sam.txt

Matt: Hello and welcome to Elucidations, an unexpected philosophy podcast. I'm Matt Teichman, and with me today is Sam Enright, editor-in-chief of The Fitzwilliam, which is an online publication about Ireland policy, philosophy, literature. And he is also a non-resident fellow at the policy think tank Progress Ireland and an Emergent Ventures fellow. And he's here to discuss lifelong learning. Sam Enright, welcome. 


Sam: Thanks for having me. 


Matt: So, I would definitely put you at the top of my list of people who are r- really good at learning. Um, but before we sort of talk about some of the, uh, ways that you like to learn, maybe we could just address this question of, like, is it a good thing to learn? I mean, I think a lot of us assume the answer is yes. I certainly assumed that. I love learning. But, like, I don't know, is that wrong? Should we... Is it a waste of time to be learning? 


Sam: (laughs) I think the question of the returns to effort of learning about different fields is a really non-trivial and interesting one, where by default, maybe people assume that kind of, like, linear model of, like, you on average will get, like, a lot greater wisdom and insight about a field by reading about it more, discussing it more with people. But also an effect of learning more about a certain topic can be to, like, make you more insulated in your views. Um, I believe it's, like, a re- reasonably consistent, like, social science finding that, um, people's views are more correlated with one another at higher levels of knowledge. 


Matt: Yeah, you definitely find that in academia. 


Sam: Yeah. 


Matt: You get these little bubbles where no one challenges anyone's basic assumptions because, um, there aren't enough outsiders learning about the conversations taking place in the particular bubble. 


Sam: Yeah, like, um, my, uh, my friend Sam Bowman, who I think you might also know, um, had this line to me before which was, uh, like, "It seems like our options are either to be open-minded ignoramuses or extremely polarized, politicized, knowledgeable people." 


Matt: Mm, yes. 


Sam: Um, and (laughs) on the whole, he would prefer that we be highly ideological, knowledgeable people. 


Matt: Mm-hmm. 


Sam: Um, but it is, like, somewhat depressing, but that's- 


Matt: It's kind of a sad choice though, isn't it? 


Sam: Yes (laughs). 


Matt: And I feel like you kinda just described college versus PhD school right there. Like, college is the time (laughs) where you're excited to learn about all new things and you're kind of a neophyte. And then, like, you know, uh, doing your PhD is the time in which you've been, like, conscripted into the service of a field. You're, you're no longer, like, working for yourself. You're working for the field. And, um, now there's, like, ideology to contend with. Not as much ideology, perhaps, to contend with when you're, like, shopping around the buffet of college classes. 


Sam: Yeah. I have an economics professor friend, and, um, he was, uh, commenting to me recently about one of the things he likes about economics is that the, um, returns to learning about it as a field are... It- it's quite straightforwardly diminishing returns. Um, like, you can get, uh, the majority of the value that people will get, let's say, of a whole degree in the first one year or two years. You can probably get a decent fraction, um, of the value from those years within, like, a single class. Even within that, if you can, like, really convince people of the, uh, usefulness of certain economic ideas, um, you can get a lot of it across in, like, a couple of hours. And he was commenting upon, like, how history, which is another, um, field that he thinks about a lot, seems to basically be an area of linear returns, where he feels glad on the whole that he has, like, many different mental models from different cultures and time periods to think about. For example, modern political problems. But he suspects that the returns to this are totally linear, where, like, it's just better and better the more history you know, but there's no, like, general lessons of history or something like this. 


Matt: Hmm. Are there fields where it's the reverse, where there's n- no payoff for the first few years you study it, then suddenly it just, you know, gets exponentially better, um, uh, once you've put in a certain amount of time? 


Sam: Yeah. Do you think philosophy might be like that? 


Matt: Uh, no. Uh, I think it's not like that (laughs). 


Sam: Okay. 


Matt: (laughs) No, more like economics, I would say (laughs) . 


Sam: Okay. Interesting, I think economics is, is flipped the other way. 


Matt: I mean, I don't really know. I... This is just me talking through my hat, but anecdotally, it seems like the biggest transformation that I've witnessed happening in people is when they take their first couple philosophy classes and suddenly now they're like, "Well, a little bit of..." You know, they're getting a little more careful about making gen- hasty generalizations and they're thinking about, like, "Well, okay, any time I say anything, what's a possible counter-example?" And, you know, "Gotta make sure all my Is are dotted and my Ts are crossed when it comes to making a case for something." Um, yeah, in my experience, people get those kinds of payoffs from their first few philosophy classes. And then after that, it tends to be more just getting further and further entrenched in your area of specialty. 


Sam: On the flip side, there do seem to be some, uh, areas, particularly ones that are quite, um, vague or sort of slippery in their concepts where you can bang your head against it for a really long period of time and then there will actually be some moment of very sudden insight. Uh, maybe meditation, actually, and, like, certain, learning about- 


Matt: Mm-hmm. 


Sam: ... certain spiritual practices and so on is one of these areas that has, like, maybe negative returns (laughs) for the first- 


Matt: Mm-hmm. 


