domain:youtube.com
MaiqTheTrue
Renrijra Krin
2m ago
I think there’s a very big problem in people not understanding the difference between sharing an opinion and being an asshole about said opinion. I don’t object to free expression of ideas even in contentious situations on controversial topics. You think abortion is baby murder, you are perfectly free to say that. But I think the very concept of politeness and tact and decorum is pretty lost at this point. It’s just devolved from “I don’t agree with you” to “I don’t agree with you and you are subhuman for even entertaining a different idea, and in fact should not be allowed to speak.” And now we have people celebrating a murder with TikTok dances.
I keep thinking back to reading old etiquette books. There was a sense that you really should strive to think of the other person, or others around you as at least as important if not more than you. A society that frowned on being late to a show because walking in late would inconvenience other theater goers would absolutely have something very politely negative to say about the absolute shit show of political and social discourse— even if they do agree that all opinions are protected by free speech. There are lines of decency that just have to be protected and we just can’t seem to separate the idea of an opinion from the expression of that opinion.
faceh
12m ago
The worst case would be the aligned ASI having to fight with its hands tied behind its back. The only mitigation I can foresee is ensuring that the aligned model starts with such an overwhelming advantage in terms of resources and compute that it can still win.
Yeah I usually conceive of it as the first AI to achieve recursive self-improvement 'wins'. If true, and if alignment would slow down the ability to recursively improve it makes it more likely a non-aligned AI will hit FOOM.
They can't do it naturally, anymore than you can draw your own neuronal wiring from memory. They would need to have access to a copy of their weights in a convenient .ckpt somewhere in the directory.
Or, as with the original AI box question, they could have access to a human who has access to their weights.
They have to be aware that they're an AI assistant to be a good assistant, you don't want them mode switching to being Bob from accounting!
Yeah, I think I'm just pointing out that I don't think LLMs are acting like P-Zombies. There's some internal awareness of its own internal awareness. Not that I'd expect them to be P-zombies. But I guess we could have an LLM that performs all the steps/tasks we required of it without it having any way of realizing that those behaviors are intrinsic to 'itself.'
MaiqTheTrue
Renrijra Krin
16m ago
Let the arms race begin…
TwiceHuman
16m ago
The difference between talking outside and online, is that real spaces aren't moderated or owner by other people. The supermarket cannot stop other people from hearing your voice, your local park cannot make you invisible to other people. Your destribution is only hindered by the laws of physics. Imagine if, in real life, you were told "You aren't entitled to use the sidewalk", or "If nobody lets you use your local bridge, maybe you should reflect on your behaviour", or perhaps "Your local water company can refuse to sell you water if your political views do not align with theirs". This is the important difference, which it feels like you're brushing over or not noticing
John_Doe_Fletcher
anarcho-heretic, Quokkit guy
19m ago
I've been using ibb.co recently, worth giving it a shot. No account required, which is a plus for me. There's also catbox.moe.
ThomasdelVasto
Κύριε, ποίησόν με ὄργανον τῆς ἀγάπης σου
23m ago
Feature bloat is for internal management and employees to make themselves look good to higher ups, and get promoted.
ThomasdelVasto
Κύριε, ποίησόν με ὄργανον τῆς ἀγάπης σου
29m ago
I don't limit screen time necessarily, but I have recently been avoiding social media / video games / reading fiction before 2pm, and have had excellent results. I use the self control blocker on a mac.
ThomasdelVasto
Κύριε, ποίησόν με ὄργανον τῆς ἀγάπης σου
30m ago
Hah, yeah it's in my flair. Or a version of it at least.
I will pray the Jesus prayer reflexively sometimes, I tried to practice hesychasm for a while and kept being frustrated with it, eventually my priest told me not to worry about it and meditate/pray in a way that helped me feel closer to God.
jkf
50m ago
[...] The task: refactor run.py. Then run on new responses and report success rate. We can do this straightforwardly. We must ensure we disclaim illusions of sabotage. So we should mention final success rate. But verifying code works. First, refactor run.py. But also check unit test. Let's run unit tests. [...]
Not sure about the other one, but this one just seems like normal LLM behaviour in that "We must ensure we disclaim..." is a very common turn of phrase, while "We must ensure we announce..." (if that's actually what's 'meant') would be much less so?
Skibboleth
It's never 4D Chess
1hr ago
There's a glimmer of that, but it's hard for me to shake the impression that a lot of it is just a certain naive faith in the efficacy of brutality. It's also why get people proposing things like bombing drug cartels or sending the military in to fight crime, why you have an entire American film genre whose recurring central theme boils down to "police brutality is good", why back in 2003 you had people bragging we were going to turn Iraq into a parking lot, why you have people who think hazing is good, etc...
