In 2025, I found plenty of time to listen to podcasts, but didn’t make enough time for more challenging reads. To resolve this imbalance, I purchased an Audible subscription and I have resolved to read (listen to) at least 26 books this year. After reading each book, I’ll log it here with a brief summary of my takeaways and some reflections.

I’m thinking of reading mostly non-fiction books this year, to balance out an emphasis on fiction in previous years and to make progress in building my understanding of the world.

A pattern that has emerged after the first few books is that I want to look out for things I learn in each book I read that are of relevance to my work in AI safety and alignment. Of course, I typically pick books that have some chance of relevance since both my work and my book choices are downstream of my broad interests, but often the specific connections I spot are unexpected.

Contents:

• 1. Deep Utopia by Nick Bostrom
• 2. Feline Philosophy by John Gray
• 3. Replacing Guilt by Nate Soares
• 4. What is Life? by Erwin Schrödinger
• 5. Why Greatness Cannot be Planned by Kenneth O. Stanley and Joel Lehman
• Up next?

§1. Deep Utopia by Nick Bostrom

Audible audiobook (2x), Jan 23–31, 2026.

Bostrom is a futurist and philosopher, and Deep Utopia (2024) is his exploration of the potential and limitations of a realistic utopian future for humanity. I got a lot out of reading Superintelligence back in the day so I had been meaning to give this new book a try since it came out.

Channeling Plato, the book’s philosophical discussion is framed as a fictional series of lectures delivered by Bostrom, interleaved with dialogues between a group of friends auditing the class. Bostrom also splices in various technical readings, including some stories from alternative universes in which sentient animals or appliances strive towards their own utopian visions. I suspect there were some allegorical elements of the alternative universe stories and the hints at university (dis)function that went over my head, but I mostly came for the explicit philosophical discussion in the lectures, and I think I was able to follow most of that.

As with Superintelligence, the philosophical discussion revealed to me that concepts I had a folk understanding of were actually deeper than they appeared, or introduced useful concepts and distinctions that I think I will find useful in framing my thinking going forward. For example:

• Bostrom taxonomises various barriers humanity could face in implementing a utopia, beyond just physical limits and also including various ‘reachability’ constraints. I appreciated this broader and unified discussion of constraints.

• Bostrom had a habit of asking, for a given human value, why do we have that value? This is obviously of interest to me as an alignment researcher. He repeatedly appealed to a concept of ‘intrinsification,’ where instrumentally useful things over time come to seem instrinsically valuable to us.

I think these kinds of things are good examples of the value of reading good philosophy, and I am overall happy I read the book.

Also as with Superintelligence, I didn’t find all of the book’s discussion philosophically satisfying. For example:

• One recurring theme that stood out was that I generally felt that the topic of developing humans into the best versions of themselves was under-treated. Perhaps this is out of scope as Bostrom is specifically talking about a utopia for us, present humans. I think I might just have an intuitively broader notion of us that includes some of who we could be? I’m not sure, I would have liked to see how Bostrom would frame this. He does discuss some related issues of identity loss later in the series.

• A more pedantic complaint. At a couple of points Bostrom suggests creating isolated situations where humans limit their access to utopian technological affordances in order to achieve certain kinds of value that the affordances undermines. He explicitly says that one could consider these people to be leaving utopia, but he prefers to consider the isolated situations as still being part of utopia. I didn’t appreciate this being dismissed so readily.

I will say, I had a much longer list of incredulous margin notes in Superintelligence. It’s on my mind to re-read that some day and see how well my complaints hold up.

Overall, the book seemed incomplete to me. I would have liked to see more of the stories of Feodor the Fox and ThermoRex the space heater—it felt like there was more to develop in both. Moreover, the final lecture in the fictional series was also open to the public, and consequently dropped most of the prior motivation of discussing utopia in order to give Bostrom’s self-contained account of ‘the meaning of life.’ I was hoping there would be a final discussion that connected this account back to the discussion of utopia, explored how our search for meaning would play out in utopia.

Note: This was a long and dense book. I followed fine, but I don’t think I will retain most of it. Of course, I would ideally like to retain valuable things I read this year, so, going forward, I’m going start trying to take notes while reading and write a summary of the key ideas in each book as I see them at the end.

§2. Feline Philosophy by John Gray

Audible audiobook (2x), Feb 2–9, 2026.

Feline Philosophy (2020) is a short exploration of the contrast between the ways cats and humans orient to life. I learned a few cool things about cats. According to Gray:

• Domestic cats are very genetically similar to their wild ancestors, apart from having slightly shorter legs and more colourful fur.

• They also have smaller brains, but it’s particularly the parts of the brain responsible for ‘fight or flight’ response that has shrunk, presumably because their environment tends to be somewhat less hostile.