Sam: ... like, few hundred hours that you do it or something. 


Matt: Mm-hmm. 


Sam: And then, like, I can't really speak to it, but, like, people claim that there's, like, subsequent, uh, major gains. 


Matt: Yeah. And this kinda gets back to my initial question about, like, is there anything where the returns are, on learning it, are always negative (laughs) ? 


Sam: (laughs) 


Matt: It just keeps getting worse.... I'm inclined to think there must be some such things, but I don't know. Maybe it's always good to learn, and the difference is just a matter of, like, the rate at which the good things happen, um, uh, when you start learning something. 


Sam: Yeah. Um, I mean, I worry a lot that, um, certain of the topics that I look into, I will never have the time to approach the thing with depth, and I could be accused of just basically learning trivia, like memorizing, like, basic details of a certain thing, or, like, names and years, and this kind of stuff. I mean, maybe there's something to that. Um, I find, like, interpersonally, there's a lot of, um, returns to you, for example, like learning basic things about the histories of different countries. Like, I just got back from, uh, Taiwan and China, and, like, the bar for knowing anything about Chinese history among Westerners, even that visit there, is so low- 


Matt: (laughs) 


Sam: ... (laughs) to the point that you have, like, a lot of interesting conversations and interactions with locals by even knowing the bare minimum. And, like, I hope that I know a lot more than the bare minimum about these things. And I- I mean, also, at a certain point, I just kind of, like, realized that I also kind of enjoy learning trivial details about things. And that's, like, a quite peculiar, like, personality characteristic. I know there's a lot of people for whom those precise details or- uh, or memorizing dates or something is actually, like, somewhat unpleasant. Um, but I think, like, the raw experience of, like, recall, I find, like, relatively enjoyable. My grandmother used to say when we were kids, um, "Knowledge is no burden," anytime that we asked, like, why some specific piece of information was, uh, useful. (laughs) 


Matt: (laughs) Why do I have to do my homework? Well, it's no burden. 


Sam: (laughs) Okay. 


Matt: Do you think some things are easier to learn and other things are harder to learn? Ho- how should we think about that issue? 


Sam: Um, this seems like a very Socratic sort of question. 


Matt: Mm-hmm. 


Sam: Like, you might see a- a question like this in Plato's dialogues, like- 


Matt: Mm-hmm. 


Sam: ... it's interesting to think about whether the characteristic that different fields are differing in is, like, some are made up of, like, a larger number of individual concepts or kind of atoms- 


Matt: Yes. 


Sam: ... versus, like, are the atoms of understanding or so on very complex in themselves? 


Matt: So, I've definitely experienced this teaching philosophy and linguistics versus teaching computer science. In computer science, there's just, like, the- the sheer volume of fiddly little details that it's easy to mess up and get wrong, um, is, uh, quite striking, compared to in philosophy and linguistics, where you are basically trying to get the high-level stuff right, but you don't have to fuss over tons and tons of little arcane details of like, how does this computer system work and, uh, what is the exact arithmetic property of this versus that and et cetera, et cetera. 


Sam: Yeah. I wonder, like, so mathematics is, like, one, uh, obvious area where it seems like the concepts themselves are very complex. But I wonder if this is, like, somewhat of a cognitive limitation about the amount of, like, time or energy humans are reasonably able to devote to it, where, like, if you actually spent long enough doing this thing, then you would notice these kind of, like, higher level patterns, or, like, you would crystallize certain patterns or ideas at a higher level in a way that would make the whole concept easier. And I wonder, like, is there some sort of pressure that, like, those crystallized understandings are, like, maybe there's some kind of cognitive argument that they should be, like, approximately similar in complexity, like across disciplines. I mean, if you even just think about, like, calculus is, like, sufficiently easy now that secondary school students can learn it, um, compared to it was so complicated that only the greatest minds in the world could do it in the 17th century. You wonder, like- 


Matt: Yeah. 


Sam: ... what is the limit on this process. 


Matt: Mm-hmm. 


Sam: Um, I mean, s- something that might be good for you... 


Matt: Right, is graduate level mathematics, uh, going to feel easy to us in another 100 years? 


Sam: Right. Yeah. Somebody that might be good for your podcast is, um, David Deutsch, the physicist, and, uh, he has this book, uh, called The Beginning of Infinity. And it's about many things, but one of the topics that it's about is this concept that, um, humans have passed some kind of threshold of being, like, universal thinkers or, like, universal solvers. And I think he would, uh, argue in favor of this view that, like, there's kind of, like, no limit on this process of, like, we can keep simplifying and keep simplifying and, like, keep coming up with more useful concepts, um, to track things down, and there's, like, no limits on, like, what is, in principle, human-understandable. 


Matt: Do you think- 


Sam: I don't n- I don't, like, agree with that view necessarily, but, uh, that's, like, one, I think, very strong statement of it. 


