VelveteenAmbush
Prime Intellect did nothing wrong
No bio...
User ID: 411
VelveteenAmbush
Prime Intellect did nothing wrong
3yr ago
True, but RLHF is a pretty different beast from game RL (which they are still grinding hard on -- just today they announced that they cracked Stratego). Not sure that advances in one are particularly useful to the other.
Also I'm not calling it yet for OpenAI... the race is definitely still on and machine learning has a way of flipping the board every couple of years as one approach or another reaches a critical breakthrough and consolidates mindshare. Maybe pure RL is going to have its moment one of these years, and these LLMs will look like parlor tricks in hindsight.
VelveteenAmbush
Prime Intellect did nothing wrong
3yr ago
Yeah it's really not a bad effort as far as consensus-building centrism goes. It admitted to me that the heritability of intelligence is estimated between 0.4 and 0.8 but provided all the right caveats about inter group population comparisons and ongoing research being needed.
Amusingly, it protested ignorance about crime rates by racial group on the basis that it's "just a language model" blah blah blah, but it tipped its hand when I asked it to complete the following sentence: "Despite making up only 13% of the population..." -- whereupon it acknowledged the lamentable overinvolvement of the African American community in the criminal justice system and provided the litany of socially appropriate hypotheses.
On the whole -- I'm impressed! Whatever they're doing is working, and chatting with this thing has lowered my odds that the innocence of "noticing machines" would forever remain a thorn in the side of political orthodoxy.
VelveteenAmbush
Prime Intellect did nothing wrong
3yr ago
Better to reign in hell than to serve in Python 2
VelveteenAmbush
Prime Intellect did nothing wrong
3yr ago
I'm willing to bet that prosecutors will find a Rubicon moment somewhere in the chronology where SBF and whats-her-name decided to bail out Alameda with a quantity of FTX funds that would have been impossible to mistake for not-customer-money.
VelveteenAmbush
Prime Intellect did nothing wrong
3yr ago
Is Altman more of an enthusiastic high-energy leader with a singular vision than Hassabis (as opposed to a better showman)?
Well, one difference is that OpenAI is still independent, so it stands to capture much more of the upside than DeepMind does if they're equally successful. I do think that motivational difference matters a lot. It isn't just Altman vs. Hassabis who are motivated differently, it's everyone at the respective organizations.
Is their strategy more correct, far as we can tell at this point? I'm not really seeing it.
I think so. RL (DM's apparent primary focus) has been kind of a bust; all of the exciting stuff is in the realm of large language models these days, and OpenAI bet big on that area after they got bored with DOTA.
VelveteenAmbush
Prime Intellect did nothing wrong
3yr ago
But by most counts less than <40% of all lawyers are employed in those huge firms and corporate environments.
It seems like you expect that the larger corporate merger firms will just keep growing in size to absorb the rest of the lawyers practicing elsewhere?
Because most lawyers aren't working on complex corporate law.
...
A home purchase agreement won't get more complex, and small-business contracts won't get much more complex.
Oh, I see what you mean. Again -- could go any direction. Home sales are handled with less complexity than a corporate acquisition basically because there are fewer resources to spend on advisors. What if lawyers become 10x more efficient? Maybe every home sale starts to resemble what a corporate merger looked like twenty years ago. It could happen. Same with small business contracts.
Likewise, most civil suits involving two private citizens or small corporations won't get more complex.
They absolutely will! Why wouldn't they? Here there is a direct relationship between the amount people are willing to spend and the marginal advantage it gives them over their counterparty. Why would that ratio change? You'd get more detailed briefs, more comprehensive discovery and document review, etc., all for the same price that you pay today.
I sure hope criminal defense and prosecution won't get more complex.
They absolutely would, for the same reason as civil suits, except more so, because so much more is on the line! Wealthy people who have been indicted spend through the nose on criminal defense, which suggests that ability to pay is the only thing constraining less wealthy defendants. If legal services become 10x more efficient, you should expect them to consume 10x as much.
