Learning Library

← Back to Library

Reading in the Age of AI

Key Points

  • The rise of AI has sparked worries that knowledge creation is stagnating, but the real issue is that we lack clear methods for reading and learning in an information‑overloaded era.
  • Reading—whether physical books, Kindle articles, or audio content—remains essential, yet our traditional habits were built for a selective information age and must be adapted for today’s flood of data.
  • The speaker proposes three reading frameworks for the AI age: “awareness reading” for quick, surface‑level updates (e.g., news feeds, skim‑reading to pass a test).
  • The second framework, “information retrieval” (also called domain completion), leverages tools like Google or AI assistants to locate and extract specific knowledge on demand, turning the reader into a more efficient, targeted learner.

Sections

Full Transcript

# Reading in the Age of AI **Source:** [https://www.youtube.com/watch?v=1Q8CjX0SP0o](https://www.youtube.com/watch?v=1Q8CjX0SP0o) **Duration:** 00:19:03 ## Summary - The rise of AI has sparked worries that knowledge creation is stagnating, but the real issue is that we lack clear methods for reading and learning in an information‑overloaded era. - Reading—whether physical books, Kindle articles, or audio content—remains essential, yet our traditional habits were built for a selective information age and must be adapted for today’s flood of data. - The speaker proposes three reading frameworks for the AI age: “awareness reading” for quick, surface‑level updates (e.g., news feeds, skim‑reading to pass a test). - The second framework, “information retrieval” (also called domain completion), leverages tools like Google or AI assistants to locate and extract specific knowledge on demand, turning the reader into a more efficient, targeted learner. ## Sections - [00:00:00](https://www.youtube.com/watch?v=1Q8CjX0SP0o&t=0s) **Reading in the Age of AI** - The speaker examines concerns that AI‑generated summaries diminish traditional reading, argues that our reading practices are ill‑defined for today’s information flood, and urges a rethinking of how we consume and retain knowledge. - [00:03:28](https://www.youtube.com/watch?v=1Q8CjX0SP0o&t=208s) **The Cost of Deep Reading** - The speaker explains how deep reading consumes significant brain energy, argues for selective use amid AI-driven retrieval tools, and calls for better filtering mechanisms. - [00:07:10](https://www.youtube.com/watch?v=1Q8CjX0SP0o&t=430s) **Different Reading Strategies for AI Books** - The speaker explains three distinct approaches—retrieval, conneto (deep narrative), and awareness reading—using examples of AI‑focused titles to illustrate when each method is appropriate. - [00:10:18](https://www.youtube.com/watch?v=1Q8CjX0SP0o&t=618s) **AI Overload and Quality Filtering** - The speaker likens the flood of AI‑generated content to a resume glut, arguing that without strong filtering mechanisms we risk drowning in low‑quality information and must curate trusted resources for both beginners and advanced users. - [00:14:08](https://www.youtube.com/watch?v=1Q8CjX0SP0o&t=848s) **Beyond Token Metaphors: Human Insight** - The speaker contends that true innovation stems from deep, iterative reading and reflection—far richer than the token‑based AI model—and believes AI can serve as a catalyst to amplify this human‑driven creativity. - [00:17:23](https://www.youtube.com/watch?v=1Q8CjX0SP0o&t=1043s) **AI-Enhanced Interdisciplinary Learning** - The speaker urges passionate, multi‑domain learners to be discerning readers, use AI as a tool to link diverse fields—while avoiding inflated claims of AI‑generated breakthroughs—and follow curated learning pathways to turn their interests into meaningful contributions. ## Full Transcript
0:00So, this one is by popular request. I've 0:02gotten asked a ton. Nate, what do I 0:05read? Look at the books off. What do I 0:07read in the age of AI? How do I know I'm 0:09reading the right things? It feels like 0:11reading is becoming increasingly 0:12irrelevant. The Wall Street Journal came 0:14out this week with an assessment that 0:16basically suggests we are rotting out 0:18knowledge in the knowledge economy 0:20because AI is repeating knowledge and 0:22regurgitating knowledge and 0:24recalibrating knowledge and there's no 0:26actual new knowledge being produced 0:28because people are just asking for 0:29summaries. I don't think that's actually 0:31true. I think that's panicinducing. But 0:34I do think we need to talk about it and 0:35I want to talk about it in the context 0:37of how we read in the age of AI. And I 0:40think this is a big big deal because 0:42even if you don't read books, right? 