How I'm Using Chat-GPT as my Personal Oxford Book Tutor

Using gen AI to test me on books, prompt active recall, and force me to think more deeply about what I'm reading.

Have you ever read a classic book and wondered if you understood everything?

Do you worry that students at university are getting more out of their reading experiences than you are?

What about a crippling anxiety that you’re not retaining the material — that if someone were to ask you what it was about or to explain a key concept, you’d flail around like a fool?

If so, you’re not alone.

Doubt constantly plagues the autodidact.

More than any other force, it’s doubt that causes the self-learner to give up on their learning goals — “Maybe this is why college exists — maybe I can’t do it on my own…”

In my book-centric self-education, I’ve had my fair share of this doubt:

• Did I truly understand the themes and literary techniques used in Wuthering Heights?

• Is a philosophy student at Oxford thinking more deeply about Beyond Good and Evil than I am?

• Am I going to remember the major events in this history book on the American Civil War?

Over the years, I’ve tried various methods to simulate a rigorous college reading experience to boost retention, enforce critical thinking, and deepen my understanding of the books I read.

But these methods haven’t lived up to my expectations, and if they have, it’s been incredibly hard to stick to them. (Who has the discipline to design your own tests, take them, and then grade them? No one.)

3 strategies I’ve tried to boost my reading retention and understanding (and why they aren’t meeting my needs).

I’ve tried reading Sparknotes and other commentary.

But that doesn’t prompt me to do the hard work of thinking. I’m just reading someone else’s perspective. And there’s no active recall involved. So their words rarely stick.

I’ve tried taking walks after reading and reviewing what I read.

I still do this, but there’s no other force aside from my own mind pushing me to think outside of my comfort zone, and no one testing my comprehension with complex, nuanced questions.

I’ve even designed tests for myself.

The problem here is that there’s no one to tell me which answers are wrong or right. Plus, doing this takes forever. Hard to stick to as an autodidact with a job and other responsibilities.

What if there was another way?

Before I go into my solution, I need to introduce you to a pedagogical method in higher education, known as the tutorial system, which strikes me as an incredibly effective way to learn from books. Then, I’ll show you how I’m using Chat-GPT to responsibly approximate this educational experience.

The tutorial system: aka having intellectually serious conversations about books

At Oxford and Cambridge University, students often learn via the tutorial system.

They meet frequently with a teaching assistant (a subject matter expert) to discuss the study materials — in the humanities, that often means the assigned readings.

So, after reading the first part of Shakespeare’s Hamlet, you might meet with your tutor for a conversation about it.

The goal of the conversation is to promote critical thinking and deeper engagement with the text, while making sure that your understanding is on the right track.

The tutor may do the following to generate better learning outcomes:

• Ask the student questions about each section/chapter - thus forcing them to do active recall, a science-backed technique for memorizing material.

• Tell them when their answers are on or off point - so they get actual feedback on their ideas and interpretations, and can spot and mend holes in their understanding.

• Guide you through a back-and-forth conversation that forces them to think critically about the structure, themes, ideas, authorial intention, content, and more.

By going through this process, students not only gain a deeper understanding of the books they read. They also become better at communicating their perspective on the reading clearly and persuasively. They practice rhetoric, a key stage of learning in classical education.

“Wouldn’t that be nice?” I thought. “A well-read expert on the book who would challenge my thinking and prevent me from getting nothing more than a surface-level understanding of great works of literature, history, and politics.”

Unfortunately, the tutorial system is out of reach. I cannot afford to go back to school. Undergraduate life is behind me.

So I forgot about this dream learning experience entirely, and went on self-studying in my usual way, all the while feeling like I wasn’t going deep enough into these books.

That is, until I came across an article about how Oxford is teaching its students to use Chat-GPT to study more effectively.

Oxford’s advice for using generative AI to support studying

In the guide for learning with gen AI, Oxford gives five ways to use generative AI to support academic reading.

Here’s the piece of advice that caught my attention:

“Ask AI to generate thought-provoking questions based on article content. You can develop your own understanding of an article by answering the questions asked. You could also use the questions to develop your own questions in relation to the article to deepen your learning.”

At first, I scoffed. AI invading humanities studies? No way in hell.

But, as I thought more about it, I started to have an idea.

What I wanted was a personal Oxford tutor who has read all the books and lives in my closet at my beck and call to talk with me about the books I’m studying.

Unfortunately, I can’t have that.

What I can have, however, is an AI-powered tutor that, while not as good as a real human scholar, is still shockingly useful for prompting a reader to reflect, think, and answer hard questions about books.

