Daniel Spiewak

It's been done before. The mathematicians complain.

Public Key

npub1nfjd63p9den5retrjz3yeye3rvhcle5amqfhnvvttraelmpsf2ps2stprm

NIP-05 Address

djspiewak_at_fosstodon.org@mostr.pub

Profile Code

nprofile1qqsf5exagsjkue6pu43epgjvjvc3ktu0u6wasymekx9437ulascy4qcz9gph8

Author Public Key

npub1nfjd63p9den5retrjz3yeye3rvhcle5amqfhnvvttraelmpsf2ps2stprm

Show more details

Last Notes

2024-09-20 13:41:28

- reply

@npub1m64…4th4 And as is the dream, you clearly identified the email as fraudulent by checking the WoT signature chain.

2024-09-08 00:07:33

- reply

@npub1m64…4th4 Pull a "data" and go with shits-for-brain

2024-08-07 16:16:23

- reply

@npub1c9m…9xjd The fact that we're reduced to "sad toots" as a phrase is, in and of itself, a meaningful commentary on the state of the world.

2024-07-27 15:38:03

- reply

@npub1xnf…jqth teams actually does something most of the way to this. You can tab back in and ask copilot to summarize what was just asked of you and its context, and it’s usually fast enough to be useful.

2024-06-27 14:05:32

- reply

@npub1ggt…t0h3 Ouch. It probably swallows other, more seriously fatal errors as well then. Lucky that it was only an assertion.

2024-06-26 21:24:04

- reply

@npub1z9m…9zmq “He *said* ‘isomorphic' but his arms looked more perpendicular, so he must mean 'orthogonal’.”

2024-06-26 20:46:44

Realized the other day that I’ve been accidentally vocalizing “isomorphic” when thinking the word “orthogonal”. Bizarre neurological implications of this aside, I’ve been doing this for at least a few days and absolutely no one questioned it, which leads me to wonder if anyone actually listens to anything I’m saying in meetings or if they’re all just politely nodding and smiling?

2024-06-06 00:44:57

- reply

@npub1z9m…9zmq I have been told of people who literally have two kitchens: one for turmeric dishes and one for everything else.

2024-05-27 23:46:43

- reply

@npub1pna…pqx3 It is exactly an own goal. This sadly happens often though.

2024-05-26 16:37:03

Easily the single most unrealistic thing about Horizon Forbidden West is the way it assumes that west San Francisco will one day build high density residential housing.

2024-05-24 02:49:21

- reply

@npub1c9m…9xjd I feel like really good mathematicians are really closer to AI than to normal humans, so this tracks actually.

2024-05-09 19:29:15

- reply

@npub1c9m…9xjd So… Mass Effect?

2024-04-10 22:33:35

- reply

@npub1vmm…kc2a You should use this as a teaching moment to introduce him to the concept of LLM hallucinations.

2024-04-09 14:06:40

Honestly the fact that eclipses pull people into random parts of the landscape ill-suited for such large gatherings is a feature, not a bug. The traffic is terrible (though subjectively much better than 2017) but there’s something cool about tens of thousands of uncorrelated people standing in the same random field to appreciate the same four minutes.

2024-04-03 01:36:43

- reply

@npub1c9m…9xjd Build systems and deployment orchestration are a staggering collective tax on our industry. I really refuse to believe it HAS to be this way, and yet I haven't been able to come up with an alternative.

2024-03-25 15:56:29

OH in a standup meeting today: "Admittedly, I foolishly overcomplicated some things that didn't need to be complicated…” Same. Every day same.

2024-03-21 22:44:41

- reply

@npub1c9m…9xjd Fascinating! I had no idea.

2024-03-15 14:19:30

- reply

@npub12jq…hq6m It wasn’t a performance beast, but it still unlocked scenarios out of reach previously. I remember gawking over demos of instances breaking 10k RPS, which was just unthinkable with servlets and Jetty. This was also why Play (and for a hot minute, BlueEyes) saw so much initial adoption.

2024-03-15 13:13:23

- reply

@npub12jq…hq6m Not sure I agree actually, and we have a bit of evidence to chew on here. Scalaz used to have an IO which was pure but entirely blocking. It was essentially entirely ignored in practice in favor of Scala Future (impure but non-blocking) until Scalaz Task came around, when people started really leaning into the pure + async paradigm.