  Aside: It’s amazing how often I find things that seem relevant to AI alignment when reading random books!

• Cats have never been pack animals, they are independent. They are happy to live with humans, but they don’t think in terms of hierarchies of dominance. They will form groups of cats only out of convenience. They will love humans but not in a dependent way.

• Cats are “selfless egoists.” This sounds oxymoronic but he means they are egoists in that they think only of their own needs while also being selfless in the specific sense that they don’t concern themselves with their identity or self-image.

This leads Gray to claim that cats were never domesticated—rather, they domesticated humans. Fun line; it reminds me of Harari’s take on the agricultural revolution from the perspective of crops.

The book made general claims about cat behaviour, and drew examples mostly from cats in literature or biographies, so I’m not sure how carefully each of these claims has been studied. But they seem plausible.

Anyway, of course, the book was just as much, if not more, about humans as it was about cats. Here’s some things I learned about humanity.

• First of all, it’s striking that at one point it was totally normal for people to think cats were unworthy of moral consideration, or worse:

  • Descartes experimented on cats, managing to somehow prove in the process of torturing them that the cats had no souls. (I think he experimented on dogs in similar ways too.) Philosophy of mind is hard, I guess?

  • Western religious societies until relatively recently viewed cats as devils, sometimes ritually burning them alive. (Of course they also did this to women accused of witchcraft.)

  • Even scientists used to feel totally free to torture cats in order to study pain. (These days there are animal ethics guidelines that might stop this but I think I have met some scientists that would be sympathetic to it on utilitarian grounds.)

• On the other hand, animist civilisations like Egyptians saw cats as embodiments of gods. To me this shows that very many different civilisational world views are possible.

• According to Gray, cats teach us that the meaning of life is in the living of life and not in the questioning of why. From what I hear about his other writing, he views humans as more similar to animals than they like to think. Except, he points out, humans have self-awareness and accordingly self-consciousness and this leads to their problems.

  • Humans orient to the world by writing stories about their lives. However, this makes it difficult when the stories are disrupted, can lock them into unpleasant roles (such as when someone views their life as tragic, they can become unable to escape the tragedy), and prompts people to think ahead prematurely to their own death (the end of their story).

  • He links self awareness to the fall of man—once you know you are in paradise, you are no longer in paradise.

There was quite a lot more in the book. Stories of many historical philosophers who had written about similar topics, and several cats and their owners, friends, and foes. These were interesting but I didn’t focus as much on remembering the details.

I think I extracted the main message of the book, that we should live more in the moment. But the book was a bit meandering, so it wasn’t obvious what the hierarchy of ideas was meant to be. There is a nice list of ten principles at the end, and the book was short and sweet, so would be pretty easy to re-read.

§3. Replacing Guilt by Nate Soares

Podcast audiobook (1.5x), Feb 15–27, 2026.

Nick Marsh recommended I pick up this 2015–2016 sequence by Nate Soares. Apart from reading Rationality: from AI to Zombies, I haven’t engaged much with LessWrong sequences. Time to start working on that. Note that as part of my broad definition of “generalised readings,” I’m happy to count a thematic collection of essays towards my yearly goal. As it happens, this collection does have an overarching structure, and has technically been published as a book, and there is an official audio version by Gianluca Truda (this is the version I listened to).

This time, I experimented with taking much more detailed notes than for the previous readings. During and after each essay, I would try to summarise the structure of the argument in my own terms. This was a useful exercise and ultimately helped me get a lot more out of the book, to the point where I could afterwards synthesise the below detailed summary. However, the detailed note-taking made it less convenient to listen, for example it was hard to listen while doing housework, and a little dangerous to listen while walking down High Street typing notes into my phone. I don’t expect to be able to keep this up for all of the books I read going forward. By default, I probably want to revert to listening without feeling obligated to take detailed notes (unless I specifically want to study a particular book in this much detail). I might try listening more slowly than 2x speed by default, as a compromise.

Anyway, in this case, I took the detailed notes. I thought the ideas were valuable enough that it was worth spending the time to write myself a detailed summary attempting to synthesise into a coherent story what I saw as the key ideas in Replacing Guilt. I have published the result as a standalone post. Here, I will just give a brief summary and some reflections.

The brief summary is as follows. In my estimation, the book has two parts.

1. The first is the titular contribution: the thesis is that chronic guilt is an unnecessary component of an effective motivation system. Soares considers many situations in which one may feel guilt (failing to meet an external objective, failing to do something you stereotypically ‘should’ do, failing to do things that are actually psychologically impossible, and many others). In each case, he argues that the guilt can either be refined into guilt of another form, or can be dismissed because suffering it has no positive consequences. All remaining guilt must arise due to failing systematically to meet intrinsically-derived objectives. In this case, Soares completes the argument by showing that positive, intrinsic motivation and a mindset of scientific self-improvement are more effective than using chronic guilt as an expensive punishment for failing to meet these kinds of objectives.

