The original dmoz/open directory project/Yahoo directory database is being hosted (but not maintained; I don't think it's being updated) at odp.org, which also includes links to other directory projects at http://odp.org/Computers/Internet/Searching/Directories
I've seen it mentioned somewhere, probably here on HN, that the original thin scrollbar was implemented in Gnome, and that it only applied until you moved your cursor to it. If true, this sounds like a great innovation. Unfortunately, nobody bothers to expand the scrollbar when you get near it. And of course, scrollbars also didn't start to disappear until long after the square-ish monitors that made them so annoying were mostly replaced by widescreen monitors that didn't have that issue.
Anecdotally, I do recall using an extremely obnoxious viewer application (PDFs or ebooks or some such) on the Windows 8 store that reversed scroll entirely. The scrollbars were still there (though I can't remember if they had handles), but scrolling the mouse wheel down moved the document up.
The best UI is no UI--the program anticipates what you need and does it when you need it. Unfortunately, this is impossible in practice for anything other than the simplest systems (eg. pause music playback when the phone rings).
MDI in browsers isn't enforced though, you can still use each tab in it's own window, and (depending on your browser) you can open other sites in a sidebar (? Used to be possible) or split-tab (I think I've seen this in some desktop browser? I know chrome does it on mobile). Office documents though? SDI is now enforced on office. You can't hide that taskbar entry for every single document you have open, even though there's only 2 important ones each open in their own window and the rest could all share a window for all you care.
MDI wasn't a failure, the requirement was a failure. SDI requirement for office is also a failure. More so than the escapable (multiple window support) MDI of browsers on desktop.
Considerate? Yes, I'll consider leaving you alone. If it seems that you're having an actual problem, I'll even leave you alone. But if you're being rude because you haven't considered that others might not be interested in listening to your tiktok videos repeat endlessly, then when I take that into consideration it will likely have a negative effect on how I decide to treat you.
I'll also take into consideration whether or not you appear to be homeless if you stink (choosing not to shower and deciding not to shower are two different things). I'll take into consideration who it sounds like you're talking to when you're telling on the phone ("are you ok" vs interminable conversations on speakerphone in the break room/restaurant/other public places). If you're playing loud music in your car, before calling you out I'll consider whether the doors are open (you might not realize) or if you have the windows open (sharing your music choices with the world).
Consideration is the entire point. It doesn't mean letting people do whatever they like, it means judging (thinking about) then before doing anything. You have every like right to play your video, and I have every right to get annoyed at you for doing it without being considerate yourself.
Considerate:
1. Having or showing regard for the needs or feelings of others. synonym: thoughtful.
Similar: thoughtful
2. Characterized by careful thought; deliberate.
3. Given to consideration or to sober reflection; regardful of consequences or circumstances; circumspect; careful; esp. careful of the rights, claims, and feelings of others.
GPT2 at 774m is considered a LLM. I wouldn't say there's much difference between that and 700m, or even 123M.
Having said that, looking up small language model these days returns tons of results calling 7B models small language models.
------
My understanding of small language models is that they're generally intended for specific purposes, like analysis and classification (whatever you'd call the text equivalent of image interrogation with clip models), translation, etc; that there small because they don't need to be big to do their intended functions, not because they're just smaller versions of bigger models.
When I hit Win+L to lock my screen and come back 4 hours later to input my pin, I turn on my monitor (that I turned off because every 5 minutes Windows turns it on and off again), push esc or Ctrl a few times to clear off the image, and start typing in my PIN. 90% of the time by the time my monitor displays the picture, it's sitting at the unlock screen with the last 2 digits of my 4-digit PIN
When you barely need said device, or you occasionally need many of said device. Not to mention you're renting more than just a GPU from them, you're getting the rest of the server too.
I rent from runpod a handful of days a year (maybe a week or so total?) and I've estimate that I've spent at most $300 total in the last 2 years.
- I get a 3090 or 4090, sometimes a 5090 or a pro GPU with 40+gb vram.
(I have no dGPU).
- Usually they have >100gb ram (my new PC has 64gb, the old one had 32gb)
- 20-30 cores available (my new PC has 20, old was 4gb ddr2(!))
- 1-2tb of disposable nvme storage (my old PC didn't have nvme at all, my current PC has ~800gb free of 2tb)
- fast network. I don't always test it, but I'm usually done downloading ~700gb of models in under 2 hours; locally this would take around 15 hours.
- electricity is far cheaper in random east European/Asian City than it is in Los Angeles. Some GPUs cost less to rent than they do to run locally, never mind the cost of buying the GPU itself.
Paying runpod or vast or just about anyone else is _far_ cheaper than buying this hardware myself.
Some of it may be unintentional. I've created issues directly through the `gh` CLI before, and issue templates don't show up there at all. If I haven't actually cloned the repo (or, more frequently, if I have a sparse checkout/deleted the files to save space), I have no way of knowing that there's a template and I'll just `gh issue -R some/thing create`.
I finally moved on to the official GitHub app on mobile, but before that I used fasthub and other clients that had no idea about issue templates.
GitHub really needs to add permissions to issues, so that users can't create issues without the template; any kind of failure in creation is a sign that you're doing something wrong. The ability to add tags to issues when creating via CLI would also be helpful.
reply