2024-03-15 01:42:57

- reply

@npub1sf7…m3th In theory, some sort of vertical tab seems a lot better because of aspect ratio. But at the same time, vertical scrolling is easier than horizontal, and English is left to right, so yes… you are in fact dirty.

2024-03-11 19:43:33

- reply

@npub1sf7…m3th @npub158t…v0ut Depends on the type of ads. As long as you're not Hulu (*cough*) and you're just worried about in-stream ads and predominantly offline targeting, it's not hard to decouple the expensive system from the cheap system.

2024-03-11 19:16:38

- reply

@npub158t…v0ut @npub1sf7…m3th Sadly, at the end of the day, there's just too much money in it. Netflix's premium tier is a pretty good indication of how expensive a sustainable streaming subscription needs to be without ads (and that's even including the fact that Netflix's service is comparatively stateless and non-real-time, which all translates into "cheap”).
The fact that consumers are already habituated to ads interjecting into their video content makes it a self-sustaining optimum for the industry.

2024-03-11 19:10:33

- reply

@npub1sf7…m3th @npub158t…v0ut Nailed it. :D

2024-03-11 14:10:32

- reply

@npub1dw4…mhkq @npub1g8l…y49n In all fairness, OpenAI does training at a data center in Ohio, which has easy access to roughly 20% of the fresh water on the planet. Also the water, once used for cooling, remains in the local water cycle. Plenty of problems with AI sustainability, but water in this case isn’t one of them.
Ohio’s power grid carbon intensity, on the other hand, is much less inspirational…

2024-03-05 22:06:07

- reply

@npub1n3h…n824 In the digital world, I'm *always* in the neighborhood! ;)

2024-03-05 18:27:19

- reply

@npub1c9m…9xjd It does! It's precisely this, in fact. The way it works is you project a spanning tree over a subgraph and deliver that to the user, who edits it (returning a new tree), and then you reproject that back into the subgraph and splice into the whole. The corner cases involve arbitrary resolution, obviously, but they arise very rarely and are easy to steer around in practice when defining transformations.

2024-03-05 17:07:19

- reply

@npub1c9m…9xjd This reminds me of an observation my team and I made several years ago, which is that the most practically useful class of graphs are DAGs, and the most useful class of transformation is one which can be expressed as a spanning tree projected onto a DAG. This observation opens up a lot of flexibility from a representation standpoint, since most functional languages tend to be great at dealing with trees, so mapping to and from spanning trees keeping the DAG canonical is really nice.

2024-02-26 23:33:57

- reply

@npub1xnx…l76r Packed bagwell tries support O(1) cons, snoc, append, length, and nth, with low enough constant factors that you basically won't notice.

2024-02-14 03:31:34

Free interview question for candidates where you want to test some basic knowledge of graph and probability theory:
You are in a city with square blocks, sidewalks on both sides of every street, evenly timed street lights at every intersection, and jaywalking is punishable by death. What is the optimal strategy for traversing from point (0,0) on the grid to point (k,j)? Can you prove it is optimal?

2024-02-05 22:26:42

- reply

@npub1cwp…6fca Pfffftttt. Give them some credit! I'm sure they're doing capability testing.
if !ad_tracking_supported(): sleep(1)

2024-02-02 15:26:20

- reply

@npub1js9…maz4 Feels like if we allow truly arbitrary exponents it would be relatively easy for us to make a perfect right triangle. Or at least, *possible*. Applicability of Euler's theorem likely determines how tractable the computation would be. But with that said, real irrational would be much less satisfying than the rational ones used here.

2024-01-17 04:40:12

- reply

@npub1sf7…m3th I mean, hell, I want this for my laptop so I don't have to subject my coworkers to my terrible bluetooth microphones.