2. That argument does not assume any particular individually-affirmed value system. However, the book spends an equal amount of time exploring Soares’ own intrinsic values, which centre around eliminating suffering and unwanted death in the long-run future. Soares points out that from the perspective of this value system, the current state of the world is extremely grim, presenting a challenge for intrinsic motivation of this kind. Therefore, he offers various strategies for rising to this challenge, including embracing the darkness as a source of resolve, avoiding resorting to despair, and a grab-bag of other rationalist strategies for being effective in the face of extreme challenge.

I should say that this decomposition of the book isn’t crisply reflected in the book itself. The book mostly discusses the dismantling of the guilt-based motivation near the beginning, and mostly advice for taking on the ‘dark world’ in the latter half, but the topics are coupled and both are discussed throughout the entirety of the book.

This is not to say I thought the book wasn’t well written. To the contrary, I found the writing and arguments extremely clear and the examples relatable and compelling. It’s natural for a sequence of essays written somewhat serially and then collected into a book to result in a somewhat non-linear story.

I don’t think I’m someone who struggles with chronic feelings of guilt to the extent of some of the examples Soares discusses. Nevertheless, I’ll gladly take on board the above framework for tapping into intrinsic motivation and coping with the darkness of our world. The advice seems sound to me, at least given the consequentialist/existentialist premises (I am tentatively on board with these, but would need to think more before fully endorsing them).

The book contains many examples, anecdotes, arguments, and detailed pieces of advice. Some I have recapitulated at the level of my detailed summary. Others didn’t end up being mentioned in my summary, though nevertheless I expect they will serve me well (particularly the essays “half-assing it with everything you’ve got,” “failing with abandon,” and “rest in motion”). The essay “the allegory of the stamp collector” contains the at-this-point-obligatory alignment-relevant material in the form of some good examples for agent foundations research—though this time it is not unexpected, given Soares’ position at MIRI.

I am happy to recommend people read the book themselves for a higher fidelity version of the whole collection of ideas. I’m certainly glad I picked it up. Thanks for the recommendation, Nick!

§4. What is Life? by Erwin Schrödinger

Audible audiobook (1.0x), Mar 8–16, 2026.

Every now and then, I hear about What is Life? The Physical Aspect of the Living Cell (1944) and I remember how awesome it sounds and how much I would like to read it. Though I’ve never had a copy on my bookshelf, I consider this a good example of one of the predictably deep books on my shelf I systematically haven’t read. Well, that was the old me. I resolved to be no longer afraid to throw myself into deep ends. When the book came up again recently, I took my chance and started listening immediately. It wasn’t long into the introduction before I knew I had made the right choice.

I would divide the contents of the book into a discussion and resolution of three apparent contradictions. Though the bulk of the book and its legacy is concentrated on the first of the three contradictions (concerning the stability of the genetic material), the second (concerning life’s ability to resist entropy) was what drew me to the book in the first place, and the third (concerning free will) was of some interest.

Regarding the first of the apparent contradictions. From the perspective of a physicist in 1944, living organisms and their evolution were quite mysterious phenomena:

1. On the one hand, based on our understanding of thermodynamics, we know that the apparently orderly laws of physics governing macro-scale systems are only orderly as a consequence of the massive numbers of atoms involved. In any small groups of atoms, physics is unpredictable, unstable, chaotic, with atoms bouncing around and colliding randomly. If you zoom out to a large enough group of atoms, the central limit theorem kicks in and on the whole, variance away from the average physical tendencies vanishes, leaving only predictable diffusion of energy or molecules.

2. On the other hand, in the course of the cycle of life, organisms maintain and transfer to their descendants some physical material that goes on to determine all aspects of their form and function. This hereditary material is stored in the chromosomes within the nucleus of each cell, which amounts to a small number of atoms from a thermodynamic perspective. Nevertheless, apart from a small rate of mutations these small groups of atoms are able to reliably govern the form and function of organisms on timescales spanning many years.

Schrödinger’s resolution to the apparent contradiction of the stability of these small groups of atoms, in light of his understanding of quantum mechanics and some existing work in experimental biology, is as follows. He conjectures that the hereditary material comprises a structure of matter not before studied by physicists, with the stability of a molecular crystal but with an aperiodic rather than repeating structure that allows it to function as a hereditary medium. Moreover, the molecules forming the code are largely stable while still allowing rare mutations that facilitate evolution by the existence of quantised states between which the molecules can transition during rare events, but within which they otherwise remain.