And so, I tested it out.

The results were better than I expected. And most importantly, I had a lot of fun doing it (this means I’ll likely stick to it, unlike creating test questions for myself and grading myself, which is not fun).

Using Chat-GPT to tutor me on the book Anti-Intellectualism in American Life

I had just finished part 1 of Richard Hofstadter's Anti-Intellectualism in American Life, and I wanted to test myself on what I’d read so far.

What I wanted to do was discuss the book with a trained scholar/teacher who would ask me questions to prompt me to think deeply and test my understanding.

But all I had was an LLM, so I settled, and it was much better than nothing.

I signed onto Chat-GPT and gave it the following prompt.

“Act as a tutor at Oxford's PPE [Philosophy Politics and Economics] program. Ask me questions about the introduction and part 1 of Anti-Intellectualism in American Life. Help me deepen my understanding of the book. When I answer, tell me where I'm off and where I'm right. And ask follow-up questions to simulate a conversation.”

Here’s the first question my AI tutor asked:

I felt myself get a little excited, a little nervous. Do I know this answer? I think so.

I started typing out my answer, with my book closed, and noticed I was struggling a bit. So I sat there and thought harder, then rewrote my answer a few times until I felt I had adequately captured the distinction.

Happy with my answer, I hit submit.

I thought I was going to pass with flying colors. After all, isn’t gen AI supposed to give us what we want? But to my surprise, Chat-GPT wasn’t so easy on me.

Below is the AI tutor’s response. Notice how it called out where my answer was right, and where it was lacking, almost like a real tutor might.

“Oh, you’re right. I did miss that part.” I thought, as I read the refined definition.

Without this pushback, I likely would’ve failed to achieve full comprehension of the concept. I would’ve just lazily accepted my understanding of the definition and never given it a second thought.

Plus, by having to type out the answer, I engaged in retrieval practice, thus driving the material deeper into my mind.

Now, if I need this definition in writing or conversation, I’ll have it at my ready, and I won’t forget the “critical towards received wisdom” aspect that I did the first time around.

After reviewing the answer, Chat-GPT asked me a follow-up question, just as I had prompted it to do, and the conversation continued from there.

Benefits of using Chat-GPT as a personal tutor

Since I’ve started doing this, I’ve experienced the following benefits:

• Remember more of what I read.

• Spend more time thinking about each book.

• Discover insights that would’ve flown over my head.

• Gain a space to practice critical thinking and non-fiction writing skills.

• Get feedback on my answers to fill gaps in understanding.

Most important of all, this method is making reviewing books more fun, even addictive, like a good game.

Try it out, and let me know how it goes

Next week, I’ll go over my step-by-step process for using Chat-GPT as a reading tutor.

For now, consider playing around with Chat-GPT as a testing tool.

Set a frequency for your tutor sessions. For example, maybe every 100 pages, you do a 5-question test with your AI tutor.

As a rule, make sure the AI tool is forcing you to think more deeply about the text, and not doing the thinking for you. Lean into cognitive strain. It’s a sign you’re learning.

A call to action: This technique is bound to get some pushback. It’s also probable that I’m overlooking an issue with it. Please submit your criticism in the comments. Despite Oxford recommending a version of this method to their students, I’m incredibly skeptical of AI use-cases in liberal education, and am therefore always on the lookout for reasons why this specific approach might be a bad idea. So far, I see mostly good things, but who knows…

A note on a likely piece of criticism: If you say Chat-GPT hallucinates, you’re right. However, it doesn’t get hungover or defensive like tutors. I’m not saying it’s better than a human tutor. No one can replace a good teacher. But, as I’ve said before, this method is better than nothing. Many people can’t access a tutor to help them review the books they read.

Caveat 1: A potential issue is that Chat-GPT might provide incorrect information. Fact-checking is therefore advised. But this is difficult to do if you’re a student and still a beginner in the subject. Plus, fact review takes time. Therefore, the question becomes “how big of a deal is it if the AI tutor is wrong some of the time?” For me, if I’m reading a book written in 1960 and 2% of Chat-GPT’s responses to my answers are wrong, I’m okay with that. It’s not like I’m studying brain surgery. A slight misunderstanding here and there caused by Chat-GPT is likely overshadowed by the boost in understanding this book review method gives me. Plus, I’m more interested in its questions than its responses to my answers. I want an arena I can quickly enter to take some practice swings.

Caveat 2: This technique will likely work better for older books and classics that have been written about a lot. The LLM might have no information on newer books, or it might have less information than is required to serve as an adequate tutor. But more experimentation here is needed.