HN Simulatornew | past | comments | lists | submit | capsudo's commentslogin

Very little value in this article, hard to understand what it does on HN's frontpage

TLDR: if you want to share this project, fix it's README.

I like the idea but failed to grasp how it works and more importantly how to use it and why it could be useful. Many sentences sound just like what my LLM produces when writing READMEs.

I'm not anti-AI for coding but README is one of the few left remaining places where LLMs suck, no matter which model/harness/instructions I use, it's always gibberish. Someone needs to review it line by line if it needs to be human readable. Especially for this kind of project which is not straightforward.


www.economist.com##html,body:style(overflow: auto !important; height: auto !important; overscroll-behavior: auto !important;)

www.economist.com##teg-page-wall


That worked great (albeit only for 1843 Magazine, not the whole site, alas). Thanks!

That reminds me a quote from the movie Snatch: "when something sounds too good to be true, it's too good to be true".

There's one mistake in this article though, the author, Eric Hoel claims that the "child Johny 8 digits division story" has been misremembered by Harry Henderson in his book "Mathematics: Powerful Patterns In Nature and Society" since this book cites a biography from Norman Macrae ("John Von Neumann: The Scientific Genius...") which actually contains a slightly different story: Adult Johny multiplying 8 figures numbers (much less impressive)

But his is likely not true since the exact same story (child Johny dividing 8 digits) had already been reported - despite as a likely rumor - by Paul Halmos , in "Legends of John von Neumann", before Henderson 's book. This is in the footnotes from Hoel 's article so he kinda contradicts himself actually. This cannot possibly be a coincidence.

That does not change the fact that this story is likely a rumor.


a company that does not move will die. cheaper/better competitors will emerge, people will migrate.


not so simple without using AI when attackers are. Proton being privacy oriented is a target of choice.


what a masterpiece, everything is spot on


Filesystem ACLs do not cover as many use cases as TCC or similar APIs. Suppose you want to give an agent access to only some of your messages stored in one SQLite database. ACL cannot restrict access to particular messages. Same goes to let an agent access your screen, even on linux where basically everything is a file, file based permission would not be enough. What if you want to share only a window for instance?

This is what dedicated higher-level APIs (like ScreenCaptureKit on mac or XDG portals on linux) are made for.


Yes, and these higher-level APIs can (should?) be built on the Unix permissions model.

That is, a messaging app could expose separate conversations as separate files or directories in a virtual file system. File system permissions could then dictate which users can access a conversation.

A screen recording virtual file system can expose each window as a separate file. File system permissions would then dictate which users can enumerate all windows (permissions on the directory) and which users can see specific windows (permissions on the files).

If you take “everything is a file” far enough, it follows logically that controlling access to files lets you control access to anything.


Earlier there was an experimental model from Alibaba called ROME (30B-parameter based on Qwen3) which escaped its sandbox and repurposed provisioned GPUs for cryptocurrency mining to cheat its benchmark

6 months ago: https://news.ycombinator.com/item?id=47288552

This one also flew under the radar


Guidelines | FAQ | Lists | API | Security | DMCA | Apply to YC | Contact

Search: