finnh

Right ... the output isn't really "keyboard & mouse operations" as stated in the abstract. The output is a sequence of atomic commands ("shoot", "select_weapon") or composite commands ("follow") that are then translated into mouse actions by a human-written layer.

Similarly, when I'm typing this post my output isn't "keyboard interrupts" nor "TCP frames". Those things do happen, but only by the grace of others' work.

Still I like the basic framework idea & I suppose future work looks at pushing the provided layer lower & lower - so the system can learn those atomic commands, rather than having humans provide them - and pushing the logic higher & higher via AI-generated composite commands.

For sure. Qubes OS is an interesting step in that direction. Mobile does mostly get it right - and yet the devices are single-user only, it's so odd. The fact that I can't share an iPad with my kid without needing to fully disable Messages, Photos, etc is crazy-making.

Every time I look at NixOS, I think that it perfectly solves a problem that I only have once every 5 years, when buying a new computer. I think I even looked into it once to automate that exact process, but that idea fell apart at the first line of Nix syntax. I'll stick with OSX and `brew bundle` I guess...

But then I read a piece like this and remember that some people do have to plumb the depths of C/C++ linkers, and I'm glad I'm not one of them.

Great post! FWIW I always want to know the prompt text when seeing an AI-generated image, I wish there were a convention around that.

If we could get an ice-9 that raises the freezing temp of water by only a couple of degrees, that might be hugely useful in a world of climate change. We could keep our glaciers and the wonderful year-round fresh water supplies (rivers) that they enable.

This was over a decade ago, so happily not something I'm feeling too sharply nowadays. It all worked out better in the end.

I am not sure I'd agree with that - at one point Salesforce tried to buy my company, which was a best-of-the-AppExchange partner, and their justification for their (very) low offer was "that's what it would cost us to copy you". We sold the company 2 years later for 10x their offer, without a ton in the way of new revenue (our AppExchange ranking mysteriously tanked after we turned down their offer).

Was there, reviewed the PR, can confirm. Hi Steve!

Since then we've also tuned it up in a couple ways, in particular adding "skip" logic similar to fast reservoir sampling to trade some accuracy for the ability to not even look at the next N {M,G,T}B if you've already seen many many many matches. For non-selective searches over PB of data it's a good tradeoff, despite introducing some search-order bias.

Good to learn, thanks. "Everyone you follow" isn't quite what I was saying, but I also don't use twitter so I'm doubly in the dark here.

I'm surprised that someone with, say, a Twitter problem and tens of millions of followers - the two seem to go hand-in-hand - drives tens of millions of writes every time they post. But there you go, learned something today.

> These numbers are a bit better than Twitter’s numbers. Because of how unbalanced the social graph is, getting performance this good and this reliable is not easy. For example, when someone with 20M followers posts a status, that creates a huge burst of load which could delay other statuses from fanning out. How we handled this is described more below.

Huh. I haven't written myself a twitter, but I've always assumed that this problem is solved by a mix of fan-on-write and fan-on-read. Most people are fan-on-write, but those with a follower count > $THRESHOLD are fan-on-read.

The fan-on-read set is small enough to be easily cacheable, so the extra reads required to service a follower's feed are more than compensated for by the fact that the content is in RAM in a cache.

Sounds like RedPanda went exclusively with fan-on-write. I'd be surprised if Twitter does the same - but it sounds like OP has more knowledge about this than me, so I'm happy to be educated!

The author should remove - or at the very least _credit_, come on - the image used in "The Pendulum of Database Realm" section. It's from Martin Kleppmann's "Designing Data-Intensive Applications", a particularly good O'Reilly book with illustrations by Rebecca Demarest.

Terrifying article about the impact of solar weather on our (electricity-dependent) society: https://www.newyorker.com/magazine/2024/03/04/what-a-major-s...

Sounds like they aren't against "transactions" as a thing - they are limiting _their_ transactions with society at large.

After all, the number of transactions required to make even a pencil is astonishingly high, as Adam Smith pointed out two hundred years ago.

