Reading is Like Pumping Iron

There’s a book I think you should read. Wait! This is not an ad! I bought the book with my own money around 4 years ago, and recent developments have prompted me to write about it because I think more people should read it. The book in question is How to Read a Book by Mortimer Adler and Charles Van Doren.

All around me I see products adding LLM-powered summarization or note-taking features, and I feel this book becoming more important than ever. How to Read a Book lays out a state of affairs where what most of us classify as “reading” is actually a largely superficial and ineffectual activity that does little to improve our understanding of the world. Over the course of the book, Adler and Van Doren explain more fruitful ways of reading that can be employed to better our lives.

Consider that the first edition of this book (written solely by Mortimer Adler, with Charles Van Doren becoming involved in a 1972 revamp) was written in 1940. Clearly, a lack of understanding about reading (or comprehension more generally) is not a new problem. Today, this problem is only becoming more pressing as LLMs summarize, outline, and take notes for us. This has less to do with LLMs specifically and more to do with the fact that these tasks can be delegated at a scale never before seen. If people are not even engaging with text, how can they possibly be getting better at comprehending it?

Since LLMs are such a hot topic right now, let me be clear that I’m not talking about the technology as a whole. I’m only referring to specific uses of LLMs for things like summarization, outlining, note-taking, or “research” (like Claude’s “Deep Research” or NotebookLM).

There is no shortage of things to read, if for no other reason than the internet. But most of what people encounter day to day puts very little demand on them as a reader. Things like text messages, emails, reports, blog posts, and news articles, can all be read more-or-less automatically. Don’t get me wrong, I think that’s perfectly fine. Straightforward and simple communication is best in all kinds of circumstances.

However, anyone who has taken a literature class can attest to the fact that not all writing is trivial. There are works across all genres, from fiction to poetry to technical documents, that are simply not easy to understand.

This difficulty may be due to superficial factors, like unfamiliar spelling, vocabulary, or grammar. As an example, consider Of Plymouth Plantation by William Bradford, which was written in the 17th century with antiquated spelling and a style that’s unfamiliar to modern readers.. However, once you're able to get past the stylistic differences in the text, understanding the actual content is relatively straightforward. The difficulty is essentially mechanical with respect to parsing the different spelling and sentence structure.

In other works, the stylistic challenges are more severe. As one example, Ulysses, by James Joyce, contains a rather notorious 4,391 word run-on sentence in Molly Bloom's soliloquy. Or take The Sound and the Fury, by William Faulkner, which is written in a tricky stream-of-consciousness style that jumps back and forth through time and repeatedly switches narrators (check out this synopsis). In these works, the vocabulary may be familiar, but it may nevertheless be difficult to parse what the text is saying or the point the author is trying to make.

Going a layer deeper, many philosophical or technical works contain ideas that are inherently complex, such that they will always be difficult to understand no matter how plainly they are written down. Textbooks and research papers often fall into this category.

So what? Why does it matter to read stuffy research papers or artsy books? Why shouldn’t we settle for reading that’s easy?

First, developing your ability to work through difficult texts gives you access to a wider range of ideas. A lot of software engineers I’ve met don’t read papers. And while there’s nothing necessarily wrong with that, it also means they’re losing out on a huge wealth of knowledge. If you’re able to parse a research paper, you’re not beholden to someone else to summarize it on Wikipedia or their blog. Furthermore, the quantity and quality of information and argumentation in research papers is often (not always) higher. The skills required for reading research papers also transfer readily to textbooks and other technical references, which are often invaluable when facing particularly difficult problems.

Second, working through difficult texts improves your overall ability to read. Reading, especially intensive reading, is a skill. Without deliberate practice and challenges, your ability to perform that skill will not improve. Grappling with difficult ideas improves your comprehension and intellectual capacity overall. This strengthening of your mind provides returns in everything you read, not just the difficult texts.

If you want to engage with ideas deeply and directly, you have to learn how to interact with and navigate difficult written material, because that’s where most of the ideas live!

The situation is not unlike weightlifting. Assuming general good health, most people could curl a 2.5 pound dumbbell for many dozens of repetitions, if not hundreds. This light weight is like the “trivial” text we most often encounter. It’s the kind of reading that provides essentially no resistance to our minds. Heavier and heavier weights require more effort to lift, until you can barely budge it, if you can move it at all. Heavy weight requires not only the requisite preparation in terms of strength, but also the use of proper technique in order to lift the weight safely and efficiently.

