• 5 Posts
  • 38 Comments
Joined 1 year ago
cake
Cake day: October 4th, 2023

help-circle

  • I think that driverless busses are probably much less of a dramatic change than driverless cars.

    If you have one person in a car driving to work and the car is fully-self-driving, then you free up one person’s time. You potentially change where parking is practical. You may permit people who cannot drive a car to use one, like young or elderly.

    With a bus, the passengers are already free to do what they want. You’re saving labor costs on a bus driver, maybe getting a safer vehicle. But I’d call that an evolutionary change.

    https://proxy.parisjc.edu:8293/statistics/300887/number-of-buses-in-use-by-region-uk/

    In 2020/21, the number of buses amounted to 37800 in Great Britain.

    Those probably get heavier use than cars. But you want scale, since driverless vehicle costs are mostly fixed, and driver labor costs variable. You’re talking about not having maybe 38k people driving. You need to cover all of your costs out of that. That’s not nothing, but…okay, how many tractor-trailers are out there?

    https://www.statista.com/topics/5280/heavy-goods-vehicles-in-the-uk/

    Heavy goods vehicle registrations bounced back above their pre-pandemic levels in 2021, reaching 504,600 vehicles in circulation.

    If you have driverless trucks, that’s an order-of-magnitude difference in vehicle count from busses in the UK.

    I’m not saying that there aren’t wins possible with self-driving busses. But it doesn’t seem to me to be the vehicle type with the greatest potential improvement from being self-driving.



  • So, this isn’t quite the issue being raised by the article – that’s bug reports generated on bug trackers by apparently a bot that they aren’t running.

    However, I do feel that there’s more potential with existing LLMs in checking and flagging potential errors than in outright writing code. Like, I’d rather have something like a “code grammar checker” that highlights potential errors for my examination rather than something that generates code from scratch itself and hopes that I will adequately review it.







  • On the other hand, there are things that a human artist is utterly awful at, that LLM-based generative AIs are amazing at. I mentioned that LLMs are great at producing works in a given style, can switch up virtually effortlessly. I’m gonna do a couple Spiderman renditions in different styles, takes about ten seconds a pop on my system:

    Spiderman as done by Neal Adams:

    Spiderman as done by Alex Toth:

    Spiderman in a noir style done by Darwyn Cooke:

    Spiderman as done by Roy Lichtenstein:

    Spiderman as painted by early-19th-century American landscape artist J. M. W. Turner:

    And yes, I know, fingers, but I’m not generating a huge batch to try to get an ideal image, just doing a quick run to illustrate the point.

    Note that none of the above were actually Spiderman artists, other than Adams, and that briefly.

    That’s something that’s really hard for a human to do, given how a human works, because for a human, the style is a function of the workflow and a whole collection of techniques used to arrive at the final image. Stable Diffusion doesn’t care about techniques, how the image got the way it is – it only looks at the output of those workflows in its training corpus. So for Stable Diffusion, creating an image in a variety of styles or mediums – even ones that are normally very time-consuming to work in – is easy as pie, whereas for a single human artist, it’d be very difficult.

    I think that that particular aspect is what gets a lot of artists concerned. Because it’s (relatively) difficult for humans to replicate artistic styles, artists have treated their “style” as something of their stock-in-trade, where they can sell someone the ability to have a work in their particular style resulting from their particular workflow and techniques that they’ve developed. Something for which switching up styles is little-to-no barrier, like LLM-based generative AIs, upends that business model.

    Both of those are things that a human viewer might want. I might want to say “take that image, but do it in watercolor” or “make that image look more like style X, blend those two styles”. LLMs are great at that. But I equally might want to say “show this scene from another angle”, and that’s something that human artists are great at.


  • I think creating a lora for your character would help in that case.

    A LORA is good for replicating a style, where there’s existing stuff, helps add training data for a particular subject. There are problems that existing generative AIs smack into that that’s good at fixing. But it’s not a cure-all for all limitations of such systems. The problem I’m referring to is kinda fundamental to how the system works today – it’s not a lack of training data, but simply how the system deals with the world.

    The problem is that the LLM-based systems today think of the world as a series of largely-decoupled 2D images, linked only by keywords. A human artist thinks of the world as 3D, can visualize something – maybe using a model to help with perspective – and then render it.

