this is just constrained generation? I thought the latest crop of decision models (inspired by Jev) do something fundamentally different in the architecture; they're not simply off-the-shelf models with a token mask
Indeed. Token masking just limits the costs associated with using an LLM. It therefore also limits the accuracy by limiting the amount of compute available.
But LLMs can mimick decision models, and I wouldn't be surprized if some labs are doing it this way at the moment.
> It's a very energetic intensive process, it cracks ribs.
I feel like lately this is becoming more common knowledge - but still something most people don't realize.
Part of it is probably the fact that it's impossible to depict "real" CPR in popular culture (movies, TV shows, etc) unless the production goes to extreme lengths to use a fake dummy. Even on The Pitt (which seems to make a point of being hyper realistic) I've seen them do "fake" CPR with shallow compressions.
so on Lost when Jack is really upset about Charlie, and he beats the shit out of his lifeless body, ... and it worked, did he do real or fake CPR? These comments make it sound very real.
Fake. You can't have fire without oxygen, but that doesn't mean you can't have oxygen without fire. Breaking someone's ribs doesn't necessarily mean you're performing CPR.
Yeah, akin to the Gell-mann Amnesia Effect, we notice a few things where we're experts but then forget everything else is likely just as bogus. Apparently one reason "Queen's Gambit" was a big deal was that most pop culture chess isn't just not very good chess (as you might innocently assume), it's literally nonsense. Like, pieces on the wrong squares, illegal moves, even simple continuity errors where pieces move between camera shots. So QG begins scoring points for chess fans when it remembers stuff like the White Queen starts on a White square...
I've found most of the differences to come not from the socket API, or the logical behavior, or CLI differences. But instead from assumptions Docker makes, that it's running rootful, when Podman will not (by default).
As such, most of the fixes for Podman/Docker incompatibilities is just addressing that assumption with a few extra flags on the Podman commands to change how the user namespace maps between the container and the host, etc etc.
This has been my experience too. And that conclusion, I've found that if I really don't want to deal with such quirks, I can indeed literally just run the same docker compose file in rootful mode, and it works.
Making the changes required to run under rootless is often very simple.
There are also different CLI flags (I recall something related to building and pushing multi-arch containers) and different features supported in compose.yml. Even rootful Podman won't be fully Docker-compatible.
I live moments away from this roundabout. It's confusing. Single-lane roundabouts are really straightforward. You're either clear to enter, or you're not.
Adding lanes makes it far more confusing. I consider myself, you know, pretty smart. Not stupid, at least.
But I almost sideswiped someone in this roundabout the other day. Years of driving experience gave me an intuition that the middle lane would not cross over the outer lane. E.x. a car in the inner lane would not pass through the outer lane (except at the very end). So when I saw an oncoming car in the inner lane I thoguht I was safe to enter the outer lane. Not so. The inner lane car was actually lane-changing to the outer lane (at the exact point I was about to enter the roundabout) in order to exit.
The lanes are quite narrow, and the outer curb is deceptive - there’s a 1” edge with a curved curb behind it making it look wider than it actually is. Scraping along that edge will push cars into the center of the road. There will be a lot of minor accidents here due to the road design.
One other second order effect - people are getting used to roundabouts here now, but nearby are ‘traffic circles’ that are roundabouts with stop signs on some entrances. People are now ignoring those stop signs (because it’s a roundabout!). I almost hit 3 cars in 2 intersections as people ignored their stop sign.
It's more complex than that. I live in Seattle near a street that has these traffic circles. The design is:
* There is a round obstruction in the middle of the intersection like a roundabout.
* The street going north-south through the intersection does not have a stop sign.
* The street going east-west has stop signs on both sides.
* But the north-south street which doesn't have to stop also has speed humps on it to slow drivers down.
On its face, this seems like a totally bananas design. The street that should be efficient by not having to stop has traffic calming speed humps on it anyway, negating the efficiency. The cross streets get none of the efficiency of a roundabout because they have to stop anyway. And the combination of roundabout and stop signs is very confusing to drivers.
It makes no sense... if you assume the intersection is designed entirely for cars.
But it isn't, it's a "neighborhood greenway"[1]. The north-south street is designed to improve bicycle traffic. The speed humps don't slow cyclists down. The roundabout middle and stop signs on the cross streets make it safer for cyclists to cruise through the intersection without stopping.
If you ever bike commute, you quickly learn how lethal a lot of stopping and going is for cycling. The effort and efficiency really only make sense if you can go a fairly steady speed for much of your commute. Accelerating a bike is a lot of work.
Once you factor in bikes, the design of these intersections makes more sense. At least in theory. In practice, though people are consistently confused by "roundabout + stop signs" and I see drivers blow through those stop signs more than I've ever seen any other traffic violation by a large margin. Because of that, cyclists and drivers going north-south still have to be paranoid going into the intersection. Even though they have the right of way, there's about a 25% chance the other driver won't stop anyway.
