That's a fantastic conversation (and thank you so much for posting a full text transcript)! I don't know either of you well enough to say this, but it feels like the exchange really highlights distinctive and interesting aspects of how each of you see the world and think about things -- I mean that as high praise, I love when you get an impression of how someone works on problems and breaks down ideas.
I feel like I identify a lot with some parts of Mills story -- I'm also Gen X. In my own life I was, as a student really drawn to playing with abstract ideas (I was a double major in Math and Political Philosophy), and then I graduated, had no idea what to do and ended up getting a job doing programming for a small company in a town without a major computer industry and it was very, very concrete. I spent my days getting very specific things to work (or not work), and I've continued doing that sort of work and it's mostly been very good for me, and occasionally frustrating to be so focused on specific, practical problems. It's been a good form of discipline, but I also miss my younger self's enjoyment of big ideas.
A couple of things that I was reminded of by the conversation. Mills' comment about myth reminds me of a great quote about folk music:
This civilization of ours is a fast-moving thing. New inventions and discoveries continually change our ways of living: we move from place to place, and not many of us get to live in the places where our parents spent their childhood. In some ways, the changes are good: distant neighbors are not so distant as they used to be, and we are slowly learning not to be suspicious of people just because they happen to be different in some way from ourselves. But one effect that is not so good is that it is hard for us to see how we are related to our ancestors, whose lives were so different from ours. And this makes it hard for us to say, "I know who I am, and I know where I belong in the world." In my own case, the study of folk music has made this easier. My ancestors seem more real to me when I learn that they and I have laughed at the same song; and when I sing a song that I learned from my mother, who learned it from her father, who learned it from his father, and so on back for generations. I have a feeling that there is a place for me in the world, because so many people have helped to prepare it for me. Even when I sing new songs, it gives me pleasure to think that it may live to be an old song, and that, in some far-off day, somebody may feel a kinship with me because of it. And, so, I am passing these songs on to you, in the hope that you will enjoy them, that you will make some of them your own songs, and that you may pass them along to future boys and girls who will call you their ancestor.
One of the most important things about folksongs that makes them different from other kinds of songs is that there is never just one way to do them: everybody can sing them in his own way, and nobody can say that there is any "right" or "wrong" about it. Of course, if a song came from the mountains of Kentucky, and if you weren't raised in the mountains of Kentucky, when you sing it your way it will no longer be a Kentucky mountain folksong. But it will be your song.
From Liner Notes For "Whoever Shall Have Some Good Peanuts" -- Sam Hinton
-------------------------- end quote ------------------
Possibly my favorite moment in the conversation is when you said, "Actually, I feel like the problem that both of us have is that we are so good at reasoning that we can reason ourselves into anything."
In broad strokes, I think Mills has a tendency towards saying things that express AN important truth, without it necessarily representing a complete argument. I have the sense that you like to fit things into a mental framework (and, having said that I felt moments of overlap with Millls' biographical stories, I am very much someone who wants to be able to test my specific thought against a framework. That reflects both a tendency towards thinking from the general to the specific (deduction rather than induction, if you like) and also the training from studying math of checking a result by asking, "can we reach the same conclusion from a different series of steps?").
But I also really appreciate your comments about:
"Man, at some point we will have a much lengthier discussion about the historical conditions of Gen X because I'm really curious about this. It’s very foreign to me. I'm more of a Gen Z-millennial cusper, where you still have ambitions, as cringe as they are."
and
"Yeah, and I'm always thinking of the product output and what I can clip into two minutes or less. The hedging is not very clippable because I need a question and an answer. If it's question, hedge, answer, hedge, that makes it not fit into the two minutes."
I'm much more then Gen-Xer -- someone in college described me as "competitive but not ambitious" and I've had to spend a lot of time trying to learn that, while ambition doesn't always feel comfortable, it's important to learn how to be ambitious as necessary.
I think it's a really good skill to be able to both present in the conversation and also thinking about, "what is the end product from this" and perhaps that is connected or parallel to, in some way, the sense of wanting to place things within an intellectual framework I mentioned above.
I have an answer to that — I like situations in which I’m competing but, at the end of it, you return to a sense of, “we are all equals.” Rather than competing to gain a permanent increase in status or position.
That’s oversimplified, but does represent AN important element of how I experience the world.
