David Kriesel

And even if you do want to track people, there are ways to ask a lot nicer than the vast majority of cookie banners. This guy does it well I think: https://www.dkriesel.com/ The request to allow cookies is reasonable, genuine (I believe) and unintrusive. It does use a bit of a dark pattern by making the refuse button more faded, but I can empathize.

When you're designing a codec or some image processing pipeline, having a human in the loop is great. Designing codecs to have human-acceptable artefacts is as much art as science. Otherwise you end up like Xerox that had scanners which tended to compress "6" as "8", since they looked close-enough to their compression algorithm[1] [1]: http://dkriesel.com/en/blog/2013/0802_xerox-workcentres_are_switching_…

There is a talk about that issue [1]. During my PhD this issue came up amongst those in the group looking into compressed sensing in MRI. Many reconstruction methods (AI being a modern variant) work well because a best guess is visually plausible. These kinds of methods fall apart when visually plausible and "true" are different in a meaningful way. The simplest examples here being the numbers in scanned documents, or in the MRI case, areas of the brain where "normal brain tissue" was on average more plausible than "tumor". [1]: http://www.dkriesel.com/en/blog/2013/0802_xerox-workcentres_are_switch…

I think my favorite "you put cleverness where it didn't belong, and look what it got you" is the Xerox "oops we stored your scan compressed and our compression matched some characters as identical that weren't so the thing we scanned has characters substituted in the image". [1] Honorable mention goes to: - the Dell switch with the changelog note "if you had a device with a MAC starting with these bytes plugged in, the last N ports on the switch will stop working" [1] - https://www.dkriesel.com/en/blog/2013/0802_xerox-workcentres_are_switc…

Thanks for the complementary links! Regarding Xerox scanner compression issues, compare this great CCC-talk by David Kriesel, "Traue keinem Scan, den du nicht selbst gefälscht hast" [0] – Sorry, German only. [0] https://www.youtube.com/watch?v=7FeqF1-Z1g0 [1] http://www.dkriesel.com/en/blog/2013/0802_xerox-workcentres_are_switch… (Didn't MS's PDF-viewer have similar issues?)

When my Xerox Story was hit last year (http://www.dkriesel.com/en/blog/2013/0802_xerox-workcentres_are_switch… I got about 200k hits in two days, at peaks reaching about 10k per hour. All numbers are from google analytics so real values might be higher. Also, it's not only hn, consider also the follow-ups (in my case, shortly after hn, the front pages of slashdot and reddit hit me as well).

Xerox also explicitly recommends the lossy/lousy/normal setting if you need to send the scan over a network. http://www.dkriesel.com/_media/blog/2013/colorqube.jpg

Previous discussion: https://news.ycombinator.com/item?id=6156238 Follow-up blog post about a conference call with Xerox: http://www.dkriesel.com/en/blog/2013/0806_conference_call_with_xerox

Lossy compression does make things up. We call them compression artefacts. In compressed audio these can be things like clicks and boings and echoes and pre-echoes. In compressed images they can be ripply effects near edges, banding in smoothly varying regions, but there are also things like https://www.dkriesel.com/en/blog/2013/0802_xerox-workcentres_are_switc… where one digit is replaced with a nice clean version of a different digit, which is pretty on-the-nose for the LLM failure mode you're talking about. Compression artefacts generally affect small parts of the image or audio or video rather than replacing the whole thing -- but in the analogy, "the whole thing" is an encyclopaedia and the artefacts are affecting little bits of that. Of course the analogy isn't exact. That would be why S.W. opens his post by saying "Since I love collecting questionable analogies for LLMs,".

There's an amazing talk about DB's punctuality from the 2019 Chaos Communication Congress: BahnMining, David Kriesel. Available in German, and interpreted to English: https://dkriesel.com/blog/2019/1229_video_und_folien_meines_36c3-vortr… My favorite anecdote from riding a German train in Switzerland was a journey from Interlaken (Switzerland) to Frankfurt in Germany. Now if you're not used to train schedules: Stops are quick, and trains are punctual. It's fairly common for stops to just be 2-3 minutes. Connections between trains are often around 5 minutes, sometimes shorter. So, this Intercity Express departed Interlaken perfectly on schedule, made a bunch of scheduled stops in Spiez, Thun, Bern, and continued on to the border city of Basel. Everything running perfectly on schedule, the train arrived at the Swiss station in Basel, continued (on time) to the German train station that's still in the same city, just a few minutes farther. Having had 3-5 minute stops all the way from early morning to lunchtime, the train then sat in the German station for about 15 minutes, departed with only a minor delay, and just a few yards after exiting the station proceeded to stop, sit on the tracks for half an hour ("to catch up with the prevalent delay conditions on the German rail network"), before starting to move again. Waiting in a comfy seat on a train with working AC isn't the worst of fates, and we made the final destination within half an hour of the scheduled…

