I still like both Opus and Sonnet 4.5 for their speed, but if I’m ever stuck again I’ll try GPT 5.2.

Both Gemini 3 Pro and Opus 4.5 was stuck fixing the same area of the code, while GPT 5.2 discovered that to get what I wanted I also needed to restructure a completely different part of the code. (I didn’t use Pro)

Some new features coming in version 3:
- Crop tool
- Audio track routing and deletion
- Timecode override
- VLCKit preview player for files not supported natively by Apple APIs.
- Timecode input/display in the trim preview player
- JPEG XL screenshot support (including alpha channel)
- Audio meter

For context; The initial crop tool that supported square pixel aspect ratios was implemented in like 10 minutes by vibe coding. It worked flawlessly.

It’s only non-square pixel formats that are problematic for me. Still determined to get this to work. Think I’m getting close.

To limit negative effects of errors one needs correction mechanisms. And the correcting mechanisms likely need to be at a similar level of power as the entity that is being checked. If not the correction could easily be ignored or overlooked.

That’s not to say that machines think/reason the same way humans do, but there are similarities. As long as there are similarities I would not expect the output to be without error. Humans (myself included) make mistakes all the time.

Perhaps better integrated sets of narrow AI would be safer.
But the benefit of deeply integrating generalized AI models into an OS —understanding how to navigate visually and through system APIs— is that you would hypothetically never need to make another integration again.

The problems of integration, inconsistent quality, security and privacy will likely be attempted to be solved at the same time. If that’s the case I would limit its use to non-sensitive tasks/systems for the foreseeable future.

Unlike the other two, my name was not mentioned in the news after the deadly crash.

But I suppose there is a chance that Apple will release a new Apple TV device that isn’t called Apple TV, freeing up the name for the streaming service. Perhaps ‘Apple HomePlay’. Signifying that it is both a home hub and a multimodal entertainment device?

To be clear; I don’t know the technical details of how online ID checks happen today.

Maybe make it so the verification center doesn’t know which site the user is registered at, but just has a signed token associated with the ID? And if you needed to legally confirm something you would need a key from both the verification center and the company that requested user verification?

Is it possible to do online ID verification in a way that is privacy preserving? For example through an independent non-profit/state verification center, that confirms if the user fulfills the age requirement or not. To prevent spreading of social security number, sex and exact age.

I’m addicted to the constant flow of information myself, but I try to remind myself that I usually feel most satisfied when paying attention to a single thing for 20 minutes or more. Attempt to not let your focus be stolen.