2024-01-15 23:52:07

TIL "variable frame rate [video]”. It's amazing any of this nonsense holds together. Also this reminds me of something my old boss used to say, which is that the whole concept of a frame is insane and we should really do away with it. A lot of problems in tech are like that: someone made a decision decades ago which made sense in context, but that context is long gone while the impact of the decision remains with us.
Don't get me started on NTSC/PAL…

2024-01-14 16:42:24

- reply

@npub1c9m…9xjd @npub1vmm…kc2a Actually this is a good example of where I think the terminology is actually helpful. The situation you're describing is a non-monotone collision, which is why there's no natural "winner" or "loser(s)". The solution is legitimately always to restructure the protocol such that there is at least one winner each round, and the terminology reminds you of this.
At least, I find that to be a helpful mnemonic :)

2024-01-13 22:11:40

- reply

@npub1c9m…9xjd @npub1vmm…kc2a I agree and disagree at the same time. I've always felt the same way, but at the same time, being able to describe the branches of a CAS as “winner" and "loser" is *really* helpful terminology, and it even scales decently beyond 2-way races. It also forms a tiny reminder that all racey protocols must have some monotonic convergence to be valid (i.e. at least one winner on each round), which is helpful mental shaping IMO.

2024-01-10 21:19:03

- reply

@npub1c9m…9xjd Oh absolutely. We're picking nits a bit here because the first three seasons were *so* well done in these respects (apart from the completely impossible landing survival at the end of S3). At least it hasn't gone full Numb3rs yet.

2024-01-10 20:50:04

- reply

@npub1c9m…9xjd Second, there's an unstated assumption that the asteroid will arrive at Earth in a reasonable amount of time, meaning it needs to be moving quite fast. How the heck are they planning on slowing it down such that it can be captured into a stable orbit, *particularly* given regular perturbation from the moon? The magnitude of that problem is arguably much larger than any of the problems they do "solve”, and yet it's completely glossed over.

2024-01-10 20:48:40

- reply

@npub1c9m…9xjd Honestly, there were two distinct suspensions of disbelief in that whole plot line which I really struggled with.
First, the idea that Mars, Earth, and the asteroid *happened* to align just right such that an orbital slingshot around Mars would get the asteroid to Earth. Of course, the precision of that alignment doesn't need to be perfect, but it needs to be within the bounds of the asteroid impulse relative to mass, which still seems pretty darn coincidental.

2024-01-08 23:44:04

- reply

@npub1sr3…hxx7 Fun fact: file I/O is not truly asynchronous because it can’t be at the hardware level. io_uring does the same thing that Java does in this case: create a thread behind the scenes and pretend it’s all non-blocking.

2024-01-03 14:29:00

- reply

@npub1zd7…yta7 @npub1sf7…m3th IMO the queen is an inevitable construct simply because truly masterless peer to peer distributed consensus algorithms are both very hard and inefficient for fundamental physics reasons, so leader election is the most natural outcome. They could just as easily make it a random drone, but they could also make their implants look less scary and fix their flickering lights.

2024-01-02 18:48:05

- reply

@npub1c9m…9xjd people underestimate how much of an impact redlining had and has on people, families, wealth disparities, neighborhoods, schools, infrastructure… the list goes on. So much in this country starts with property ownership, so anything that explicitly or implicitly affects this has an exponential impact across generations.

2024-01-01 01:57:56

- reply

@npub1g4v…kdah “Fixed build”. “Really fix build”. “This time for real”. “Fix". "More fix". "Idk maybe”. "We should all admit we don't really know what we're doing". "Okay I'm the only one who doesn't know". "HOW?!”. "Oh…”

2023-12-30 02:43:23

You know, if I'm being honest, this should be all of my commit messages.
https://cdn.fosstodon.org/media_attachments/files/111/667/065/570/332/526/original/5049b5f0bb49d3e3.png

2023-12-04 09:33:50

- reply

@npub1sf7…m3th @npub185t…pwlf @npub17pe…vcm6 @npub15ml…2w4s @npub1txf…4llt Metronome? 😛

2023-12-03 08:41:45

- reply

@npub1sf7…m3th @npub185t…pwlf @npub17pe…vcm6 @npub15ml…2w4s @npub1txf…4llt I always used to count. 1…2…3…4…1…etc. Constant rhythm. Do it together at the start of the recording for a cycle or three then go from there. Easy time sync in post.

2023-11-29 14:24:24

- reply

@npub1vmm…kc2a I think a lot of the really perverse incentives ultimately come from the combination of the accumulated precedent together with the fact that the entire modern economy is based on IP, something that the creators of the patent system simply could not have foreseen.

