Tuesday, February 24
Daily News Stuff 24 February 2026
Hardly Working Edition
Hardly Working Edition
Top Story
- Is age verification a trap? Yes. (IEEE Spectrum)
Online age verification intrinsically damages user privacy, while failing to work in both directions:False positives are common. Platforms identify as minors adults with youthful faces, or adults who are sharing family devices, or have otherwise unusual usage. They lock accounts, sometimes for days. False negatives also persist. Teenagers learn quickly how to evade checks by borrowing IDs, cycling accounts, or using VPNs, and posting old Altered Images and Mental As Anything videos to their blogs that they've been running continuously since... 2003.
Okay, I'm maybe not so young that I need to worry about that problem.The appeal process itself creates new privacy risks. Platforms must store biometric data, ID images, and verification logs long enough to defend their decisions to regulators. So if an adult who is tired of submitting selfies to verify their age finally uploads an ID, the system must now secure that stored ID. Each retained record becomes a potential breach target.
One more quote:The age-verification trap is not a glitch. It is what you get when regulators treat age enforcement as mandatory and privacy as optional.
This is not an accident.
- But how big is the problem really? Surely nobody is out there putting terabytes of age-verification data in unprotected databases accessible to anyone on the internet oh that just happened again. (Tech Radar)
IDMerit, an AI-powered age-verification service, had three billion user records exposed in an unprotected MongoDB database.
Hey, at least they weren't running Elasticsearch.
Tech News
- Not so fast on that last story, says Technowize. (Technowize)
This was clearly the work of black-hearted Russian hackers trying to extort money from a wise and noble age verification service, says an article that reads exactly like an AI press release put out by a desperate company that has just lost control of three billion user records.
And down at the bottom of that page: ID Verification powered by IDMERIT.
Huh.
A very similar article appears at The HR Digest and again, down at the bottom: Powered by IDMERIT. Written - or "written" - by Diana Coker, whose work also features at... Technowize.
- A user who just wanted to control his DJI Romo vacuum cleaner inadvertently found himself in complete control over a global army of 7000 robots. (Tom's Hardware)
And the servers at DJI that store all the user data.
Because there was no security enabled at all.
You know how you sometimes scoff at film scenarios where people don't take the most basic common-sense security measures and the thieves (or the good guys, depending) just make off with everything?
Yeah. I've stopped scoffing.
- The new CEO of Microsoft Gaming - formerly in charge of AI Slop at Microsoft - says she has no tolerance for AI Slop. (Variety)
I have stopped scoffing in that one particular scenario.
- Why AI can't read PDF files. (The Verge)
Because PDF is closer to an image format than a document file. All the words you read are in there, but they might be broken into individual letters depending on the precise layout of the document and the program that generated it.
Solution: Render the PDF to a proper static image format and then use OCR to read it back.
Does that actually work?
No. Well, mostly, but as is so often the case it makes the easy parts easier and the hard parts explode.
- AI news app Particle listens to podcasts so you don't have to. (Tech Crunch)
I don't know. Couldn't I just... Not listen to the podcast?
Musical Interlude
Disclaimer: That song is so 80s and so Japanese it makes me want to accidentally blow up a planet.
Posted by: Pixy Misa at
06:51 PM
| Comments (4)
| Add Comment
| Trackbacks (Suck)
Post contains 618 words, total size 6 kb.
1
Posted the following on Ace, but might as well repeat it here, I suppose.
I had to extract some data from PDF files just a couple weeks ago. It was originally from a spreadsheet, but instead of being saved as .xlsx or even .csv, someone had done a "Print to PDF" operation and saved it as PDF. Can confirm that PDF isn't a text format: it's a bunch of "these letters are in this position on the page" data. If you copy and paste, text from different columns of the table end up copying next to each other with no spaces or anything between them.
I did actually find an AI tool that did a fairly reasonable job of looking at the PDF's structure and extracting the columns from the data. It made a lot of mistakes, getting some columns swapped and some rows swapped, but even fixing the mistakes it made ended up being less tedious than typing all that data in myself.
But this was an AI tool that was specifically trained on PDFs. Most tools aren't.
I had to extract some data from PDF files just a couple weeks ago. It was originally from a spreadsheet, but instead of being saved as .xlsx or even .csv, someone had done a "Print to PDF" operation and saved it as PDF. Can confirm that PDF isn't a text format: it's a bunch of "these letters are in this position on the page" data. If you copy and paste, text from different columns of the table end up copying next to each other with no spaces or anything between them.
I did actually find an AI tool that did a fairly reasonable job of looking at the PDF's structure and extracting the columns from the data. It made a lot of mistakes, getting some columns swapped and some rows swapped, but even fixing the mistakes it made ended up being less tedious than typing all that data in myself.
But this was an AI tool that was specifically trained on PDFs. Most tools aren't.
Posted by: Robin Munn at Tuesday, February 24 2026 07:37 PM (0QQsP)
2
As for the "listen to podcasts so you don't have to", transcribing a video (or an audio feed) in text format is actually one of the places where using an AI tool does make sense. I could take 15 minutes to watch some informative video, or I could read the video's transcript in five minutes or less. Guess which one I would prefer? (Though in some cases I'll watch the video anyway: if the guy is doing a demo of some piece of software, then actually seeing on-screen what he's doing has value that the transcript can't reproduce).
But finding the "most interesting" clips from a podcast, and just feeding you those? Yeah, that's dumb. You're going to miss context and misunderstand what's being said, and that's assuming the AI tool can accurately identify the actual most-interesting parts (it probably can't).
But finding the "most interesting" clips from a podcast, and just feeding you those? Yeah, that's dumb. You're going to miss context and misunderstand what's being said, and that's assuming the AI tool can accurately identify the actual most-interesting parts (it probably can't).
Posted by: Robin Munn at Tuesday, February 24 2026 07:41 PM (0QQsP)
3
Music Interlude - not a subscriber there, so it's not playing. Guess if I really wanted to, I could find it elsewhere.
Posted by: Frank at Tuesday, February 24 2026 08:50 PM (zCiG7)
4
I spent some time reading on python toolchains that allege they can handle PDFs. I was feeling optimistic about those handling stuff, but am less so now. (My use case may or may not be reasonable and sane.)
Posted by: PatBuckman at Wednesday, February 25 2026 12:51 AM (rcPLc)
56kb generated in CPU 0.0146, elapsed 0.1313 seconds.
58 queries taking 0.1214 seconds, 365 records returned.
Powered by Minx 1.1.6c-pink.
58 queries taking 0.1214 seconds, 365 records returned.
Powered by Minx 1.1.6c-pink.