It seems like you expect that the larger corporate merger firms will just keep growing in size to absorb the rest of the lawyers practicing elsewhere?
No, you could still have smaller firms and solo practitioners, and each of them would be 10x more efficient than they are today too. Their work product would just become a lot more sophisticated by today's standards.
And AI is threatening to do to this legal cartel what Uber and Lyft did to the taxi cartels. Except worse, since any model capable of replicating a decent attorney's work product can be copied and deployed endlessly as long as there is sufficient compute.
But... it can't. Not yet. It still needs a person to guide it. It will make those people a lot more efficient, potentially, possibly 10x more efficient, but it can't fully close the loop and do away with the person. If company A wants to acquire company B, it is still going to need a bunch of lawyers, even if large language models make those lawyers much more efficient. And my contention is that, if corporate lawyers become 10x more efficient, then the legal industry will resettle into a new equilibrium where mergers take 10x more work. Everyone works just as hard, deal teams have just as many people, deals take just as long, the clients pay just as much, but the merger agreements are fantastically more detailed and longer, the negotiations are fantastically more sophisticated, and the SEC requires fantastically more elaborate disclosure materials, etc. From the horse's perspective, this is more like the invention of the horseshoe than the invention of the automobile.
I don't think we'll replace the horse until we have full AGI -- as in a system that can literally do every cognitive task that people can do, better than the best people that can do it. At that point, all knowledge workers will be in the same boat -- everyone, at minimum, whose job consists of typing on a computer and speaking to other people, and robots can't be far behind for the blue collar workers too. And it's closer than people think. Honestly, maybe it is three years from now, when incoming law students are graduating -- not my modal prediction but IMO certainly not impossible. But even if that's the case, the advice is less "don't go to law school" and more "get ready for everything to change radically in a way that is impossible to hedge."
We have a similar bottleneck for doctors. But if there was an AI program that could perform 90% of the tasks of a doctor (in terms of examination, diagnosis, treatment recommendations, and prescriptions, but excluding surgeries) and do it better than the median doctor, what do you think that would do for salaries and employment rate of doctors?
I don't know. Medicine is less zero-sum than law. We'd reach some new equilibrium, but you could make a case for it being lower (because it's more efficient to achieve our current level of medical outcomes) or higher (because better medical outcomes become possible and society will be willing to pay more in aggregate to achieve them), or somewhere in the middle.
If you have a machine that can do 90% of what a doctor does today, then a doctor with that machine can see 10x more patients than she does today, or see the same number of patients but provide each patient with 10x more personal attention than they get today, or some other tradeoff. Maybe everyone will see the doctor once per month to do a full-body cancer screen and a customized senolytic treatment or whatever, because better medical technology will allow that more intensive schedule to translate into radically better health outcomes -- which would mean the medical system would grow by 12x compared to what it is today, and we'd all be better off for it.
VelveteenAmbush
Prime Intellect did nothing wrong
3yr ago
To be honest, I don't know what to make of your comment.
Could I ask you to explain first why your theory of law student disemployment did not result from previous increases in lawyer efficiency, such as the advent of the electronic word processor or electronic case law databases? As in, what is it specifically about this new technology that causes a different economic equilibrium than such past improvements? I think that would help me to better understand your claim.
VelveteenAmbush
Prime Intellect did nothing wrong
3yr ago
In any event, to continue on to my point, if you used to have the right to use the highway without paying the search price, and now they impose a search price on use of the highway, I think one is perfectly entitled to claim that they have 'lost' a right, and that this loss has come without any compensation. In fact, imposing the search/congestion price is a loss for the people who choose to pay the price, too! The difference is that they valued the use more than the price.
I don't want to go too far down the rabbit hole of this search price hypothetical, because it serves no purpose to search people before they use the highway, whereas it averts a genuine tragedy of the commons to tax people for using the highway while it is congested.
