David Kriesel

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-vortrags_bahnmining.

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…

> 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_switching_written_numbers_when_scanning

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.

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_switching_written_numbers_when_scanning

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.

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_switching_written_numbers_when_scanning

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.

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_switching_written_numbers_when_scanning 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,".

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_switching_written_numbers_when_scanning

"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.
"