LessWrong 2.0 Reader
View: New · Old · Top← previous page (newer posts) · next page (older posts) →
← previous page (newer posts) · next page (older posts) →
I don't think it is sensible to model humans as "just the equivalent of a sort of huge content window" because this is not a particularly good computational model of how human learning and memory work; but I do think that the technology behind the increasing context size of modern AIs contributes to them having a small but nonzero amount of the thing Steven is pointing at, due to the spontaneous emergence of learning algorithms. [LW · GW]
evan-anders on Sparse autoencoders find composed features in small toy modelsHi Ali, sorry for my slow response, too! Needed to think on it for a bit.
Got it, makes sense, agreed.
tomcatfish on My Interview With Cade Metz on His Reporting About Slate Star CodexHm. I think we like Slate Star Codex in this thread, so let's enjoy a throwback:
It was wrong of me to say I hate poor minorities. I meant I hate Poor Minorities! Poor Minorities is a category I made up that includes only poor minorities who complain about poverty or racism.
No, wait! I can be even more charitable! A poor minority is only a Poor Minority if their compaints about poverty and racism come from a sense of entitlement. Which I get to decide after listening to them for two seconds. And If they don’t realize that they’re doing something wrong, then they’re automatically a Poor Minority.
I dedicate my blog to explaining how Poor Minorities, when they’re complaining about their difficulties with poverty or asking why some people like Paris Hilton seem to have it so easy, really just want to steal your company’s money and probably sexually molest their co-workers. And I’m not being unfair at all! Right? Because of my new definition! I know everyone I’m talking to can hear those Capital Letters. And there’s no chance whatsoever anyone will accidentally misclassify any particular poor minority as a Poor Minority. That’s crazy talk! I’m sure the “make fun of Poor Minorities” community will be diligently self-policing against that sort of thing. Because if anyone is known for their rigorous application of epistemic charity, it is the make-fun-of-Poor-Minorities community!
I’m not even sure I can dignify this with the term “motte-and-bailey fallacy”. It is a tiny Playmobil motte on a bailey the size of Russia. (from https://slatestarcodex.com/2014/08/31/radicalizing-the-romanceless/)
Can your use of "Journalism" pass this test? Can we really say "the hostility to journalists is not based on hyperbole. They really are like this. They really are competing to wreck the commons for a few advertising dollars." and expect everyone to pay close attention to check that the target is a Bad Journalist Who Lies first?
jeffjo on The Solution to Sleeping BeautyLet it be not two different days but two different half-hour intervals. Or even two milliseconds - this doesn't change the core of the issue that sequential events are not mutually exclusive.
OUTCOME: A measurable result of a random experiment.
SAMPLE SPACE: a set of exhaustive, mutually exclusive outcomes of a random experiment.
EVENT: Any subset of the sample space of a random experiment.
INDEPENDENT EVENTS: If A and B are events from the same sample space, and the occurrence of event A does not affect the chances of the occurrence of event B, then A and B are independent events.
The outside world certainly can name the outcomes {HH1_HT2, HT1_HH2, TH1_TT2, TT1_TH2}. But the subject has knowledge of only one pass. So to her, only the current pass exists, because she has no knowledge of the other pass. What happens in that interval can play no part in her belief. The sample space is {HH, HT, TH, TT}.
To her, these four outcomes represent fully independent events, because she has no knowledge of the other pass. To her, the fact that she is awake means the event {HH} has been ruled out. It is still a part of the sample space, but is is one she knows is not happening. That's how conditional probability works; the sample space is divided into two subsets; one is consistent with the observation, and one is not.
What you are doing, is treating HH (or, in Elga's implementation, H&Tuesday) as if it ceases to exist as a valid outcome of the experiment. So HH1_HT2 has to be treated differently than TT1_TH2, since HH1_HT2 only "exists" in one pass, while TT1_TH2 "exists" in both. This is not true. Both exist in both passes, but one is unobserved in one pass.
And this really is the fallacy in any halfer argument. They treat the information in the observation as if it applies to both days. Since H&Tuesday "doesn't exist", H&Monday fully represents the Heads outcome. So to be consistent, T&Monday has to fully represent the Tails outcome. As does T&Tuesday, so they are fully equivalent.
