3 ms·

In AI finetuning, there's a theory that the model already contains the right ideas and skills, and the finetuning just raises them to prominence. Similarly in p

In AI finetuning, there's a theory that the model already contains the right ideas and skills, and the finetuning just raises them to prominence. Similarly in philosophic pedagogy, there's huge value in taking ideas that are correct but unintuitive and maybe have 30% buy-in and saying "actually, this is obviously correct, also here's an analysis of why you wouldn't believe it anyway and how you have to think to become able to believe it". That's most of what the Sequences are: they take from every field of philosophy the ideas that are actually correct, and say "okay actually, we don't need to debate this anymore, this just seems to be the truth because so-and-so." (Though the comments section vociferously disagrees.) And it turns out if you do this, you can discard 90% of philosophy as historical detritus. You're still taking ideas from philosophy, but which ideas matters, and how you present them matters. The massive advantage of the Sequences is they have justified and well-defended confidence where appropriate. And if you manage to pick the right answers again and again, you get a system that actually hangs together, and IMO it's to philosophy's detriment that it doesn't do this itself much more aggressively. For instance, 60% of philosophers are compatibilists. Compatibilism is really obviously correct. "What are you complaining about, that's a majority, isn't that good?" What is wrong with those 40% though? If you're in those 40%, what arguments may convince you? Repeat to taste.

margalabargala 1y ago
> And it turns out if you do this, you can discard 90% of philosophy as historical detritus This is just the story of the history of philosophy. Going back hundreds of years. See Kant and Hegel for notable examples.
- FeepingCreature 1y ago
  Sure, and I agree that LW is doing philosophy in that sense.
  - margalabargala 1y ago
    Sure, I just object to the characterisation of "actually correct", as though each of those ideas has not gone back and forth on philosophers thinking that particular idea is "actually correct" for centuries. LW does not appear to have much if any novel insight; just much better marketing.
    - FeepingCreature 1y ago
      I think philosophy has gone back and forth for so long that they're now, as a field, pathologically afraid of actually committing. The AI connection with LessWrong means that the whole thing is framed with a backdrop of "how would you actually construct a mind?" That means you can't just chew on the questions, you have to actually commit to an answer and run the risk of being wrong. This teaches you two things: 1. How to figure what you actually believe the answer is, and why, and make sure that this is the best answer that you can give; 2. how to keep moving when you notice that you made a misstep in part 1.
margalabargala 1y ago
Additional note: compatibilism is only obviously correct if you accept that "free will" actually just means "the experienced perception/illusion of free will" as described by Schopenhauer. Using a slightly different definition of free will, suddenly Compatibilism becomes obviously incorrect. And now it's been reduced to quibbling over definitions, thereby reinventing much of the history of philosophy.
- FeepingCreature 1y ago
  I think free will as philosophically used is inherently self-defeating and one of the largest black marks on the entire field, to be fair.
  - margalabargala 1y ago
    Why is that? Here's what we know: - we appear to experience what we call free will from our own perspective. This isn't strong evidence obviously. - we are aware that we live in a world full of predictable mechanisms of varying levels of complexity, as well as fundamentally unpredictable mechanisms like quantum mechanics. - we know we are currently unable to fully model our experience and predict next steps. - we know that we don't know whether consciousness as an emergent property of our brains is fully rooted in predictable mechanisms or has some decree of unknowability to it. So really "do we have free will" is a question that relies on the nature of consciousness.
    - FeepingCreature 1y ago
      No, I disagree with this conclusion. The problem is very much solvable if one simply keeps the map/territory split in mind, and for every thing asks himself, "am I perceiving reality or am I perceiving a property of my brain?" That is, we "experience free will" - this is to say, our brain reports to us that it evaluated multiple possible behaviors and chose one. However, this does not indicate that multiple behaviors were physically possible, it only indicates that multiple behaviors were cognitively evaluated. In fact, because any deciding algorithm has to evaluate a behavior list or even a behavior tree, there is no reason at all to expect this to have any connection to physical properties of the world, such as quantum mechanics. (The relevant LessWrong sequence is "How An Algorithm Feels From Inside" https://www.lesswrong.com/posts/yA4gF5KrboK2m2Xu7/how-an-algorithm-feels-from-inside https://www.lesswrong.com/posts/yA4gF5KrboK2m2Xu7/how-an-alg... which has nothing to do with free will, but does make very salient the idea that perceptions may be facts about your cognition as easily, if not more easily, as facts about reality.) And when you have that view in mind, you can ask: "wait, why would the brain be sensitive to quantum physics? It seems to be a system extremely poorly suited to doing macroscopic quantum calculations." Once the alternative theory of "free will is a perception of your cognitive algorithm" is salient, you will notice that the entire free-will debate will begin to feel more and more pointless, until eventually you no longer understand why people think this is a big deal at all, and then it all feels rather silly.
TheAncientGeek 1y ago
>actually, this is obviously correct Nobody know what's actually correct, because you have to solve epistemology first, and you have to solve epistemology to solve epistemology...etc.etc. >And it turns out if you do this, you can discard 90% of philosophy as historical detritus Nope. For instance , many of the issues Kant raised are still live. >The massive advantage of the Sequences is they have justified and well-defended confidence Nope. That would entail answering objections , which EY doesn't stoop to. >Compatibilism is really obviously correct Nope. It depends on a semantic issue , what free will means.

