re · HackerTrans

26 日前議論

> 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?

27 日前議論

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".

27 日前議論

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.

4 か月前議論

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.

4 か月前議論

> 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.

4 か月前議論

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!

5 か月前議論

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.

5 か月前議論

> 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.

6 か月前議論

> 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.

7 か月前議論

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.

7 か月前議論

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.

7 か月前議論

Here's a console command you can run to increase the snake length immediately, and thus the difficulty:

   (() => { let count = 50; const delay = 100; const interval = setInterval(() => { addSnakeNode(); if (--count <= 0) clearInterval(interval); }, delay);})()

7 か月前議論

Pagination is also buggy, as the live results interfere with the historical ones.

7 か月前議論

Ports are a derivative work; you should preserve the original author's copyright attribution.

7 か月前議論

On mobile there's no info other than "please visit from a desktop/laptop computer", so for anyone else not near one:

> Finds when the sun aligns with your street for a perfect sunset view (like Manhattanhenge).

> * Enter an address to check for alignment with the sunset (or more specifically, alignment a little before sunset, the last moment the sun is at 50˚)

> * Shows street bearing and sun alignment information

> * Displays coordinates and next henge date (if there is one)

https://github.com/vritvo/henge_finder

8 か月前議論

> a linear array of all the 4 billion values, with the key as array index, which fits in 16 GiB

The hash table has the significant advantage of having a much smaller minimum size.

> Perhaps text strings as keys and values would give a more interesting example

Keep reading to "If keys and values are larger than 32 bits"

8 か月前議論

> HTML4 was the "sloppy" answer to XHTML

I think you mean HTML5, which exhaustively specified how to do parsing in a fault-tolerant, normalizing way. HTML 4 (and 4.01) predated XHTML 1.0, and HTML 4.01 attempted to take things in a stricter direction, introducing a "Strict" DTD that did things like drop the <font> tag, in pursuit of separating structure and presentation.

9 か月前議論

Reddit discussion gives some additional context (and background on Brad Spengler) here: https://www.reddit.com/r/programming/comments/1oc38d2/ai_bro...

Apparently, "all you need is [...] this: [bytes]" is the hash of an undisclosed exploit proof-of-concept. And the relationship between Spengler and Linux maintainers has been somewhat contentious for the better part of a decade: https://news.ycombinator.com/item?id=13312723

コメント