Hack:

    long-running-command &
    pid_of_long_running_cmd=$(pidof long-running-command)
    tail --pid=$pid_pf_long_running_cmd -f /dev/null && forgotten-command

Tail quits when $pid_pf_long_running_cmd disappears, and kicks off the forgotten-command. (It won't know however the exit code of the original cmd, so that's a drawback)

Yeah, the problem here is it requires preparation and preemptively adding that '&' to the first command. It doesn't work for the "oh, right" situations. Running all commands this way by default wouldn't make life easier.

But along those lines yeah, it's basically the same as the "wait for it in another terminal" approach and I guess a solution would contain something similar.

So you want to host it yourself, or you want to use bittorrent? The former is how people distributed cat videos before the Youtube era.

The latter is not hosted by you, but by all users participating the swarm. The first page for "bitternt stream online" (sic) keywords in your favorite search engine should should give some ideas about that (I use ecosia now, which gives at least usable results for this).

Well I would seed the torrent and my hope is other people would participate in the swarm at some point. If it was a bittorent user, hopefully they would find it somehow by browsing the list of new videos on piratebay or something...

The goal is for visitors to visit my website read about the video in question, click on the video and get it streamed to their web browser from the torrent itself without a local BitTorrent client.

Personal opinion: 1000 users sounds realistic, but only with a service that appeals the masses, some generic good. I find Japanese learning to be a fairly niche thing, however. Now saying that it's a bad idea, but probably you should readjust your success criteria based on the size of your target market.

    If we delivered a bad quarter, it is evidence there's an AI bubble. If we delivered a great quarter, we are fueling the AI bubble,

While I think the bubble pop will be spectacular, he is not wrong here. You can't do right, if others want to see bad only.

It's a followup on https://news.ycombinator.com/item?id=45960893 from a few hours ago

Thanks! Macroexpanded:

Core Devices keeps stealing our work - https://news.ycombinator.com/item?id=45960893 - Nov 2025 (110 comments)

Every living creature evolves into crabs eventually.

Every software evolves into Electron apps eventually.

Python serves as an established, mature glue language that's relatively easy to learn. But of course even in Python, if some heavy lifting needs to be done, that happens in C/C++, and precompiled as a native binary object, with a Python layer on top of it for the users.

Am I supposed to feel bad for Adobe? The only better news this week was that someone wants to buy EA, with leveraged buyout. I'm not praising today's AI, but some companies just can't go down fast enough.

They tag 25-30 GH accounts in each issue, looking for applications. Apparently they ran out of candidates.

Financial software is the one of the last ones that I want to be written by LLMs, as long as they claim that loading a 10GB file into memory instead of reading it line by line is memory efficient

LLMs can be used to make software more robust and more secure. You can ask the very best LLM like o3 Pro, give it a lot of time, to look for potential security issues and bugs in the code. You can ask them to write extremely stringent unit tests.

And finally, humans should obviously review all the code in any critical system.