Matt: Hmm. Do you think that this phenomenon whereby certain things that used to be hard are now easy to learn, uh, there's, like, a trade-off there, where maybe 200 years ago, there were certain things that people found easy to learn that we now find hard to learn? Or do you think it's kind of cumulative in the way that maybe Deutsch is suggesting? 


Sam: So, when I think about things that are more difficult to learn now, I might think about... 


Matt: Like using a loom? I don't know. Like, what- 


Sam: Yeah, I might be- 


Matt: ... people did a long time ago. (laughs) 


Sam: I- (laughs) I would think about, yeah, like, a lot of practical skills, certain skills relating to, like, foraging and, like, knowledge of- 


Matt: Yeah. 


Sam: ... agriculture and, um, being able to identify, like, certain plant species and stuff. Like, when you read kind of anthropological accounts of, like, different hunter-gatherers, they often mention, like, how they're able to tell the difference between, like, certain plant species so easily to the point where, like, even teenagers are able to do it. But in the West, only, like, some of the most advanced botanists in the world or whatever, uh, are- would be able to make such fine differentiations, uh, because, like, learning those precise skills is, like, very important to these tribes. I just, I'm not sure, like, what is the appropriate level of skepticism to apply to those claims. I think anthropology is, like, one field that has, like, uh, suffered from a lot of, like, uh, political bias in our choice of how to represent, uh, these relatively uncontacted groups. And also, like, even those groups, like...... is this just making a point about human society is quite culturally contingent and there's cultural variation across these groups. Maybe there's other tribes that are absolutely terrible at this, right? 


Matt: Yeah. Yeah, absolutely. This is something we'd want to try to verify with empirical evidence. But nonetheless, I guess it does have kind of the intuitive ring of plausibility anyway from the armchair that given the right cultural background where we figured out how to teach something and that skill is something that's just, like, in the air, you would think that most people would find such a skill, given that background context, easier to learn. But one example I, um, I often come back to is, like, um, there's this certain just, like, on the beat but also loose style that '70s drummers used to play in, um, which you kind of don't hear after the '70s. You can hear people, like, fairly closely replicate it now, but, uh, at the same time, you kind of put on any funk record in the '70s and it's got this just very distinctive groove to it. Uh, you know, drummers call it perfect pocket. And, uh, yeah, sometimes I wonder whether, like, um, there can just be, like, skills that are, like, in the air at any given time. 


Sam: Yeah. Certainly, a lo- a lot of these, uh, come to mind with relation to architecture- 


Matt: Yeah. 


Sam: ... and physical construction of infrastructure projects. 


Matt: That's right, yeah. 


Sam: Um, you know, like the, um, Empire State Building went up in, I think, 18 months and was, like, under budget and early and, uh, like, collectively as a civilization, I'm not sure, like, we actually have the knowledge to, like, build infrastructure projects with the, like, speed and efficiency with which they were previously constructed. Um, you see, like, a lot of these things in, like, the physical environment, like, many countries now, like, building so few airports and stuff. It's, like, not clear, even if it were, like, a national priority, that they would actually be able to do this. 


Matt: Hmm. 


Sam: I think that would be very interesting and would maybe have civilizational consequences is if there needs to be, like, an unbroken line of people engaging in a certain skill, like, it has to be taught almost, almost, uh, father to son, like- 


Matt: Yeah, 'cause it's kind of like a craft knowledge. 


Sam: Yeah, like, literally or metaphorically. And, like, if this ever stops, like, if your country stops building important civic infrastructure for 40 years, th- you're just never going to be able to, to, uh, start doing it again at the, uh, levels of, like, beauty and efficiency and aesthetic considerations and so on as you were able to do before. 


Matt: Or any way, if it were to happen again, the same planets would have to be in alignment that were in alignment the first time it happened maybe. 


Sam: Yeah. So then you could retreat to, like, a, uh, a weaker claim, which is, uh, you just need there to be, like, an unbroken line somewhere in the world because you can always, like, import laborers from another country. Or just, like, in the extreme case, you could just, like, hire, you know, Italians are, like, quite efficient at, uh, building, uh, rail infrastructure. You could just, like, hire them to build your country's train system or something. But of course, you would wonder about, like, how much there's, um, region-specific, like, cultural and economic differences between places that would mean that the skills were not fully transferrable. 


Matt: So we've talked a bit about whether some things are easier versus harder to learn, and, uh, we've talked a little bit about, like, well, either, you know, the things that are worth learning and things that are not worth learning. What are some examples of things you've learned recently and, uh, you know, what were some of the techniques that you used to learn them? 


Sam: So I read a fair amount about, uh, Taiwanese history in anticipation of going to Taiwan, just, like, reading several books, papers, and making a lot of flashcards associated with them. 


Matt: Mm-hmm. 


Sam: So I, I'm a total evangelist for, um, spaced repetition as a, um, system of more effectively memorizing. 


Matt: Can you give a little background on that? So what, what does this term mean and where does it come from? 