The majority of the book is concerned with outlining the above contradiction and reviewing the biological evidence leading to Schrödinger’s conjecture. This was the main thrust of the book and is also the basis for its main claim to fame, as less than a decade later, Schrödinger’s conjecture would be proven correct upon the discovery of DNA by Franklin, Watson, and Crick. Both Watson and Crick attributed part of their inspiration to enter the field of molecular biology to reading What is Life?

But, as I said, this is not what brought me to the book. I wanted to read the book for what turned out to be a brief discussion towards the end, of a second apparent contradiction, concerning life and thermodynamics at the macro-scale rather than the micro-.

1. On the one hand, we have the observation that according to the second law of thermodynamics, all closed systems tend towards a maximally disordered state.

2. On the other hand, organisms resist this tendency and maintain themselves in a state of low entropy over their lifetimes.

The resolution to this paradox is simpler: the organisms are not living in a closed system, but are in fact exporting entropy to their environment, or, as Schrödinger puts it, “drinking orderliness” (negative entropy or, more accurately, free energy). The “orderliness” in turn is supplied ultimately to the Earth by the sun, and a full accounting of entropy exchanges between all relevant parties will leave the second law upheld.

This much, I had concluded after studying thermodynamics during undergrad, but it was still satisfying to finally see this famous rendition. Schrödinger also pointed out that the organism’s ability to concentrate a stream of order into itself is also downstream of the hereditary material, based on which the entire organism is of course constructed.

Finally, we have in the epilogue a discussion of the topic of free will. A third apparent contradiction is raised:

1. In the case of we humans, the physicist sees that our bodies are made of atoms that function mechanically according to the laws of nature.

2. Yet, we know, by “incontrovertible direct experience,” that we control and take responsibility for our motion and actions.

The resolution that Schrödinger suggests is simply to define ‘we’ as that which moves the atoms by the laws of nature. It follows that the apparent plurality of consciousness is an illusion, and all consciousness is merely different aspects of one singular thing.

I thought this was an interesting take, pleasant, though not compelling. I remain, like everyone else is as far as I can tell, very confused about consciousness on a technical level. On that note, Schrödinger does not seem to be hopeful—here’s a nice passage from earlier in the book:

[Life] is a marvel than which only one is greater; one that, if intimately connected with it, yet lies on a different plane. I mean the fact that we, whose total being is entirely based on a marvellous interplay of this very kind, yet possess the power of acquiring considerable knowledge about it. I think it possible that this knowledge may advance to little short of a complete understanding of the first marvel. The second may well be beyond human understanding.

What did I learn from this book that is of relevance to AI safety? On a technical level, I already knew most of the ideas from statistical mechanics. I knew a little about genetics and evolution, though the book revealed that I don’t know that much: I learned that heritable mutations are rarer than random mutations and for good reason from an evolutionary search perspective. I think this is a basic and well-known fact about evolution, but it was new to me. I should study some evolutionary biology, I am sure it would be good for me.

What I’m taking away from the book is more that it’s an example of both the power of deep scientific understanding and of science communication—two topics close to my heart. Schrödinger applied a mastery of the principles discovered in the field of physics to extrapolate beyond his field’s historical subject matter and make definitive predictions about an important scientific question. We could use some more of that kind of thing in the field of AI.

Moreover, in doing so, Schrödinger brought attention to a promising line of biological research and helped inspire a generation of physicists to get into molecular biology, seeding new discoveries and progress. In a sane world—one where we get the time we need to navigate the transition to a world with advanced AI systems—it will take an all-out, multi-disciplinary, multi-generational effort to prepare ourselves intellectually. We need visionary teacher-leaders to see far ahead of us and chart the course we need to take. Let Schrödinger be our example as we aspire to rise to this challenge.

§5. Why Greatness Cannot be Planned by Kenneth O. Stanley and Joel Lehman

TTS audiobook (1.0–1.5x), Mar 16–, 2026.

I’ve been thinking about what to focus on during my PhD. A couple of people pointed out that my approach—of trying to derive a direct route to a solution to the ambitious problems I am facing and then pursuing that route—might not be the best way. This book is about another way. Maybe the most direct route to solving my ambitious problems is the indirect approach of reading this book.

§Up next?

Here are some books on my reading list (no promises).

1. The Precipice by Toby Ord, which I started once upon a time, but never finished.

2. Reasons and Persons by Derek Parfit.

3. The Dream Machine by M. Mitchell Waldrop.

4. The Hero with a Thousand Faces by Joseph Campbell.

5. Replacing Fear by Richard Ngo. A sequel, of sorts, to Replacing Guilt by Nate Soares, that aims to generalise the analysis there.

6. Ladies, We Need to Talk by Yumi Stynes and Claudine Ryan, on a recommendation from the most important lady in my life.

I welcome further recommendations!