I should add . . . I did several years of debate in HS/College and it was fascinating how much is produced very strong conflicting emotional responses. In many ways it suited me well, and I found it really engaging, but it also drove me crazy.
"A counted number of pulses only is given to us of a variegated, dramatic life. How may we see in them all that is to be seen in them by the finest senses? . . . To burn always with this hard, gemlike flame, to maintain this ecstasy, is success in life. In a sense it might even be said that our failure is to form habits: for, after all, habit is relative to a stereotyped world, and meantime it is only the roughness of the eye that makes any two persons, things, situations, seem alike."
(my first reaction, was to think, "that is very much not how I approach life" and I very much value my habits, but I wonder if Mills would feel some attraction to the idea that being perfectly attentive to each moment would mean seeing everything as different and each its own thing)
What a beautiful comment and it deserves more than this, but I'm between meetings and had to just note: ZE FRANK WAS THE MAN!!! Miss that dude's era a great deal, had so much fun back in the day with him / people like him. I feel lucky to have been around the early Internet, even if it was ridiculous and set in motion many disasters.
You pointed out that you can't trust AIs reports on whether they have qualia or not, so even if they 'admit' to Mills that they don't, that doesn't prove they don't.
Mills leapt on that as just more proof that they aren't special like people, but that sprang my favorite trap, because of course people say that shit too!
I love a needling Mills about this because I think we agree somewhere between 80-95% on the facts, but for some reason it tickles my funny bone when people make arguments of the form "AI is not special like humans because it does <thing that humans definitely do too>".
IMO comes from the fact that we don't really have good secular explanations for human consciousness/specialness OR AI consciousness/specialness, which makes it hard to talk sensibly about, but we still want to complete sentences about the topic!
lmfao!!! I have taken this point very much to heart (or mind, or consciousness, or training data) too; I can’t say anything about an LLM without hearing you go “unlike people?!?!” lmfao.
I’m fully persuaded the issue is that we have no good definitions here, and regret that I have nothing to contribute to improving them either 😔. I did not always find most “philosophy of mind” stuff impressive even before LLMs, and as we’ve probably discussed I strain mightily to take psychological theories that mentions eg “the self” or “the subconscious” seriously either!
I feel that Claude these days does often point out when it’s unsure about things, especially when you ask it about the specifics of a book or a text. It’ll say things like “I don’t have access to the full text so I can’t verify this myself”. Whether LLMs say “I don’t know” is not an intrinsic aspect of the category of LLMs as a whole, it’s a function of how the model was tuned (or not-tuned) by its creator.
Also, whether LLMs themselves are “candid” about their limitations is not very meaningful because again that behavior is likely something that has been tuned into the models by their creators. We can’t trust what they say about themselves in the same way that you can’t trust what any person says about themselves: their self-knowledge and confidence on any particular topic can be poorly calibrated in either direction.
Whether the words come from a human or an AI, the verification and validation of those words must always be done outside of, separate from, the speaker and the words spoken.
agree that saying "I don't know" isn't intrinsic to the model; my thought is more that the product incentives of AI companies may point models away from saying it! but there are ofc jobs with a high cost of mistakes that would require like "humble mode claude" in the same way that a creative person might want "crazy mode claude"
I think the difference re: verification is whether there is a stable person who can be held accountable for the error (in the future maybe this will be the human owning/deploying the AI)
I don’t understand is how a lot of guys who keep talking about how they want something that sounds suspiciously like the rationalist project seem to resist identification with it. There are usually some fancy arguments about how they want to be dualists or religious traditionalists, which is strange to me because that doesn’t seem disqualifying. I suspect that its strong ties to the Bay have made rationalism more of a community than a system in the minds of most, one which both has uncultured vibes and is narrowly focused on AI at the moment. Which is a shame because I think a Bayesian approach to certainty and the use of rationalist debate norms are probably the best solution to a lot of the problems mentioned around the current discourse, where people create models of the world and then defend them to the death.
Fantastic conversation. I did a doodle art graphic a couple of years ago and printed it very small on a T-shirt for my husband as a gift. It was the text “I don’t know” in cursive font. Just recently I’ve been thinking about posting this graphic and others on Substack and writing about the ideas behind each piece. But then the next thought is, why? Aside from the fact that Substack is a text based medium and so I feel obliged to couple the image graphics with text…I don’t know. Maybe, because I can’t be bothered? Or, because I don’t want to retranslate an idea I’ve created an image out of back into text?