0:44Maybe you're on a Kindle, maybe you are 0:46reading articles, you are still 0:48consuming Hunter, maybe you're in this 0:49video. I think if you are listening to 0:51this video, it's like an audible book. 0:53It counts as reading. Maybe you're 0:55you're feeling like that's being 0:56generous, but it's true. It counts as 0:57reading. I want to suggest that the way 1:00we read is not well enough documented. 1:03And that's part of why we're suffering 1:05is we don't know how to be readers in an 1:08age when we're drowning in information 1:10because reading evolved for a selective 1:13information age. We learn to read 1:16because we needed to remember things and 1:18remembering by writing down works pretty 1:20well. And to retrieve it, you have to 1:22read it, right? That was the fundamental 1:23idea. That's why a lot of the initial 1:25records we have for writing are commerce 1:27records. You are trying to read and 1:29understand. Court documents are also 1:30very early examples of writing. You're 1:32trying to read, understand, remember. We 1:34have since evolved and we read for all 1:36kinds of reasons. In the age of AI, I 1:38want to give you three ways that I think 1:41we read and learn. And I want to give 1:44you frameworks for how you think about 1:46this when you have like let's assume you 1:48have an intelligent AI counterpart, a a 1:51buddy, a chat GPT in your pocket that 1:53you're reading with. How do you use 1:56that? Well, so you are not just wasting 1:58your time and information goes in one 2:01ear and it goes out the other ear. So 2:03here are the three frames that we're 2:05going to walk through. Number one, I 2:07think it's the lightest is what I would 2:09call awareness reading. So that's the 2:10idea that you just need to catch up on 2:12the information of the day. This might 2:14be the person who is scrolling X. This 2:16might be the person who is trying to 2:18understand what's in the news. This 2:20might be the person who is just skimming 2:22through a book in the bookstore or maybe 2:24skimming through a book that they're 2:25going to get a test on and they just 2:26need to have enough awareness to pass 2:28the test. Right? That's that lightest 2:30layer. The second layer is what I call 2:32information retrieval. You also hear it 2:34called domain completion. It actually 2:36maps pretty well to a lot of what we do 2:38with Google. I think it's underthought 2:40of that we Google as readers. In a 2:44sense, Google is a form of reading 2:46because we're trying to go out and 2:48retrieve information that has been 2:50written down somewhere. That is the act 2:51of reading. And when we do a lot of our 2:53sort of intermediate level reading, what 2:55we're really doing is we're saying fact 2:57A is in a book. Fact A is on a web page. 3:00I need to go and get it and retrieve it 3:02for a particular purpose. I'm going to 3:03use it. I'm going to repurpose it to 3:06borrow token architecture. I'm going to 3:07stick the token in and I'm going to 3:09produce a new token. Right? That's 3:10retrieval reading. And then three, and 3:13this is the one that people get like 3:14lots of big feelings about. It's what I 3:17there's not been a good name for it. I'm 3:19calling it conneto reading. Other people 3:21call it deep reading. I like to call it 3:23conneto reading because my thinking is 3:25that, you know, our brains are plastic. 3:26Our brains are forming all the time, 3:28evolving all the time, pruning pathways, 3:30etc. We are deepening the conneto in our 3:32brains when we read deeply. And this is 3:35the kind of reading that teachers get 3:37very excited about. This is the kind of 3:39reading that everyone worries is 3:41disappearing, but it's only one part of 3:43this larger reading landscape. And by 3:46all accounts, it's a very expensive form 3:48of reading from an energy perspective. I 3:51was doing some research on sort of 3:52reading and the science of reading, 3:54apparently, if you're doing that kind of 3:56deep reading, you are burning a 3:58tremendous amount of glucose in your 4:00brain because you're trying to process 4:02so much. And so as much as we may try 4:04and lament and sort of put the 4:06responsibility for reading