Published on November 3, 2025 3:52 AM GMT
Cross-posted from my website.
Last year, I wrote a list of things I've changed my mind on. But good truth-seeking doesn't just require you to consider where you might be wrong; you must also consider where you might be right.
In this post, I provide some beliefs I used to be uncertain about, that I have come to believe more strongly.
My belief: Evolution is true.
Why I believed it originally: I learned about the theory of evolution in school. I had the impression that it was a popular but unproven hypothesis ("just a theory").[1]
What made me more confident: When I was maybe 10 or 12, I read an article in some science magazine (National Geographic, maybe) about evolution. It said, "evolution is a theory in the same way atoms are a theory." I probably put too much credence in this one sentence in one article, but in my mind, this was definitive proof that evolution is true.
Later, when I was 14, I started getting interested in the specifics of the theory of evolution and learned much more about the supporting evidence. (My motivation was mostly that I wanted to argue with creationists on the internet.)
I went through a similar trajectory when learning about quarks. I was taught that a quark is a hypothetical particle that exists inside atoms, but has never been observed. Later I learned that the existence of quarks is well-established, and it became well-established nearly three decades before I was born.
On the subject of outdated pedagogy, this is a bit of a tangent but in 5th grade I was taught the five kingdoms of life: monerans, protists, fungi, plants, and animals. Recently, I learned that not only do biologists no longer use this classification system, but that it was already obsolete when my 5th grade teacher was in 5th grade.[2]
(My 5th grade teacher was pretty young, but still.)
My belief: Value investing works.
Why I believed it originally: I read about Joel Greenblatt's magic formula investing and its strong historical performance.
What made me more confident: I read more research on value investing, including the seminal paper The Cross-Section of Expected Stock Returns[3] by Fama and French, and more in-depth research showing value investing has worked across the world and across asset classes[4], and on older data going back 200 years[5].
My belief: Peaceful protests can be effective.
Why I believed it originally: I actually went back and forth on this one. In school I learned about Martin Luther King and how he was a hero of the civil rights movement, the Montgomery Bus Boycott that he helped organize, Gandhi's protests against colonialism, and implicit in all this was the idea that these tactics were effective.
Eventually I learned about GiveWell, which was the first time I'd ever encountered the notion that just because a charity says it's effective, doesn't mean it's actually effective. I started thinking critically about protests in the same way, and I realized that I'd never actually seen good evidence that MLK or Gandhi were responsible for the positive changes that coincided with their activism.
Then I started thinking, well, there's not strong evidence that protests work, but there's at least some reason to believe they work. That's about where I was at in 2024 when I donated to PauseAI—I thought, I don't really know if this is gonna work, but it's worth trying.
What made me more confident: I wrote Do Protests Work? A Critical Review, in which I carefully investigated the strongest evidence I could find. I found that the best evidence was better than I'd expected, and it pointed toward peaceful protests being effective.
My belief: Seed oils are good for you; seed oils don't cause obesity.
Why I believed it originally: I had never heard of the seed oil-obesity hypothesis until I read Dynomight's article on the subject, which argues against the hypothesis. Dynomight presented some evidence that seed oils are harmful and then ultimately concluded that they're not. I didn't think much about the evidence the article gave, but its conclusion seemed reasonable to me.
What made me more confident: I researched the issue in more depth while writing Outlive: A Critical Review, specifically the section on saturated fat. I looked through the literature and presented what I believed to be the strongest evidence on the matter: meta-analyses of RCTs that directly compared dietary saturated fat with unsaturated fat (which usually meant seed oils). The experimental evidence finds that seed oils are, if anything, healthier than saturated fat, which contradicts the seed oil-obesity hypothesis.
I read some writings by proponents of the seed oil hypothesis, and their arguments seemed incredibly weak to me.
(Later, I re-read Dynomight's article and found that it cited the same evidence I had looked at while writing my review of Outlive, which I had completely forgotten about.)
Dynomight presented the seed oil hypothesis as reasonable but ultimately probably wrong, so that's what I believed at the time. After examining the evidence in more depth, I don't think the seed oil hypothesis is reasonable. Dynomight admirably followed Daniel Dennett's principles for arguing intelligently, in which you present your opponent's case as strongly as possible. But this gave me impression that the seed oil hypothesis is more plausible than it actually is.
My belief: Absent regulation, we aren't going to solve the AI alignment problem in time.
Why I believed it originally: I've vaguely believed this since I first learned about the AI alignment problem (in 2013, if I remember correctly). The problem seemed to involve some thorny philosophical problems of unknown size, like the outline of an enormous beast under a murky ocean. But at that point, humanity had collectively only spent a few hundred person-years on AI alignment, and I thought, perhaps there will be some breakthrough that makes the problem turn out to be much easier than expected. Or perhaps as superintelligent AI becomes increasingly imminent, humanity will rally and pour the necessary resources into the problem.
What made me more confident: In this case I haven't much changed my interpretation of the evidence, but I've become more confident as new evidence has come out. Namely, AI has gotten extraordinarily more powerful; alignment work has not kept up with the increases in AI capabilities; even though alignment work gets more attention now, the problem still seems about as hard as ever.
Beyond that, almost all alignment work is streetlight effect-ing, focused on solving tractable but mostly-irrelevant problems; and the frontier AI companies mostly don't engage with, and are sometimes even actively hostile to, the idea that solving alignment will require major philosophical breakthroughs and it can't be done using the sorts of empirical methods that they're all using.
My belief: Most studies on caffeine tolerance are not informative.