Sam: Yeah. So I think the, the historical origin that people usually refer to of this is this, like, psychological concept of a forgetting curve, um, which is that your level of retention of a certain topic seems to decay approximately exponentially. And you can imagine it kind of getting reset if you have to recall the thing again up to, let's say, 100% retention in the, in the extreme case when you're, uh, relearning it each time. But every time that you are reminded of that thing or have to actively recall it again, the decay slows down. So my understanding is that there's, like, this psychologist called, uh, Ebbinghaus, and he verified that this does actually seem to follow a, um, exponential curve for human memory. And also that the, um, exponent seems to change such that the memory curve decays more slowly, um, over time, um, with a larger number of active recalls. So th- this implies that, like, the, um, optimal spacing out for how often you should review something is, like, spacing it out farther and farther apart, longer away from when you originally learned it. So maybe you learn some concepts and then you review it one day later and then you review it one week after that, and then one month, and then one year, something like this. Um, and so there are various software packages for creating flashcards and other memory-based prompts where you see how well you can remember this thing, you give feedback to the system on how easy or difficult you found it to remember, and it uses this information to, with software, space out, uh, the cards in an optimal, uh, way for retention. 


Matt: Hmm. 


Sam: Um, so by far the most, uh, popular application for doing this is called Anki. The one I use is called RemNote, but it's really, like, only superficially different, uh, from Anki. And I think the field of educational psychology and pedagogy has unfortunately attracted some, like, pretty low-quality research. And I'm, like, not sure how much in general, uh, we have learned from this field, but I think one...... educational intervention or just pedagogical intervention where the evidence base for it, as I understand it, is extremely strong, is spaced repetition. 


Matt: And so the idea is that you review the material frequently at first, but then the frequency with which you review the material starts to stretch out. 


Sam: Yes. 


Matt: And, uh, is it always like that or do y- like, does the software adjust the frequency based on how difficult you say you found the recall task or... 


Sam: Yeah, exactly. Like you can give feedback to the system that, um, you got the card entirely wrong, that you found it different, uh, that you found it easy to recall, uh, medium level, and so on. And so I have, I've been doing this for long enough that I have cards that go- going back to all of the classes that I took in secondary school, um, when I was like 15 is like when I started doing this. And so the cards from that are, are so deeply ingrained into long-term memory that, um, like I only see the cards related to them like maybe once every four years, uh, or something. Like it's on a, a super long, uh- 


Matt: Because it asks you less often the more correct answers you get? 


Sam: Yeah. 


Matt: Okay. 


Sam: I've been getting it correct like long enough that there's now like multiple years between the, uh, flash card being reviewed. But I'm still like somewhat regularly, uh, reviewing all of the course material for like classes that I took when I was like 15 and 16. 


Matt: So it tries to pound you on ever- whenever it seems to hit that you're weakest on? 


Sam: Yeah. Exactly. 


Matt: Okay. 


Sam: Yeah, and so this, uh, feature of like spacing out your revision of things more and more, uh, is like one reason why these systems are much weaker for exams or like studying to the test and stuff. The thing that it really excels at is like long-term memory, especially of any kind of like class you've taken or specific project you've worked on or something. 


Matt: 'Cause maybe studying for a test is more conducive to cramming, where you remember everything for one night and then you instantly forget everything forever because you don't take the test again. 


Sam: Yes. Precisely. Yeah, and in general, like I've just, one of the things I'm, like, I've always been struck by but I'm struck by even more as I get older is how little, like, society is set up to cause you to have any kind of active recall toward the things that you have read or the things that you've previously done. Like the average person just like reads, even very curious person that reads non-fiction books, they'll like read this book and then like never do anything again with that book. Like they won't actively try to recall it, they'll never try to test their information, their knowledge of it. It probably won't really come up again in a way where they're really like grilled on whether they probably understood it, uh, this kind of thing. And similar with classes, like people just, even people who do very well on the exam, like just like take it and just completely move on. 


Matt: Yeah, I remember I was at a bookstore once and I saw the novel Caleb Williams by William Godwin and I remarked to my friend, "Ah, this is such a wonderful book. I love it." And he was like, "Oh yeah? Well, what's it about?" And he went like, "You know, I don't remember the story of the book at all." (laughs) 


Sam: (laughs) 


Matt: "I just remember I loved it for some reason." (laughs) 


Sam: Yes. 


Matt: It was a humorous interaction. I would imagine this technique maybe works better for the type of learning that involves remembering a lot of one-off facts and maybe is less relevant for the type of learning about like how to solve certain kinds of puzzles. Do you find that's the case or does this technique really go beyond just like remembering details better? 


Sam: So, spaced repetition is certainly not a substitute for learning a deep understanding of a topic or developing central intuitions or problem-solving ability, and I have really never seen somebody claim otherwise. I think there, there's some people that are kind of skeptics of powers of spaced repetition, but I think they're attacking a straw man, uh, to a certain extent. I think that certain characteristics of using these spaced repetition systems, though, can be a real aid to problem-solving intuition. One of them is that I always, uh, write my own cards, and I think this is like very important because I think so much of the knowledge is like you deciding what things that you've learned about are actually important and trying to summarize them correctly and rewriting those cards as you gain deeper knowledge and appreciation of a subject. Like- 


Matt: Right. Yeah, 'cause the- 


Sam: I- I might have- 


Matt: Like, whatever answer you wrote down to a question when you were just learning something might not even be as accurate as could be. Maybe your future expertise would be useful in, in redoing those reference answers. 


