
   4. The Siri app may fail to save history records sometimes (bug).
   5. Siri’s voice/voice profile currently cannot be customized.
3. Photos AI editing
   1. Image expansion can only extend a tiny bit around the edges. It cannot expand just anything, and the model recognition itself has some issues. It does not look like Banana 2’s expansion capability.
   2. Reframing has very little room for horizontal movement, and it also cannot shrink the subject. It can only expand the reframing. It uses depth information to create a basic depth-based 3D model (note: this is not a 3D model), then moves that model to recompose the image.
4. Scheduling optimization
   1. I cannot feel it getting faster. After all, it is a beta version, and it is laggy by itself.
   2. Search optimization keeps indexing, so I cannot test it yet (to be added later).
5. Small changes
   1. The keyboard feel is different, with slightly adjusted haptic feedback.
   2. Spotlight search can now ask Siri directly, but it does not record the conversation.
   3. Wallet payments are more lively.
6. Bugs
   1. When searching in Apple Music, the top bar shows strange padding.
   2. The Siri app may fail to save content sometimes.
   3. Saving a screenshot does not save the edited content.
7. Other
   1. Battery drain is worse than iOS26 Developer Beta 1 (possibly because of index rebuilding) (about 7% per hour on a 16PM, battery health 91%).
   2. Smoothness is not much different from the official iOS26 release (is scheduling doing the work?).
   3. There is slight heating, probably because of index rebuilding.

So can anyone explain why Google’s system elements are plastic?

Can I say Google really sucks?

Leaving aside that Gemini 3.5 Flash’s context understanding is far worse than 3.1 Pro,

as an updated version of veo 3.1, omni should at least be stronger than 3.1, right? But on world understanding, veo 3.1 absolutely runs circles around omni.

ChatGPT CodeX GPT-5.5 xhigh Fast weekly token limit is around 3.2 billion

(tested personally)

Because OpenAI is just that good, I seriously bought x20 Pro.

Then I found that Codex seems able to make a lobster. The reason I did not keep updating my experience with that lobster (OpenClaw) in real time before

was that the lobster deleted itself.

All I can say is that it was very 🐮, and there was no follow-up at the time.

Now I have used codex cli myself to make another framework similar to that lobster.

Trying it out.

From Antigravity shrinking in February to yesterday’s GitHub Copilot billing change.

I still think OpenAI is the one…

Gemma 4 has been released, let’s test it out.

Sonnet 4.6 is so silly, it’s ridiculous. Sonnet 4.6 has followed in ChatGPT’s footsteps.

And to be honest, I still feel that the original Claude 4.5 is the strongest 🤔