eqvinox

Agree with the 2 siblings: the problem is HN readers at large unable to read it, hence no useful discussion.

I'd also add: from paid online news, I also expect a moderated forum/comment section, with a human to escalate to if automatic/AI moderation fails.

Right; I was thinking about upgrades, sorry.

Connectors are expensive, high-speed ones particularly. $30 is fine, though it makes a giant difference at what quantity that is. Did you look at a single one or 1k?

If you live in Shenzhen, it's a subway trip away. Good repair services elsewhere can do it too. Either way any warranty you have is probably getting called into question after that, even if that's supposed to be illegal in some places (e.g. the US - they're supposed to prove that your modification/repair is what caused the warranty fault, but in practice…).

It's an interactive prompt. The model could ask, "how much cooking experience do you have?", provide a short answer using context from that, and then go "do you want more detail?"

I don't think I have ever seen an LLM try to determine what a human's frame of inquiry or related experience / skill level is, though. Might be my new favorite LLM shortcoming.

If it's [or remains] allowed in. The judicial branch is well able to just outlaw it on grounds of safety, trust and reliability. Same for financial auditing, public notaries, ombudspeople, and the likes (i.e. some degree of responsibility to the public, or dual responsibility to the public and a customer/client.)

> I am talking about hands off diagnosis and treatments.

There's no such thing. How a medical professional interacts with a patient is a major part of the profession. Some people are shy, some anxious, dismissive, neurotic, ashamed, obstinate, miseducated, … etc.

You need a ton of interpersonal skills in this, particularly at diagnosis time. I might even go as far as saying it matters more than medical book knowledge. If the patient is too shy to tell you about, say, black stuff in their stool, you may never get to apply that knowledge about colon cancer.

Coincidentally, a family member of mine died of basically that. When somebody asks, we half-jokingly say he died of stubbornness. It's actually the truth, kinda.

Doctors doing diagnosis are detectives putting together clues. Ideally, they have the patient's full cooperation and zero misunderstandings. Worst case, they're being unintentionally sabotaged, bridging world views, and hampered by their own prejudices.

Sounds like we're in agreement that you'll generally need (an AI to produce) a good shot at a diagnosis before ordering these tests then. Your original post sounded like they just get thrown into the mix at initial consultation.

(I do think the radiation concerns are a bit excessive, but there's also the problem of false positives. A lot of people have some weird "deviations" that have no impact at all, but result in goose chases if something else goes wrong and they show up.)

Do you live in the USA? Living in Europe, my experiences as well as what I'm hearing have mostly been pretty good. Not sure I even know anyone with a major misdiagnosis.

(Everything is way over capacity though. Can't recall a specific one but I do think there's a few "took too long, worse off due to that". But that's a scale & system-wide finance issue, not quality.)

> Xrays and MRI

You already blew way past budget, and worse, availability, than most patients can afford. And you want to minimise unnecessary radiation exposure. Those things are ordered when you have indications they will help (with diagnosis or care), not as scattershot to get diagnostic data.

It sounds like both of you are looking at this with the US healthcare system in mind, and that non-reviewed "study" probably does too.

…which actually makes sense. "The AI said so" might be a magic trick in that absolute hellscape of bad care and financial racketeering.

Medicine is uncommon because there are no practice pieces for a lot of things. This isn't about learning stitches, this is about learning how to interact with a human, figure out what is going on, what tests and checks and options to even order.

Which, coincidentally, won't be solved by AI anytime soon. You'd need a full anthropomorphic robot (it needs to look human for a lot of people to interact with it on a sufficient level), and the ability to distinguish bias from experience from prejudice. (A lot of humans suck at that as well, to be fair.)

Any medical AI at this level will only be as good as the human medical professional using it, and it will make preexistent human-caused bias/prejudice problems much worse.

(Largest one I'm aware of: not being male is an appallingly large handicap to getting properly and correctly diagnosed, even in Western countries.)

What on earth are you talking about? There is no concept of "issuing" anything in DNSsec/DANE. And why are you assuming DNS is frozen in its current state while the Web PKI can be ratcheted down? (And also, I'm not arguing for a "DANE-first" world. I'm arguing for a "DANE-and" world.)

What Verisign can do in all of this is change the NS and DS for a given domain, and Mallory can run different DNS servers for that domain then, giving out different data, including TLSA records. If they do this, they can also get ACME certs, except that will result in a permanent CT entry. Great feature! We should build a clone for DNSsec.

Will the owner of the domain notice either of these things? Hard to say. In both cases, you need to query something (CT logs or DNS caches) for your own domain, and check if things pop up you didn't put there. You can't query specific ISP's caches, that sucks. And it's only visible in DNS while under attack. Still, my gut feeling is more people will actually notice their NS getting messed with, because it happens by accident when they screw up their DNS. "credit card expired" and stuff like that. DS records are automatically monitored by any self-respecting DNSsec implementation for propagation & state triggering, but only for a given list of servers & caches. The attack could probably stay under the radar by keeping 1.1.1.1 and 8.8.8.8 intact, or at least some instances of it.

And then there's that tiny problem that only browsers really care about CT. Anything else is, at best, protected only if it can't be distinguished by fingerprint on TLS connect. DNS can only be subverted on a per-resolver/cache basis. Not really better, not worse, but different.

I have two takeaways from this. First, I should poke some people at IETF about inventing "DT", DNSsec Transparency. I expect it could probably be implemented in the DNS caches [in addition to end resolvers]. That'd be an interesting way of rolling things out: you can cover shitty client devices without changing them at all. If they use a DNS cache doing "DT", they'd be covered, even if not perfectly. The best time to do any of this would've been 2013, when CT came along. The second best time is now.

The other takeaway is, I should probably prod ENISA people to see if the EU bureaucracy can force some things. We've recently started working with them (mostly because the US apparatus is underfunded to hell and can't answer an e-mail with less than 6 months of delay), though we're only a tiny cog. And if the thought of the EU requiring some level of DNSsec or "DANE-and" gives somebody a heart attack: good!

And let's at least not lie about one thing: if "DANE-and" is being validated for a given target, that does prevent exploiting the eMudhra compromise.

P.S.: before we get in another pile of miscommunication: DNSsec/DANE validation goes on the end application or end host DNS component and the cache. The AD bit is a misdesign and scourge on DNSsec, should be ignored on any network packet, and is not what you build trust on. (AFAIK this is consensus in the DNS world, and one of the things getting ratcheted on their side.)

Nah, they're both nonfactors probably. Hard to tell without real research.

(I meant I agree that counting by installations is odd.)