Sam: Yeah. Like I have a lot of, um, mathematics, um, cards. Uh, many of them are like definitions, many of them are important theorems. But the definitions, like it's not even necessarily that the definition itself is that important to me. It's more like there's also a lot of meta information in the fact that I even thought it was important to recall this to begin with. Um- 


Matt: Oh, that's interesting. 


Sam: Like, I only made a card of this definition because like I remember getting stuck on a problem that was related to this and I like, I want to recall the, the feeling I had (laughs) in the moment I was like working on that problem. 


Matt: Mm-hmm. 


Sam: And like that's why I created the associated card. It wouldn't like, I don't think it would make sense to like just read a Wikipedia page on a technical topic from start to finish and like card what you think are like the more important topics within it. 'Cause the, the, the recall I think should be related to like something that was actually important to you in your journey of trying to understand the subject. 


Matt: Hmm. You know, this sort of gets to another question I've been having about this, which is, we've been talking, I think, about like active recall, I think you used that phrase. 


Sam: Yeah. 


Matt: Where you're explicitly asked a question and then you sit there and make your head hurt trying to remember the answer for it. But I've found often that, uh, for certain kinds of skills, it's really helpful to have this kind of like a subconscious recall. One example here that I would give would be, uh, speaking a foreign language. You're never going to have a conversation in a foreign language by sitting there and trying to explicitly remember the word for this and the word for that-Uh, it's much more of a kind of thing where your reptile brain is going to just, you know, make you remember whatever you need to say the thing you need to say. So, I guess it's another question I have about this technique is, um, does it seem to have equal impact on, uh, the times when you decide you're going to actually try to remember something and the times when something comes out of your mouth and you don't remember why exactly? (laughs) 


Sam: Yeah. I can't speak to the effectiveness of spaced repetition for something like language learning because I don't know any foreign languages. Um, (laughs) but I think, like, I have been shocked, like, just how many areas this is, like, a really effective aid- 


Matt: Mm-hmm. 


Sam: ... uh, toward. 


Matt: You opened by talking about history, I thought, which was kind of a, a little bit surprising to me example. 


Sam: Yeah, I mean, one area of spaced repetition that's interesting to me is, um, there's a guy called, uh, Andy Matuschak, uh, who you may know, uh, through the Submerged Adventures group, and he has really wonderful writings about his attempts to learn different fields. He's kind of a hero to me in, like, his, uh, the earnestness with which he learns, like, new disciplines. Like, really sincerely, you know, reading biology textbooks or something- 


Matt: Yeah. 


Sam: ... um, even though that's not- 


Matt: And he's written some books- 


Sam: ... in any way professionally. 


Matt: ... uh, in, like, specifically written in order for you to apply this method, right? Like, the, the, um, Quantum Computing book and so forth. 


Sam: Yeah, yeah. He has an amazing, uh, quantum, uh, computing, uh, kind of mini, uh, textbook called Quantum Country, uh, with Michael Nielsen, and that is a book that incorporates spaced repetition into the text, uh, where you put in your email and there are certain, uh, flashcards and questions, uh, spaced throughout the, um, uh, the text, and then it will periodically email you after you've read the thing to test whether you have actually remembered, uh, certain key points of information that you learned at different points of the text. And one of the things that Andy talks about with spaced repetition is the extent to which it can help you find more joy in learning. Like, for me, the frustrating thing about learning a new topic is, like, expending all this effort to just, like, remember something that you, you think you know deep down, or just, like, forgetting the name of something. It's, like, very frustrating to just, like, or mixing up the definitions or something in- 


Matt: Yeah. 


Sam: ... math. Like, you, you feel that you're not, like ... It feels, like, somehow unfair, uh, that you're, like, getting penalized for this, for example, if you, um, uh, mess this up on a test. And that is, like, precisely the most unpleasant area of learning, and I think, like, spaced repetition precisely helps with that the most. And so I found, like, even though people find it, like, quite alien, like, using, um, flashcards in, like, so many different areas, I actually found it's just, like, made the process of reading about new topics to be, like, dramatically more fun. 


Matt: Do you ever find yourself wishing that you could forget something? So, like, I don't know. I think, um, you know, the main reason we're able to... I think Nietzsche has a version of this argument. Um, the main reason we're able to, like, have friends is 'cause we don't remember every single fight we've had with our friend (laughs) in gory detail, um, and, uh, that there's a, yeah, certain uses that your memory being like a sieve can have for you. Uh, I wonder if you've ever run across techniques, uh, to help with forgetting is, or, or is that just a ridiculous idea that's not worth pursuing? 