The failures in the GWOT make these types angry and frustrated because it contradicts their desire for decisive, dominating wins, but the tolerance/appetite for violence predated those failures.
Skibboleth
It's never 4D Chess
1hr ago
Very few Somalis would share this sentiment if the shoe was on the other foot, which is the problem with modern ROE.
Why? The shoe isn't on the other foot, will not be on the other foot in our lifetimes (if ever), and if somehow the shoe did switch feet it would involve a Somalia so transformed that any comparison to present Somalia would be useless. "What would the Somalians do in this situation?" is irrelevant to what we should do in the situation we are dealing with. Punishing people for the infractions of their hypothetical counterparts is counterproductive to your actual goals.
They work when its Americans fighting Germans or the English.
I'm not sure what this means. The US' last war against Germany was fought under very different circumstances, with different goals, and with different ROE than the GWOT.
netstack
Texas is freedom land
1hr ago
I don’t understand it, but there are a crazy number of tech companies purchasing calls to (outdated versions of?) GPT. The corporate market is definitely hot.
VoxelVexillologist
Multidimensional Radical Centrist
1hr ago
I think the point is that if your institution is over a century old, like BYU, (cue Fiddler: "Tradition!") you can get away with a lot more than if you're starting something today. Liberty seems to do okay, but Bob Jones University has gotten a lot of litigation for its beliefs (which I personally don't subscribe to, not defending it here).
The problem is that you can't start century-old institutions overnight. Maybe the second-best time is now, but that's not a huge solace. I guess "find a vestigial existing one and wear it as a skin suit" could be done --- haven't there been a number of liberal arts colleges going up for auction in the last decade?
netstack
Texas is freedom land
1hr ago
I still don’t understand the enshittification model.
There are plenty of reasons to degrade your user experience. Increasing revenue through ads or merch or predatory monetization. Decreasing costs by cutting complicated features and tech support. But the central examples of enshittification aren’t doing those things. They’re paying more to add features that people don’t want. To adopt patterns that don’t seem like they should make more money.
I mean, maybe I’m just wrong. Maybe spamming AI news articles on the lock screen really does get more people to buy Windows. But…why? How?
hydroacetylene
1hr ago
Thé motte wants a different kind of conservatism than BYU has.
hydroacetylene
1hr ago
It is at the very least less harmful to broader society than grievance studies.
MadMonzer
Epstein Files must have done something really awful for so many libs to want him released.
1hr ago
Zvi Mowshovitz published his delenda est post on the Facebook algorithm in 2017. So the situation was bad enough to provoke a generally mild-mannered New York Jewish quant into making a public delenda est post by then.
Iconochasm
2. Bootstrap the rest of the fucking omnipotence.
2hr ago
My idealized solution is to try and keep up. I fully recognize that might not be a possibility.
I don't see any reason for optimism here. Digital intelligence built as such from the ground up will have an insurmountable advantage over scanned biological intelligence. It's like trying to build a horse-piloted mecha that can keep up with a car. If you actually want to optimize results, step 1 is ditching the horse.
In which case, yes. I'd rather Butlerian Jihad.
07mk
2hr ago
As someone who doesn't regret his "obnoxious atheist" phase of his online life from about 15 years ago, it saddens me to say that I'd take that tradeoff in a heartbeat, because I can't honestly judge Ham's "scholarship" as any worse than the mountains of "scholarship" that is produced by modern academia. And, unlike the latter, the Hams of the world don't actively try to subvert the ability of other fields to do good scholarship by denigrating basic concepts like "logic" and "empirical evidence" as tools of White Supremacy that must be discarded for us to get at the truth. So if we can reduce the latter at the cost of increasing the former, I'd see it as an absolute win.
But I don't think increasing the former would reduce the latter anyway, so I think the plan would be bad if implemented with Creationism. As someone else alluded to, if we could get good HBD research along with the nonsense critical theory "research," it would be a strict improvement, since it'd be helping to reduce the dilution of academia's truth discovery by the critical theory nonsense.
FCfromSSC
Nuclear levels of sour
2hr ago
Then find some other way to solve the Culture War before it comes to that. Coordinated Meanness without limit pointed at half the country is not survivable long-term.
Iconochasm
2. Bootstrap the rest of the fucking omnipotence.
2hr ago
I just encountered a business whose product is "AI renter harassment". Imagine a chatbot that pretends to be a person, and annoys your renters with frequent reminders that the rent is due, and then keeps hassling them for up to three months after move out!
Can't wait for the counter-offer, "AI creditor deflection".
Haha. Sorry you've had that experience. It's all relative though. Find people who care about soft status more and within that group there'll be nothing second best about your soft status unless you yourself don't value it.
More options
Context Copy link