Indeed. Most LLMs seem to act like 12-year-old children. They can get things right with a script, but they're easily fooled by intelligent malicious adults. You wouldn't put a 12-year-old child on the front line of your customer service. Why would you put an LLM there?

Ever since those robotaxis started roaming, I've been have Waymo sex in moving vehicles!

Most of these deep neural network systems seem to lack the equivalent of a mental model. There was an article a while back with a line that has really stuck with me (emphasis added):

When engineers do peer into a deep neural network, what they see is an ocean of math: a massive, multilayer set of calculus problems that—by constantly deriving the relationship between billions of data points—generate guesses about the world.

That ocean of math doesn't correspond to any of the kinds of abstractions about the world that humans use in their conscious thinking. And since language is created by humans for the purpose of conveying information about human conscious thought, it's all about mapping words onto those abstractions... abstractions that simply don't exist in a modern AI system. So it's little wonder that they have trouble mastering it.

I hear you, but it all comes down to how you much it costs to have control over your own destiny. Building your own data center and scaling it at just the right rate is difficult and expensive. Companies like Amazon and Microsoft that invest billions in that and sell it as a service have to be careful to protect their reputations in order to protect that investment. This arrangement makes it cheap to do lots of stuff but doesn't translate well to controversial activities, since the world is full of snowflakes (I count myself among them) and lots of them have a problem with calls for violence and will vote with their dollars (as we expect in a capitalist society). If you want to do something that upsets the apple cart, you can expect the First Amendment to keep the government out of your hair, but you'll still need a lot of people, a lot of money, or both to come together to take the risk.

For all you URL hackers out there, fastcompany's image server hands out an arbitrarily high resolution version of the first snowflake (the highest I tried without a server error was 9100x5119: Ice Queen), but only 1280x960 versions of the other two: Yellowknife Flurry and No Two Alike.

I'm sure MS characterized the use of Google in their web browser as an edge case not needing to be tested.

Given the discussion about how visual cortex defects can predict schizophrenia, it sounds to me like it's like being born without an appendix (not saying whether that's possible, but just suppose): you can't get appendicitis. Born without sight, the visual cortex doesn't develop, so it can't get "sick" and cause schizophrenia.

Oxygen fasting!

Also, the first Penny Arcade: https://www.penny-arcade.com/comic/1998/11/18

Not that actual logic is likely to make a difference, but... From the summary:

Malady says Verizon will "not call our 4G network a 5G network if customers don't experience a performance or capability upgrade that only 5G can deliver." But that isn't the same thing as saying "we won't label our network 5G unless it's 5G." In fact, if you turn that sentence into a positive statement, it says "we will only call our 4G network a 5G network if it delivers a 5G-like experience."

No, that positive statement is not the same thing. If you say you won't use the 5G label unless you're providing a capability that only 5G can deliver, then an upgraded 4G network (that's still 4G) can't qualify by providing a 5G-like experience: if 4G can provide it, then it's not something that only 5G can deliver.

I wish I could believe that the president has been doing this as some sort of Corbomite Maneuver.

The short answer is that devices are designed for use in multiple markets with varying plug arrangements. To reduce cost, the form factor of the conversion hardware is unchanged between markets; only the prongs are changed. That which is inconvenient in Europe may be perfectly fine in the US. It's not evil. It's a natural consequence of global commerce without global power connection standards.

And what happens when a paid flirter interacts with other paid flirters? Can an entire relationship be developed by paid proxies?

The Gentleperson's Guide To Forum Spies has been around for a while but it's still worth a read. Not that I'm saying any of this is actually happening. Just something to keep in mind.

Reminds me of an exchange from the movie Deal Of The Century involving a totally automated fighter aircraft: Haven't ya ever heard of RAIN?