more on 4:07ourselves or on our peers, maybe part of 4:09it is recognizing that this is a very 4:11energetically expensive task and we need 4:13to make the choice to use that reading 4:16skill where it matters the most. And 4:18that is the part that I would argue has 4:20gotten harder in the age of AI. And 4:22that's what I want to address. I think 4:24that we need better tools for filtering 4:27and simplifying reading tasks so that we 4:30can truly build understanding at the 4:32levels we need. We can build awareness 4:33where we need it. We can retrieve 4:34information where we need it. Uh which 4:36also, by the way, I would argue LLMs are 4:38retrieval reading. You're getting 4:40answers there. There's a popular term 4:41now called answer engine optimization 4:44because people needed an answer for 4:45search engine optimization in the age of 4:46AI, so they made it up. But the idea is 4:48the right. You're doing retrieval and 4:50you're coming back with information. And 4:52then there's conneto reading, right? The 4:53the three I want to suggest that there 4:56is a way for us to understand in advance 5:00what kinds of books we need to dig into 5:03in a conneto sense, in a deep reading 5:05sense, what kinds we can just skim, what 5:08kinds we can do some retrieval on. And 5:10it has never been more possible to do 5:14that kind of learning across multiple 5:18book types, multiple retrieval types 5:20using the assistance of AI. Yes, I 5:22unlike the Wall Street Journal, I do not 5:25think that AI is bad for reading. I 5:28think people who passively use AI are 5:31bad readers. And that's a that's a 5:33different thing. So what do I mean here? 5:35Let's go to a few examples. Let's for 5:38example look at AI literacy 5:40fundamentals. The author is Ben Jones 5:42and the pitch is pretty simple, right? 5:44If if you're overwhelmed by AI, this 5:46will help you, right? You can join in 5:47the conversation, etc. It is clearly a 5:49beginner grade book. I would say just 5:52looking at it and this is again I want 5:53to talk this through so you get this 5:55into your head. This is an example of a 5:57book where there are lots of facts that 5:59you want to retrieve and you want to 6:01understand. This is a good example of a 6:03book where you should have a copy and 6:06then you should read with AI as an 6:08information retrieval system. So you 6:10should say this is my current level of 6:13fluency in a prompt. This is the book 6:15that I'm holding and looking at. By the 6:17way, I don't advocate not getting the 6:19book. I think you should get the book 6:20and I want to read only the sections 6:23that I'm missing on. Maybe I want to 6:25read about how token architecture works. 6:28Maybe I want to read about how next 6:29token prediction works. Maybe I just 6:31want to understand how AI conversations 6:34happen in a chatbot. That's the level 6:36I'm at. You can then go in and talk to 6:39the AI and say, "What parts of the book 6:41do I need to read? What elements of this 6:44book speak to that knowledge gap, the 6:46thing I'm trying to learn? And maybe 6:48what other books do you recommend?" And 6:50by the way, I tested this on AI to see 6:53how much hallucination I would get as 6:55far as madeup books. It is remarkably 6:57accurate. I was working with Perplexity 6:59for this. I find that Perplexity is a 7:01great search engine here. I only had one 7:04instance in about 30 books where 7:06Perplexity made up the book entirely. 7:08And I mean, is that great, right? You 7:10have to check it. I caught it. But the 7:13other 29 books were legit and they were 7:14good books and there was a great find. 7:16Saved me a ton of time kind of sort of 7:17pulling the list together and making 7:18sure that I had a fluent understanding 7:20because as much as I read, there's books 7:22coming out on AI all the time and I 7:24wanted that broadstroke view. So I think 7:26AI literacy fundamentals is a good 7:28example of the kind of retrieval motion 7:31I'm talking about where you need to 7:32acquire new facts. Let's talk about 7:35something where I think you have a 7:36different kind of reading. Let's talk 7:37about co-intelligence living and working 7:39with AI. It's a bestseller by Ethan 7:41Mllik. People have heard about it. As 7:43someone who has read that book, that 7:44book is literally over here on my shelf. 