The problem is that doctors are making elementary errors, failing to verify, and putting ego and large numbers of consultations a day over and above the wellbeing of patients.

That, to me, is gross malpractice.

The correct answer is not necessarily more AI, but that might well be the end result. The correct answer is to require doctors to recertify through such test cases and withdrawing a license to practice if the success rate is under 90%.

AI is, ultimately, just using differential diagnosis, because that's how AI works. Differential diagnosis is what doctors are supposed to use. If they were, and took the time needed, their scores should be identical to AI. If it is less, then they are taking shortcuts that are medically unjustified and that should be grounds for dismissal.

Instead of higher throughput, doctors need to have higher success rates. Don't give them a choice. If that makes medicine more expensive, oh well. It's better than a bunch of dead patients.

The US elected the US as world police.

That's how you cast yourselves, that's the world you created by becoming a superpower, in fact that's precisely what a superpower is.

Don't like it? Go back to an agrarian society.

Intel's stock is crashing and their reputation is through the floor. They're on the verge of being taken over, and their latest CPU is a dud.

This is when AMD needs to push hard on both R&D and QA, to capture the marketshare that no longer trusts Intel, but is still wary of AMD because of their failings.

The two major alternatives - ARM64 and RISCV - are also chewing the scenery. Anything AMD doesn't grab will go to these.

As for the AI market, AI needs SIMD. Basically, you want processors that can digest a vector or a matrix in one go. nVidia basically started with a GPU and then bolted on bits that would boost performance for yesterday's algorithms. Meh.

A more sensible approach would be to look at SIMD and vector processing in general, and look for low-ganging fruit that nVidia hasn't considered. Bubbles burst and AMD won't conquer anything with an architecture that's overly specialised - - a mistake Intel made with their iWarp and Itanium lines, and a mistake that basically destroyed Transmeta.

All that data? A miniscule fraction of the genome, of which only a small part relates to anything useful?

If you're worried, there are ways to become a stakeholder in your DNA, obligating companies that use it to pay you a percentage.

23&Me collect a tiny number of markers, most of which are heavily studied. Personally, I wouldn't worry too much.

However, if you're nervous, there are sites that will make you a stakeholder in your DNA data, thus obligating companies using it to share in profits gained from using it.

There's no evidence that I'm aware of that 23&Me provides data to insurance companies, in violation of federal law. Do you have such evidence?

Far as I know, there's no evidence 23&Me provides data to insurance companies. Do you have such evidence?

....then purrs and requests belly rubs, then it had better damn well be a cat. Broadcom does not qualify.

Sorry, but my level of trust here is not merely low, it's a wholly imaginary number.

Not that it's been high since I tried a dual-core Broadcom MIPS64 board and discovered that it would do a hard reset if you tried to do any serious work with it.

I want things to work. If I pay a little, that's fine. But it needs to work. And it needs to work well. I use Linux and FreeBSD because they work well. OpenIndiana and Haiku are also free, but I don't use them, because they don't work well.

With the prevelance of SSDs and the death of physical media in most data centres, you're virtually always going to be better off with total emersion cooling.

Neither. You NEVER have the developer test their own code and you NEVER rely on certification authorities.

The cryptography lounges are maintained by the cryptography community and include each and every paper ever published on the strengths and weaknesses of every cryptographic algorithm.

From these papers, many such lounges will place any given algorithm into one of four categories - currently secure, low risk, high risk, and broken.

You never, ever use any algorithm that isn't marked OK.

But number 7 will shock you.

Quantum computing is coming under pressure as genetic algorithms and neural nets (but not LLMs) are starting to prove faster than their quantum counterparts, even at quantum mechanical problems.

We need to think not only about quantum computing but other areas of computing that may potentially discover vulnerabilities. After all, cryptography cannot genuinely be indistinguishable from a random oracle except with one time pads.

That isn't to say they will or can, but rather that the playing field is larger than anticipated, so we need cryptography that is robust against multiple types of attack and hybridised attacks.

There are some "Hello World" programs out there.

You shouldn't assume anything. NIST produces decent stuff and, yes, compromised stuff, as does the IETF, as does virtually every organisation.

You can't judge compromise by the label. Rather, you monitor the lounges, the cryptographic mailing lists, arXiv, and the testing sites.

You trust nothing that is new and shiny, you trust only that which is tested and found robust.