Why I believed it originally: Prior to writing my post Does Caffeine Stop Working?, I reviewed some studies on caffeine tolerance and I thought to myself, these studies aren't even testing the hypothesis they claim to be testing, surely I must be missing something?
What made me more confident: I read the studies more carefully and spent more time thinking about them, and read a few contrary papers by other scientists who study caffeine. My more careful analysis only reinforced my initial belief that most studies on caffeine tolerance are, indeed, not useful.
My belief: I am smart.
Why I believed it originally: In elementary school, I knew I was the smartest kid in my class. But my class only had about 20 students, and I figured I wasn't that smart in the grand scheme of things. Like, not as smart as scientists and people who go to Harvard and stuff.
What made me more confident: The first big piece of evidence came after I took the PSAT in 10th grade and my score was good enough that I realized I had a good shot at getting into a top university.
Then I actually attended a top university and realized that many of the people there were not that smart compared to me. College was still a big step up from elementary school: I went from always being the smartest person in the room to being only in the top 1/3 most of the time, and I sometimes found myself in the bottom third.
This trend of repeatedly up-rating my own intelligence reached its peak when I started taking advanced computer science classes, where I was close to the 50th percentile. And nowadays I'm about average within my social circles, and often below average.
(If you're reading this, there's a good chance that you're smarter than me.)
Another canon event happened when I saw the data on the distribution of my school's SAT scores. The school's average score was just over one standard deviation above the population mean.[6] I went through high school thinking my average classmates were average, when in reality they were considerably smarter than average.
My belief: When I first got into lifting weights a decade ago, I learned a lot of conventional wisdom like:
- Low reps are better for strength, and high reps are better for hypertrophy.Compound exercises are better for strength, and isolation exercises are better for hypertrophy.Long rests are better for strength, and short rests are better for hypertrophy.If you want to bulk or cut, you should eat at a 500 calorie surplus/deficit to gain/lose about a pound per week.
Why I believed it originally: It was the conventional wisdom—people generally agreed that these things are true, even though nobody talked about why.
What made me more confident: I started paying more attention to scientific literature on resistance training and I learned that the conventional wisdom pretty much had it right, at least on these points.
(The first three pieces of advice are all explained by a unifying factor: to build strength, you want to lift as much weight as possible, and to build muscle, you want to do as much volume as possible. High reps, isolation exercise, and short rests all enable you to wear out your muscles while lifting lighter weights, and the lighter the weights, the more volume you can do. These three bits of advice aren't overwhelmingly important—you can still build muscle doing compound exercises at low reps—but they're useful as guidelines.)
My belief: Exercise is good for you.
Why I believed it originally: Everyone says exercise is good for you, right? But I didn't know how you'd demonstrate scientifically that that's true. I thought perhaps it's reverse causation (sick people can't exercise) or confounded by socioeconomic class or something.
What made me more confident: I learned more about the scientific evidence on exercise.
- Many randomized controlled trials show that exercise improves short-term health markers—it reduces blood pressure, improves blood sugar regulation, etc.A smaller number of long-term trials show long-term health benefits to exercise.
Spoilers for Game of Thrones / A Song of Ice and Fire. I don't see a way to do spoiler tags in LessWrong Markdown, so I will just put some empty space before giving the spoilers.
My belief: R + L = J. That is, Jon Snow's parents are Lyanna Stark and Rhaegar Targaryen.
Why I believed it originally: This had long been a popular fan theory. I didn't figure it out on my own, but I was reasonably convinced by the evidence in this article. I thought it sounded right, but I was uncertain because the textual evidence wasn't conclusive.
What made me more confident: I watched an interview with David Benioff and Dan Weiss, the creators of the TV show. They told a story about how they met with George R. R. Martin to get him to agree to adapt his books. At some point in the meeting, he asked them: Who is Jon Snow's mother? They gave an answer, and he didn't say whether they were right, but he gave a knowing smile, and he agreed to let them make the TV show.
They didn't say what their answer was. But I found this story to be pretty much decisive evidence for R + L = J because what it proved was that the answer was knowable. If David and Dan could know it, then the rest of the fan base could, too.
Later I became even more confident when the TV show revealed that R + L = J. (Rarely in life do you get definitive confirmation that your theory is correct!)
When I was young, I thought the way evolution worked was that a group of apes were born about 500,000 years ago, and these apes lived for hundreds of thousands of years, over which time their bodies slowly morphed to become more and more humanoid, until they became fully human, at which point they birthed human offspring and then died.
One time I told my dad that I wish I could've gotten to evolve because I wanted to live for 500,000 years. That's when I learned that that's not how evolution works. ↩︎
Carl Woese defined a six-kingdom taxonomy using evidence from ribosomal RNA in 1977, at which time I believe my 5th grade teacher would've been in 2nd grade.
Lest I sound like I know what I'm talking about, the only reason I can talk coherently about ribosomal RNA methods for taxonomic classification is because I just read those words off Wikipedia 15 seconds ago. ↩︎
Fama, E. F., & French, K. R. (1992). The Cross-Section of Expected Stock Returns. ↩︎
Asness, C. S., Moskowitz, T. J., & Pedersen, L. H. (2012). Value and Momentum Everywhere. ↩︎
Baltussen, G., Swinkels, L., & van Vliet, P. (2019). Global Factor Premiums. ↩︎
And someone who gets an average score on the SAT is above-average intelligence, because taking the SAT at all already screens off the lower end of the bell curve. ↩︎
Discuss