margalabargala 1y ago

> And it turns out if you do this, you can discard 90% of philosophy as historical detritus This is just the story of the history of philosophy. Going back hundreds of years. See Kant and Hegel for notable examples.

FeepingCreature 1y ago
Sure, and I agree that LW is doing philosophy in that sense.
- margalabargala 1y ago
  Sure, I just object to the characterisation of "actually correct", as though each of those ideas has not gone back and forth on philosophers thinking that particular idea is "actually correct" for centuries. LW does not appear to have much if any novel insight; just much better marketing.
  - FeepingCreature 1y ago
    I think philosophy has gone back and forth for so long that they're now, as a field, pathologically afraid of actually committing. The AI connection with LessWrong means that the whole thing is framed with a backdrop of "how would you actually construct a mind?" That means you can't just chew on the questions, you have to actually commit to an answer and run the risk of being wrong. This teaches you two things: 1. How to figure what you actually believe the answer is, and why, and make sure that this is the best answer that you can give; 2. how to keep moving when you notice that you made a misstep in part 1.

margalabargala 1y ago

Additional note: compatibilism is only obviously correct if you accept that "free will" actually just means "the experienced perception/illusion of free will" as described by Schopenhauer. Using a slightly different definition of free will, suddenly Compatibilism becomes obviously incorrect. And now it's been reduced to quibbling over definitions, thereby reinventing much of the history of philosophy.

FeepingCreature 1y ago
I think free will as philosophically used is inherently self-defeating and one of the largest black marks on the entire field, to be fair.
- margalabargala 1y ago
  Why is that? Here's what we know: - we appear to experience what we call free will from our own perspective. This isn't strong evidence obviously. - we are aware that we live in a world full of predictable mechanisms of varying levels of complexity, as well as fundamentally unpredictable mechanisms like quantum mechanics. - we know we are currently unable to fully model our experience and predict next steps. - we know that we don't know whether consciousness as an emergent property of our brains is fully rooted in predictable mechanisms or has some decree of unknowability to it. So really "do we have free will" is a question that relies on the nature of consciousness.
  - FeepingCreature 1y ago
    No, I disagree with this conclusion. The problem is very much solvable if one simply keeps the map/territory split in mind, and for every thing asks himself, "am I perceiving reality or am I perceiving a property of my brain?" That is, we "experience free will" - this is to say, our brain reports to us that it evaluated multiple possible behaviors and chose one. However, this does not indicate that multiple behaviors were physically possible, it only indicates that multiple behaviors were cognitively evaluated. In fact, because any deciding algorithm has to evaluate a behavior list or even a behavior tree, there is no reason at all to expect this to have any connection to physical properties of the world, such as quantum mechanics. (The relevant LessWrong sequence is "How An Algorithm Feels From Inside" https://www.lesswrong.com/posts/yA4gF5KrboK2m2Xu7/how-an-algorithm-feels-from-inside https://www.lesswrong.com/posts/yA4gF5KrboK2m2Xu7/how-an-alg... which has nothing to do with free will, but does make very salient the idea that perceptions may be facts about your cognition as easily, if not more easily, as facts about reality.) And when you have that view in mind, you can ask: "wait, why would the brain be sensitive to quantum physics? It seems to be a system extremely poorly suited to doing macroscopic quantum calculations." Once the alternative theory of "free will is a perception of your cognitive algorithm" is salient, you will notice that the entire free-will debate will begin to feel more and more pointless, until eventually you no longer understand why people think this is a big deal at all, and then it all feels rather silly.

TheAncientGeek 1y ago

>actually, this is obviously correct Nobody know what's actually correct, because you have to solve epistemology first, and you have to solve epistemology to solve epistemology...etc.etc. >And it turns out if you do this, you can discard 90% of philosophy as historical detritus Nope. For instance , many of the issues Kant raised are still live. >The massive advantage of the Sequences is they have justified and well-defended confidence Nope. That would entail answering objections , which EY doesn't stoop to. >Compatibilism is really obviously correct Nope. It depends on a semantic issue , what free will means.