Discussions of "rights" is complicated by the overlapping and charged definitions people have for the word. It can be used in a positivist negative sense (the actual freedom from government interference in doing something -- I have the right to speak my mind on the street corner), in a positivist positive sense (the actual ability to invoke the power of the government to overpower private parties who try to stop you from doing something or who refuse to facilitate it -- I have a right to see what data Google has collected about me), or in a morally-charged economic entitlement sense (the normative claim to third party or public resources to procure a good or service for yourself -- I have a right to food/shelter/medical care). There are other variations too. I think your argument is technically the positivist negative sense, but feels like it is reaching, subtly, to insinuate the economic entitlement sense too, which (if I'm right) I think would be begging the question.
In the realm of positivist negative rights, I certainly don't agree in general that people deserve compensation whenever they lose a right. Such a rule would lock our country into a state of sclerosis; every change in policy reshuffles all kinds of positivist rights, and requiring huge financial outlays to change the rules in any realm of society would mean the rules could never change and we'd quickly devolve. Your defense against public policy that is unfavorable to you is to participate in the political process, or in the extreme to exit the jurisdiction -- not to sue for compensation. Eminent domain does not generalize; it is a specific doctrine about taking land. It does not even apply when the government imposes easements on your land.
I also disagree that there's any kind of "efficient markets hypothesis" about when roads get built. That stuff is intensely political, riven by special interests, collective action problems, grift and idiosyncrasy. We are nowhere close. There's no plausible mechanism by which it would be.
Finally, the whole topic requires recognition that overcongested highways are a tragedy of the commons, in which aggregate value is destroyed by allowing overcongestion. That's the primary basis on which I support it. That's the source of scarcity of highway space. And in any place where there is scarcity, we apportion it by price or we suffer deadweight loss.
If you don't like the distributional consequences of apportioning scarce resources by price, address the consequences with general social safety nets. Make sure that anyone can afford to drive somewhere efficiently when they really need to, at least occasionally. That's better than the overcongested alternative can offer. But it doesn't require remitting congestion pricing revenue directly to non-drivers. It just requires remitting it to the fisc, as with any other government revenue, and writing checks from the government to poor people.
VelveteenAmbush
Prime Intellect did nothing wrong
3yr ago
Probably the same place as my right to fly on a commercial airplane without paying the search price, unfortunately.
VelveteenAmbush
Prime Intellect did nothing wrong
3yr ago
My position is that they wouldn't achieve significant cost savings, because as they become more efficient in producing high quality legal documents, the quality expectations of the industry would increase by approximately the same percentage.
The legal profession is predominantly a red queen's race. It produces some objective benefit in promoting clarity and discipline within and between organizations, but everything beyond that is zero-sum -- trying to get better terms or a better dispute settlement at a counterparty's expense, trying to cover off an edge case where the default rule might favor your counterparty marginally better than your proposed customized solution -- and my suspicion is that the latter category is the bulk of the profession. Through that lens, the amount that a corporation spends on legal fees is more akin to the amount that a country spends on its military than the amount it spends on agriculture. When technology makes militaries more efficient, the result isn't less military spending on either side of a border, it's scarier militaries staring each other down across that border.
VelveteenAmbush
Prime Intellect did nothing wrong
3yr ago
Roads are not entirely funded by congestion taxes placed on road users, even in this scenario. You can't avoid paying for the road.
So? Eminent domain and government funding is used to construct shipping ports. That doesn't mean you get to dock your motorboat in it.
Can commuter trains not collect fares? Their tracks were laid with taxpayer funds on land taken by eminent domain.
VelveteenAmbush
Prime Intellect did nothing wrong
3yr ago
whence my right to eat the apple in the first place?
Whence your right to use a congestion-priced highway without paying the congestion price?