It was a good idea, and maybe the execution will work out once people get more educated. But right now it's a mess. I walk along that street often and I spend a lot of time gesticulating wildly at drivers when they blow through those stop signs.
I live by a bunch of these, but worse IMO is the "four way intersection with no signs or circle or anything" where you just kind of hope no one comes fast and hits you (I imagine on a bike where you wouldn't stop if there was a sign this is even more dangerous). Seattle is weird. Traffic circle with stop signs is fine by me, they don't really function well as roundabouts for most cars, anyways, some of them are too cramped to actually go all the way around smoothly.
> worse IMO is the "four way intersection with no signs or circle or anything" where you just kind of hope no one comes fast and hits you
Agreed, totally. That's what this street used to be before they converted it. I hated driving it because I had to crawl through each intersection because you never knew if someone was going to just blow through it or not. It was even slower than a four-way stop.
Hah! Yeah I also live on a neighborhood greenway street with a traffic circle in Seattle and the odd stop sign layout. It confused me for a year until I learned about the greenway program.
Seattle has plenty of traffic circles just laying around that predate the greenway program though.
I almost never see Seattle drivers blow through stop signs though, maybe people in my neighborhood are just more chill? :D
Mine was converted to a greenway pretty recently, so it may be that drivers just have old habits from before when those intersections didn't have any stop signs.
Good to hear they are adding more green ways! I love the program and I feel they are safer (and a lot more affordable) than "bike lanes" on arterial streets that are often just a painted line.
Seems reasonable. Of course there are really good (true) roundabout designs that are great of cycling (and walking), but without understanding the space it's hard to know if they would fit / work well.
What they're talking about is basically a small concrete circle (sometimes it's not a circle, but a round-ish shape vaguely near the center) plopped down in a standard residential intersection (sometimes with a lot of stuff in it that blocks the view in case you were worried about knowing if someone is crossing the street on the other side). Better designs likely wouldn't fit without raising the cost a huge amount.
> Single-lane roundabouts are really straightforward
Like the Arc de Triomphe roundabout? :)
If it had lane marking there would apparently be 10 tracks around the circle.
What makes it seem crazy is cars entering have right of way.
The reason it works seems to be French attitude. Cars entering do their thing, and cars already going around do their thing (and just have to avoid anyone on their right).
The top of the Arc is open space. Standing up there watching the traffic is amusing. I once saw two buses, side-by-side, exit the roundabout, while the motorbike sandwiched between them attempted to continue further around the circle. Luckily, everything happened at fairly slow speeds, so collision was avoided.
> What makes it seem crazy is cars entering have right of way.
The thing that’s bananas is that this is a City of Paris rule. In the rest of France, traffic entering a roundabout must yield. Not inside Paris. Better make sure you know this!
And even more crazy, the Periphérique, the controlled-access ring road around Paris which tends to move at 40-50mph, is a city street and so traffic entering that highway-like “street” also has right-of-way.
> What makes it seem crazy is cars entering have right of way.
I saw that on either Top Gear or The Grand Tour and was convinced they were just making fun of the French. It's really odd to not have changed it when everyone else learned that lesson already.
Not hitting anyone is a skill some people seem to lack. Sometimes the rules say one thing, but you have to do another, mostly give way even if you're not supposed to yield.
tl;dr - within the container, the exploit works, and elevates to root (uid 0) within the container - BUT because that namespace actually maps to uid 1000 (the user) outside the container, the escalation does not flow up to the host.
But… does this escape the container? If not (the author seems to indicate it does not) then does it matter if you are in Docker or rootless Podman, right, since the end result is always: you have elevated to root within the container. If the rest of the container filesystem isolation does its job, the end result is the same? Though I guess another chained exploit to escape the container would be worse in Docker? Do I have that right?
This is a problem and most people hadn’t considered it before because the caching is done to speed up build pipeline performance:
“ While rootless containers prevent the attacker from escalating to host root, the page cache is still shared across the host. Containers that re-use the same base image layers share the same cached pages for those layers — if a malicious CI job corrupts a binary in the page cache, other containers launched from that same image could end up executing the poisoned version.”
I'm no expert, but the kernel is shared between all containers and the host.
I don't believe the kernel maintains separate page caches for each container; a malicious CI job could corrupt a binary from any container, or the host.
If any security relevant file from the host is mounted into the container this could be exploited quite easily. It is definitely a viable tool for escaping containers but it would require a bit of an attack chain and some containers may not be vulnerable.
Not exactly. One can be charged with stalking, even though the offender only went to places in public that the victim also went to. If combined with a pattern of behavior that, in aggregate, infringes upon the rights of the target, it can become a crime.
The interesting part, for anyone who actually reads the article - the change was fixed in an RC and then reverted in the final release.