7:46I think that benefits more from conneto 7:49reading. I think that's an example of a 7:50book you should read end to end because 7:53the narrative builds over time. You may 7:56not have a paper book. Maybe it's on the 7:57Kindle, but you should read it and you 7:59should let it sink in. The good news is 8:01this particular book is written very 8:03clearly. It's easy to understand and 8:04digest. As much as you're doing conneto 8:06reading, it's not complex and hard to 8:09read. Now, an example of awareness 8:11reading. I actually think a fantastic 8:13example of awareness reading is the 8:16Washington is the I talked about at the 8:18beginning of this video. The one that 8:21talks about knowledge rot. I think you 8:23just need to be aware of it. It's a 8:24great example of something that frankly 8:26perplexity could summarize for you and 8:28just let you know what the state of the 8:29conversation is. It is not something you 8:31need to spend time digging into because 8:34frankly the level of effort and 8:36investment by the authors in a newspaper 8:39article is something that varies a lot 8:42and you have to use your judgment about 8:44whether there's something new and 8:46noteworthy there from an AI perspective 8:48that is going to get you value for the 8:50time you're putting in to read that 8:52article end to end. And I can sort of 8:54hear and feel the journalists that might 8:56be listening to this sort of having 8:57heart palpitations because like that 8:59feels like it disintermediates the the 9:01journalist and and the print experience 9:03a little bit. I love good journalism. I 9:05think there's a place for good 9:06journalism and I've been pretty honest 9:08about the fact that the number of good 9:10articles that deeply understand 9:13artificial intelligence, I can count 9:15them on like one hand. And I think 9:16that's a larger issue that I would love 9:18to see addressed and I'd love to see a 9:20larger journalistic conversation about 9:22better quality newspaper articles, but I 9:24got to be honest about where we are. I 9:26think a lot of the articles that I see 9:28are not well enough researched to 9:31deserve deep reading at this time. And 9:33so I think they go in the awareness 9:34bucket. So when you think about it that 9:36way, my thesis is that this dramatically 9:39simplifies the whole problem. I think if 9:42we think of knowledge rot as a problem 9:45and we obsess over the the the gap the 9:48the so-called rot in the space, we're 9:50focusing on the wrong thing. I think we 9:53should instead focus on reading as a 9:55skill, recognize that reading is a skill 9:58issue that needs to be updated for the 10:00age of AI and that the primary way it 10:02gets updated is by figuring out how to 10:05work with this new co-intelligence to 10:08borrow Ethan's phrase to read well to 10:11read in a way that we remember to save 10:13room for deep reading where it matters. 10:15And I think if we think about it as wow 10:18AI is enabling us to produce so many 10:21more tokens than we had before. 10:23Therefore our filtering problem is 10:24really the issue that simplifies a lot 10:26of this discourse because then it's 10:28really not our fault. Like if you think 10:29about it this is analogous to the job 10:31market situation where as most people 10:34who are in the job market know we are in 10:36a world where recruiters are snowed 10:37under with resumes that have been 10:39published prepared by AI. the it's a 10:42terrible experience for the applicants 10:44because the applicants can't get any 10:45attention because there's so many 10:47perfect resumes now. It's a terrible 10:49experience for the recruiters and for HR 10:51because they can't figure out how to 10:52sift for quality anymore and the whole 10:54system has broken down. In that world, 10:56AI producing tokens has led to a crisis 10:59of sifting and businesses are reading 11:02candidates quote unquote by bringing 11:04them on site by doing anything they can 11:06to get an actual candidate experience. 