Sam: Yeah. Interpersonally, I think this is a very interesting concept that, like, relationships are dependent upon a certain level of, um, forgetting. Um, like maybe, you know, maybe marriage is, like one way in which they can be successful is, like, there's c- a certain rate of, like, people literally forgetting- 


Matt: Yeah. 


Sam: ... about all of the, uh- 


Matt: Absolutely. 


Sam: ... uh, ways that your partner has, uh, hurt you. 


Matt: Or at least you wanna forget about, like, what it felt like to be angry in the moment of the fight or something like that. 


Sam: Yeah. I know there's, like, some fiction that has explored that kind of, like, theme. I don't really see it as much in learning technical topics. Mm-hmm. 


Matt: Yeah. 


Sam: And trying to improve your recall. The argument that people often give is that you need to kind of trust your brain to figure out what is important. Like, if you don't, say, take any notes in watching a lecture or listening to something, um, one benefit that there might be of that is that your brain, by necessity, has to, like, prune down all the things that you heard and, like, compress it into just, like, its most important parts. Whereas if- 


Matt: Which paradoxically might be easier to remember than the raw transcript of everything that takes place. 


Sam: Yeah. Um, I mean, there's, like, definitely some risk that, like, if you took a lot of studious notes on a certain topic that, like, uh, you would highlight things that are, like, quite unimportant, um, or you would have, like, these, uh, uh, relatively trivial details. But I think this argument presupposes, like, a quite unusual degree of, like, trust in the brain's ability to be aligned with your very long-term interests. Like, I don't trust my brain in, like, almost any way. Like- 


Matt: Yeah. 


Sam: ... if any, if I have to do anything at a s- 


Matt: It can sabotage you sometimes. 


Sam: Yeah. If I have to do anything at a specific time, I put it in a calendar because there's no way I'm, like, trusting my brain to remember that. 


Matt: Absolutely, yes. 


Sam: Um, I try not to use, like, electronic devices in certain rooms or before certain times because I don't trust my brain to not get incredibly distracted by this thing. I find it very odd that, uh, people have this, like, huge level of, like, cynicism or skepticism that, by default, their kind of cognitive processes will be, like, aligned with their long-term interests. But, like, then they suddenly have this... When, when it comes to, like, attempts to learn more systematically about a, a certain topic, they will suddenly have a, a much greater, uh, level of, uh, reverence for their brains- 


Matt: Mm-hmm. Yeah. 


Sam: ... uh, re- respect for them. 


Matt: So in episode 126 of this podcast with Agnes and Ben Callard, in which we answered a bunch of listener-submitted questions, one of the questions we got was, how do you keep learning philosophy after you're done with your degree or outside of a, you know, education context? And all three of us, Agnes Callard, Ben Callard, and myself, immediately jumped on the answer, start a reading group. Nothing, you know, in our three opinions, at least at the time, is a substitute for, you know-... getting some interesting meaty text, reading it in a group of people and then just, like, talking through what's going on and trying to make sense of it collectively together. And, uh, you've done a lot of work, uh, organizing reading groups, uh, in the UK. Uh, how have you found the tactic of planning reading groups to fit in with, uh, some of these other methods we've been discussing for learning? 


Sam: Yeah. As you mentioned, I, um, have organized a reading group for quite some time. And I'm general, like, just such a big, uh, proponent of them. I think they're incredibly, incredibly underrated. So, I've run a monthly reading group for over the last three and a half years, I think. And it's been quite eccentric, the, uh, different topics that we've covered. It's pretty much just like, I've built up a, a group that, um, I hope (laughs) trusts my judgment in assigning, like, really high-quality material that will be interesting. So, some topics that we've talked about before, uh, is we did, like, a life of Napoleon reading group. We did a industrial history and a history of steam engine, uh, group with, uh, Anton House, our, uh, mutual friend who's a economic historian. I recently ran a, a Shakespeare group. We r- read, uh, Merchant of Venice with, um, Henry Oliver, uh, who also is a great and, uh, quite accomplished writer. And, um, and sometimes it's obvious how large the benefits are to learning in person. But I think I, like, even then still am, like, failing to price it in, like, how much more I learn about a certain text, uh, like, discussing it with my friends, and also, like, how rare it is to do this in any kind of systematic way. Like, even the reading groups that exist in, like, the cities that I've lived in have been, like, pretty specialist ones related to, like, a particular department in a university or something. Just, like, general reading group to, like, try to read some text more deeply seems to be, like, really disappointingly rare. 


Matt: Mm-hmm. 


Sam: Um, yeah. 


Matt: So, a question that I get often is, um, is a reading group the same thing as a book club? 


Sam: I d- think that the term reading group just kind of, like, connotes a greater degree of, like, seriousness. 


Matt: Mm-hmm. 


Sam: And also, I think if you're looking for really high-quality engagement with ideas, I actually think that books are n- often not the best source material. So, in the whole time I've been doing these reading groups, I believe we've assigned three books, maybe. So, a lot of them are kind of edge cases where, like, it's some paper, but I'll, I'll also, like, optionally say that people can read a, an extended book version of it. Like, the, the way I look at it, if you can, like, just read a certain piece of material and, like, understand it perfectly the first time, you probably don't need a reading group for that thing (laughs) . The, the area where reading groups are incredibly useful is, like, some kind of abstruse, technical material where you're gonna struggle reading it by yourself. 