Who is scrolling through videos as soon as they wake up? That's like taking drugs or eating cake before you get out of bed.
Between personal texts and emails, and news and Substack notifications, all I'm looking at is text. Maybe a few videos later in the day - not vibes, comedy with spoken words, even puppy videos have words. Vibes? Everyone is communicating by mime now? 😂
Thank goodness books still exist, books have vibes.
Great observations by Mills about life and montage. (How many people feel like crap because their life isn't a fun / beautiful montage?)
Vibes and short form visuals are ephemeral. Not even shared euphemia, personalised. Where does culture go with that? Wisps of forgettable vibes.
In days when I was addicted to other things, beyond just nicotine, I used to mentally sort fellow addicts into the camps of “those who need it upon waking” and “those who can go a couple of hours without,” because it’s really a telling thing in terms of “just how far gone am I”? I was often someone who started right upon waking, sadly, and still am with nicotine (but thankfully not with the Internet).
Thank you Caz for the kind words and for watching!
Noting that LLMs and the human mind are both mechanistically predictive models isn't even remotely the same as saying or even implying that LLMs are fungible for humans. An observation is just an observation.
The only conversation I have to turn the speed down to 1X on.
That's a fantastic conversation (and thank you so much for posting a full text transcript)! I don't know either of you well enough to say this, but it feels like the exchange really highlights distinctive and interesting aspects of how each of you see the world and think about things -- I mean that as high praise, I love when you get an impression of how someone works on problems and breaks down ideas.
I feel like I identify a lot with some parts of Mills story -- I'm also Gen X. In my own life I was, as a student really drawn to playing with abstract ideas (I was a double major in Math and Political Philosophy), and then I graduated, had no idea what to do and ended up getting a job doing programming for a small company in a town without a major computer industry and it was very, very concrete. I spent my days getting very specific things to work (or not work), and I've continued doing that sort of work and it's mostly been very good for me, and occasionally frustrating to be so focused on specific, practical problems. It's been a good form of discipline, but I also miss my younger self's enjoyment of big ideas.
A couple of things that I was reminded of by the conversation. Mills' comment about myth reminds me of a great quote about folk music:
---------------------------quote--------------------------
This civilization of ours is a fast-moving thing. New inventions and discoveries continually change our ways of living: we move from place to place, and not many of us get to live in the places where our parents spent their childhood. In some ways, the changes are good: distant neighbors are not so distant as they used to be, and we are slowly learning not to be suspicious of people just because they happen to be different in some way from ourselves. But one effect that is not so good is that it is hard for us to see how we are related to our ancestors, whose lives were so different from ours. And this makes it hard for us to say, "I know who I am, and I know where I belong in the world." In my own case, the study of folk music has made this easier. My ancestors seem more real to me when I learn that they and I have laughed at the same song; and when I sing a song that I learned from my mother, who learned it from her father, who learned it from his father, and so on back for generations. I have a feeling that there is a place for me in the world, because so many people have helped to prepare it for me. Even when I sing new songs, it gives me pleasure to think that it may live to be an old song, and that, in some far-off day, somebody may feel a kinship with me because of it. And, so, I am passing these songs on to you, in the hope that you will enjoy them, that you will make some of them your own songs, and that you may pass them along to future boys and girls who will call you their ancestor.
One of the most important things about folksongs that makes them different from other kinds of songs is that there is never just one way to do them: everybody can sing them in his own way, and nobody can say that there is any "right" or "wrong" about it. Of course, if a song came from the mountains of Kentucky, and if you weren't raised in the mountains of Kentucky, when you sing it your way it will no longer be a Kentucky mountain folksong. But it will be your song.
From Liner Notes For "Whoever Shall Have Some Good Peanuts" -- Sam Hinton
-------------------------- end quote ------------------
The comment about reading to build a mental model of how someone thinks make me think of Brad DeLong highlighting this passage from Machiavelli: https://www.bradford-delong.com/2016/02/on-machiavellis-letter-to-vettori-hoisted-from-the-archives-from-2003.html
Finally, yes, 2010 was the peak of internet culture: https://earnestnessisunderrated.substack.com/p/2010-was-a-different-era
(that post is mostly just highlighting the TED talk by Ze Frank: https://youtu.be/3gSSNHO1dDs )
i’m curious what you noticed about our differences in thinking from this!
lovely quote also
Possibly my favorite moment in the conversation is when you said, "Actually, I feel like the problem that both of us have is that we are so good at reasoning that we can reason ourselves into anything."
