re

I don't know where you get "really difficult" and "fighting against macOS" from. You only have to hit the disclosure arrow next to "Where" in the save dialog one time and it remembers it and shows it by default from then on. (Okay, maybe it's once per app? I don't remember for sure, but I can't remember the last time the save panel didn't show me the folder browser by default.)

> Claude Code can decrypt summarized reasoning traces sent by the API.

Can you cite specifically what in the linked article or discussion leads you to say that?

No, it looks like the domain was taken over by squatters after CA went defunct in 2018, and they're currently using it for AI-generated "content".

Like others here, the content is something that I'm interested in, but I too have a reaction to the LLM-isms. I think part of it is that, once I recognize I'm reading LLM output, I have no way of knowing if it wrote the whole post based on a one-or-two sentence prompt or not, and if it did, how much can I really trust the output to not be hallucinated. If I'm going to read LLM output, I'd rather do so in the context of my own chat with it so I can understand the inputs to it and ask follow-up questions.

Can you put the pre-LLM version of your post on your site somewhere? I (and presumably many others) would be much more interested in reading that, even if the English is a bit awkward.

It's much worse than it used to be. Before it was only really a problem with apps with a lot of menus, and you could access the items by switching to an app with fewer of them. Now, the notch takes up a lot of space, and you hit it really soon on a 14" display—I can only have maybe three third party menu applets on top of my collection of built-in ones before they disappear into the notch.

> getsupport.apple.com.phish.xyz

I notice that a lot of scam texts use domains that start with a TLD followed by a hyphen, like:

  https://wa.gov-phish.fit/dol
  https://seattle.gov-phish.cc/dmv

(Real examples, with "phish" replacing a string of 3-4 random letters)

In some ways, it's a more convincing fake URL, since even if you're used to reading the domain right-to-left, your brain wants to start from the hyphen since it's a different character following a familiar TLD. But that type of domain also seems a lot easier for spam detection rules to catch.

Not too sure of the utility of this. It's not an easy sentence to remember, because while grammatical, it's nonsense—it would take some effort. So if I'm trying to memorize a static IP, setting up a DNS name is likely to be easier. And also if I'm going to be using this to memorize IPs I'd like the algorithm to be open source.

All that being said, I think it's a neat idea and a cool tool!

I didn't get that sense from the prose; it didn't have the usual LLM hallmarks to me, though I'm not enough of an expert in the space to pick up on inaccuracies/hallucinations.

The "TRAINING" visualization does seem synthetic though, the graph is a bit too "perfect" and it's odd that the generated names don't update for every step.

> it really makes me doubt that what is written there is actually true at all

Indeed, the whole "Ironically, switching from Apache 2.0 to AGPL irrevocably makes the project forkable" section seems misguided. Apache 2.0-licensed software is just as forkable.

> I wanted to see if, with the assistance of modern AI, I could reproduce this work in a more concise way, from scratch, in a weekend.

I don't think it counts as recreating a project "from scratch" if the model that you're using was trained against it. Claude Opus 4.5 is aware of the stable-diffusion.cpp project and can answer some questions about it and its code-base (with mixed accuracy) with web search turned off.

I have had issues with resizing Quick Look windows with their rounded corners on macOS for the last several major versions, well before Tahoe. The resize cursor indicator there also doesn't seem to appear at the correct location for the actual resize handles.

Because I learned JS before ECMAScript 6 was widely supported by browsers and haven't written a ton of it targeting modern browsers. You're right that it's unnecessary.