Reading is similar in that everyone has a threshold of difficulty at which they can comprehend and engage with the text, but only with great effort. As you practice engaging with difficult texts, your capacity to handle them increases. Furthermore, a certain amount of technique is required for your time with a book to be productive. One can read a book as an act of literacy and still not gain any understanding from it. Letting the words of a book pass through your eyes and into your brain is only the most basic part of reading.

There is a superficial similarity between all kinds of reading, because they all involve the same process of converting physical symbols into ideas. However, this similarity can easily hide the fact that there are many ways, and indeed more sophisticated ways, to approach reading.

I will speculate that sometimes people encounter a difficult text and begin to read it just like anything they read in day-to-day life. When the text proves challenging, they may unfortunately assume they are not "smart enough" for that particular book, essay, paper, etc. In reality, it has much less to do with some innate talent and much more to do with practicing intensive reading as its own skill. We simply cannot approach difficult texts with the same mindset, techniques, or level of effort that we employ when reading trivial text. Reading a text message is a nearly automatic process. In contrast, reading a technical paper or dense scientific or philosophical work is an active, effortful process that will often leave you feeling drained.

The automatic part of reading is most of what we learn in grade school, and at this level of reading we answer the question “What does this sentence say?” in its most basic form. Adler and Van Doren call this level of reading “Elementary Reading”.

Beyond this foundational level of reading, Adler and Van Doren identify three more levels of reading that build on one another. These additional levels are all classified as “active reading” and are successively more demanding on the reader. All kinds of active reading involve significant effort and external activities like asking questions about the text, or outlining, summarizing, and writing about what you’re reading.

Now, my purpose here is not to regurgitate How to Read a Book. If anything, I encourage you to go get a copy and read it for yourself. Beyond that, my purpose is to point out that this distinction between elementary and active reading exists, and to strongly encourage you to engage in active reading.

I can explain my urgency in advocating for active reading by returning to the topic of LLMs. It is all too easy to conflate the value of the summary with the value of the summarization process. When you yourself read a text, turn it over in your mind, and produce a summary in your own words, you are engaged in learning. The summary itself is of some superficial value. But really it stands as evidence that you engaged in a learning process with the material, and can perhaps help you recall what you learned. In the gym, the movement of the weight in space is not what’s valuable, it’s the effort you expend in moving it. When you summarize or outline a text, the effort you put forth to create that summary or outline is the most valuable thing, not the written artifact itself.

Using an LLM robs you of this opportunity to develop your own mind, and then tricks you into thinking you understand more than you do. Sönke Ahrens, on page 85 of How to Take Smart Notes, writes:

“Reading [Ahrens refers to ‘elementary reading’ here], especially re-reading, can easily fool us into believing that we understand a text.”

And then continues on page 86:

“While it is obvious that familiarity is not understanding, we have no chance of knowing whether we understand something or just believe we understand something until we test ourselves in some form. If we don’t try to verify our understanding during our studies, we will happily enjoy the feeling of getting smarter and more knowledgeable while in reality staying as dumb as we were. This warm feeling disappears quickly when we try to explain what we read in our own words in writing. Suddenly, we see the problem. The attempt to rephrase an argument in our own words confronts us without mercy with all the gaps in our understanding.”

We cannot escape the fact that becoming smarter and wiser requires deliberate hard work. There are better and worse ways to go about improving yourself, but there are no shortcuts. An AI generated summary, outline, or “deep research” report, will trick you into thinking you understand something when you actually haven’t changed at all.

Yet, I see AI note-takers, summarizers, or “researchers”, proliferate more and more. You could argue that this proliferation is evidence of their effectiveness, but effectiveness in what? Certainly it is not effectiveness in elevating human beings, because no one is learning by delegating these tasks to a machine. This is true in just the same way that you will not grow bigger muscles by having someone else work out “for you”.

So if they’re not valuable for promoting understanding, where does the value come from? At the end of the day, these tools are “valuable” at work because corporate culture treats them like they’re valuable. Every time I see someone post Claude or NotebookLM’s “research”, it is another mark of a culture that is fundamentally superficial, and that cares more for appearances than substance. It makes me sad, I wish things were different, and things can be different.

Corporate culture is the product of the people who engage in it, and if those people stop engaging in LLM summarization, it loses its legitimacy. So, I urge you to pursue real learning and real understanding. Remember there are no shortcuts. You are the only one who can develop your own mind. Don’t let anyone take that away from you.