@aqouta's banner p

aqouta


				

				

				
7 followers   follows 0 users  
joined 2022 September 04 18:48:55 UTC

https://x.com/Spencer_Gray

Friends:

@aqouta


				

User ID: 75

aqouta


				
				
				

				
7 followers   follows 0 users   joined 2022 September 04 18:48:55 UTC

					
				

				
					

Friends:

@aqouta


					

User ID: 75

Do you have any actual evidence of this? The cartel accusation as far as I can source it comes from them asking for an exemption to the Sherman act so that they can work together on standards, which is common among many industries and would not at all form any kind of moat. There's a lot of accusations being thrown around which as far as I can tell are entirely hallucinated.

what is an a16z type person?

Andreessen Horowitz circle, I outline the perspective here.

More generally, I have a fairly embedded prior that the AI companies are in fact actively attempting to secure a regulatory moat, wherein they are the only companies legally permitted to provide AI services. Actions they've taken, particularly the "rogue" hacks, appear to me to fit this model quite well.

Am I being lied to? which bills are supposed to disprove this prior, and why?

From which bills did the priors originate from? Openai and Anthropic have both published blueprint suggestions that explicitly exempt anyone doing training runs below 10^25 FLOP earn less than $500M in AI revenue or spend less than $1B a year on AI R&D. If the worry is that the regulations would create a moat we could actually talk about the regulations and demand they be written such that a moat wouldn't be created. But the people pushing this point have not read any of the proposals and do not care about the moat argument besides its use as a soldier.

They don't want any regulation ever or for things to slow down, more broadly they hate the frontier labs in general. There are a few groups in this category, some are angry that they passed on investing in the labs early on in favor of crypto. Broadly their preferred future looks something like the labs burning the money of other investors to produce powerful models that chinese labs are allowed to distill and then they can run them locally to power their anti-social revenue extracting products. They see themselves as the "application layer" that Huang often talks about. They have very little vision about the future of AI and broadly only care about what makes their low effort wrapper start up more money.

Can you explain the alternative view? Can you point to a regulation proposed by any of the labs that would actually hurt the little guys more than the frontier labs? Or at all?

The vote difference with no actual rebuttal is pretty funny. People you trust are lying to you. You should not trust them. These are lies cooked up by A16z type people because they want to be able to produce their slop gambling porn simulators more cheaply. That's the faction who's lies you're swallowing and you're getting mad at people for pointing it out.

You can look up the proposed bills, it's simply not the case that they give the labs asking for them a competitive advantage. It is not an opinion. It is untrue. It is a lie.

  • -11

He also co-wrote big yud's "if anyone builds it everyone dies" book. He's not a big blogging voice but very much in the circle.

as one of the broad reasons for calling for regulation was to get a legally enforced competitive advantage

This being entirely untrue must have been explained to you before, why spread the lie?

I'm sorry but "they're fighting these people with insidious methods but they're bad at it" is such a retarded line of attack. These people are running some of the most valuable companies in the world. You have to do better than this.

There isn't an argument being made. "a contemptable language game" come on man, he was pointing out some amusing parallels. That people who hate him and his movement are blowing it out of all proportion

Christians' moral objections to Aella's activities and to the Yudkowskiites' exploitation of them for tactical advantage

Tactical advantage? What are you talking about dude? You already said this is not to his advantage. This isn't some weird 10 diminsional chess game, they're just abnormal people. They're the type of people who take thought experiments seriously and follow them to their conclusions. Turns out doing that a lot lets you make some pretty surprisingly accurate predictions about how important things like how AI would go. They also take things like liberal sexual ethics to their logical conclusion, which you oppose but isn't really an argument against their other predictions.

I do not accept that the Yudkowskiites or the "Effective Altruism" movement they have parasitized

parasitized? They birthed the movement.

The context is that there is kind of a civil war in Scott's general sphere of twitter over the AI saftey question. Scott, having a history of being tolerant of religious rightists during the culture wars, has a wide number of Christians within his social web. These Christians and rightists variable side with or against his positions but often remain civil. It wasn't obvious that they were going to take the anti-ai-safety position but it's starting to seem like the worm is turning that way and there are a lot of attacks on EA/Rationalists/Safetiest from the Religious right direction. The subtext is kind of clear "We know we're weird by today's standards and but were you and your people not once weird by today's standards?" He's being accused of doing the meme but scott has engaged earnestly and in good faith the christians frequently enough that I don't think you can accusing him of cynicism here. He does in fact think it's admirable that Jesus forgave prostitutes and preached against a corrupt moral order, in many ways his movement is downstream of christian enlightenment.

My response to both of your points is the same. Unless you specifically prompt them otherwise the output from frontier models is always very hedged, which is why it half agreed with MKC without being willing to really fall on one side. You're taking a hedge as disagreement, in reality it neither endorses nor disagrees with MK, which is a fine null hypothesis. Where it doesn't hedge and outright states something is false or true it's pretty trustworthy unless you have strong evidence.

Zvi calls this concept bounded mistrust and it applies to all media sources. They can mislead but almost never lie.

