If you scan a thousand British faces and compare them to a thousand criminals, you will do 1,000,000 comparisons. (that's the birthday paradox part).
If your error rate is 0.8%, you'll get roughly 8,000 false positives and negatives.
That's bad enough if they are all false positives: people get arrested, then released.
It's way worse if they are all false negatives: 8,000 criminals get ignored by the police dragnet.

That was Britain: false positives are life-threatening in countries where the police carry guns.
0.8% is a good error rate. 34% wrong is typical in matching black women. See
https://www.aclu-mn.org/en/news/biased-technology-automated-discrimination-facial-recognition#:~:text=Studies%20show%20that%20facial%20recognition%20technology%20is%20biased.,published%20by%20MIT%20Media%20Lab.

At a certain company long long ago, managers had a mainframe-based planning app that looked like a sort of spreadsheet.

The company did a study to see how much it improved the manager's team's productivity...

Oops! Use of the tools was correlated with declining productivity.

Alas, if a 16-year-old is doing the buying over the net, Bob's out of luck. Bob can't even tell if the customer is a dog (:-))
The parent who buys their kid a phone is a really good source of the kid's age. It's not leak-proof, but most other approaches leak like sieves.

Once upon a time, lint would critique your work, at the expense of often being wrong. I tried using LLMs for that, and they work only a little less well than lint.
I was insulted, though, when It misquoted my sentence and then said I had a grammar error in the misquoted part (:-))

Steven Rostedt wrote -
"I played a little with [Rust] in user space, and I just absolutely hate the cargo concept... I hate having to pull down other code that I do not trust. At least with shared libraries, I can trust a third party to have done the build and all that..."

The various crate-like things are a fad. The arguably correct way of using shared libraries was reinvented independently by the Gnu libc team and by Solaris, from a first use in Multics. You remember, Unix's papa and Linux's grandpa?

Give it a few years, the hype bubble for importing static libraries will burst, and shared libraries with updaters and downdaters will be re-re-invented.

From the Canadian Government page cited below:

A person given a "quit or return to the office" has been fired, and can sue the pants off the employer. The lawyer involved may well offer a good price on a suit to everyone the employer fired, thus increasing the risk to the employer.

See https://www.canada.ca/en/emplo..., or google for "lawyer constructive dismissal" if you're not in Canada

The company is testing humans for their ability to do something they are inherently bad at.
Filtering programs, such as the one at spamcop.net, do it well:
- I haven't had a false positive for about three years.
- I get a false negative about once a month.

Whenever I get an email at a customer's, I run it through the spamcop filter. That reliably identifies the phishing-test emails,

I prefer to report those on the equivalent of the IT slack channel, so others aren't caught out by them (;-))

That's a lot like character device drivers in CP/M, which would fill/clear their queues before returning. The queues were short.

Need I say more?

More than 10 years ago, my company tried to do facial recognition in an airport in Europe, for their security service. Alas, they didn't know about the "birthday paradox", and tried to match about 1,000 criminals against several thousand passengers. They shut it down when the system identified someone's grandmother as a male member of the Baader-Meinhof gang.

The (birthday) paradox is caused by trying to match each passenger against 1,000 criminals, not just one. Even with only a 1% error rate, there will be 10 false positives and negatives per passenger. And we don't have anything like a 1% error rate.

We need a 1/infinity error rate (:-)) Otherwise innocent grandmothers will be pulled aside, while actual criminals will breeze on through.

Its security is deliberately hard to engage. Instead, it's an "attractive nuisance", a fine way to communicate your complaints about a certain government to your friends, and be sure they are brought to the attention of that country's security service (:-))

Alas, HR departments will wonder why they're getting fewer and fewer applications. Only if hiring manager need to sign off on ads for their positions will one have a good chance of getting a valid skills list and job description.

My company used to own a headhunter. We had to have a co-op student train them on AND and OR before we dared use them for our own job postings. Their training doesn't prepare them for hiring people with skills.

Previous folks announced they had AI, talking about
- rule-based systems written in lisp and prolog
- machine-learning systems written in math and stats
- large language learning-models written in harder-to-evaluate math

All but the last was massively different from the previous.

I'll therefor be waiting to see another massive change before I believe they've made an advance toward AGI. Right now they're at MachineLearning++ running on ArrayProce$$or$++.

Consider a few examples of successful forks:

• - LibreOffice -formerly OpenOffice, formerly StarOffice
• - OpenStreetMap -growing Google Maps replacement
• - OpenTofu - fork of Terraform
• - KeyDB - Redis alternative
• - OpenSearch (AWS-backed) - Elasticsearch alternative

The scholarly publishing ecosystem runs largely on unpaid academic labour. Give that, I wonder if a university press or three should pick up the task of publishing academic journals, and incidentally using some of those profits to improve the journal's quality. By paying editors.

The effects were predicted, and the proponents went on and did it anyway.
The result? it's exactly what they planned.