I’m a broken record on this, but almost all of the most ~thoughtful~ people I know (or have interviewed) have memorized a lot of dates, places, names, numbers etc. And not just memorized, but care about their ability to recall specifics.
Seek truth from facts. It’s the only way. — Joe Wiesenthal
So many people reason through heuristics, like who is aligned with whom, what personal financial interests they might have at stake, what peer group pressure they get. And I get it, because these things are complicated.
But the way out, as always, is to seek truth from facts. — Joe Wiesenthal
“You open the newspaper to an article on some subject you know well. In [Murray Gell-Mann’s] case, physics. In mine, show business. You read the article and see the journalist has absolutely no understanding of either the facts or the issues. … [Then you] turn the page to national or international affairs, and read with renewed interest as if the rest of the newspaper was somehow more accurate about far-off Palestine than it was about the story you just read. You turn the page, and forget what you know.” — Michael Crichton
Many who read about Diamond’s geographic determinism, Fukuyama’s end of history, Morris’s energy-capture thesis, Acemoglu and Robinson’s inclusive and extractive institutions, or Turchin’s structural-demographic theory would have been better served learning basic empirical facts about historical or present societies.
There are a bunch of reasons:
With basic knowledge, you can reason from “first principles” rather than merely by weighing arguments or perspectives based on the brains and motives of their speakers.
Within an intellectual community, opinions are less correlated when people reason more on their own rather than defer. Otherwise it’s possible for the appearance of a universal opinion to hide its origins in a small number of independent thinkers.
If you learn some facts, then you can quickly evaluate people’s general theories against the examples that you know well. One way to limit vulnerability to the Gell-Mann Amnesia Effect is just to have immediate sanity checks on hand for more areas of knowledge.
Holden Karnofsky describes getting that last benefit as a result of his minimal-trust investigations—where he can determine which experts to trust more on a given large question based on whether they share his conclusions or generally exhibit good reasoning on the subquestions that he has evaluated in more depth—but basic empirical facts are much easier to accumulate than the complicated evaluations he talks about in that post, and so might make for a better sanity check that can discredit the propounders of many theories. Another benefit shared by the two practices is the ability to, as he says, get intuitions about “what sort of thing usually checks out” in social scientific theories or theories about human behavior or politics in general.
As an example, if you are interested in AI timelines you could, instead of reading others’ arguments, ask Claude to draw charts of how much compute was used to train each model, or what revenue trends are over time, or how expensive it will be to build more compute, or other basic facts. If you are interested in a political topic like abortion, you might ask an LLM how many abortions are performed and at what gestational ages, what the distribution is of lifetime abortions across women, the age distributions thereof, etc. If you are interested in why the Democrats won the midterms in a landslide, you might ask for the trends in turnout and different voting patterns among different demographics, or polls about what voters in different demographics viewed as their top issues, and what candidates overperformed versus underperformed the overall shift, etc.
An interesting Tanner Greer post, “A Study Guide for Human Society,” advocates for studying in some depth three historical periods that vary in time and space, including one modern, one ancient, and one non-Western culture. So you might study Republican Rome, the Protestant Reformation, and 20th-century India. My intuition is that industrialized societies and more recent societies in general are more relevant than older ones, but the value of diversity might dominate.
Your goal here is to build up a fairly granular knowledge of a particular time or event than can be called on to test and assess theories and narratives that will be thrown at you. “Famous scholar X proposes that y leads to z, but did y lead to z in each of the eras I am most familiar with?” You will know you have the background knowledge to do this right when you can answer questions like the following for a given era of expertise: “What are some of the biggest disagreements historians have about this era/event? What are the main sources historians or archaeologists use to try and understand the era, and how might they bias this understanding? If you had to pick one small incident or detail about the era that seems insignificant at first, but is actually very revealing example of the way this society/event worked, what would it be?”
You don’t need PhD levels answers to these questions. Just something more insightful that you would get from the Wikipedia page.
He gives this reading regimen as a prerequisite to more general histories or grand theories. Ben Landau-Taylor says “If you want to understand how society works, you should spend more time reading about specific events and case studies than you spend reading about theories and explanations. Somewhere around 4:1 is roughly the right ratio.” I don’t know if that ratio is necessarily correct, but empirical facts seem overwhelmingly more important than general theories.
More generally, I think that the idea of rationality in general is somewhat overrated, especially in places like LessWrong, compared to the advantage of knowing a larger number of empirical facts (not only about history but also about natural or social sciences, philosophy, and so on). Without those foundations, it’s difficult to reach more accurate conclusions with lower cognitive bias alone.
Many who read about Diamond’s geographic determinism, Fukuyama’s end of history, Morris’s energy-capture thesis, Acemoglu and Robinson’s inclusive and extractive institutions, or Turchin’s structural-demographic theory would have been better served learning basic empirical facts about historical or present societies.
There are a bunch of reasons:
Holden Karnofsky describes getting that last benefit as a result of his minimal-trust investigations—where he can determine which experts to trust more on a given large question based on whether they share his conclusions or generally exhibit good reasoning on the subquestions that he has evaluated in more depth—but basic empirical facts are much easier to accumulate than the complicated evaluations he talks about in that post, and so might make for a better sanity check that can discredit the propounders of many theories. Another benefit shared by the two practices is the ability to, as he says, get intuitions about “what sort of thing usually checks out” in social scientific theories or theories about human behavior or politics in general.
As an example, if you are interested in AI timelines you could, instead of reading others’ arguments, ask Claude to draw charts of how much compute was used to train each model, or what revenue trends are over time, or how expensive it will be to build more compute, or other basic facts. If you are interested in a political topic like abortion, you might ask an LLM how many abortions are performed and at what gestational ages, what the distribution is of lifetime abortions across women, the age distributions thereof, etc. If you are interested in why the Democrats won the midterms in a landslide, you might ask for the trends in turnout and different voting patterns among different demographics, or polls about what voters in different demographics viewed as their top issues, and what candidates overperformed versus underperformed the overall shift, etc.
An interesting Tanner Greer post, “A Study Guide for Human Society,” advocates for studying in some depth three historical periods that vary in time and space, including one modern, one ancient, and one non-Western culture. So you might study Republican Rome, the Protestant Reformation, and 20th-century India. My intuition is that industrialized societies and more recent societies in general are more relevant than older ones, but the value of diversity might dominate.
He gives this reading regimen as a prerequisite to more general histories or grand theories. Ben Landau-Taylor says “If you want to understand how society works, you should spend more time reading about specific events and case studies than you spend reading about theories and explanations. Somewhere around 4:1 is roughly the right ratio.” I don’t know if that ratio is necessarily correct, but empirical facts seem overwhelmingly more important than general theories.
More generally, I think that the idea of rationality in general is somewhat overrated, especially in places like LessWrong, compared to the advantage of knowing a larger number of empirical facts (not only about history but also about natural or social sciences, philosophy, and so on). Without those foundations, it’s difficult to reach more accurate conclusions with lower cognitive bias alone.