@ThenElection's banner p

ThenElection


				

				

				
0 followers   follows 3 users  
joined 2022 September 05 16:19:15 UTC
Verified Email

				

User ID: 622

ThenElection


				
				
				

				
0 followers   follows 3 users   joined 2022 September 05 16:19:15 UTC

					

No bio...


					

User ID: 622

Verified Email

He appears to have lived and breathed a world where bank fraud charges are routinely brought and routinely won by the government.

The root issue here seems to be too much government power; give me six lines written by an honest man etc. It always starts out unobjectionable: of course we want to be able to battle terrorists/drug cartels/child pornography rings. But once the tool is established, its scope of use expands to things that I find unobjectionable (in this case, I have no deep objection to private organizations investigating other private organizations).

And ultimately it ends in endless lawfare, with whoever is in power deploying those tools against their enemies: and everyone is guilty of something, so any complaints can be met with "the law is the law," and defendants can only respond with weak arguments and accusations of hypocrisy. It is, of course, rich for Democrats to complain about this, but it's nevertheless an unfortunate situation.

Is your complaint that the model is training on tokenized language data? I.e. hypothetically, if you had a model that instead learned human language by being trained purely on tokenized audiovisual data taken from some embodied robot with motor controls, would you see that model as an intelligent being? I don't see why that should matter and we're nowhere close to that, but I want to understand your objection better.

Or is it the very idea of using prediction as a self supervised learning objective that's objectionable?

You can "pretrain" with RL:

https://arxiv.org/abs/2506.08007

The model it builds--what types of people do certain things, how it relates to their perspective on the world--is in the weights, not in the preamble. The memory bank is only for identifying which type of person is prompting the LLM; the actual model of P(vote|person likes horror fiction, fine press, math) doesn't reside in the bank.

There's some kind of model building going on. E.g. I never discuss politics with Claude. Yesterday I was filling out a ballot and tested it by asking it to predict my votes (based purely on its memory bank of our previous interactions; I confirmed there are no political memories in it). Out of what was over a dozen real choices, it only got one incorrect. That's a pretty effective model of my political preferences (which wouldn't be accurately modeled by aggregate polling of the normie voter).

Maybe the issue is that "interpolation and limited extrapolation paired with a reward signal" is really what intelligence is, and although it's all LLMs are doing, it's also all that humans do too.

Gary Marcus is a hack who has been saying deep learning can’t possibly work for years

Gary Marcus has plenty of issues, but fundamentally he doesn't reject DL, or even that DL will one day enable AGI. When pressed he thinks we will get to AGI in 8-15 years. Which is an entirely reasonable timeline, if a touch conservative.

He's a smart guy who's discovered a niche playing up current limitations to midwits in a way that seems he's making much stronger claims than he actually is.

I like this variation: same exact framing, but if you press blue, a random person dies, unless 50% or more press blue.

Morally, it seems like it has most of the same forces involved: there will be idiots or psychopaths who press blue, but if society can coordinate to get to 50%, no one dies. It adds some skin in the game for red: voting red doesn't give you safety, though it increases your chances of survival infinitesimally.

But I think this would drive more people to choose red, because it makes the sacrifice of voting blue not yourself but instead another person (extremely likely).

This is weird too me: if all lives are equally valuable, then which of the two scenarios shouldn't change whether you choose red or blue.

Her good poll numbers among Democrats is just name recognition: Trump bad, she anti-Trump, therefore Harris good. But, in practice, she has been an incredibly weak candidate in every election she's run in, for two decades. Whatever advantages she does have are matched and exceeded by Newsom in particular.

Unfalsifiable seems a bit strong, here: we just don't have the data collected to falsify the hypothesis. But that data isn't in principle uncollectible.

Though, it's an interesting question from a philosophy of science point of view: if data that no one will collect anytime soon is necessary to falsify a hypothesis, is that hypothesis unscientific?

Why should I pay attention to something that I have literally zero power over, as opposed to focusing on building a rich, fulfilling personal life? Does that make me stupid?

Men are expected to construct romance. It's a labor of love, but labor nonetheless. And part of that construction is hiding the labor, making it seem effortless and even magical.

To share that breaks the illusion. This reveals character traits about the man: either he is socially incompetent enough to not understand his role, or he understands it but resents it enough that he's nevertheless going to throw it in women's face. Both of those possibilities are unattractive, for different but very good reasons.

A third possibility is that he's just a deep systemizer and is looking for another deep systemizer and wants to fly his freaky systemizer flag wildly to filter out nonsystemizers. This is more sympathetic, but women who reject him for it are doing him a favor: they're incompatible, and he doesn't want to date a nonsystemizer anyway, so everyone wins.

And it's easy not to discuss it: it's a shared understanding of reality similar to "the sky is blue." If I'm dating a woman, obviously she already knows that e.g. women in general strongly prefer height. Bringing it up isn't going to lead to enlightenment on anyone's part, and that makes it come off as more begging for sympathy or validation than an interesting topic of conversation, which is obviously unattractive in a man.

I know some couples that qualify, but it tends to be an above average in attractiveness Indian guy who is reasonably professionally successful. That doesn't seem to be the case here.

I don't think the Motte is excluded. People here have done truesight tests with recent frontier models, and they're able to identify their handles.

The people who need to hear women are attracted to dominance aren't the ones beating their girlfriends. Typically, they are very passive and milquetoast in their interactions with women, leading to their failures in dating.

Hypothetically, you might imagine one of them wildly overshooting, but in practice that's not an issue: "dominance" isn't close to their default state, and even minimal movements in that direction are very uncomfortable.

Make life socially a lot tougher for male children. This will result in 45th trimester self-abortions, concentrated among the bottom quintile of men.

I would guess that Aboriginal Australians have a high percentage of genes from whatever hominid species was in Australia when Homo Sapiens showed up

Denisovans. IIRC the Denisovans were never actually in Australia, but Aboriginal Australians (along with Melanesians) picked up their genes in transit.

You left out the "borderline retarded 16 year old guy looking for a friend."

"A lot of money" here is probably $500: https://www.thecut.com/article/how-to-pitch-the-cut.html .

It's more frustrating to me that thecut.com can parade a mental illness around for engagement (and I'm indeed engaged) and make five to six figures in revenue from a single viral article/carnival act.

Selection effects are likely significant here. If high income people are disproportionately likely to get married, it could simply be a case of "the top 1% of the top 50% has a higher average than the top 1% of the whole."

Probably it's a combination of both.

Even in San Francisco, I would bet that the average male nurse would get more opposite sex matches online than the average male SWE, which is as good a proxy for status as any. The only place that wouldn't hold is if you work for a number of prestige companies you could count on one hand (and, no, Meta and Google aren't there).

if your recovery requires other people to orient their bodies around your triggers, your recovery is not going well

Ortega's recovery clearly isn't going well, but the essay raises a question for me: has anyone investigated GLP-1s and their effects on disordered eating? A WeGovy Rx might genuinely be what Ortega needs (since she broke the dam of getting overly involved in others' medical decisions). It silences exactly the kind of compulsive addiction patterns that might be involved in eating disorders. If it helps with alcohol, nicotine, and gambling addiction, why not eating disorders too?

Americans get access to ridiculously opulent luxuries, like air conditioning and closets.

Beneficial alleles might not coincide with the values or goals someone has for Homo sapiens. You could very well see e.g. rule following and neuroticism as good (either in themselves or in what function they perform in society), but they make carriers susceptible enough to behavioral defects to pull them out of the gene pool.