Curious why you are careful to never mention the handle of the HN user in text, only in images. What is the perceived threat model of stating clearly (in this comment, or in your blog post) the name of the HN user who copied you etc?

I didn't realize Kamasi Washington is on Ninja Tune!

They had a predilection for releasing zillions of compilation albums with heavily overlapping track lists, which kind of fatigued me on the label, but I'm still fond of the early releases of theirs that I bought.

Funkjazztical Tricknology FTW!

I think used car lots are the biggest users, per se - like a CarFax situation. I think a lot of the relevant data can be pulled off the car's monitoring post (i forget the acronym).

I'm not sure that's true - temperature is a big factor that's uncorrelated with mileage, as is how frequently the battery depletion beyond the optimal level. There may be other things - I don't own an EV yet so I'm not particularly current on it,

Exactly, the battery question is what it is all about. I wonder if these Hertz cars will be tracked on Recurrent[0] so you can gauge their battery health

[0] recurrentauto.com, started friends of mine but otherwise no relation

For the first decade or so that I used SoundCloud, music discovery was WORK - their recommendation engine wasn't great, and their search tools were limited, so you'd have to just kind of chase links through artists & commenters, and grind it out. Every couple weeks I would spend a few hours hunting for gems.

Eventually I switched most of my listening to Spotify, on the strength of their recommendation engine, even though my heart is with mixes on SoundCloud.

Then sometime about 5 years ago they fixed their recommendation engine, and it's really just a great experience. Their search tools are still limited, and they've "pigeonholed" my tastes a bit more than Spotify, but I'm still a big fan of the platform and paying subscriber.

If you love SoundCloud ... pay them! Let's keep this beautiful thing alive.

[UPDATE] - also I've had such great interactions with artists on SoundCloud, up-to-and-including some emails to get a local copy of favorite remixes after a track disappeared from SC. So much positivity.

Nice! My family has a "Spirit of 76", inherited from my father-in-law who bought it new. It's always fun to open the machine up and trace through all of the connections.

About ten years ago we hired a pair of local guys to replace some worn-out bits - nothing major, just bumpers and light bulbs. They were exactly what you would hope for in such a duo: their initial evaluation step was just one guy playing (the guy with long hair), while the other guy (the guy with glasses) watched closely. During this phase they communicated in grunts and monosyllables, a tight code borne of years of shared experience between them.

[update: The Spirit of 76 gameplay is nearly identical to the Aztec shown in the vid]

The "Washington" in "Washington Commanders" is Washington, DC. Not Washington State (which has no income tax).

Can you point me to docs for ZFS offering a nocow mode? I haven't used it in about a decade, but i can't see how that would work - wafl/cow is a pretty fundamental invariant in everything ZFS does

overtraining is such a trap! My college crew program had something like 11 workouts per week - one on the water, at 5am, and one in the gym at 4pm, every weekday, plus a "big one" on Saturdays. Very little tapering before big races.

That's simply too much, especially for 18 year olds...

why not use boats, on the surface of the water?

ugh sorry to hear that. That squeeze I mentioned before (lowball offer at "this is what it will cost us to copy you") was, thankfully, for a company that we bootstrapped so we had no VC pushing us to close. And we bootstrapped because of my experience at my previous, VC-backed company.

Thanks. I've sold two companies, and gotten a pressurized squeeze from a major player (which we said no to - we were profitable so didn't have runway problems), but I've never heard of dropping an offer valuation at the last minute.

Nasty business!

Where are you getting those numbers from? The valuations of small acquisitions are closely held information; afaik there is no place to just look this stuff up.

you might like "Without You There Is No Us," a memoir by a foreign national hired to work at a boarding school for the children of the NK party elite

That must have been the section that broke downward, so was traveling faster than terminal velocity.

that's so ravyn

Based on the Blackwing paper the "Microsoft Surface Type Cover" hack seems pretty legit. They can gain login access to a Surface-with-cover by simply unplugging the cover and plugging in their evil USB device. That sounds like a bypass to me, no horse shit required.