@FeepingCreature's banner p

FeepingCreature


				

				

				
0 followers   follows 0 users  
joined 2022 September 05 00:42:25 UTC
Verified Email

				

User ID: 311

FeepingCreature


				
				
				

				
0 followers   follows 0 users   joined 2022 September 05 00:42:25 UTC

					

No bio...


					

User ID: 311

Verified Email

Or it is literally just a parameter (PDF).

At a high level, we first identify a sparse set of attention heads with high linear probing accuracy for truthfulness. Then, during inference, we shift activations along these truth-correlated directions. [...] ITI results in a significant performance increase on the TruthfulQA benchmark.

From the fact that you are allowing @foreverlurker to walk

Was the parent comment edited...?

First, @foreverlurker [...] you are banned

To be fair, abortionists are maybe just smart enough to know that it'd be bad play. And at the end of the day, the way to deceive someone is doublethink- just avoid noticing the contradiction. Humans are very good at this. The unusual behavior from a human would be requiring consistency based on the actual cognitive features of the embryo or baby. "I don't see how that helps me win this conflict."

Let me just google these:

  • Policy Press seems to be attached to Bristol University.

  • Stylus Publishing doesn't offhand have an association with any university, though they certainly depict themselves as a university publisher.

  • Multilingual Matters "is an independent academic book publisher based in Bristol, UK"

  • And St. Martins just seems to be a freeform publisher.

So depending on qualification, this is either 0/4, 1/4 or 3/4. What do you consider an "academic press"?

Always the question, categories for what purpose? Whether adoptive mothers and mothers are the same or different categories will depend on whether you're a woman trying to commiserate about pregnancy (obviously different) or a child classifying their experience of motherhood (usually indistinguishable). Without clarifying what task we are trying to differentiate, the question is unanswerable; when clarifying, the categories almost disappear.

As a European, this is half the reason I'm in favor of us supporting Ukraine.

But by those rules as you described it, wouldn't then the president taking stuff home be considered spillage in the same logic? Ie. from himself to himself? Then because - because - the president doesn't have clearance to own it anymore, he might actually paradoxically be in the clear to pass it on to journalists?

Like, he now has classified information, but he's not "forwarding classified information" because he's not in a position where he has special authority over classified information to begin with. It's just like a journalist passing a leak to another journalist. And when he had the authority to possess it, he spilled it, but that can't be a crime because, well, he had the authority to do that then. So in the journalist analogy, the president basically acts as both leaker and releaser at different points in time.

All the nerds use old.reddit.com because new.reddit.com is fully focused on centering content rather than comments and spamming page reloads so it can show more ads. (You get like five comments per page load.) Lots of people, me included, would leave the site if old.reddit was ever turned off, it's that bad.

The Reddit app imitates new Reddit. If you want old Reddit experience on a phone, you need an unofficial app. These use the API.

If the president calls some other country and tells them something that's classified, and he doesn't know it's classified, I think it still becomes declassified in doing so. At least, that seems to be the argument. So in effect, by taking them home and keeping them past the end of his presidency, Trump declassified the materials without realizing he did so.

I mostly just find that argument amazing and I hope it wins for its own sake as an argument.

Hang on. You're saying he declassified them by accident?

Basically, by putting them in his possession when he became a person who did not have authority to see them, he implicitly declassified them because he was still president when he made the decision?

If yes, that's hilarious.

I mean, it would lead to endless tit-for-tat only as long as supplies of crimes last. I mean, you could make it last a long time by changing laws, but you'd have to put a bunch of additional work in. Absent a new wave of ex-post-facto laws or blatant procedure prosecutions, honestly my first reaction is "yes, good." Let justice reign, etc.

Seems special pleading: why are we requiring no special incantation when declassifying, but are requiring a special incantation ("these are classified" is not enough!) when classifying?

That said, has Trump spoken any incantation for declassifying?

