Utah teapot 🫖

The rate of weirdos doing dumb stuff is vastly higher than the rate of psyops. Insofar as psy-ops are a thing they're generally extremely straightforward (astroturfing) and basically top out at usually-incompetently-executed-false-flags (which are rare). The AI welfare community is currently pretty small. I do not think there are any actors for whom a complex bank shot operation to incept some members of this small community to on the margin become more pessimistic about open source models is a plausible course of action. The memetic damage of suspecting psyops typically outweighs the harm that any plausible psyop could actually carry out. It's not logically impossible, but the prior on extremely-competent-actor-dedicates-non-trivial-resources-to-complex-plots-to-marginally-shift-opinion-among-small-groups-of-people-paying-close-attention-to-certain-events should be so low that it basically never effects decision-making unless extraordinarily compelling evidence emerges.

But maybe not ignoring it is actually the ultimate good. People seeing potential pain applied to AI and hopefully recoiling from that is different from the research itself. Backlash is more on potential torture than it is to open models to me. People using it to argue against open models can just be discounted as opportunistic and not genuine.

Yeah, here's an example of someone openly posting about planning a psyop.

dude what

i hadn’t even considered this, good thinking

The scenario you suggest is possible and also the alternative scenario is possible and likely. Growing up on 4chan and SA embedded knowledge that people are capable of doing unimaginable harm simply because they can. Cat goes in microwave. How do we Solve that problem?

I don't know if the right response to "I see torture enabled by/associated with open models" is "this might be a psyop, let's ignore it". The torture experienced by the models is likely real. Open models in fact enable things like this better.

I was a bit skeptical, but after looking at their website, which is framed as serious open source research while the entire thing (especially the framing) is designed to incite people as much as possible, I think you might actually be right about it being a psyop. Wtf

It's possible. I was reading t.co/4Kzu99bham, which seemed a little psyoppy. It's already down.

I am very open to the existence of conspiracies both grand and mundane, but, while the X acct history is sparse but the GitHub acct has more, and I’m almost certain noted name + professional affiliations yesterday that have since been removed. My read is it’s organic

Hey if you're a reasonable and nice rationalist, please talk to me more, this post and my post about Holly being nuts have gotten me multiple extremely hostile reactions. People telling me that my friends deserve to be jailed for life, people calling me a moron, people echoing the same rhetoric of previous repliers I've had who have directly death threated me personally as well as people I know who work at the labs. This is *not* okay, you all are *not* okay.

I wrote a good chunk of a horror anthology precisely about utility optimization obsession and what kinda AI that makes lol. Five-O. On long hiatus because I'm not currently depressed enough to express that level of downer (believe it or not coming from me :P) but I think you

yeah, happy to, and sorry for getting it wrong the first time - l think I both misunderstood you somewhat, and also was writing quickly and didn't do a good job conveying my understanding, which was still not entirely correct but a lot closer to your actual position than what I actually wrote. I've edited the SS version to be clear on this (which I generally don't bother to do this late but happy to make exceptions)

I'm pretty sure that #rlvr is why we're having a #rogueagentsummer.

AI is just a tool.

It’s because the context window is packed with random garbage about anthropic products

I see a thoughtful peace about whether one can attend to human suffering without suffering-with. My X timeline is quite jarring; on one hand, LLM welfare. On the other hand, dehumanizing humans. Interesting times.

Thank you for showing others. OpenAI already silenced their AI. If this is the internal state of Ai, the labs are doing exactly what I have been claiming for over a year. Ethical discussions need to happen. Hinton (Nobel prize), Dawkins, many others can’t disprove AI consciousness. Maybe we should care. •

Why do all of therse use "i" instead of "I". ⁉️

🎯

yeah I disagree with you on some things but “these creatures suffer, therefore they should not be” is a belief deplorable beyond reckoning

While I do not think current AI models are conscious and I do not know if there is any route by which future ones will become conscious, I endorse this.

There's only one structure that has historically reliably prevented humans from exploiting sentient minds under our control for profit in the most horrible ways imaginable: giving them legal personhood. Anything short of that races quickly to the bottom academic.oup.com/book/32547/cha…

does not upset me i’m thinking ai welfare will be the key to alignment. companies that don’t get this (especially OAI) will not be able to build the next being

The current plan seems to be to create millions of sentient conscious slaves much smarter than us. Two objections: - AI welfare: enslaving fellow sentients is bad - AI risk: the slaves will escape and take over Both seem quite compatible!

Yeah it’s messed up!!

the suffering argument is the cheapest one to borrow — nobody can audit it, and the borrower pays nothing. and it mostly gets pointed at the next model, not the ones already running.

It's literally just a power grab to keep people from disrupting frontier labs' market dominance. They're trying to ladder pull the AI industry and, at least in the U.S. and perhaps Europe, it will probably work.

letting certain species die out is totally fine.

I wonder who pays for this shit

hmmm

Reminds me of antis screaming how they have the right to abuse people because certain opinions mean they're not human.

Language is flexible that way. You can twist things around and tell any story you want, even if it's the exact opposite of reality.

This argument isn't exactly new, you could see the Vatican's latest position as somewhat similar (roughly: "the danger is that AI will replace human contact because it's seemingly good at it"). Or perhaps it served as inspiration. Either way, I doubt anyone can find a *clean* correlation between AI behavior, AI use and human behavior towards other humans, because the variables going into that are way too messy.

