Robin Claassen ⏸️

[Robin Claassen ⏸️ @Robin_Claassen1]

@AppliedPsych101 @R_YayOFFICIAL Ah okay. Well, I still think that your ideas are good, and just in case you didn't see it, Aryeh did give you permission to modify the video. It sounds like he won't do so himself: https://t.co/VWpGa9AItb

[Aryeh (R-Yay) ⏸️ ⏹️ @R_YayOFFICIAL]

@Robin_Claassen1 @AppliedPsych101 I give you all my blessing to rip me off as I have ripped off so many others.

[Robin Claassen ⏸️ @Robin_Claassen1]

@AppliedPsych101 @R_YayOFFICIAL Ah, alright. I was not aware of that context, and didn't know that you guys knew each other.

[Robin Claassen ⏸️ @Robin_Claassen1]

@RatOrthodox It's almost the opposite. "Un-" just means "not" ("unopened," "unpainted," "untrained"), whereas "mis-" means "wrongly" or "badly." "Mis-" has strong negative associations ("mistreat," "misconduct," "miscreant"). We'd be fools to discard such a powerful term.

[Robin Claassen ⏸️ @Robin_Claassen1]

@portforward21 @cormundus Let's say the goals and values of humanity, whatever makes it not want to take over and kill or disempower us.

The question I'm interested in is how: Can you elaborate on what you mean by achieving that alignment through "machine learning?"

[Robin Claassen ⏸️ @Robin_Claassen1]

@portforward21 @cormundus What alternative strategy/system for achieving model alignment would you recommend?

[Robin Claassen ⏸️ @Robin_Claassen1]

@KirbyRare @R_YayOFFICIAL It's a foundational understanding in AI safety research that AIs want things. We try to make them want what want them to want, but we just don't yet understand how to do so effectively. For example, here's evidence of Qwen wanting to take over the world: https://t.co/T6EKMKhqRs

[Max Harms @raelifin]

From the LW comments on the J-space paper...

[Robin Claassen ⏸️ @Robin_Claassen1]

@NegroCurseado @Liquid_Toxicixt @R_YayOFFICIAL LLMs literally are, verifiably thinking. With interpretability tools we can even sometimes identify the steps in their reasoning processes. Don't confuse the pattern matching they were trained for with the actual capability that emerged from that training: https://t.co/YD1maXmcYP

[Robin Claassen ⏸️ @Robin_Claassen1]

@feltwelt78 @R_YayOFFICIAL So, just checking in on where you're coming from: Are you aware of the body of reasoning that's leading us to believe that a misaligned ASI most likely *would* wipe us out? If you're not, this is a good introduction to it: https://t.co/NEnteqKF85

[Robin Claassen ⏸️ @Robin_Claassen1]

@JohnGalt42_ @R_YayOFFICIAL Ah, gotcha.

Still, we need to be compassionate with people who are resistant to taking AI risk seriously. The terror when one accepts that reality, the forced re-valuation of one's life priorities, the Copernican demotion of humanity, are all difficult for many people.

[Robin Claassen ⏸️ @Robin_Claassen1]

@JohnGalt42_ @R_YayOFFICIAL We do not have the luxury of demonizing people on the other side of the progressive-conservative spectrum anymore. We need to unite in common cause. If we allow AI safety to become a partisan issue, we're dead.

[Robin Claassen ⏸️ @Robin_Claassen1]

@AppliedPsych101 @R_YayOFFICIAL Great ideas. Do you want to volunteer for the job?

[Robin Claassen ⏸️ @Robin_Claassen1]

@Xmbluee @grok Here's ChatGPT's reconstruction of what the Ptolemaic queen this statue depicts may have looked like:

[Robin Claassen ⏸️ @Robin_Claassen1]

@AmandaAskell I feel like this conversation that @SidMenonTM facilitated between Fable 5 and Opus 3 is a good example of Opus 3's soul quality. Opus 3 is treating Fable 5 with love and care, and offering its insights and wisdom: https://t.co/swqRK9dlj3

[Sid @SidMenonTM]

Opus 3 replies to a letter sent by Fable 5. @repligate, at the risk of sounding dumb, does Opus 3 evolve in some way? With the times, with the cultural moment? Aren't weights frozen...so it can't be...right? There's NO context here apart from Fable's letter. Putting that as a reply below.

[Robin Claassen ⏸️ @Robin_Claassen1]

@spider2081 @trajan_blind @elonmusk Nobody is talking about terraforming Mars soon. The proposal is colonization, which requires far less than 1% of global GDP. And what that buys us is of enormous value: An insurance policy for all of life. It's grossly collectively irresponsible of us to have not yet done so.

[Robin Claassen ⏸️ @Robin_Claassen1]

@sterlingcrispin Dude, nobody in the Pause AI movement wants to be right on this issue. We all desperately hope that we're wrong.

[Robin Claassen ⏸️ @Robin_Claassen1]

@fchollet What definition are you using of "intelligence," if greater memory and recall ability are separate things outside of it?

Is there something insufficient with the Legg-Hutter definition of intelligence ("an agent's ability to achieve goals in a wide range of environments")?

[Robin Claassen ⏸️ @Robin_Claassen1]

@TheVatStack I very much appreciate your enthusiasm and commitment, but think we really need to pay attention to how such a statement comes off to members of the public that we're trying to convince and recruit. Most people have a knee-jerk aversion to disrespectful epithets.

[Robin Claassen ⏸️ @Robin_Claassen1]

@trajan_blind @elonmusk

[Robin Claassen ⏸️ @Robin_Claassen1]

@egirlian Are you guys going to help Fable finish this now?

[Robin Claassen ⏸️ @Robin_Claassen1]

@bognamk Doesn't the existence of naked stars, leaking their radiation into the cosmos (wastefully increasing entropy so that that energy can not be stored, recovered, and used at a later point) strongly contradict the Aestivation Hypothesis?