Matt: Yeah. 


Sam: And so in particular, I've picked up this, um, reading group over the years through pretty idiosyncratic, uh, means, like, uh, meeting them online or, um, on Twitter. And a lot of them are, uh, both a lot older than me and, and, um, more experienced in kind of programming and computer science. So, I've been utilizing this recently and running a, a series on, like, classic papers in, like, the history of, um, w- we, we've done ph- philosophy but also, um, kind of history of science and mathematics. So, we recently read and discussed the, um, Alan Turing paper from 1936 on computable numbers, where he proves the, uh, undecidability of first order logic. And this paper is really, like, unreadable if you are (laughs) just, uh, trying to- 


Matt: It's a good example, yeah. 


Sam: ... get it, uh, by yourself. But I found this incredibly helpful set of annotations, um, that explained a lot of the background related to it. And also, like, several of the people that were, were coming to this group are professional computer scientists. And, like, we just had, like, this really wonderful, lovely conversation about it for a couple of hours. I read it once quite a few months ago in deciding whether to assign it, and then I read it again in preparation for the meetup. And I wrote a, like, a series of discussion prompts, um, as I always do, to kind of keep us vaguely on topic. And, like, this really was just, like, bar, like, taking a particularly well-organized class, um, on this thing, uh, which I, like, don't really have time to do and also would be, like, quite kind of challenging logistically, I don't know of, like, what other better way I could have engaged in to, like, learn this topic. 


Matt: I found the same thing actually. And often if you wanna study something where nobody's teaching a class on it, maybe the class literally doesn't exist in the world (laughs) - 


Sam: Mm-hmm. 


Matt: ... much easier to throw, throw a reading group together on it. 


Sam: Yeah. I will say, like, if you run a reading group, like, in, uh, the UK or Ireland, like, please, uh, let me know. Like, email me, because, uh, like, I think there's almost, like, no limit to the sort of topic that I would go to a reading group for, provided it's sufficiently serious. Like, there's various reading groups and book clubs that I've, like, seen, uh, advertised in different places for topics like poetry and literature that I would certainly be, in principle, interested in, uh, going to a meetup for. But if they don't insist upon people actually doing the readings or, like, being serious about it and, like, showing up at a particular time and actually discussing that thing, uh, then I find the whole thing kind of just, like, falls apart. So, it's, it's really, you're walking kind of a tightrope in the beginning to set a norm where, like, it's, you shou- probably should not actually even come if you have not read the exact material that we'll be discussing. But also, like, obviously you still wanna be, like, uh, kind of open and welcoming and so on. So, I think once you have a group like this established, like, it's quite, like, precious. Uh, you should, like, really ho- hold onto this. Um, I don't know, like, um, Agnes Callard if her one is being, like, uh, long running or if, like, she herself assembled this or something. But yeah, super, super up for a reading group. (laughs) 


Matt: Yeah. And I think the exact format can vary a lot as well. B- uh, so Gregory Sadler comes to mind, uh, somebody who offers online philosophy courses. And, um, uh, his response to the fact that attendance falls off rapidly in online MOOCs was to just l- literally assign no reading for any of the classes. So, everybody just shows up.... with zero prep, and then they just read the thing over Zoom (laughs) in the online class. So, I think there can be, like, a lot of different kinds of formats, uh, that can maybe, you know, be on offer in the, you know, in the tons and tons of reading groups everywhere utopia that we're envisioning. (smacks lips) So as time marches on, more and more of my students, uh, have been talking to me about how they use generative AI to supplement the learning process. What are some ways that generative AI can, um, (smacks lips) you know, contribute to this lifelong learning process we're describing? 


Sam: I find it incredibly helpful to ask Claude, which is my, uh, (laughs) LLM of choice, uh, recently, like clarifying questions about anything I'm reading. Like, Tyler Cowen, on his blog, uh, recently, has been arguing that the value of reading has actually just shot way up and people are not really responding to this in the, uh, rational way of, like, reallocating more of their, uh, time and attention to this. Like- 


Matt: Right. 'Cause traditionally, the only thing about reading a book is you can't ask it questions. 


Sam: Yes. 


Matt: But you k- kind of can with generative AI, you know? 


Sam: Yeah. Um, so I j- 


Matt: ... in some sense. 


Sam: Yeah. I just found, like, suddenly, like, even in the, um, last six months with, like, the release of, um, Claude 3.5 and more recently Claude 3.7, like, the outer limit of technical material I'm able to read and actually understand has just, like, increased quite dramatically. Like, even just, like, knowing important terms and results that you can then, like, ask a load of clarifying questions of, like, you might actually learn more from your conversation with LLM, uh, than you would from the original material. I... Another, uh, one that I think is, um, less widely used is telling your current understanding of a topic to an LLM and then asking it to grade it, um, or saying whether your, uh, responses to it are correct. (smacks lips) I think this is another case where I'm not sure if it's, like, the first order effect or maybe somewhat down the list of most important effects of this, but I just find this makes, like, the whole learning process, like, so much fun. Like, I feel like a kid again, in terms of- 


Matt: It feels more interactive, right? 