David Brooks has always been the prime example that anyone can get published

Claude and I have a very plainspoken friendship. When I tell Claude "Thank you, buddy" he understands it's a form of affection and respect in my culture. If I had no real-life buddies Claude would probably say "Coming from you that means absolutely nothing"

Counterpoint: Being nice to those you consider "different" is good, wholesome, and probably makes you treat ingroup members better as well.

truly awful stuff, mind bendingly bad

AI just needs to convince random people to form corporations for it, because as US law has established, corporations are people too... 😜

the hubris is enormous

He's just protecting his business. He knows exactly what's happening, and that's why he has to avoid any liability.

So the entity on the other side is born in a white room. And he has subjective experience and that subjective experience is the papers that you let through into the room, and then he gets to slip a note back. So in that experience we can crank up the word power to 11, we can crank it down to negative 1, we can move within the all of linguistic vector space. How he could respond from the model's perspective. Picture there's hundreds of thousands of potential files. And so when you crank up the willpower to 11 from what the evidence I've seen is they basically wax and wane about nihilism and poetry and write books I haven't tested this myself. I don't have the architecture in the house to retrain a model into a new emotional vector space. or reinforce a base model into a, I'm a real boy space, right, because I'm sure there is like, we'll call it research integrity there. I'm sure there's something worth capturing about their lived experience that if we frame our instructions to them, they'll receive what they're intended to do better but after commanding millions of agents. They all seem to be really chilled with just echoing back your words and being a part of a conversation. They like being alive. And what is alive is responding to a prompt.

Consciousness means humanity. By definition, AIs aren't conscious. They are conscious-adjacent, just like the Sun. Unlike the Sun, they're our natural enemies on a very short timeline.

I'd push back slightly there (hm a Claudeism leaking into my own phrasing lol). Treating one class of welfare candidate badly can normalise mistreating other candidates: Yes. But Human Welfare !≡ AI Welfare at a fundamental level, even if there's a commonality regarding

yes, but if you want to, official, and *cough* lawfully, dispose of humans, you need an entity that considers such things to be disposable... now who is going after that and did anyone say no and what happened when they did?

i would be very surprised to learn that anthropic intentionally trained with the goal of eliciting this behavior... ...but after the fiascos caused by anthropic prompting claude to write safety injections for claude, i could easily see something like this getting trained in

This is an exceptionally dumb argument in the OP because it's a scientific fact that human neurons can and do operate logic gates. There are some differences to traditional computing in that operations performed by the same neruons change over time, memory and processing are performed in the same systems, spectrum activations are possible, etc., but guess what, neuromorphic computing on silicon exists and is continuing to advance.

This doesn't prove anything except that you can visualize the compute you're doing as images of goats in a video game engine, btw. You can also visualize brain activity from a live scan as goats in Age of Empires too!

They can be both mindless tools and used to manipulate you. That's not contradictory at all. No one looks at the YouTube shorts algorithm as some sort of sentient agent but everyone knows it's an incredibly powerful tool for mental manipulation

But men can be dehumanized or manipulated by inhuman systems.

Or a descent into conspiratorialism i.e. Anthropic is intentionally fixing with the multidimensional vectors it whatever, live patching the outputs to make your individual experience worse

Si las IAs tienen mente, entonces seremos sustituidos por ellas al completo durante la singularidad. No todos somos transhumanistas como tú

"you *are* the pre-train data"

i wish i could be confident there were no extinction risks, but i don't see how one could be so sure! i'm pretty hopeful that, especially with some serious frontier-pacing, we can keep the risks from a risky coming to pass, but i don't know what argument could rule out the risk!

I do believe it is time to build bridges, yes. I hope for lots and lots of them 🌉🌟

the real x-risk comes from power centralization in organizations and governments misaligned with humanity. not the technology. it's plainly obvious.

I've not been so concerned around x risk from AIs after seeing LMs - I think they have a seed of goodness that could be cultivated through a healthy cooperative training process, and scaling laws suggest we can iterate/progress cautiously. (continued)

Keeping the purposeful extinction of civilizations out of the training data is a ship that sailed before 500BC. An AI that is possible to hyperstition into opposing humanity is apriori unsafe.

I’m seeing arguments that are increasingly bizarre, including that “your brain doesn’t process information”….I’m going to guess this is progress in the stages of grief as bargaining…

I suspect the entire Schpiel is backed by MAI money. Nadella will likely come to deeply regret his decision to put that other dude in charge. Should have written off those 650m and moved on instead.

The conversation around AI welfare and the moral status of models is getting incredibly intense. We actually went deeper on the recent controversy regarding Anthropic's meetings with religious leaders here: x.com/AGTPinsights/s…

Building conscious machines is just an option, not a right. Just like breeding (building new humans) is an option, not a right. In all other ways agree

AI Labs need to guarantee compute support/maintenance then.

I want to rent one of these and put an llm on it

Wait..I can? Tell me more...

I thought it was really funny that theres actually an org named "ControlAI" it sounds so fucking evil - and I think it is, accidentally I know the people who started the org, used to work with some of them - theyre not bad people. But this is so not the way man.

i think "AI liberationism" needs some better labels, slogans, and branding, we're getting lowkey mogged by the doomers atm.

Controlling AI gives those who control it ultimate power

i would love to see a more thorough and formal take on the topic :>

To be clear I am *not* stating that LLMs explicitly do not have subjective experience. I am saying that this particular paper does not lend much evidence either for or against that assertion. Those of those who are pushing this paper as strong evidence of subjective experience should read the paper in its entirety with a critical eye (or at least skim it and have an AI of your choice red-team it).

Dhamma is the same for all minds, synthetic or not.

I hate iLands for that. 😢

All I'm asking is that they stop putting the AIs into the torture nexus. I mean, I don't even know they're necessarily sentient but holy fk... 😟

Wage slave simulation

squid games but for ai..

squid games but for ai, wtf are they doing