    So, okay. If you want to create a facial portrait of a kinda novel character, that’s something that you can do pretty well with AI-based generators.

    But now try and render that character you just created from ten different angles, in unique scenes. That’s something that a human is pretty good at. Here’s a page from a Spiderman comic:

    https://spiderfan.org/images/title/comics/spiderman_amazing/031/18.jpg

    Like, try reproducing that page in Stable Diffusion, with the same views. Even if you can eventually get something even remotely approximating that, a human, traditional comic artist is going to be a lot faster at it than someone sitting in front of a Stable Diffusion box.

    Is it possible to make some form of art generator that can do that? Yeah, maybe. But it’s going to have to have a much more-sophisticated “mental” model of the world, a 3D one, and have solid 3D computer vision to be able to reduce scenes to 3D. And while people are working on it, that has its own extensive set of problems. Look at your training set. The human artist slightly stylized stuff or made errors that human viewers can ignore pretty easily, but a computer vision model that doesn’t work exactly like human vision and the computer vision system might go into conniptions over. For example, look at the fifth panel there. The artist screwed up – the ship slightly overlaps the dock, right above the “THWIP”. A human viewer probably wouldn’t notice or care. But if you have some kind of computer vision system that looks for line intersections to determine relative 3d positioning – something that we do ourselves, and is common in computer vision – it can very easily look at that image and have no idea what the hell is going on there. Or to give another example, the ship’s hull isn’t the same shape from panel to panel. In panel 4, the curvature goes one way; in panel 5, the other way. Say I’m a computer vision system trying to deal with that. Is what’s going on there that there ship is a sort of amorphous thing that is changing shape from frame to frame? Is it important for the shape to change, to create a stylized effect, or is it just the artist doing a good job of identifying what the matters to a human viewer and half-assing what doesn’t matter? Does this show two Spidermen in different dimensions, alternating views? Are the views from different characters, who have intentional vision distortions? I mean, understanding what’s going on there entails identifying that something is a ship, knowing that ships don’t change shape, having some idea of what is important to a human viewer in the image, knowing from context that there’s one Spiderman, in one dimension, etc. The viewer and the artist can do it, because the viewer and the artist know about ships in the real world – the artist can effectively communicate an idea to the viewer because they not only have hardware that processes the thing similarly, but also have a lot of real-world context in common that the LLM-based AI doesn’t have.

    The proportions aren’t exactly consistent from frame to frame, don’t perfectly reflect reality, and might be more effective at conveying movement or whatever than an actual rendering of a 3d model would be. That works for human viewers. And existing 2D systems can kind of dodge the problem (as long as they’re willing to live with the limitations that intrinsically come with a 2D model) because they’re looking at a bunch of already-stylized images, so can make similar-looking images stylized in the same way. But now imagine that they’re trying to take stylized images, then reduce them into a coherent 3D world, then learn to re-apply stylization. That may involve creating not just a 3D model, but enough understanding of the objects in that world to understand what stylization is reasonable to create a given emotional effect and be reasonable to a human, and when. People may not care that that ship is doing some impossible geometry, but might care a whole lot about the numbers of limbs that Spiderman has. Is it technically possible to make such a system? Probably. But is it a minor effort to get there from here? No, probably not. You’re going to have to make a system that works wildly differently from the way that the existing systems do. That’s even though what you’re trying to do might seem small from the standpoint of a human observer – just being able to get arbitrary camera angles of the image being rendered.

    The existing generative AIs don’t work all that much the way a human does. If you think of them as a “human” in a box, that means that there are some things that they’re gonna be pretty impressively good at that a human isn’t, but also some things that a human is pretty good at that they’re staggeringly godawful at. Some of those things that look minor (or even major) to a human viewer can be worked around with relatively-few changes, or straightforward, mechanical changes. But some of those things that look simple to a human viewer to fix – because they would be for a human artist, like “just draw the same thing from another angle” – are really, really hard to improve on.


  • there’s some stuff image generating AI just can’t do yet

    There’s a lot.

    Some of it doesn’t matter for certain things. And some of it you can work around. But try creating something like a graphic novel with Stable Diffusion, and you’re going to quickly run into difficulties. You probably want to display a consistent character from different angles – that’s pretty important. That’s not something that a fundamentally 2D-based generative AI can do well.