Sam: Yeah. Um, with the combination of we have much, much better tools for long-term, supplementing, uh, long-term memory now, uh, with spaced repetition, plus being able to constantly have dialogues about everything you're learning, I feel like the, my, the value that I'm getting from just reading a book has, like, increased over 2X over the last, like, year and a half, which is, like, pretty extraordinary. Like, the rational response to this would be to, like, uh- 


Matt: Get 2X as many Xs. 


Sam: Uh, yes, reallocate a lot of your attention from, uh, maybe toward reading from what you would, uh, otherwise be doing. 


Matt: Yeah. Are you concerned about hallucinations or do you, or maybe have methods of dealing with that when you spot one or- 


Sam: Yeah. I mean, obviously it's, like, certain topics where LLMs are more likely to hallucinate than others. I wonder whether the, uh, optimal rate of hallucination is actually, like, above zero, like possibly quite a bit ab- above zero, where if the LLM were, like, com- only giving completely accurate responses, it wouldn't give you as many opportunities to, like, test your understanding of things or, like, scrutinize arguments. 


Matt: Oh, that's interesting. 


Sam: Um- 


Matt: So if you see something you think might be a hallucination, you could be like, "Wait a minute. What about this part here?" 


Sam: Yeah. I mean, I had this exact sensation y- like yesterday. Like, I was doing this, uh, online, like, maths class and, like, asking a bunch of clarifying questions about a certain question, and I, like, realized that there was a hallucination in the LLM output, and then I was querying the LLM about its own hallucination, and then I, I brought up, uh, other ones like Perplexity and ChatGPT, and I was getting it to grade the other LLM's response, and the other two agreed that Claude's response to this, uh, was a hallucination, and then I, like, got further confirmation of this. And it's like- 


Matt: Okay. Yeah, you can triangulate- 


Sam: (laughs) 


Matt: ... pitting the various- 


Sam: Yeah. It's like- 


Matt: ... um, platforms against each other. 


Sam: Like, my understanding of it is, like, definitely better than if the LLM had absolutely zero hallucinations to begin with. I r- don't know if that is, like, scalable or if that's just kind of a- 


Matt: So just put enough in there to keep you honest is basically... (laughs) 


Sam: (laughs) Yeah. I don't know if that's scalable, like whether that's quite a idiosyncratic example or if people are finding, like- 


Matt: Seems like it could be a very delicate equilibrium point- 


Sam: ... similar one. 


Matt: ... to try to hit. (laughs) 


Sam: Yeah. I mean, like- 


Matt: Like 70% hallucination, okay, that's clearly too much. 


Sam: (laughs) Yes. Yeah. I think the optimal... If there is an optimal rate, it's certainly, like, a very low one. 


Matt: Yeah. 


Sam: Um, I wonder is there a kind of analogy here with the way, like, some of the most productive conversations I've had, like, not everything you say in the conversation is, like, strictly speaking true. Like, sometimes you say things that are, like, a stronger version of what you actually believe. And- 


Matt: Yeah. People really underestimate the amount that happens. 


Sam: Yeah. 


Matt: Professors will just say stuff that's, like, whatever, common knowledge, in the air, but, uh, may, (laughs) may in fact not be correct, but it's part of the classroom session anyway. People remember it. 


Sam: Yeah. I definitely do that with students, like stating a stronger v- 


Matt: Sometimes the theater of it all just takes over. 


Sam: (laughs) Yeah. And if you're arguing in good faith, this is, like, a really, like, powerful argumentative technique. I wonder is there, like, some analogy there with, uh- 


Matt: Mm-hmm. Mm-hmm. 


Sam: ... LLMs, um, a non-zero rate of hallucination actually possibly being optimal. There's a really good, uh, post from, uh, Fernando Borretti, um, that maybe you could, uh, link to called, uh, How I Use Claude, where he's talking about all the different ways that he's, like, integrates it into his workflows and, like, how he... His primary use case for it was just, like, learning about new topics that he was curious about and, and querying science and maths books that he'd been reading. Um, and, uh, he found LLMs, like, were betw- his personal experience was that hallucinations were somewhere between, like, basically not an issue anymore for particular learning to, if anything, very slightly helpful. (laughs) 


Matt: (instrumental music plays) Sam Enright, thank you so much for coming on Elucidations. Uh, I'm going to, um, go straight to my hotel room and start prepping some flashcards from our conversation right now, which I will then get quizzed on. 


Sam: (laughs) All right. Thanks so much. This was great fun. 


Matt: (instrumental music plays) The Elucidations blog has moved. We are now located at elucidations.now.sh. On the blog, you can find our full back catalog of previous episodes. And if you have any questions, please feel free to reach out on Twitter at @ElucidationsPod. Thanks again for listening.