If you are observing state TH it necessary means that either you've already observed or will observe state TT.
You are projecting the result you want onto the process. Say I roll a six-sided die tell you that the result is odd. Then I administer the amnesia drug, and tell you that I previously told you whether th result was even or odd. I then ask you for your degree of belief that the result is a six. Should you say 1/6, because as far as you know the sample space is {1,2,3,4,5,6}? Or should you say 0, because "you are [now] observing a state that you've already observed is only {1,3,5}?
And if you try to claim that this is different because you don't know what I told you, that is exactly the point of the Two Coin Version.
The definition of a sample space [was broken] - it's supposed to be constructed from mutually exclusive elementary outcomes.
It is so constructed.
I've specifically explained how. We write down outcomes when the researcher sees the Beauty awake - when they updated on the fact of Beauty's awakening.
Beauty is doing the updating, not "they." She is in an experiment where there are four possible combinations for what the coins are currently showing. She has no ability to infer/anticipate what the coins were/will be showing on another day.
Her observation is that one combination, of what is in the sample space for today, is eliminated.
No, I'm not complicating this with two lists for each day. There is only one list, which documents all the awakenings of the subject,...
Maybe you missed the part where I said you can look at one, or the other, or bot as long as you don't carry information across.
You are mistaken about what the amnesia acomplishes, Once again I send you to reread the Effects of Amnesia section.
Then you are mistaken in that section.
Her belief can be based only on what she knows. If you create a difference between the two passes, in her knowledge, then maybe you could claim a dependence. I don't think you can in this case, but to do it requires that difference.
The Two Coin Version does not have a difference.
tomcatfish on My Interview With Cade Metz on His Reporting About Slate Star CodexYour comment is actually one of the ones in the thread that replied to mine that I found least inane, so I will stash this downthread of my reply to you:
I think a lot of the stuff Cade Metz is alleged to say above is dumb as shit and is not good behavior. However, I don't need to make bad metaphors, abuse the concept of logical validity, or do anything else that breaks my principles to say that the behavior is bad, so I'm going to raise an issue with those where I see them and count on folks like you to push back the appropriate extent so that we can get to a better medium together.
tomcatfish on My Interview With Cade Metz on His Reporting About Slate Star CodexI'd be amenable to quibbles over the lock thing, though I think it's still substantially different. A better metaphor (for the situation that Cade Metz claims is the case, which may or may not be correct) making use of locks would be "Anyone can open the lock by putting any key in. By opening the lock with my own key, I have done no damage". I do not believe that Cade Metz used specialized hacking equipment to reveal Scott's last name unless this forum is unaware of how to use search engines.
tomcatfish on My Interview With Cade Metz on His Reporting About Slate Star CodexI don't (and shouldn't) care what Scott Alexander believes in order to figure out whether what Cade Metz said was logically valid. You do not need to figure out how many bones a cat has to say that "The moon is round, so a cat has 212 bones" is not valid.
tomcatfish on My Interview With Cade Metz on His Reporting About Slate Star CodexThe evidence offered "Scott agrees with the The Bell Curve guy" is of the same type and strength as those needed to link him to Hitler, Jesus Christ, Eliezer Yudkowsky, Cate Metz, and so on. There was absolutely nothing special about the evidence that tied it to the people offered and could have been recast without loss of accuracy to fit any leaning.
As we are familiar with, if you have an observation that proves anything, you do not have evidence.
random-developer on Modern Transformers are AGI, and Human-LevelYeah, the precise ability I'm trying to point to here is tricky. Almost any human (barring certain forms of senility, severe disability, etc) can do some version of what I'm talking about. But as in the restaurant example, not every human could succeed at every possible example.
I was trying to better describe the abilities that I thought GPT-4 was lacking, using very simple examples. And it started looking way too much like a benchmark suite that people could target.
Suffice to say, I don't think GPT-4 is an AGI. But I strongly suspect we're only a couple of breakthroughs away. And if anyone builds an AGI, I am not optimistic we will remain in control of our futures.