Favourite fax stories: 1. Different standards, so if something doesn't go through between you and a party only, each side blames the other. "I'm receiving faxes from others, so the problem is on your end". "My faxes to others are going through, so the problem must be you". 2. Some fax machines would use some OCR tech to "sharpen" faxes. e.g. close the loop on some 6s and turn them into 8s. Exaggerated example: "I did respond to you. I sent a fax to the number on your own faxes to me. Your fax said your fax number is 1-888-888-8888 and I have the confirmation here". (turns out 888-888-8888 was an unrelated toy store's fax machine in a different city). "I have my letterhead right here and our fax number is 1-666-666-6666, are you drunk?" http://www.dkriesel.com/en/blog/2013/0802_xerox-workcentres_are_switch… 3. If you have two fax machines and aren't receiving any for a while, you can do a thorough self-test by sending a fax from one to the other. 4. Needing to cancel a doctor's appointment, but not getting through over the phone, I just sent a fax.

> and connected them realistically… Well, that's the crux of the issue that you don't seem to see. For many observers, it does not look realistic. The unnaturalness of the pixels is what makes it plausible for Apple software to have a bug. If it looked 100% realistic to everyone, we wouldn't be in this thread discussing it. E.g. Image saving algorithm with a software bug can cause pixels to be incorrectly "replaced" even though the programmers never intended it: https://www.dkriesel.com/en/blog/2013/0802_xerox-workcentres_are_switc… EDIT reply to: >, especially when the image so clearly depicts that. You still seem to be missing why this discussion of AI error was even possible. In this very thread, there are commenters who still don't see "clear depictions" of a realistic foreground leaf even after being made aware of parallax demonstration in the update... because the blurriness keeps hijacking the brain to make it look like a fake artifact. Consider that the actual person who took the original photo didn't realize it was a real foreground leaf. (Or so he claims.) In other words, saying "Am I crazy? I don't see why there's debate about the dress being blue or white when ... it is clearly and realistically BLUE" -- doesn't actually add to the discussion. Hey, it's great that the foreground leaf looks realistic to you. But your personal perception is not relevant to why others perceived it as an artifact.

I can totally see Google incorporating this in Self driving cars down the road, after training on millions of hours or dashcam footage, to augment or maybe even replace Lidar. Paper suggests it is capable of segmenting 2D picture input into logical objects and their discrete configurations/positions. Its not without its pitfalls tho, instead of Monocular SLAM generating factual, albeit fuzzy point cloud map we get overfitted (5 pictures x 2M similar scenes) magic black box generating very training set specific hallucination. This is how we get Scanners replacing numbers in scanned documents http://www.dkriesel.com/en/blog/2013/0802_xerox-workcentres_are_switch… Similar example was posted 2 months ago https://data.vision.ee.ethz.ch/aeirikur/extremecompression/ example picture (no doubt best case authors could manage) gained additional data absent from original, some of it dangerous like fake license plate numbers.

> people are upset by certain types of enhancement error, but not others, and they can't even articulate the difference to themselves People are upset by enhancement errors that produce an image that is meaningfully different from the "true image". Replacing the sun with the moon or replacing the letters with different letters (e.g. https://www.dkriesel.com/en/blog/2013/0802_xerox-workcentres_are_switc… ). The philosophical problem here is that you can't get at the true image, you can only imagine ideas like "what the human originally taking the photo saw with their naked eye".

Yeah. There was a weird Xerox printer bug that swapped digits (turning 6s into 8s) on scanned documents caused by the JBIG2 image format [1]. [1] https://www.dkriesel.com/en/blog/2013/0802_xerox-workcentres_are_switc…

The problem is when you can't tell which bits are unmangled. OCR systems will happily give you plausible but wrong readings, and even some scanners/copiers will change things: https://dkriesel.com/en/blog/2013/0802_xerox-workcentres_are_switching…

> I would definitely understand post processing, like extracting data, answering question .. etc, but why re-doing the OCR engine itself? Well... the idea seems to be (as far as I understand it, at least) that optical errors and artifacts can now be compensated as the OCR engine is now context-aware. Say, for example, some random long ass name chemical. It's not going to be in a word correction database, but a context-aware engine (ideally, one that has been supplemented with chemistry data) can now correct "bad" reads of the chemical's name. Of course, there remains the issue of how to prevent the infamous Xerox bug [1]... [1] https://www.dkriesel.com/en/blog/2013/0802_xerox-workcentres_are_switc…