11:08Some of them are trying AI interviews 11:10etc. In the same way, we need to get 11:12aggressive about sifting for quality 11:15information. And so, part of what I'm 11:17doing by sort of building this book list 11:19is I want to have a source for quality 11:22AI information. And some of it will be 11:24retrieval and some of it will be 11:25introductory and for beginners and some 11:27of it's going to be for like advanced 11:29users as well. Like one of it, one of 11:31the advanced ones that I absolutely 11:33love, it's from Stripe Press is the art 11:36of doing science and engineering. It is 11:38not directly about AI. It is a fantastic 11:42book. It's by Richard W. Hamming and if 11:45you are anywhere in the technical space, 11:47I would recommend it. It helps you 11:49understand how innovation happens in 11:52technical spaces and I think it's a 11:54highly relevant foundational text. So 11:56there's there's going to be sort of 11:57books like that as well. Please, please, 11:59please do not listen to the people who 12:03tell you that the problem is information 12:06rot per se. Because if you believe that, 12:09you are going to get into a position 12:10where you think there's nothing you can 12:12do about it. The problem is not 12:14information rot per se. The problem is 12:17that we need new skills in a world where 12:20we have artificial intelligence right 12:22next to us all the time. I think one of 12:24the key skills is learning when to read 12:26for retrieval, when to read for 12:28awareness, when to read deeply for 12:29conneto reading. The other key skill is 12:32going to be learning when and how to put 12:36out what we actually think into the 12:38world. I'll give you an example. This is 12:40a work of passion for me. I care about 12:43reading. Are you surprised? Look at 12:45these books. I care passionately about 12:47reading and supporting readers, 12:48supporting next generation readers. I 12:50see the anecdotes coming out of the 12:52education system where people are 12:54struggling to get students engaged on 12:56reading where people in high school and 12:58above are saying that students are 12:59phoning it in and just using AI to sort 13:01of put the essays in and the educational 13:03system that was predicated on the idea 13:06that people needed to write essays that 13:08people needed to demonstrate their 13:10understanding by producing words is just 13:11not working anymore. I think that the 13:13way back the way back is reigniting 13:16passion for deeply understanding a 13:18subject. Reigniting curiosity, 13:20understanding that there is too much 13:22information to process and that wrote 13:24assignments were never that effective 13:26anyway. And that we need to challenge 13:28people to grapple deeply with things to 13:30get them to read. And if you want to 13:32learn about AI, your best asset, 13:35ironically, despite this being about how 13:37you use AI to read better, your best 13:39asset is still your curiosity. Your best 13:41asset is still your passion for 13:44artificial intelligence and the subject 13:46matter. That matters. That matters a 13:48lot. And I want to remind you of that 13:52because that is empowering. If you are 13:54curious, if you are passionate about 13:56something, not only will you read 13:58better, not only will you learn better, 14:00you're going to find your voice. You're 14:02going to find something you want to say 14:04into the world. Maybe it's about AI. 14:06Maybe it's about science and engineering 14:08like Richard Hamming did. Maybe it's 14:10about something else entirely. But that 14:12voice, that desire to speak back, in my 14:15experience, only comes when you are 14:17reading enough to prime the pumps. When 14:20you are thinking and interacting with 14:22the world and taking in information and 14:24meditating on it deeply and letting it 14:25reprocess inside you. I don't think the 14:28token architecture metaphor works very 14:30well for people. I know we use it a lot 14:31because that's the predominant tech hype 14:33cycle of our age. What we do with 14:36information is deeper than that. The way 14:37we reprocess, the way we mix it, it's 14:40not something that is easily 14:41translatable in machine terms. And that 14:43magical compost pile is where the best 14:47authors have always gotten their stuff. 14:49It is where people who come up with 14:51ideas to build a business get their 14:53stuff. Except for them, they're reading 14:54customers. They are reading the market. 