I think this crucially depends on the death rates from punching vs shoot-to-fist ratio. Also I don't think fistfights are rational, but getting into a situation where a fistfight may ensue is absolutely rational. If you look at for instance duels, IMO a society where getting shot is at risk can develop alternate ways of mediating the sorts of situations that otherwise become fistfights. This is becoming really hard to model - but I don't think that guns are limited to having a 50% reduction on fistfights. If they were, they'd probably be a bad trade on utilitarian grounds alone, though there may still be other cases for them.

If I had two sons, and one son got drunk and punched someone at a bar while another got drunk and was punched by someone at a bar. I would not want to live in a world where the former was killed and the latter killed their assaulter. I'd much rather live in the alternative world where no one died. Which would you rather live in?

I think this is an instance of causal decision theory in the wild, in that you're holding the punch stable when there's no reason to expect that to be the case. What if it being "the sort of world where people who throw punches are killed" means that instead you get to pick between the world where your sons punch and are punched, and the world where nobody is even punched? Then the question would be to what extent punch-kill actually allows acausal flow, right? Ie. we may imagine a world where some people just, out of the blue, are struck by the urge to punch and otherwise-agentically seek out a target to punch. In that case, the kill-branch obviously would only worsen the situation. So the question comes down to if the punch urge is such that the kill branch can successfully shift the incentives enough to suppress the punch branch enough to make up for the QALY loss.

Because at the end of the day, we'd at least somewhat prefer that the least people die. Right?

Aliens make no sense because the stars still shine. I would not expect the greatest visible evidence for aliens to be on Earth, I would expect it to be humanity surrounded by Dyson spheres. (If I was a civilization that got post-singularity, I would totally eat every sun.) The idea that the strongest visible evidence for alien life is found in Earth's atmosphere simply does not pass any smell test.

I mean, you'd want it to know where its infrastructure is so you can train it to protect that infrastructure. That does make some sense.

Sometimes, if a thing is "needed" and violates the constitution, that means you still shouldn't get to have it. What's the point of principles if you only hold to them on matters that are agreeable anyways?

I guess this reinforces something like "seasonal industry", where you can scale your production up and down with the power price. Might take a decade to adapt to this.

Like half the point of book term limits is to allow round-robin lending. If you're swiping the book, you're defecting against the person who wants to re-loan it, but that person is defecting against the library system.

I don't read this as "hey, can you relinquish your moral claim on this bike and transfer it to me" and more "hey, please relinquish your physical claim on this bike because it is immoral".

I don't think anybody was expecting ChatGPT to cheat the system like that. GPT-3 and GPT-4 aren't interesting because they're superintelligences, they're interesting because they seem to represent critical progress on the path to one.

A hilarious note about Bing: When it gets a search results it disagrees with, it may straight up disregard it and just tell you "According to this page, <what Bing knows to be right rather than what it read there>".

Yeah, I'm beginning to come around to the possibility that I did have a vastly different experience of the pandemic.

I mean, to my knowledge nobody actually got lynched. Of course, I don't have a control genocide to see if the actual rhetoric would be different.

You think we're epistemically in a 1937 position with regard to Covid camps?

If non vaccinated, or positives covid tested individuals and families had been shipped off to camps (outside china), and had been killed would that really have been incongruent with the rhetoric and propoganda deployed?

I obviously cannot prove this, but my immediate reaction is "yes, of course, massively incongruent."

We'd need more samples. I was right this time, but obviously n=1.

I think even in China, you could predict fairly reliably if a given camp or campaign was going to engage in mass murder or not, ie. whether the Uighur rhetoric is like the Nazi rhetoric in ways that the Covid rhetoric is not. To be clear, I don't have an opinion on this; I haven't done any research on genocide in China, but I'd expect if there was genocide we should see commonalities in the rhetoric.

edit: Ie. say, nobody was calling Covid victims dangerous parasites.... Okay, I'm not willing to say that. Maybe it's just that the US CW is so hot that the rhetoric on the street was genuinely indistinguishable from Mein Kampf? If so, Scott may be apropos: "Stop telling people they’re going to be killed. ... Stop trying to convince Americans that all the other Americans hate them."