2,786 karma · joined July 29, 2010
> we can achieve this within a span of "just" 90 days using 16 A100-40G GPUs
I knew the computational power required to train LLMs was absurd, but seeing the figures of larger networks (which are just too large to intuitively understand) it didn't really register. With this one I could actually imagine the 16 machines with A100 GPUs sitting on a server room running at full blast for 90 days so it was more tangible... And now to think about the larger ones is kinda scary
Edit: Did the math and just the GPUs (at 250W each) consumed around 8.64 MWh, which is at the same ballpark of the power consumption of the average US home in one year (10.5MWh)
This is actually a common scare for parents when their kids suddenly pee in a different color, and later they find out they had a strong colored food at school, or at a friend's place.
Both buildings are conceptually very similar. I do like both.
That's fair, I did get them mixed up. The _supply_ is constrained since they introduced more strict rules for adding new emoji, the demand is still much larger than that.
> the 2023 additions produce endless varieties of human "representation", and everyone wants to be represented in emoji
Most of the 2023 additions are about horizontally flipping actions that have directionality. Only the new family pairings could be considered representation-related, but IMO they're mostly about completing the set of possible combinations.
> the new sequences depend on the new codepoints, otherwise they would be preexisting emoji, not 2023 additions
It's complicated, because Unicode publishes both the codepoints and also a list of recommended interoperable ZWJ sequences. Newer additions have been to this list.
Anyway, ultimately this is a matter of opinion, but arguing that unicode growing to encompass a lot of emoji as time goes by is not something scalable seems weird to me. One of the explicit goals of unicode was the ability to handle CJK ideograms. There are ~90,000 of those, and only ~3000 emoji. Even if they added 100 new emoji per year, which is far more than what they're doing, it would take about 900 years for the emoji set to match the size of the CJK ideograms set. This seems very scalable to me.
Also the existing system is (mostly) backwards compatible. The main issue is that it's not forwards compatible, with the exception of said ZWJ sequences.
The weird "some apps save it to Photos while others save it to Files" situation is a consequence of Files being a relatively late addition to the iOS ecosystem. A lot of apps are poorly maintained, use some cross platform framework that doesn't support the Files feature well, or the developers are simply unaware of the distinction. It will probably get better over time.
One thing Apple could do in the mean time is to also expose Photos as a folder view inside of Files (they do this on macOS, to some extent, on the file pickers. I've never actually used it)
Re: Transferring it to a PC, the one thing that won't work is the seamless copy and paste via handoff. You can plug in a USB stick into an iPad or iPhone (using an adapter for pre-15 models, or if the USB stick is USB-A) formatted as exFAT and it should just work.
AFAIK, there isn't a single button to delete all photos, probably to avoid people doing it accidentally. You'll need to manually select all photos and hit delete. Or you can also write a small script via the Shortcuts app to delete them for you.
I typically just hit Cmd+C on the Mac and long press+Paste on Files on the iPhone. If you are using the iPad with an external mouse or trackpad you can also drag and drop it directly to the Mac.
As for the distinction between random JPEG files and the Photos app, I think that's actually quite good. I don't get my gallery littered with random images, and it also supports non destructive editing, among other features. Moving between the two is also fairly easy, you can use the Share sheet or just drag and drop.
The one thing I would change is that screenshots end up in Photos.app by default, I'd rather have them go to Files.
It is implemented, but there are probably edge cases where the proper way to wrap and retain the best possible formatting is still disputed. Since a lot of people format on save, via commit hooks or during CI, having this feature enabled as stable and _changing the behavior later_ would cause massive annoying diffs in the future, or CI failures for a ton of projects.
https://docs.github.com/en/copilot/configuring-github-copilo...
Given how cheap text search is compared with LLM inference, and that GitHub reuses the same infrastructure for its code search, I doubt it adds more than 1% to the total cost.
The major OS release cadence only seems to affect major UI revamps which aren't that frequent anyway, so it's not that much of a limitation.
You can now also update Safari independently of macOS, which is nice for IT environments that restrict major macOS updates until they have been approved.
Hopefully in the future they will also allow that for iOS, so that older devices that can't update have some more longevity for web apps (and so that we as web developers don't need to support as many older versions)
Imagine with Meta Al isn't available in your location yet. You can learn more about Al at Meta in the meantime and try again soon.”
I wonder why it's region-locked?
color: media(min-width: 1001px, red, blue)
I might be mistaken, but I believe this would only apply to copyleft licenses? Permissive licenses allow redistributing under a different license, as long as the attribution and original copyright disclaimer are retained.
Sure, but written words _can_ be meaningfully quantified. We have been doing that for thousands of years. Starting with numerology and other mystical/religious beliefs, poem metrics, stylometry, crypto analysis, stroke counting, to name a few.
> Not everything needs to be.
Why not?
Yes? I don't think it's a very radical, unreasonable or difficult thing to provide: this is available as a first class feature on most blogging/forum software (RSS feeds) and wasn't even remotely controversial 10–12 years ago.
> They also seem to reinvent history such that there was some "implicit" agreement between Reddit and the users of Reddit.
There was an _explicit_ agreement: the site's terms of use have allowed free API access with reasonable quotas for 15 years, and are being changed now with very short notice. Of course the terms of service can change, but users are also free to leave, complain, or demand that they don't change. There are also strong cultural factors involved. For example: just like in the US it's customary for tap water to be free at restaurants, and people would be probably angry if you decided to charge for it, on the open web it's customary for some reasonable level of API access to be free.
> IMO, just because I wrote some shitposts on Reddit does not mean that I am entitled access to every single shitpost that has ever been written on Reddit.
There's no such asymmetry in scale: _everyone_ wrote the content, _everyone_ is entitled to access it. If you're referring to incredibly resource intensive, mass-access to download several terabytes for AI training I can see how that should cost a price. But they're blocking reasonable use-cases.
A lot of Windows applications now also behave like that behind-the-scenes, though that isn't the default behavior and requires some cross-process signaling.
- The introduction of several banners/popups/modals, either to shoddily comply with tracking consent laws or to drive up engagement (newsletter subscription, notifications, check out new feature X)
- The display of several, sometimes even nested loading indicators all over the place
- The adoption of "trendy" UI patterns that don't necessarily fit (e.g. "stories", a few years ago)
- An obsession with reducing information density and making everything very sparse
- Rejection of color as a redundant signifier for UI elements, particularly icons
- An overall extremely hostile attitude towards the user, with dark patterns, disregard to preferences and privacy