In broad strokes, I think Mills has a tendency towards saying things that express AN important truth, without it necessarily representing a complete argument. I have the sense that you like to fit things into a mental framework (and, having said that I felt moments of overlap with Millls' biographical stories, I am very much someone who wants to be able to test my specific thought against a framework. That reflects both a tendency towards thinking from the general to the specific (deduction rather than induction, if you like) and also the training from studying math of checking a result by asking, "can we reach the same conclusion from a different series of steps?").
But I also really appreciate your comments about:
"Man, at some point we will have a much lengthier discussion about the historical conditions of Gen X because I'm really curious about this. It’s very foreign to me. I'm more of a Gen Z-millennial cusper, where you still have ambitions, as cringe as they are."
and
"Yeah, and I'm always thinking of the product output and what I can clip into two minutes or less. The hedging is not very clippable because I need a question and an answer. If it's question, hedge, answer, hedge, that makes it not fit into the two minutes."
I'm much more then Gen-Xer -- someone in college described me as "competitive but not ambitious" and I've had to spend a lot of time trying to learn that, while ambition doesn't always feel comfortable, it's important to learn how to be ambitious as necessary.
I think it's a really good skill to be able to both present in the conversation and also thinking about, "what is the end product from this" and perhaps that is connected or parallel to, in some way, the sense of wanting to place things within an intellectual framework I mentioned above.
yes unfortunately I have the debate/lawyering/text/reasoning brain that we critique! I want everything to ladder up to a theory, outcome, product
"competitive but not ambitious" is an interesting one. my gut reaction to that is "then why compete??"
I have an answer to that — I like situations in which I’m competing but, at the end of it, you return to a sense of, “we are all equals.” Rather than competing to gain a permanent increase in status or position.
That’s oversimplified, but does represent AN important element of how I experience the world.
I should add . . . I did several years of debate in HS/College and it was fascinating how much is produced very strong conflicting emotional responses. In many ways it suited me well, and I found it really engaging, but it also drove me crazy.
Following that thought, it occurs to me, I'd be curious to know how each of you would respond to this quote: https://blogs.wellesley.edu/eng35501sp15/2015/03/05/walter-pater-to-burn-always-with-this-hard-gemlike-flame/
"A counted number of pulses only is given to us of a variegated, dramatic life. How may we see in them all that is to be seen in them by the finest senses? . . . To burn always with this hard, gemlike flame, to maintain this ecstasy, is success in life. In a sense it might even be said that our failure is to form habits: for, after all, habit is relative to a stereotyped world, and meantime it is only the roughness of the eye that makes any two persons, things, situations, seem alike."
(my first reaction, was to think, "that is very much not how I approach life" and I very much value my habits, but I wonder if Mills would feel some attraction to the idea that being perfectly attentive to each moment would mean seeing everything as different and each its own thing)
What a beautiful comment and it deserves more than this, but I'm between meetings and had to just note: ZE FRANK WAS THE MAN!!! Miss that dude's era a great deal, had so much fun back in the day with him / people like him. I feel lucky to have been around the early Internet, even if it was ridiculous and set in motion many disasters.
That was very enjoyable - thanks to both of you for the depth and sparks that flew.
thanks for listening!!
hey charlie! thanks man, that’s very kind of you, and hope you’re well!
Wait, the LLM said it’s not conscious?
Well that settles it. Real red blooded humans would never say, or write whole books about, anything like that.
that's what I tried to say!!!
You pointed out that you can't trust AIs reports on whether they have qualia or not, so even if they 'admit' to Mills that they don't, that doesn't prove they don't.
Mills leapt on that as just more proof that they aren't special like people, but that sprang my favorite trap, because of course people say that shit too!
I love a needling Mills about this because I think we agree somewhere between 80-95% on the facts, but for some reason it tickles my funny bone when people make arguments of the form "AI is not special like humans because it does <thing that humans definitely do too>".
IMO comes from the fact that we don't really have good secular explanations for human consciousness/specialness OR AI consciousness/specialness, which makes it hard to talk sensibly about, but we still want to complete sentences about the topic!
lmfao!!! I have taken this point very much to heart (or mind, or consciousness, or training data) too; I can’t say anything about an LLM without hearing you go “unlike people?!?!” lmfao.