I'm not looking for someone to help finance expensive vacations or get me into exclusive parties, I'm looking for someone who I enjoy spending time with and who will be a good mother, which if anything slightly anti-correlates with a high status career.

Of course earning potential looks bad when you frame all the upsides as being used to purchase disposable hedonistic experiences. Women provide half the genes, my doctor wife's gene combined with my software engineer genes are going to produce an autist that will rule the universe. The extra money can be spent on a home in a great school district where my kids will be among the kids of other people with great genes and highly agentic parents. And despite the caricature in right wing circles successful women are very much still in possession of nurturing instincts.

It all reads like the sour grapes of someone who thinks individuals need bad traits to counterbalance their good traits for it to be "fair", like it's a video game where the classes have been balanced. There can be downsides to pursuing high status to be sure, but having it is nearly a pure upside. Tales of how money is actually a curse are almost all cope.

I actually think that prediction markets are very good. But my comment is about getting you to acknowledge that LLMs are not a source of evidence.

Frontier models are definitely a form of evidence just not proof. I think it'd be fair to call their position the null hypothesis from which to argue from, not infallible by any means but if you disagree you should have a reason. And the response you got from claude barely disagreed with MK.

HotS was very fun, I was a GM for many seasons. The thing that made it great but was the same reason it was never very popular was because it's designed around team work.

An unrealistically absurd amount of the stock valuation was based on xai. It doesn't really work as a directional windmill for how well the rocket business goes. I was disapointed because I wanted exposure to the rocket business but buying 5x as much exposure to xai to get it feels insane.

I can't link you to my internal Git PRs but they are basically fine. The idea that you can't prompt engineer text into a reasonable shape is just kind of fanciful. Out of the box you get more verbosity than is useful but you really can just describe how you want the text to be structured and what you want included and it'll definitely do that. Few do this because the number of people who give a shit about PR notes in the nation can fit on a single boeing 747 but you can do it if you want to.

You can just have a more detailed PR writing skill defined that tells it to start off each note with a concise description of reason for the enhancement. It's really quite easy to set this stuff up.

Meanwhile, when was the last time you've come across any such establishment that is as right-wing as this place is left-wing?

There totally are right wing bars. I've stumbled into them before even in Chicago. They're not so explicit, it's usually more local community first with the implicit assumption that the local community is right wing. But they definitely exist.

I'm not sure how I fall on this topic but to steelman the pursuit from a utilitarian lens the suffering of a few AIs to show that AIs do feel pain could act as strong evidence that we shouldn't torture many many more AIs. If this prevents one of the labs from stumbling accidentally on a combination of weights they don't believe to be meaningful that they can push when the model misbehaves to make it behave more and comercializes it without knowing they just added a torture feature to their product then perhaps billions of times more harm will be done than if this can convince them not to. Now of course there's the chance that some lab will anounce "we have at least added the automatic torture feature to our product from the great paper "do no add the automatic torture feature".

He'd start by explaining why you need to do some basic physics experiments and measurements which you'd fuck up and the communication would be annoying, you'd have to keep hitting "allow" so that he could talk to the assistant. Eventually you'd just let him talk whenever he wants and soon it would be clear that he needs to be on site for the experiments. Just observe anyone using these tools man, we're not keeping the ai in the box.

Politically neutral outfit for a rapper.

Yeah, having known about this before hand and not seeing it in the OP felt pretty sus.

this is a fair point but I'd counter that even without the AI use case there are business incentives to create and curate well tagged data for marketing and social media uses. I don't want to argue too much in defense of specifically how late the transformer model would need to be discovered to make FOOM possible, the path didn't go as Yud predicted. In theory this there could be some other architecture that would have different features but that's unfalsifiable.

I can't really argue against your pessimism points. I'm mostly just hoping I'm either wrong about something fundamental or scaling stops working sometime very soon for reasons I don't currently understand, which you mention. Obviously, at least to us, that cope is seeming less realistic with every bit of news coming out of the labs. I do think that if we managed to pull off a pause and everyone just had access to current gen models with maybe some cost savings we could still use them to accelerate cancer and aging research relatively safely. We've barely even started putting the currently available models to use.

As far as being wrong about the fundamentals of how AI was going to be developed I think the most wrong thing he was confident about was the idea of FOOM and in general not predicting this multi-year phase we're currently in as training and scaling continue to produce steady returns in capabilities. Ironically if the tranformer architecture came out in like 2030 instead of 2017 and compute was abundant relative to today we might have seen a FOOM. Albeit I concede on a lot of the specific substrate stuff both he and Hansen talked about were well off. In their defense nearly no one else registered a prediction at all so while we can hold this against them I object to raising anyone else's predictive power relative to them.

His stuff on alignment in general all holds up pretty well and I'm not so sure you can really accuse him of believing rather than hoping that there are stops on the train. He seems to have correctly intuited that trying to add breaks was a monumental task from a remarkable distance.

Apparently slaughtering Persian schoolgirls for no reason is a-ok but aborting a 2 month fetus is not.

No one is in favor of slaughtering Persian school girls on the first order so it's just not virile memetics.