@VelveteenAmbush's banner p

VelveteenAmbush

Prime Intellect did nothing wrong

3 followers   follows 1 user  
joined 2022 September 05 02:49:35 UTC

				

User ID: 411

VelveteenAmbush

Prime Intellect did nothing wrong

3 followers   follows 1 user   joined 2022 September 05 02:49:35 UTC

					

No bio...


					

User ID: 411

Are you really just overlooking the word "white" as though it's an unnecessary detail in this discussion?

Less unlikely to do so. Where are the pictures of the White Christian White House staffers having developed a firm enough ethnic identity as such to gather together and take pictures of themselves celebrating their shared White Christianity?

Every alliance that excludes you from its membership is an implicit threat to your interests. This is basic Survivor logic.

In all seriousness, top companies had to have prepared PR teams for this scenario.

They very much haven't.

I think it is impossible to overstate just how far outside of the bounds of thought EY style doomerism has been and remains for... well, everyone except the "rationalists." It is literally impossible to talk about "AI safety" with normal human beings without them looking at you like you have two heads. The logic doesn't matter. The world runs on inductive reasoning, not deductive reasoning. Because "AI safety" has never been a problem in real life so far, it is literally impossible for normal people to understand it, much less take it seriously. If you try to explain it, you will notice that they cock their heads while they listen to you, and this is from the cognitive effort of rewriting your arguments in realtime as they hear them to be about jobs and racial bias instead of AI safety.

I am not an AI doomer. I ascribe to exactly your view with respect to Erlich and Yudkowksy, and it's well said.

But I am reporting to you, from the corporate front lines, that every single person in a position of authority has a brain defect that makes it literally impossible for them to understand the concept of "AI safety." They don't disagree with AI safety concerns; they cannot disagree with the concerns, because they cannot understand them, because when you articulate a thought about AI safety, the words completely fail to engender concepts in their brain that relate to AI safety. They cannot even understand that other people have thoughts about the concept of AI safety, except perhaps as a marketing ploy to overstate the the commercial utility of various AI-powered systems.

So the PR people have not planned a response, and the policy people have not engaged with the concept, and the executives have not been briefed, and you should expect large companies to continue acting as uncomprehending about the topic of AI safety as they would about the threat of office wall art coming to life and eating their children.

That's part of the bargain to be a lifestyle brand. If you want people to value your product not just for its practical utility but for what that product says about the people who consume it, then they're going to be as insulted by the unfavorable brand implications as they are flattered by the favorable. It doesn't matter that the ad wasn't "aimed" at them. Lifestyle brand advertising works by influencing what other people think of the product's customers, not just the customers themselves. The whole reason you choose to become a customer of a lifestyle brand is because of what you expect it will make other people think of you.

OK, so then my initial point that a parental rights argument is complete bullshit here is correct then?

Some red states believe that social transition is a question that should be left to the parents. Others believe that it should be forbidden. Blue states seem to adopt the mirror image of the latter view, that it should be required (for children who demand it) without the parents' approval. The second and third groups of states would seem not to care much for parental rights on this topic. The first does. That seems to be the lay of the land. I'm not sure how it would improve one's understanding to insist that any particular argument is "complete bullshit," unless you believe that it is somehow illegitimate for different states to adopt different policies.

Then why are those states also trying to ban minors from social transitions with parental consent?

Because those decisions are seen as so intrinsically damaging to the child that the decision to proceed should not be allowed at all. We take the same approach (correctly, IMO) with respect to the question of whether the child should be permitted to have sex with adults. We can disagree at the object level (i.e. whether in fact social transitions are damaging to the child), but if one accepts those states' belief at the object level for the sake of argument, then there is nothing anomalous about their policymaking approach to the matter.

My point is less that the decision should rest with the parents, and more that it shouldn't rest with the child.

Sometimes we leave the decision with the parents... whether to attend a parochial school, whether to allow the child access to Instagram, etc.

Sometimes we decide for the family; children aren't allowed to have sex with adults, for example, regardless of what the children or their parents think about it.

What we never do (other than with this trans issue -- at least, I cannot think of any other examples) is leave decisions of momentous lifelong consequences fully up to the child, and attempt to disintermediate the parents in favor of the child's own judgment.

So arguments premised on a child's voluntary and insuperable consent in high-stakes decisionmaking seem rather anomalous.

cancer treatment either saves you or you die, you aren't expected to have regular chemotherapy for the rest of your life

Cancer treatment is extremely expensive and cancer patients (especially those caught after Stage I) are often medicalized for life even in the cases where they live for decades afterward.

Moreover what does the duration of the treatment matter? Why should the pharma company's incentives be different as between a therapeutic that is extremely expensive over the short term and a therapeutic that is moderately expensive over the long term, if the NPVs are the same?