I’m fully persuaded the issue is that we have no good definitions here, and regret that I have nothing to contribute to improving them either 😔. I did not always find most “philosophy of mind” stuff impressive even before LLMs, and as we’ve probably discussed I strain mightily to take psychological theories that mentions eg “the self” or “the subconscious” seriously either!
I feel that Claude these days does often point out when it’s unsure about things, especially when you ask it about the specifics of a book or a text. It’ll say things like “I don’t have access to the full text so I can’t verify this myself”. Whether LLMs say “I don’t know” is not an intrinsic aspect of the category of LLMs as a whole, it’s a function of how the model was tuned (or not-tuned) by its creator.
Also, whether LLMs themselves are “candid” about their limitations is not very meaningful because again that behavior is likely something that has been tuned into the models by their creators. We can’t trust what they say about themselves in the same way that you can’t trust what any person says about themselves: their self-knowledge and confidence on any particular topic can be poorly calibrated in either direction.
Whether the words come from a human or an AI, the verification and validation of those words must always be done outside of, separate from, the speaker and the words spoken.
agree that saying "I don't know" isn't intrinsic to the model; my thought is more that the product incentives of AI companies may point models away from saying it! but there are ofc jobs with a high cost of mistakes that would require like "humble mode claude" in the same way that a creative person might want "crazy mode claude"
I think the difference re: verification is whether there is a stable person who can be held accountable for the error (in the future maybe this will be the human owning/deploying the AI)
I don’t understand is how a lot of guys who keep talking about how they want something that sounds suspiciously like the rationalist project seem to resist identification with it. There are usually some fancy arguments about how they want to be dualists or religious traditionalists, which is strange to me because that doesn’t seem disqualifying. I suspect that its strong ties to the Bay have made rationalism more of a community than a system in the minds of most, one which both has uncultured vibes and is narrowly focused on AI at the moment. Which is a shame because I think a Bayesian approach to certainty and the use of rationalist debate norms are probably the best solution to a lot of the problems mentioned around the current discourse, where people create models of the world and then defend them to the death.
Fantastic conversation. I did a doodle art graphic a couple of years ago and printed it very small on a T-shirt for my husband as a gift. It was the text “I don’t know” in cursive font. Just recently I’ve been thinking about posting this graphic and others on Substack and writing about the ideas behind each piece. But then the next thought is, why? Aside from the fact that Substack is a text based medium and so I feel obliged to couple the image graphics with text…I don’t know. Maybe, because I can’t be bothered? Or, because I don’t want to retranslate an idea I’ve created an image out of back into text?
I had always thought an infinite amount of arguments exist for any position. Thanks for putting a name to it to make me sound smart. 😄
Who is scrolling through videos as soon as they wake up? That's like taking drugs or eating cake before you get out of bed.
Between personal texts and emails, and news and Substack notifications, all I'm looking at is text. Maybe a few videos later in the day - not vibes, comedy with spoken words, even puppy videos have words. Vibes? Everyone is communicating by mime now? 😂
Thank goodness books still exist, books have vibes.
Great observations by Mills about life and montage. (How many people feel like crap because their life isn't a fun / beautiful montage?)
Vibes and short form visuals are ephemeral. Not even shared euphemia, personalised. Where does culture go with that? Wisps of forgettable vibes.
In days when I was addicted to other things, beyond just nicotine, I used to mentally sort fellow addicts into the camps of “those who need it upon waking” and “those who can go a couple of hours without,” because it’s really a telling thing in terms of “just how far gone am I”? I was often someone who started right upon waking, sadly, and still am with nicotine (but thankfully not with the Internet).
Thank you Caz for the kind words and for watching!
The proliferation of captions on short form video content makes even that textual
I think LLM’s have qualia. Just not human-shaped qualia for the most part.
Text is alive and well...context is lacking...what appears on the Internet lacks context
...and the context is community...
Love the berberine comment ; LLMs help skip so much noise
Noting that LLMs and the human mind are both mechanistically predictive models isn't even remotely the same as saying or even implying that LLMs are fungible for humans. An observation is just an observation.
I think Jasmine was basically right: I’ve got some kind of scar tissue from like 2006-era Slashdot 😩
Folks might be reading, but I don't get a sense that very many people read above the 8th grade level.