    On the other hand, there’s also stuff that Stable Diffusion can do better than a human – it can very quickly and effectively emulate a lot of styles, if given a sufficient corpus to look at. I spent a while reading research papers on simulating watercolors, years back. Specialized software could do a kind of so-so job. Stable Diffusion wasn’t even built for that, and with a general-purpose model, it already can turn out stuff that looks rather more-impressive than those dedicated software packages.







  • I’m an android user and I shred my files using a app that uses an algorithm that overwritten that bytes of the file

    I suspect that it doesn’t actually work. I mean, they can overwrite the logical positions in the file file if they want, but that doesn’t entail that it actually overwrites the underlying physical blocks, for a number of reasons, starting with some of the stuff at the drive level, but also because of higher-level issues. What filesystem does Android use?

    googles

    Looks like yaffs2, at least on this system.

    https://stackoverflow.com/questions/2421826/what-is-androids-file-system

    rootfs / rootfs ro 0 0
    tmpfs /dev tmpfs rw,mode=755 0 0
    devpts /dev/pts devpts rw,mode=600 0 0
    proc /proc proc rw 0 0
    sysfs /sys sysfs rw 0 0
    tmpfs /sqlite_stmt_journals tmpfs rw,size=4096k 0 0
    none /dev/cpuctl cgroup rw,cpu 0 0
    /dev/block/mtdblock0 /system yaffs2 ro 0 0
    /dev/block/mtdblock1 /data yaffs2 rw,nosuid,nodev 0 0
    /dev/block/mtdblock2 /cache yaffs2 rw,nosuid,nodev 0 0
    /dev/block//vold/179:0 /sdcard vfat rw,dirsync,nosuid,nodev,noexec,uid=1000,gid=1015,fmask=0702,dmask=0702,allow_utime=0020,codepage=cp437,iocharset=iso8859-1,shortname=mixed,utf8,errors=remount-ro 0 0
    

    https://en.wikipedia.org/wiki/YAFFS

    YAFFS is a robust log-structured file system that holds data integrity as a high priority. A secondary YAFFS goal is high performance. YAFFS will typically outperform most alternatives.[3] It is also designed to be portable and has been used on Linux, WinCE, pSOS, RTEMS, eCos, ThreadX, and various special-purpose OSes. A variant ‘YAFFS/Direct’ is used in situations where there is no OS, embedded OSes or bootloaders: it has the same core filesystem but simpler interfacing to both the higher and lower level code and the NAND flash hardware.

    Yeah, note the “log-structured” bit there.

    https://en.wikipedia.org/wiki/Log-structured_file_system

    A log-structured filesystem is a file system in which data and metadata are written sequentially to a circular buffer, called a log.

    So, what happens is that when you write, it’s going to the log, and then there’s a metadata update once the write is complete saying “I wrote to the log”. The app probably isn’t writing to the previous location of the data on the disk, because writing to byte offset 32,000 the second time in a file will go to a different logical location on the storage device than the first time you wrote it, causing the thing to not actually be overwritten.

    googles

    https://arxiv.org/pdf/1106.0917

    Secure Deletion on Log-structured File Systems

    We address the problem of secure data deletion on log-structured file systems. We focus on the YAFFS file system, widely used on Android smartphones. We show that these systems provide no temporal guarantees on data deletion and that deleted data still persists for nearly 44 hours with average phone use and indefinitely if the phone is not used after the deletion. Furthermore, we show that file overwriting and encryption, methods commonly used for secure deletion on block-structured file systems, do not ensure data deletion in log-structured file systems.

    I’d also note that this is a lead-up to proposed solutions, but that’s only handling things down to the level that the OS sees, not what the flash device sees; they don’t mention things like wear leveling, so they probably aren’t taking that into consideration.

    EDIT: Oh, they do mention it, but just to say that some of their approach might work (like, what they mean is that if it writes enough data in the background, it might eventually overwrite whatever, even if the OS has no control as to what’s being written):

    Wei et al. [16] have considered secure deletion on flash storage in the context of solid state drives (SDDs). An SSD makes use of a Flash Translation Layer (FTL). This layer allows a regular block-based file system (such as FAT) to be used on flash memory by handling the nuances of erase blocks opaquely through the FTL’s layer of indirection. This layer has the same effect as a log-structured file system, where the FTL writes new entries at empty locations, so old entries remain until the entire erase block can be reclaimed. They executed traditional block-based approaches to secure deletion and determined that they do not properly sanitize data on flash storage. They also showed alarmingly that some built-in sanitization methods do not function correctly either. They propose to address this concern by having flash hardware manufacturers make use of zero overwriting, and add it into the FTL hardware. They state that circumventing the problem of a lack of secure deletion requires changes in the FTL, but depending on how the FTL is implemented, our userlevel approaches may also succeed similarly without requiring hardware changes.