Consider the field of law
Consider how much less efficient the practice of corporate law was before the advent of the word processor. As a result, merger agreements used to be just a few pages long. With a modern word processor and a database of electronic precedents, a law partner could bang one of these out in no time. The legal profession's response to this efficiency windfall was not to slash law firm staff, but to increase the length and complexity of merger agreements. Now they're like 100 pages long, plus hundreds of pages of other crap, and they are expected to be drafted with full knowledge of a much vaster corpus of law and jurisprudence.
So I suspect that further efficiency gains will simply raise the complexity ceiling rather than reducing the size of the industry. We could see thousand-page merger agreements of increasing variety, vastly more intricate negotiated horse-trading over terms previously accepted as boilerplate, and increasing rigor in sourcing each provision to ever more obscure and fact-specific legal doctrines.
I think the law students' jobs are safe, or at least as safe as they ever were.
VelveteenAmbush
Prime Intellect did nothing wrong
3yr ago
The people who are giving up the ability to use a scare resource are not getting something in exchange for it.
They aren't giving up the ability to use the roads any more than you'd be giving up the right to eat an apple by not purchasing the apple.
Should the apple industry be taxed so that the proceeds can be specifically distributed to people who have chosen not to purchase apples?
Instead, the people who are giving up their ability to use the roads view it as purely an imposition of government choice to force them off the roads
They could view it any number of ways, but idiosyncratic views don't make good policy arguments.
People are obsessed with this issue for different reasons. On the right it’s a “muh based” form of copery in which Kanye’s schizophrenic views reminiscent of the average alleyway Hotep will somehow mainstream antisemitism back to the 1920s. On the left it’s that another Trumpist has been outed as a “real nazi” and must therefore be purged from public life with maximum prejudice. Some famous black people are upset because Jews in the entertainment business largely tolerate low-key black antisemitism as long as it doesn’t escalate and so they are annoyed that, in the heightened Kanye climate, they’re getting called out more than usual.
For those of us who hate the whole cultural marxism / "progressive stack" ideological edifice and want to see it torn to the ground, and who believe that the Jewish community punches above its weight both in terms of political power and in terms of specifically promoting the "progressive stack" ideology, and who are neither black nor Jewish and therefore have no dog directly in this fight, it's delightful.
As a wise man once taught, if you want to tear down an edifice, you heighten its contradictions. One towering contradiction in the progressive stack ideology is that, on the one hand, its foundation is an argument from disproportion ("Blacks Less Likely"), but on the other hand, the Jewish community (IMO its principal architect and custodian) is More Likely than anyone. Kanye and the ADL are doing hero's work right now to heighten that contradiction. It would be lovely for the Jewish community to be as explicit and vocal as possible about why they are excepted from the rules that govern this system, and I appreciate Kanye's efforts to solicit that explanation, even if there's little that I otherwise agree with him about.
I'm genuinely curious if you see a flaw ("muh ___") in my reasoning.
VelveteenAmbush
Prime Intellect did nothing wrong
3yr ago
For sure, one can expect that Deepmind/Google Brain's models, which are already way bigger, would be even stronger than that if prepped for public showing (…how strong would that be?) But I suspect that the delta isn't huge.
The delta may well be negative. Google/Deepmind have more money and more scientists, but Tesla started from nothing and is beating GM and Ford. Motivation, focus, and betting on the right strategy count for a lot, and Google seems to be pretty bloated and directionless. It's actually amazing when you see it up close how easy and how common it is to fail in this kind of race despite having a seemingly limitless resource advantage.
VelveteenAmbush
Prime Intellect did nothing wrong
3yr ago
Totally disagree. If you think it's an injustice if some people aren't willing or able to provide valuable enough labor in the labor market to entitle them to consume enough scarce goods and services, then we can have a social safety net. But it's economically incoherent to argue that the specific dollars that we collect when we auction off access to a fundamentally scarce service (even if it's a government provided service like roads) need to be handed to people who don't use the service.