My first, second, and third instinct here is to say this is pretty obvious and sloppy fraud. But it did remind me of the famous case discovered by David Kriesel where Xerox scanners changed documents in surprising ways. The caption on the YouTube video linked here is entertainingly accurate. https://www.dkriesel.com/en/blog/2013/0802_xerox-workcentres_are_switc… "On the scale of things too horrible to contemplate, "document-altering scanner" is right up there with "flesh-eating bacteria". Since 2006, Xerox scancopiers literally are making stuff up. They, for example, replace digits with others in scans. The replacement digits are layouted perfectly into the page, so the errors are hard to see. Sounds unbelievably insidious, but it's true. Drug prescriptions, construction plans, just anything can be affected. "

Maybe another old example of this scanner bug? https://www.dkriesel.com/en/blog/2013/0802_xerox-workcentres_are_switc…

This is both good and bad. Good ASR can often understand low quality / garbled speech that I could not figure out, but it also "over corrects" sometimes and replaces correct but low prior words with incorrect but much more common ones. With OCR the risk is you get another xerox[1] incident where all your data looks plausible but is incorrect. Hope you kept the originals! (This is why for my personal doc scans, I use OCR only for full text search, but retain the original raw scans forever) [1] https://www.dkriesel.com/en/blog/2013/0802_xerox-workcentres_are_switc…

If I recall it was an artifact of the compression algo. Full context and details: https://www.dkriesel.com/en/blog/2013/0802_xerox-workcentres_are_switc…

>> sometimes replaces characters with what it thinks is a more likely character from context Like the JBIG2 algorithm used in a zero click PDF-as-GIF exploit in iMessage a while back: https://projectzero.google/2021/12/a-deep-dive-into-nso-zero-click.htm… The vulnerability of that algorithm to character-swapping caused incorrect invoices, incorrect measurements in blueprints, incorrect metering of medicine, etc. https://www.dkriesel.com/en/blog/2013/0802_xerox-workcentres_are_switc…

Whenever I read about Xerox, it reminds me of the story that their scanners would randomly change numbers on prints https://dkriesel.com/en/blog/2013/0802_xerox-workcentres_are_switching…

Quite some years ago there was some data backed evidence of this I would say. https://www.dkriesel.com/blog/2019/1229_video_und_folien_meines_36c3-v… Here is the link to his talk directly https://youtu.be/0rb9CfOvojk?si=7EImZU9x4zFb6LSf Its true that the network is quite dense and used by also cargo trains, but there is no denying that things got worse and worse. I constantly experience delay to do some stuff not working. I forgot even the minute threshold when a train is still punctual according to DB. I believe it's 10min by now, which can be deadly if you need to switch trains :)

> To make this even worse, it may even produce Y just enough times to make it seem reliable and then it is unleashed without supervision, running thousands or millions of times, wrecking havoc producing Z in a large number of places. Reminds me of the famous Xerox scanning bug: https://www.dkriesel.com/en/blog/2013/0802_xerox-workcentres_are_switc…

Please no. This is what jbig2 does for images and it’s a nightmare in my view, you can’t trust the result is not something totally different from the original [1] [1]https://www.dkriesel.com/en/blog/2013/0802_xerox-workcentres_are_switc…

The case you might be thinking of is the JBIG2 implementation bug [1, 2] in Xerox photocopiers where the pattern-matching would incorrectly treat certain characters as interchangeable, leading to numbers getting rewritten in spreadsheets. [1] https://www.bbc.com/news/technology-23588202 [2] https://www.dkriesel.com/en/blog/2013/0810_xerox_investigating_latest_…

I'm scared of something like the Xerox number-corruption bug [0], where some models will subtly fuck everything up in a way that is too expensive to recover from by the time it's discovered. [0] https://www.dkriesel.com/en/blog/2013/0802_xerox-workcentres_are_switc…

>Unlike traditional OCR systems that fail obviously when uncertain, LLMs make educated guesses that appear plausible but may be entirely wrong. Except for a very special kind of bug: https://www.dkriesel.com/en/blog/2013/0802_xerox-workcentres_are_switc… >Xerox scanners/photocopiers randomly alter numbers in scanned documents

Could this be a repeat of the Xerox image duplication bug? https://www.dkriesel.com/en/blog/2013/0802_xerox-workcentres_are_switc…