This idea is old as the hills (1940s), and is what led to neural networks. But, what actually happened? It turned out that the algorithms known collectively as neural networks have very little to do with the brain, and have much more to do with statistics, probability distributions, sophisticated mathematics, and clever algorithms leveraging that mathematics.

So, this is just a play to try to leapfrog LLMs.

Are you talking about the scientific method? Asking other humans to do the same test to see if they get the same answer?

AI can do that too. They just need an api for doing the test (whatever it is), and they can easily have multiple separate instances of a model use the api and do the test, and then compare notes.

Most humans are not well-positioned to be running large hadron colliders, however. Mostly all they can do is read up on the information available online. And....AI can do that too!

Humans don't have some magical ability to just intuit truth from falsehood. And neither does AI.

So, my point stands.

There IS a difference in how humans are much quicker to state that they don't know something, whereas AI models tend to just make stuff up. This doesn't seem to be the exact same thing as being able to determine truth from falsehood, however. Incidentally, humans also sometimes make stuff up when they don't know the answer (key evidence: widespread belief in supernatural beings that want 10% of your income).

Some of those things are probably profitable if they can gut most of the staff and centralize the maintenance work while cashing the cheques from old people who signed up and keep paying. AOL has been coasting on that for decades already, so it's little wonder they and similar sites have wound up on the island of misfit internet companies. They're not making very much money for the valuation they're after though. The notion that the dozens of millions in profit will be able to turn into hundreds of millions or billions is a pipe dream. This IPO is the private investors trying to get out before the customer base dies off and the cheques stop clearing.

Maybe this is better than the Google approach of killing it off entirely even if it could eke out a meager existence, but it's a business that has no future. Dead or dying internet platforms don't come back. This company is just a hospice for websites and tech companies some people remember from years ago and haven't used in almost as long. There might be money to be made there, but it's not a billion dollar company.

These are different issues. Dunning Kruger was about estimation of competence not whether or not one actually knows whether they know the answer to specific questions. LLMs have no meta-cognition.

Even in terms of general competence reality does not match the Dunning Kruger Internet meme.

"Our results supported the third hypotheses by confirming that (a) peoples' self-assessed competence generally accords with their demonstrated proficiency and (b) peoples' frequencies of self-assessed underestimation of their competence are similar to their frequencies of overestimation."

https://digitalcommons.usf.edu...

An animals core algorithm, is buried deep within its animal instinct. No matter how harmless Fluffy McBeans appears sitting there in Grandmas lap, make no mistake. Poor Grandma misses a feeding or three due to her untimely death, and Grandmas face might become Fluffys next meal due to that core instinct kicking in.

Not sure what the worlds largest store manager is doing rooting around our base primal instincts, but that sounds a lot like a brainwashing tactic for mass control of a civilian army wholly addicted to your drug.

Also known to the friends-with-enemies-closer crowd as "Tik Tok".

He's probably using the term more loosely, but it is impossible. The human brain works in ways that no Turing Machine and therefore no algorithm can replicate. I think that it's possible to build something that can function that way using existing computer hardware that's really just simulating a human brain if such can't be directly implemented in hardware, but we're a long way off from being able to do that even if the current crop of LLMs have fooled a lot of people into believing they can already replace a human.

Assuming that he or anyone else does succeed, they'll have no better understanding of how what they made actually works than we do of how our own brain actually works. Whatever it is that's going on up there, it certainly isn't algorithmic in nature.

I already know everything worth knowing about the Dunning-Kruger effect.

That's a fantastic idea. Thanks!

Human employees cost a lot. AI is the holy grail of eliminating labor costs. As with any investment, it will cost a lot more in the short run than having human employees would, but once the next threshold is surpassed, the savings and profits will more than make up for it.

In theory, anyway.

The slightly less cynical take is that AI can empower humans to achieve more than they otherwise could. Rather than eliminating jobs, we get a lot more productivity from the same number of employees, thus producing everything that everybody wants even faster and in greater abundance.

Either way, it's clearly valuable to many people for many reasons, so it will naturally garner a lot of investment money.

Incidentally, humans aren't very good at telling facts from lies either.

I don't think this makes the next level of AI any easier to achieve, but I don't think "the ability to tell facts from lies" can be postulated as a defining quality of a human, nor something that differentiates humans from AI.

It takes less effort to provide the raw uncensored footage to law enforcement.

In your scenario the acceptable answer would have been "We no longer have the footage". They didn't do that. They actually had the footage requested. And they censored it.

IANAL but I'm guessing if a civilian did this, they would be charged with multiple crimes, to include evidence tampering and obstruction of justice.

This creates a covert listening device, which a crime to operate it in many places and a crime to own in some. That includes the US under 18 U.S.C. 2511. But rest easy, the maximum penalty is only 5 years in federal prison and a $250'000 fine. Nothing to worry about.

You are missing a nuance here in the Federal law.

This is with reference to INTERCEPTING conversation / communication. Federally, it is considered a one-party consent rule situation....as long as one person in the conversation knows, it's ok....so if YOU are recording it is presumed YOU know you are and consent is satisfied.

This law is primarily considered for wiretaps, bugging places, etc.

"make" it the grid's problem?! Exactly how the hell is it not "the grid's problem" right from the very beginning? Since when ensuring quality and reliability of a service (such as power delivery) not the responsibility of provider of said service but its receiver? Do the proposed solutions such as flywheels and whatnot strike you as "yus, that's definitely the kind of stuff that's appropriate for software companies to be building, and definitely not the power company"?

The problem with this approach is, it only works as long as someone does the checking. In practice everyone turns on 'safe update channel' and nobody actually tests the bleeding edge, ten days elapse, and the malicious code flows into the 'safe channel'.

It is like sending for help in a first-aide situation, you need to point at someone specifically, make eye contact and say "YOU! go get help" if you just shout 'someone get help' and go back to recuse breathing or whatever you're occupied with everyone will stand around on-looking assuming someone else is doing something.

I love Ruby, it is an elegant language and it has made great performance gains in resent years, but but bundler and the drama around rubygems is a really problem, for anyone trying to make commercial use of it. I hate to say it but if Ruby is going to survive it probably needs to find another major patron besides Shopify, that is willing take some ownership and investment in the outside the standard library supply chain. Bandages like this are not going to cut it, and pure community lead effort isn't likely to be able to keep up with the evolving threat landscape. Unless your project is Linux, Samba, Bind, Apache, level deployment scale it just does not work with the degree of attack surface something like package/module repository offers.

Take a guess as to where a Yoga studio reporting a B&E with no injuries stands in the daily priority list at a California police dispatch, in a state that requires a business to carry an insurance policy.

Also known to law enforcement as the Get Out Of Work Free card.

"Waymo" is the only reason we're reading about this here. No one gives a shit about a thief who got away with it in a state that turns a blind eye to it otherwise.