I think it has more to do with the fact that there has to be some term to describe the concept of "model which has weights that are openly accessible"
Like, that's just a logical thing to call it. I don't believe anyone is making a judgement on the intelligence of the reader to call it "open weight" when it refers to weights that are openly available.
"Open source" would be a more appropriate term to describe a model which also includes the training source.
> People have tried to make businesses out of this stuff but it has rarely worked
The only two startups I'm aware of who have managed to make compchem-as-a-service a viable business are Mat3ra and Materials Square. Both are a little rough in their own ways (to be expected from startups), but it's still interesting to see some of this move beyond university clusters.
> it appears computational chemistry is stuck in batch processing era
I wouldn't call it "stuck," since it's one of the few fields where batch processing makes sense.
You generally run your jobs on a supercomputer that's being shared by a few hundred other people. You need a scheduling system to allocate resources (e.g., 5 jobs requesting 10 nodes for 8 hours, 1000 jobs requesting 0.5 node for 5 minutes, etc).
It's hard to tighten the workflow when the average calculation takes 2-3 hours of time.
While they can, drones are still limited by physics.
Sure, the aircraft can pull a ton of G's to defeat a missile or achieve a firing solution. Maybe it even succeeded at that.
It still pulled 9+ Gs and has burned an incredible amount of kinetic energy doing do. And the harder it pulls, the quicker it bleeds speed. It might not be able to defeat the next missile or outmaneuver the other plane's buddies because it is now slow and very likely outside its optimal turning speed.
Even today, missiles can pull dozens of G's yet still are defeatable by humans who can only pull about 9.
They also succeeded in the monumental task of making VR look boring.
VR platforms are an escapist's dream: you can be anything you want doing whatever you want. And how did they show off their fantasy world machine? They did office meetings in avatars of their real life selves.
Just spend one night in VRChat and everything Meta did will look like Plato's cave shadows.
Any performance gains caused by the internal bandwidth of the card will evaporate once you spill into system RAM, because now your bottleneck is probably a slow PCI lane.
And if your jobs do fit onto a 24GB card, then you are not the target user for the "AI mini PC" niche that these guys are trying to carve out
Exapunks and TIS-100 were a huge influence on my career trajectory.
I was always scared of assembly and low level stuff as a kid / college student, who mostly was trying to learn from random sites that assumed a lot of CS background.
Even though they're not near the complexity of x86, these games made me realize that assembly isn't really that scary. I still don't daily drive x86, but they gave me the confidence to go through a few Advent of Code and Project Euler problems. Having a really stripped down assembler was a fantastic learning tool!
Without them, I'd probably still only be working in Python (which is a great language, but abstracts a lot)
It looks like git after 2.22 was dropped because it took an LLM commit. Same with ghc.
If I have to choose between this or git and the latest ghc, I think I'm going to just wait for someone to fork annex.
I don't even feel strongly one way or the other on AI stuff; pragmatically, I'm just not going to stop using the most widely used version controller, or Haskell, just for some guy's (forkable, AGPL licensed) hobby project.
> This will probably prevent git-annex from taking advantage of most new improvements to the Haskell language going forward. That is deeply unfortunate. This is the main reason why git-annex is not guaranteed to never change to depend on LLM generated code, because cutting it off from all future Haskell language improvements may be worse than the alternative.
Looks like they are aware, and git-annex has been around for decades written by one of the best Haskellers. “Some guys hobby project” is not fair
UX work is often also a lot heavier and more subjective than the plumbing.
I might open a pull request to support some new video code, and that might only require a few dozen lines over a few files. That's easy to review, and it either works or it doesn't. Worst case they say "our convention is to register codecs as a subclass of X class, but you subclassed Y class" or something equally straightforward.
Let's say instead I wanted to change the workflow to register an account. Now I'm changing a bunch of JavaScript, CSS, templates, I'm adding pages, and I also need to update the backend. Even if someone is that into frontend work, it might take forever to even get reviewed by the maintainers because it's a massive PR.
Plus, now we've moved into subjectivity land: "I'm used to the old workflow," (because they designed it) "The last one was really easy" (for an engineer), "I think we should focus on the backend before we work on the UI," "I don't like this font because the license isn't free as in freedom" etc.
Even if you just mockup something on Figma or whatever, unless you're a maintainer it's probably going to just get ignored as a feature request. Because there's also the psychological aspect of basically being told that the UI you wrote is implicitly bad, if you're the maintainer reviewing the mockup.
Just close them. You're never going to read them. If you really think there's something you need, export the browser state to an archive file, then delete in 10 years after you've never consulted it once.
(Disclaimer: I'm aware that there may be valid reasons for this workflow, but in most cases it's just digital hoarding and the above advice is sorely needed. If you really need 1800 tabs, you know who you are and you can safely ignore me.)
Brave has a function to bookmark all open tabs. I have used that from time to time. Or just made a temp file and written all the URLs to it, in case I want to open a few/dozen of them after rebooting.
I think all browsers have that now. I use it 3-4 times a year with Safari on both macOS and iOS. That way I feel comfortable closing them all without worrying about “losing” anything.
I’ve been doing this, or something similar, for at least 15 years now. Dozens of mass-bookmark “folders”. I’ve never once looked at the bookmarks I made, not a single time. I even have old bookmark files archived here and there, from machines and browsers for which the bookmarks weren’t auto-backed-up to “cloud” storage like my Safari stuff is. As soon as I can get a local llm running that’s up to the task, I’ll probably have it build some kind of table with categorization out of all of them, then have it edit out entire categories of crap until I have something I might actually scan over to recover a few interesting tidbits. Finally make some use of them, now that tools can take enough of the pain out to make that a less-daunting prospect.
The irony is that now, in the LLM age, it wouldn't be that hard to feed a few thousand browser tabs and/or bookmarks into a vector database or whatever and then if some fuzzy memory tickles your brain about something you saw a year ago, you can just query it.
It's the same logic as keeping RAW digital photos. Lightroom is already gaining "find the keepers" AI features. Sooner or later it'll be possible to feed a bunch of burst shots into an AI that just weren't worth the trouble to manually sort through. The AI can do the drudge-work of digging up any gems in the rough...
how do you see how many tabs you have open, an extension? i have tab session manager and it shows i have 80 tabs. about 60 of those are ephemeral, and the other 20 i'd have open on a new browser anyhow (email x3, goog cal, caldav cal, nextcloud files, router, local and remote proxmox, navidrome, the documentation server, etc) everything else is superfluous. although i'd probably be a bit sad if i lost all my tabs right now; hence tab session manager.
Like, that's just a logical thing to call it. I don't believe anyone is making a judgement on the intelligence of the reader to call it "open weight" when it refers to weights that are openly available.
"Open source" would be a more appropriate term to describe a model which also includes the training source.
reply