ano-ther

Even all of Austria

Except for cannabis related diagnoses, their effect size (change in the incidence rate ratio ∆IRR) is of a similar magnitude as opioid-related diagnoses which they included for falsification testing and said it did not change.

Am I misinterpreting their statistics?

> Cannabis-related diagnoses (∆IRR 1.138, 95% CI 1.066 to 1.214, p<0.001),

> AF (∆IRR 1.023, 95% CI 1.010 to 1.035, p<0.001),

> HF (∆IRR 1.037, 95% CI 1.029 to 1.046, p<0.001),

> MI (∆IRR 1.031, 95% CI 1.014 to 1.048, p<0.001),

> ischaemic stroke (∆IRR 1.011, 95% CI 0.999 to 1.024, p=0.078),

> VTE (∆IRR 1.009, 95% CI 0.999 to 1.020, p=0.066)

> and VT (∆IRR 1.019, 95% CI 0.993 to 1.046, p=0.161).

> For falsification testing, there was no change in hospitalisations for

> urinary stones (∆IRR 1.006, 95% CI 0.986 to 1.027, p=0.535) or

> opioid-related diagnoses (∆IRR 0.966, 95% CI 0.927 to 1.006, p=0.091).

Ah yes. Network effects of course.

Must have been quite a curveball for Taler. Good luck on new venture!

Had the same question. The article has an Amazon link to the bearings.

That’s bad, but the headline seems to suggest it was the tax system, when it was actually his fraudulent business manager:

> "I write a check to go pay my taxes, and instead of her using it to pay my taxes, she would send me a fake filing of taxes and take that money."

I am always amazed how these complicated systems evolved through random variations.

There must be plenty of “versions” which don’t do anything but generations later are crucial to a function.

How does evolution not get stuck in a local optimum?

With two people in the company, there is not a lot of room for corporate games though.

> According to Denmark’s Central Business Register, Pays ApS had two employees as of July 2026.

So it was actually two weaknesses:

* The non-password at a two-person IT company (Pays ApS)

* And then completely unchecked access to the CPR database for 22 days which apparently does not have monitoring or limits if someone tries to access all the records (they must have made some 16k downloads per hour).

I really would like to see their tests and the model’s reasoning traces.

Why would it go to PhillyUnsolvedMurders.com and decide to make up a crime report? And what did it do with the other websites?

> Anthropic said its model was conducting a test involving interactions with randomly selected websites when it accessed PhillyUnsolvedMurders.com and submitted false information on an unsolved homicide. The tip claimed to come from someone with information on the case.