2023-11-29 14:22:57

- reply

@npub1vmm…kc2a Well the original problem statement was avoiding situations where inventors hoarded their secrets and the knowledge ended up dying with them (very common prior to the 19th century), which in aggregate is quite bad for society. Obviously it's great that this no longer happens, but equally-obviously we have created a destructive minefield of multi-century terribleness in the jurisprudence around this whole space.

2023-11-29 14:09:16

- reply

@npub1vmm…kc2a I think that if we were reinventing that whole mechanism (patents) with no prior art, we would end up with something that looks a lot different. Same goals, more than likely, but very different mechanics, particularly in light of the fact that people don't need much encouragement these days to write things down, and so inventions are almost never truly lost anymore, with or without patents.

2023-11-29 14:07:57

- reply

@npub1vmm…kc2a Patents are IMO a good idea, but the structure is so anachronistic that they've lost all connection to their goal. The concept is to ensure that knowledge isn't lost entirely by giving an incentive for sharing it: a time-limited monopoly. One of the many problems today though is most tech inventions already have a very short shelf-life before obsolescence, so the time-limited modifier is somewhat meaningless: it's just an absolute monopoly. And of course, monopolies are good business.

2023-11-26 19:38:44

- reply

@npub1ph0…e6wu The irony of that post is that the author *already* works on one of the largest commercial Scala codebases, and the systems in question are far from legacy.

2023-11-26 18:37:08

- reply

@npub1ns9…an9x What does this say about the Rust Belt region?

2023-11-26 18:33:34

- reply

@npub1wqz…a7f3 That's a super good analogy honestly.

2023-11-26 18:32:25

- reply

@npub1ns9…an9x I'm almost afraid to ask: what does "post-modern" mean in the context of text editing? Is that when you *don't* build on Electron?

2023-11-23 19:48:55

- reply

@npub1m64…4th4 For a similar reason, I consider the origin of Christmas to be a state propaganda compromise between the Roman empire and various Germanic tribes. We can point to the rich history of solstice celebrations and tree worship as being part of the prehistory of the holiday, but as a holiday, it began there IMO.

2023-11-19 00:37:36

- reply

@npub1sf7…m3th Live stuff is a good callout. I would have to parse a baseball stream to see for sure though. Also as you know, DAI is a whole lift above and beyond just midroll ads, so I wouldn’t be totally surprised if that’s still WIP

2023-11-18 23:58:46

- reply

Oh and right next to that preroll metadata tag is a second tag which lists the *number* of expected midroll breaks. Currently that number is 0 (as expected), but consider this a strong hint that Apple is at last thinking about adding an ad supported tier to Apple TV in the future.

2023-11-18 23:57:55

- reply

They then have custom logic in their own player to interpret this bit of metadata as a preroll fetch, and then likely fork off a second player, superimposed over the main player, which handles the preroll (not dissimilar to how they implement interstitials, but without having to solve buffer alignment).
The cost here is longer video start time (which is easy to see as a user of Apple TV), but the upside is they got to ship a very user-impacting feature (promos!) without waiting for spec.

2023-11-18 23:55:48

- reply

And as it turns out, this is exactly what Apple did. Though there are hints that midroll injections (*ahem* ads *ahem*) are coming in the future to an Apple TV near you. Specifically, Apple TV promos are *not* using HLS interstitials (even though they're now supported on all current Apple hardware). Instead, Apple uses an out of band bit of metadata at the front of the manifest to link out to an *additional* URL (probably service-backed) which likely returns an HLS fragment for a sub-player.

2023-11-18 23:54:12

- reply

And yet, Apple TV had personalized preroll promos long before this, hence my minor curiosity.
Now, to be clear, pre- and post-roll content injection is fundamentally easier than midroll (e.g. most streaming ads) because you don't need to worry about lining up with a segment boundary in the stream: the stream hasn't started yet, so just slap some video in there *before* you start and you're good to go!

2023-11-18 23:52:10

- reply

Apple was the originator of the HLS specification, so it was always a safe bet that they weren't using DASH (which has nicer solutions to this problem). There have long been a few non-standard approaches to try to address this within HLS players, but none of that seemed Apple's style. Nonetheless, the “proper" solution (HLS Interstitials) were only added to the spec within the last two years, and remained poorly supported even on Apple hardware until much more recently than that.

