Watch time is pretty important on YouTube afaik, initial clocks themselves don’t count for that much

What? Since when does Valve prohibit companies from redirecting customers to non-Valve purchasing flows? Because that’s what this ruling is about, it says Apple can’t prohibit apps from telling users to go buy off-platform for lower prices. Valve isn’t doing that with Steam afaik, actually I’m not aware of any other platform that does this

Hadn’t heard of that but wouldn’t surprise me

Immich might not hold up yet in every aspect to Google photos, but I was and am still blown away by how much better face detection and grouping works. I cannot believe how ridiculously bad that feature is in Google, you just have to pray that it works, and if it messes up, it’s extremely annoying to fix. In immich, it works exactly as you’d expect.

“The planning thing in poems blew me away,” says Batson. “Instead of at the very last minute trying to make the rhyme make sense, it knows where it’s going.”

How is this surprising, like, at all? LLMs predict only a single token at a time for their output, but to get the best results, of course it makes absolute sense to internally think ahead, come up with the full sentence you’re gonna say, and then just output the next token necessary to continue that sentence. It’s going to re-do that process for every single token which wastes a lot of energy, but for the quality of the results this is the best approach you can take, and that’s something I felt was kinda obvious these models must be doing on one level or another.

I’d be interested to see if there are massive potentials for efficiency improvements by making the model able to access and reuse the “thinking” they have already done for previous tokens

not a common feature in proprietary software

Just so you know, the GDPR mandates that you can at any time get a full export of all your personal data from anyone who’s processing it in a common, machine readable format. It is laudable though to have that integrated as a feature in the software, rather than jumping through hoops contacting support etc.

If you have a USB C to C cable handy, that’s also going to work to charge other devices, loses less power to the transfer, is quicker, and also more convenient since you can keep using both phones. Of course you’ll often not have a cable, but if you do it’s probably gonna be the better option

I agree that this way of displaying the data is appropriate, but it would be nice to have a very visible indicator of this. Some kind of highlighted “fold” line or something at the very bottom of the chart, maybe. If I can deduce the units from context, and the trend is more interesting than absolute numbers, then I’m not going to look at the axes most of the time

One country cozying up to Putin is hardly a reason to call the entire EU divided

This is not at all relevant to the comment you’re responding to. Your choice of password manager doesn’t change that whatever system you’re authenticating against still needs to have at least a hash of your password. That’s what passkeys are improving on here

I’m German, and I’ve never heard that before. I’d be seriously weirded out by someone saying that or teaching it to their kids

I’m German, and I would not want that. German grammar works differently in a way that makes programming a lot more awkward for some reason. Things like, “.forEach” would technically need three different spellings depending on the grammatical gender of the type of element that’s in the collection it’s called on. Of course you could just go with neuter and say it refers to the “items” in the collection, but that’s just one of lots of small pieces of awkwardness that get stacked on top of each other when you try to translate languages and APIs. I really appreciate how much more straightforward that works with English.

You need both ends of the cable connected, so the phone is out. And even on PC, I’m not sure if it would work with the USB drivers in-between the software and the actual ports

Reading the article, it seems like it will actually be opt-in for everyone

The algorithm is actually tailored to find out if/when you fall asleep while watching videos, and then recommends longer videos in autoplay when it believes you are, because they’ll get to play you more ads and cash out more.

You might be misremembering / misinterpreting a little there. This behavior is not intentional, it’s just a side effect of how the algorithm currently works. Showing you longer videos doesn’t equate to showing you more ads. On the contrary, if you get loads of short videos you’ll have way more opportunities to see pre-roll ads, but with longer videos, you’re just to just the mid-roll spots in that video. So YouTube doesn’t really have an incentive to make it work like that, it’s just accidental.

Here’s the spiffing Brit video on this, which I think you might have gotten this idea from: https://youtu.be/8iOjeb5DTZI

Edit: to be clear, I fully agree that YouTube will do anything to shove ads down our throats no matter how effective they actually are. I’m just saying that this example you’ve brought is not really that.

Oh awesome, thank you so much!

Seconding this. Legitimately better than Google photos in a lot of ways, even if you don’t care about the data ownership aspect. If you’ve ever been annoyed at how Google Photos handles face detection / grouping, you’ll love Immich.

I’d love to know what font was used for the big “Saturday” there!

Now please explain to me how C works.

That’s not what they’re asking. It’s not about how C works, it’s about how specific APIs written in C work, which is hard to figure out on your own for anyone who is not familiar with that specific code. You’ll have to explain that to any developer coming new into the project expected to work with those APIs, no matter their experience with C.

It is an algorithm that searches a dataset and when it can’t find something it’ll provide convincing-looking gibberish instead.

This is very misleading. An LLM doesn’t have access to its training dataset in order to “search” it. Producing convincing looking gibberish is what it always does, that’s its only mode of operation. The key is that the gibberish that comes out of today’s models is so convincing that it actually becomes broadly useful.

That also means that no, not everything an LLM produces has to have been in its training dataset, they can absolutely output things that have never been said before. There’s even research showing that LLMs are capable of creating actual internal models of real world concepts, which suggests a deeper kind of understanding than what the “stochastic parrot” moniker wants you to believe.

LLMs do not make decisions.

What do you mean by “decisions”? LLMs constantly make decisions about which token comes next, that’s all they do really. And in doing so, on a higher, emergent level they can make any kind of decision that you ask them to, the only question is how good those decisions are going be, which in turn entirely depends on the training data, how good the model is, and how good your prompt is.