Which implies there was some regression, some issue, some incorrect behavior or negative impact. One has to wonder… what could it have been? What could the issue with having a more accurate clickbox for the corner of the window possibly be?
For example: imagine you have 2 windows, the lower right corner of one window almost touching the upper right corner of the other, so that the bounding rectangles overlap but the graphics don't.
With the inaccurate "false square" corners, you just had to check the bounding rectangles, to know which window to resize, now you have to check the actual graphics (or more likely, a mask).
I am not saying it is the problem, but that's the kind of thing that can happen. Or it may be a simple bug, like a crash, memory corruption, an unhandled exception, the usual stuff, but they couldn't fix it in time and it is better to revert instead of leaving the buggy code or pushing an untested fix.
Just revert the code back to pre-26! This is ridiculous, it can't possibly be this hard and if it is, it just points to the degradation in the quality of Apple software! This is maddening!
This is already the pre-26 bounding box, isn't it? It's the new graphics that don't line up. (Not a great excuse, but the graphics are here to stay at least for a little while.)
> the graphics are here to stay at least for a little while
And that's the reason why I won't buy a new Mac.
Tahoe and Liquid Glass are so horrible that they're going to lose customers because of those. They should realize what they did and just backtrack: it wouldn't be the first time they admit they made a mistake [1].
The magic mouse have been there, almost unchanged, since 2009. That is a lot for a tech product, and retiring a product after 16 years is not admitting to a mistake. For example, the Logitech G5 mouse and its direct evolutions were among the most successful Logitech products, and it didn't last that long.
No, it is not just refusing to admit that the magic mouse was a mistake, it is considering that it is the best ever. That USB port on the underside is still one of the great mysteries though, maybe it is some quirk of evolution, because it is certainly not intelligent design.
In addition to vertical scrolling, the Magic Mouse can do horizontal (or diagonal) scrolling, zooming in and out, and a couple of other tricks. This makes it worthy for the people who need this for their work. There are mice that can do horizontal or vertical scrolling -- but not both at the same time.
People who do their work on large documents (pics in Photoshop, videos, CAD, music, even Excel, etc.) use these capabilities every day, and they like their Magic mice very much. If you are not one of these people (software development, for example, can be done with vertical scroll only, for the most part), it doesn't mean it's a bad product -- all it means it's a product which is not for you.
I don't use Magic Mouse but am very far from expecting Apple to admit "the magic mouse was a mistake" though.
I am using Sequoia and the windows are definitely rounded! Though the radius is pretty small (the curved region is about a quarter of the mouse cursor area), so the fact you can drag it from outside the window doesn't look ridiculous.
I think it shows how difficult it is to ship a seemingly easy thing inside the Apple machine.
I'm more interested in how or why this bug was approved up be worked on so quickly after it was surfaced, rather than other longstanding and arguably more impactful bugs.
It's because the bug got publicity. Apple marketing prioritizes what does and doesn't get built. Someone saw bad publicity on the front page of HN and requested a fix.
The answer is probably a ho-hum combination of different teams work on different issues, and this one having annoyed one of the devs who could work on it.
macOS does have weirdness with windows that span multiple screens. I bet some of that kicked in to an unacceptable level. It can create incoherent moving/snapping, for example. Has been kind of crazy-making for a while, for my set-up where screens are not joined but adjacent in a triangular configuration.
Yeah, that's something that was unambiguously better back in the "Classic MacOS" days (probably starting with the Mac II). Windows could overlap multiple screens and they were always drawn correctly.
At some point in OS X in the switch to hardware acceleration, they started rendering windows on one screen only.
I get that you hardly ever really want a window spanning two screens, but when you accidentally misplace a window it would be handy to be able to see it on each overlapping screen so you can track it down. Right now you can put a few pixels of the title bar on the wrong screen, and the rest of the window just vanishes.
These regressions are weird given that modern hardware is vastly more powerful than a Mac II.
Given that the video is fully interactive and lets you move around (in a “world” if you will) I don’t think it’s a stretch to call it a world model. It must have at least some notion of physics, cause and effect, etc etc in order to achieve what it does.
Pixel by pixel, time-slice by time-slice, in a 2D+T convolution. You provide enough examples of videos of changing point-of-view, and the model reproduces what it is given.
Yes, it reproduces what it is given by modelling the rules of physics, geometry, etc.
For example, image generators like stable diffusion carry strong representations of depth and geometry, such that performant depth estimation models can be built out of them with minimal retraining. This continues to be true for video generation models.
I’m not sure about “dumb them down”, I suspect it’s more like “subtly influence popular opinion”.
My husbands TickTock feed is full of things like “10 things Americans do that Chinese think are weird”, “10 reasons Chinese cities are in the 22nd century” etc etc.
I personally don’t think that’s propaganda - most of it is factually true and would be pushed to the front of any fair algorithm because it is engaging. But I can kinda see the concern, even though I disagree with the outcome.