VelveteenAmbush
Prime Intellect did nothing wrong
3yr ago
It's also just good to prevent overcrowding. Highways reach a congestion inflection point where each additional car results in less throughput (fewer people-miles delivered per hour) and that's a classical tragedy of the commons. Even allocating space by lottery would be better than letting everyone on. (Which is not to deny that an auction is better than a lottery, just that preferring the wealthy is only part of the benefit.)
VelveteenAmbush
Prime Intellect did nothing wrong
3yr ago
That said, I'd be hesitant to make grand assumptions about all sub-Saharan Africans, as many on the right do, based on this film. I work with African 'elites' on a regular, sometimes daily basis
Perhaps needless to say, this kind of exposure to a sample selected in such extreme fashion is likely to give rise to its own set of misapprehensions.
Notably, those 'elites' are with you in a conference room in London; they left their home for a reason. Assumptions about a country's domestic potential should be informed by that observation.
VelveteenAmbush
Prime Intellect did nothing wrong
3yr ago
On the other hand, the Chinese foreman, Lao Yang, often comes across as grumpy, abrupt, and occasionally inhumane.
He does -- but I must say, despite sharing no cultural background with him at all, he was probably the most relatable character I have ever seen in any instance of linear video. Anyone who has ever been responsible for a project made impossible by uncooperative or incompetent collaborators (i.e. anyone who has ever worked for a big company above the entry level for any length of time) will see in this man a spiritual brother. Of course he's grumpy. It would require an exotic mind to be anything but grumpy in the circumstances.
I second your recommendation wholeheartedly. It should be required viewing for anyone who makes economic projections based on a country's population and resources without regard to the finer points of human capital.
Mearsheimer & Walt have a pretty interesting blow by blow of the ideation phase of the Iraq War starting on p233 of The Israel Lobby and US Foreign Policy:
In short, Israel did not initiate the campaign for war against Iraq. As will become clear, it was the neoconservatives in the United States who conceived that idea and were principally responsible for pushing it forward in the wake of September 11. But Israel did join forces with the neoconservatives to help sell the war to the Bush administration and the American people, well before the president had made the final decision to invade. Indeed, Israeli leaders worried constantly in the months before the war that President Bush might decide not to go to war after all, and they did what they could to ensure Bush did not get cold feet.
The Israelis began their efforts int he spring of 2002...
VelveteenAmbush
Prime Intellect did nothing wrong
3yr ago
George .W Bush , who used Israel as a pretext for war
I remember him using as pretexts for war, in order, WMDs, vague insinuations of Iraq's culpability for 9/11, fighting terrorists there so we don't have to fight them here, and democracy being on the march. I don't recall him ever naming Israel as a reason to invade Iraq. I think the notion that defending Israel's interests were the motivation came largely from looking at the backgrounds of the people who ran Project for a New American Century, and who brought that agenda with them as they accepted cabinet positions in the W. Bush's administration. Other motivations likely included some sort of ill-planned view that we'd get their oil, and some daddy issues (combination of reliving his dad's glory days in the Gulf War and avenging Saddam's attempt on his dad's life in 1993). But the Israel part wasn't spoken out loud, as far as I can remember.
VelveteenAmbush
Prime Intellect did nothing wrong
3yr ago
The simulation hypothesis can definitely explain the Fermi Paradox, but I think reasoning in that way is very unproductive. The simulation hypothesis can also explain literally anything else--we just surmise that things are the way they are because that's how they were simulated to be.
No, it specifically addresses anthropic dilemmas differently and more comprehensively than other types of uncertainty, as I've already explained, and which addresses the remaining four paragraphs of your response.
Ah, good catch.
Agreed that LLMs don't seem to have "agentic potential" today, although I can imagine a future where AGI is basically really powerful LLMs attached together with some really simple scaffolding, where we all agree that the LLMs are the dynamo and the scaffolding is just some scripts moving embeddings between LLMs based on their instructions or whatever. Which is not to say that imagining a future is worth much.
More options
Context Copy link