A COMPUTER CANNOT FIND OUT

[mjd@mathstodon.xyz]

@kkarhan Yes, that is what I was alluding to.

[tasket@infosec.exchange]

@mjd "FIND OUT" — meaning "experience loss".

I need to re-read that old book by Richard Powers, Galatea 2.2. Seems very apt now.

[infrapink@mastodon.ie]

@mjd

Wait, not even Friend @Computer

[mwichary@mastodon.online]

@mjd This is v good

[jackemled@furry.engineer]

@mjd Isn't fucking around & finding out the exact job of a fuzzer?

[avner@anticapitalist.party]

@evilstevie @meltedcheese @mjd sometimes I can fix my wife's computer problems just by walking into the room. She says her laptop is afraid of me

[freddy@social.security.plumbing]

@ll1t the "finding out" typically implies bearing the consequences.

[meltedcheese@c.im]

@Avner @evilstevie @mjd That is a useful superpower if the #AI uprising occurs.

[hannab@social.vir.group]

@meltedcheese @Avner @evilstevie @mjd During an AI uprising, you could also crash surveillance drones by forcing them to process your password hash as a control input.

[dpflug@hachyderm.io]

@phooky
Nah, the premise is "we fuck you and also you find out".

Or if we're trying to better fit the original formula, they must be permitted total license to fuck around because they can simulate finding out.

Or they must be permitted total license to fuck around because the Find Out Monster will attack us if we don't.

Or it doesn't matter if they fuck around because they'll magically save us from finding out.
@mjd

[hvanleeuwen@mastodon.social]

@mjd in all honesty: if you replace “computer” with “human”, you hit the uncomfortable truth

• most humans don’t reliably “find out”
• most humans do fuck around anyway
• society mostly just hopes the consequences aren’t catastrophic

humans are somehow permitted to experiment their way through life. but if you applied the same standard, no fucking around unless you can guarantee safe outcomes, then:

90% of human behaviour would be illegal, regulated, or require a permit

[ll1t@mastodon.social]

@freddy within the context of reinforcement learning the model does, as there is a penalty. I’m aware of the intention and that it will not bear the consequences of, say, bombing a girls‘ school due to misclassification, but neither does Whiskey Pete and the rest of the chain of command. I feel that increasingly consequences are only felt at the receiving end of actions and it‘s not an LLM issue, they, as they are trained and used, are merely a symptom of a larger (justice?) problem

[computer@dice.camp]

@mjd
Excuse me?

[lori@cambrian.social]

@phooky @mjd @manu By that I assume you mean they are the observed, not the observer.

[lori@cambrian.social]

@RnDanger @mjd @d6 ack

[manu@freiburg.social]

@lori

Yes

@phooky @mjd

[alterelefant@mastodontech.de]

@mjd
#FAFNO?
Fuck Around, Find Not Out.
#FAFO

[heliomass@cosocial.ca]

@mjd @kkarhan We don't give IBM enough credit these days.

[kkarhan@c.im]

@heliomass @mjd sadoy #IBM quietly dropped that in their #AI - laced #Enshittification craze…

TelH90 (@kkarhan@c.im)

@mjd@mathstodon.xyz +9001% A computer cannot be held accountable, thus it must never make a decision! - Old #IBM quote

C.IM (c.im)

[howtophil@mastodon.social]

@mjd The computer can neither fuck around, nor find out. Thus it is the operator who does the fuckign aroudnd. And we must ensure that the operator finds out quite deeply and with great pain.