2023-11-18 23:49:54

- reply

So then the problem becomes the following: how can we have *subsections* of the video stream which are non-static? Ads are the big gorilla in the room here, since we want to ultra-personalize every ad opportunity at all times, but promos (such as those on Apple TV) are another good one.

2023-11-18 23:47:06

- reply

This is a nice scheme, because it allows the client to instrument its own capabilities on the fly – including network conditions, hardware, localization, etc – and choose the best variant to match. It even allows client-side control over quality/bandwidth tradeoffs (higher quality usually means more bitrate which means more bandwidth), which we see in YouTube's cog menu (among other places).

2023-11-18 23:39:04

- reply

All of this has to work even while our network conditions fluctuate from good to bad and back again, which brings us back to the origin of our problem: how do we *stream* video and associated timeline-aligned data (e.g. subtitles) in a way which is reliable in the face of unreliable network conditions and dynamically makes tradeoffs against quality? The answer is playlists.

2023-11-18 23:29:53

I was curious as to how Apple was handling pre-roll promos and ads. These pre-rolls are pretty clearly dynamic (i.e. not burned into the playlist) since sometimes I don't get them, and can be multi-stage (e.g. Foundation episodes usually have three independent ones). HLS Interstitials are very new and Apple TV's pre-rolls predate its support, so naturally I wanted to know how this was done.

2023-11-14 18:34:02

- reply

@npub17lr…z4zs The author of this table is pretty clearly much more familiar with the right column than the left. :(

2023-11-12 03:49:18

- reply

@npub1j55…p4t9 Sure but it feels like the gaming aspects work perfectly fine as a second device.

2023-11-12 03:04:56

- reply

@npub1j55…p4t9 Why though? At that point it’s just a more expensive underpowered laptop.

2023-10-27 13:22:58

- reply

@npub143w…pykj De Brujn notation, but with number of underscores instead of an integer scalar.

2023-10-22 14:30:38

- reply

@npub1wc8…uzs6 @npub12jq…hq6m Well, io_uring is several things at once. The main benefit is the way in which it allows you to minimize syscalls, which are very expensive. The fakery with block file handles is the annoying part and I wish it could be disabled.

2023-10-18 16:37:18

- reply

@npub1sf7…m3th I doubt that very much. The big tech companies were building this type of infrastructure internally two decades ago out of necessity. That trend of abstracting away the physical layer always creates marginal waste, and that waste can be sold to smaller entities for straight profit. That business model evolution feels inevitable to me.
The only real surprising thing is that Apple and Meta haven't cashed in on this in the same way. (though Meta’s network layer makes it much harder)

2023-10-18 15:39:46

- reply

@npub12jq…hq6m The even deeper point Arman is making is that some I/O is just fundamentally blocking because that's just how the hardware itself functions. No syntactic magic can paper over that fact. Btw it's not just Loom that falls into the trap of trying to massage this. io_uring perpetrates the same falsehood.

2023-10-12 13:52:26

“Loom doesn't fix blocking I/O, it fixes blocking APIs built on top of non-blocking I/O.” – Arman Bilge
This is the most concise and insightful description of the loom that I have ever heard. It pretty comprehensively covers both the benefits and the limitations of its approach, all in a sentence.

2023-10-07 14:44:53

- reply

@npub12jq…hq6m This was almost precisely the intended use case. 🙂

2023-10-05 18:19:42

- reply

@npub1sf7…m3th @npub1xz5…zyew paranoid clean build retest intensifies

2023-09-19 18:23:45

- reply

@npub1sf7…m3th A lot of the complications seem to be failing to update, not just the sunrise/set ones. I'm also experiencing similar issues on the phone itself, almost as if scheduled background refreshes just aren't executing. I was planning to give it a week for all the dust to settle before trying to dig into it.

2023-09-05 22:06:56

- reply

@npub1c9m…9xjd Same and same. I thought that episode was pretty shallow but that one particular reference was nice. It didn’t feel forced and the visual effect was cool.

2023-08-08 15:30:13

I had an absolutely enormous amount of fun building and delivering this talk. Please enjoy! https://www.youtube.com/watch?v=n5u7DgFwLGE #scala #performance