14:57They are reading pain points and they're 14:58thinking about it and obsessing over it 15:00and and studying it. We need that kind 15:02of passion in order to put something 15:05back in, in order to have the skill to 15:09fight the knowledge rot, to have the 15:10skill to say something new. I don't 15:13think that that's hopeless. In fact, I 15:15have never been more bullish on it 15:17because I think that properly used AI 15:19can help us to say new things. I'm going 15:21to give you an example here. I am as a 15:25project as a as a brain stretching 15:27project working on learning Swedish. I'm 15:30not good at it. Please, no one in my 15:32comments come for me and say, "Tell me 15:33about your Swedish skills. They're 15:34terrible. I'm learning." But I have a 15:36background not just in English, but also 15:39in Indonesian. There is nothing in the 15:42literature on learning English or 15:44learning Swedish if you know Indonesian 15:47already. I could not find anything. And 15:49so I worked with GPT5 Pro and I put 15:52together a way of thinking about Swedish 15:55grammar that makes sense to someone who 15:57knows Indonesian and a way sort of to 15:59walk into learning Swedish. As far as I 16:02know, that hasn't been done before. I'm 16:04sure that could be done by someone with 16:06a linguistics degree who knows Swedish 16:09and Indonesian. I don't pretend that 16:10that's something that's like completely 16:11impossible for people, but it hasn't 16:13been done. And GP25 Pro enabled me, a 16:16humble person who is not a linguist, to 16:18just go and do it. That is an example of 16:21passion working with an AI to put 16:24something out that I can understand and 16:26actually like gain a skill set on. It is 16:28using AI to be active. And so I want to 16:31challenge you. Don't be afraid to work 16:33with AI to build new and active things. 16:36It is worth trying. It is worth taking a 16:39prompt and saying, "Hey, can you please 16:41help me with this?" And I will say the 16:43advanced reasoning models do a lot 16:44better. In particular, if you are if you 16:46were sort of inspired by this and you're 16:48like, what what is the takeaway here? I 16:49don't do Swedish, but is there something 16:51I can learn? Think about it as GPT5 and 16:54other advanced reasoning models are very 16:56very good at stitching between domains. 17:00And so if you have existing knowledge 17:02domains, you will have cracks in those 17:04knowledge domains. I described one 17:05between Indonesian and Swedish 17:07linguistics. No one had bridged that 17:09before. It knows all the domains well 17:11enough that it can establish a bridge or 17:14at least the start of a bridge, a 17:16scaffold of a bridge between those 17:17domains that ends up being a new 17:19connector in the human knowledge set. 17:21That's really exciting. And so what you 17:23should do is if you're passionate about 17:24something and you cover multiple 17:26domains, be a good reader, work with AI, 17:29see if you can't connect those domains 17:32together in a way that's interesting and 17:33novel. This is not the same thing as the 17:36people working with AI and claiming they 17:38are coming up with novel physics when 17:39they are not physicists. That is a very 17:42different thing. I would say that is 17:43trying to say that the reasoning models 17:46can advance the edges of human domain 17:48experience beyond what PhD physicists 17:51can do. That is at best unproven at 17:54best. And there are articles sort of 17:56from physicists explaining why that's 17:58challenging. You have similar things 17:59with mathematics. I think physics just 18:01gets more attention because people watch 18:02Star Trek a lot. So, where does that 18:05leave us? Please be a careful reader. 18:07Please filter what you want to read. 18:09Don't blame yourself if you feel 18:11overwhelmed. Figure out what your 18:13pathway is for reading and jump into it. 18:15I'm going to sort of put something up on 18:17the Substack with like learning pathways 18:19for different goals in AI. But really, 18:21it's not about AI. It is about AI, but 18:23like use it to learn anything in the age 18:25when you have a lot of information to 18:26choose from. And then when you're 18:28passionate about something, think about 18:30ways that you can actually put your 18:31voice back out there. And maybe it's AI 18:33assisted like I talked about with GPT5 18:36Pro and Indonesian and Swedish. And 18:38maybe it's not. Maybe you have the voice 18:39inside and you already know. An example 18:41of that one, as far as I know, I came up 18:43with a sort of retrieval awareness uh 18:46deep reading thing. I don't know if 18:47other people have done it. It certainly 18:48didn't come from AI. It came from me 18:50arguing with AI and saying, "You're 18:52wrong. I think this is the way we should 18:53categorize it." Out of my own head 18:55because I had passion for it. Because I 18:57care about reading. care about reading. 18:59Learn. Put something out there.