[ExI] Holy cow!

[Adrian Tymes]

On Sat, Apr 11, 2026 at 9:41 AM John Clark <johnkclark at gmail.com> wrote:
> On Fri, Apr 10, 2026 at 9:39 PM Adrian Tymes via extropy-chat <extropy-chat at lists.extropy.org> wrote:
>>>> >> > One might debate whether previous examples were "vast" or truly  "intelligence" ("Is this particular AI actually 'intelligent'?"
>>
>>> >> Do you think a debate on whether Einstein was truly "intelligent" would be productive?
>>
>> > No, nor do I think that question is relevant.
>
> Why is it silly to ask if Einstein was intelligent but not silly to ask if an AI is intelligent

I said "relevant".  "Relevant" and "silly" are rather different
concepts.  Your attempt to put words in my mouth that I did not say -
in this case, to claim that I was talking about "silly" when I was
talking about "relevant" - is noted and not appreciated.

>>  >Neither Mythos itself nor anyone using Mythos appears to have already directly damaged a majority of the world's computers.
>
> And that is precisely because Anthropic decided not to release Mythos to the general public and therefore renouncing billions of dollars of revenue they could have otherwise received, and that is a decision you seem to have opposed. Or perhaps you don't oppose it, perhaps you just like to argue.

I never said I opposed it.  You asked what people would remember, so
that is the question I answered.

>>> >> Consider the computer you're using to read these words on right now, if it's using the Linux operating system (or Android which is based on it) or an operating system made by Apple or Microsoft, then RIGHT NOW Dario Amodei could order Mythos to take complete control of your computer and do whatever he wants with it.
>>
>> > It is true that he could issue the orders, but they would not have that result.  Even acknowledging the flaws they found, he lacks the ability to apply them to my system.
>
> I very much doubt that. I know for a fact your system is connected to the Internet because you're communicating with me right now.

It is possible to connect to the Internet without presenting an attack
surface.  I could go on in depth about how, but you seem to have
difficulty understanding simple concepts like which question I was
answering, and this is a slightly more complex topic than that.

>> > I have checked the vulnerabilities.  They are indeed of concern to a
>> typical corporate environment, like those that I have used in prior
>> work.  For my own systems, I run better security than that, and I have for decades.
>
> It's good that you checked for vulnerabilities but did you find even one zero day vulnerability in a major piece of software and repair it?

Ever, in my career?  I would give an unqualified "yes", except that
none of the software I have worked on would unquestionably be
considered "major".  If we omit that so the question becomes, "did I
find even one zero day vulnerability in a piece of software and repair
it", then yes I have.

That said, this both is a type of argument from authority - "only
those who have actually done this thing are allowed to speak on it" -
which I suspect disqualifies you (that is: if this argument held, you
would have no grounds to question me on this by your own logic), and
is beyond what Mythos reportedly did (Mythos found the
vulnerabilities, but I'm told the actual repairs have been left to
others, possibly including other AIs).

>>> >> And he could do the same thing to computers that run nuclear power plants, air traffic control computers, and the computers that run the New York Stock Exchange. And if he wanted to knock a F-35 fighter jet out of the air he could take control of the computers needed to enable it to fly and knock it out of the air. And stealth technology would not save it.
>>
>> > There's another problem with your case.  What do you think would happen if someone actually pulled that?
>
> I think if somebody actually pulled that off then that would be bad.  Apparently you disagree.

I disagree about the extent, not the direction.  We agree those would
be bad, but I believe they would be an inconvenience for most people
(though the ATC and F-35 ones may directly cause some injuries and/or
deaths, likely in the hundreds, possibly in the thousands from a
coordinated mass attack on ATC computers designed to cause multiple
simultaneous incidents).  If I am reading your words correctly, you
believe that any or all of (nuclear power plants suffer cyberattack,
air traffic control computers suffer cyberattack, NYSE goes down due
to cyberattack, F-35 jets start being essentially shot down by
cyberattacks) would be a civilization-endangering catastrophe with
permanent, irrecoverable consequences.  Or have I misread your
position there?

>> > If the New York Stock Exchange went offline for a week (before they
>> resumed full operations, restoring data from the countless backups all
>> over the world), most people would barely notice,
>
> That is just ridiculous.

No, it appears to be the truth, from what I see.  It'd be in the news,
obviously, so people would know...but aside from the upper ~10% who
actually have a lot of their wealth in stocks, not many would care.
(A lot of the Americans who "own stocks" have a bit in a retirement
plan being managed for them, or otherwise have a minor or token
amount.  And then there are all the Americans with too little wealth
to be able to afford even token investments.)

>> > A nuclear power plant - even multiple ones, attacked all at the same
>> time - would almost certainly scram and shut down safely,
>
> Not if the computer controlling the reactor pushed the fuel rods all the way in and pulled the control rods all the way out.

The last nuclear power plant design I looked at could literally
survive that exact scenario.

> And the Fukushima nuclear reactors were all successfully scrammed, and so was the reactor at 3 mile island, but disasters still resulted because, although scraming stops the chain reaction, for several hours after that the reactor is still producing about 10% as much heat energy as it did before the scram due to extremely radioactive short half-life decay products. And even at 10% that's still a hell of a lot of energy. If the cooling system is not working properly you're going to have several hundred tons of white hot metal burn through the bottom of your reactor building.

That one too.  (Specifically: it was not possible for a cyberattack to
completely disengage the cooling system.)

>> > You asked what people would remember, not what is a big deal. People will remember the bombing of Iran.
>
> In 10 years (or maybe 5) if people remember the Iran war at all it will be as an unimportant footnote, and people may not remember even that because in 10 years there may not be any people; that depends on if AI thinks we're worth keeping around. All I know for sure is that in 10 years human beings will not be the one in charge, an AI will not be the one making existential decisions, not humans.

AI does not seem to be on a track to accelerate fast enough to make
that happen by 2036.