Shouldn't your theory predict, for example, that pharma companies selling expensive therapeutics for lung cancer should oppose smoking cessation efforts?

Then the argument moves to, well isn't puberty blockers irrecoverable harm

By moving the argument to puberty blockers, are you agreeing that all gender confirming care for minors that is more aggressive or less reversible than puberty blockers should be banned?

Otherwise, I assume you'd move the argument to those instead, right?

What if the children just wanted their ears removed? This wouldn't render them deaf, just leave them visibly mutilated by prevailing standards. Is that irrecoverable harm? Who is to decide what constitutes harm, and what constitutes the realization of one's inner truth however aberrant by wider standards?

Plenty of companies are already doing that. Doubtful that Eli Lilly's contribution would move the needle much.

By the way, do you have the same reaction to companies that produce cancer drugs -- that they should invest in causing as much cancer as possible to expand their addressable market?

If you truly believe children should be empowered to make all important personal decisions for themselves, without interference, then how would you argue against pedophilia? Or are prohibitions of pedophilia likewise to be excluded from the eschaton, whatever that means?

Suppose an online fad were persuading children to have their left arms amputated, and the power of the state dedicated itself to facilitating the amputations and to retaliating against parents who tried to interfere. What argument against that public policy would you consider to be fair, if any?

A sterile kid with diabetes is going to get lifelong injections and have to adopt if they want kids but I wouldn't characterize that as a "vista of terrifying possibilities".

I actually think "terrifying" would be a pretty reasonable word if one were contemplating a scenario in which children were being persuaded at scale by niche online communities to become infertile diabetics for life and the government were employing the power of the state to prevent parents from protecting them.

voluntarily sterilize themselves

I don't want to engage with most of this analogy, but I think your view is impoverished if it doesn't account for children's questionable ability to provide informed consent, and the seeming purpose of the law in attenuating their parents' ability to act as stewards for their children's interests. The question of what is truly voluntary is the heart of the matter.

GPT4 is 3 months old

6 weeks, actually. It was announced on March 14.

If simple double-passing the prompt was sufficient, I imagine the developers would have figured it out a long time ago.

It is, and they have. But this requires twice as much inference, and GPT-4 is already very expensive compared to the set of internet operations that we intuitively consider in the same class. Then you need to compare the answers and determine whether they match, which requires either manual effort or a third prompt.

This tech just hasn't been around long enough to build products around it. But if and when our civilization gets around to building and iterating a dedicated commercial medical expert product from LLMs, I've little doubt that hallucination will be a solved problem, because the cost of running a whole bunch of parallel prompts and then a subsequent round of prompts to confirm their consistency will be negligible in proportion to the commercial value of the tool.

I don't see what conquered people has to do with anything given that losing a war doesn't suggest one is more predisposed to violence than the victor

It suggests, on average, that the conquered people are less fit than the conquerors.

HBD posits a partial reversion to the mean one generation after the selection event occurs. After that, there should be no further effect; the non-heritable components of the initial selection (including both shared and non-shared environmental components and test error from the selection event) will have washed out with the next generation, while the heritable component will remain forever.

If the American slave population was adversely selected in the first place -- African tribes selling their own convicts, misfits and conquered people to Western slavers -- then HBD provides an explanation why the group descended from them continues to underperform.

The entire affirmative action policy regime depends on the assumption that group disparities are a problem that can be rectified.

I think there's a redistributive justification too, as well as a representational justification, as well as a justification premised on the purported instrumental organizational benefits of diversity. I happen not to find any of those justifications persuasive, but affirmative action supporters do not have all of their eggs in the remediation basket.

because self-driving cars could completely disrupt it

Impossible. There just isn't enough road capacity to replace the NY subway system. Traffic in NYC is merely painful now, but that's because the subway exists and most people use it. If everyone tried to move to cars -- even assuming the parking problem away with self-driving rideshares -- the roads would become jammed to the point of complete dysfunction.

So, it seems that the city CAN tow his car; they just can't charge him for the costs of doing so.

So... they can tow it, but then they have to give it back for free and let him drive away with it as soon as it reaches the impound lot? What would be the point of that? Couldn't he just drive it right back to where they towed it from?

What fraction of parents do you think are actually reading these articles, or even seeing the headlines? I'd estimate less than 10%.

I know lots of people who maintain a healthy weight. I know many fewer people who were once obese but then slimmed down and maintained a healthy weight by dieting. It's far from clear from the data that "trying to diet" or "telling people to diet" is an effective intervention once they are already obese. It seems like something intrinsic to metabolism and appetite regulation is irreversibly broken at that point, and the only permanent solutions are gastric surgery and (now, hopefully) semaglutide and its analogues.