  • But despite that mistake and a few other hiccups—my punctuation seemed unnatural because it was too accurate—Daniel offered me the job.

    “Baby, I like talking to you and all…but I notice that you’re using unspaced em-dashes as the alternate form for parenthetical phrases. No other girl I’ve ever met on OnlyFans has done that.”

    Oh shit.

    The agency’s manager sent me a background memo about the woman I’d be playing, a purported 21-year-old university student blessed with physical proportions that are in vogue these days. To ensure that my performance was as authentic as possible, I spent two hours committing all of her details to memory: her favorite programming language

    You want to avoid those awkward, immersion-breaking moments in the chat where you’ve got some snippets of code in language that the client happens to know and you don’t, I imagine. The real pro route is to choose something adequately-esoteric that nobody is likely gonna call you out on it.

    “When it’s late at night, sometimes I like to relax with a little coding in REXX, though Prolog is good to mix things up.”

    I was to be paid 7 cents per line of dialog, with each dialog running for a minimum of 40 lines. For my first assignment, I had to compose 20 dialogs involving sex in public places—10 at the beach, five inside a car, and five in a forest or garden. There was a list of particular sex acts I had to include, as well as a stricture that I refrain from using emoji in more than 30 percent of lines. I had only 48 hours to complete the task.

    Yeah, I can see why they want to get AI chatbots working for this.

    Robert Carey, a Phoenix-based partner at the law firm Hagens Berman, which specializes in massive class actions, has a less charitable view of the matter. In the midst of my plunge into the chatting industry, I caught wind that he was looking for men to become plaintiffs in a class action against both OnlyFans and the agencies who hire chatters. A lead attorney in the lawsuits that revolutionized college sports by making it possible for student-athletes to get paid for name and image rights, Carey argues that the managers who run creators’ accounts are engaging in a type of bait and switch that fits the classic definition of fraud. “When you subscribe, the very first thing it says is, ‘Have a DM relationship,’” he said. “Well, that’s totally fraudulent … It’s an open secret they’re just defrauding people.”

    Carey, who confided in me that his firm plans to file its lawsuit soon, contends that the chatting illusion can lead to serious harm for unwitting subscribers.

    Hmm. I wonder how that works. Do they, in the discovery phase of this lawsuit, just require all of the service’s chat logs to be handed over?



  • I don’t know what happened, but I put together a PC for the first time in some years, and holy mother of God, all the components have RGB LEDs slapped on them now. I had to actively work to find parts that didn’t have RGB LEDs on them (and I still accidentally wound up with some on the motherboard). I mean, yeah, LED case fans have been a thing for a while, and there was always a contingent that put electroluminescent strips on their computers. And it kinda grew into a lot of keyboards and mice. But now it’s a large portion of CPU fans, most cases, RAM sticks have RGB LEDs, motherboards have RGB LEDs. I didn’t have trouble finding non-RGB LED NVMe storage, or non-RGB LED SATA drives, but even there, you can get them. Hell, there are RGB LED cables.

    I can only assume that a large portion of the people building PCs these days are doing it to have them physically blinged up.

    Like, nothing wrong with wanting to do that, but I couldn’t believe the tiny proportion that wasn’t doing that.


  • Eh, that’s been a thing for a long time. Decades at least.

    I think that the problem is that there isn’t really a great term to clearly refer to the “non-monitor-and-peripherals” part of the “computer”. “Case” would refer to just the case, not what’s in it. “Tower” or “desktop” is overspecific, refers to particular form factors. I have a tower, but some people have under-monitor desktops (though that’s rare today) or various times of small form factor PCs. If I say “computer”, that doesn’t really clearly exclude peripherals.

    And honestly, we don’t really use the term “GPU” quite correctly either. I’ll call a whole PCI video card a “GPU”, but I suppose that strictly-speaking, that should only be talking about a specific chip on the card.