Jacob Greif

New blog post: The third wave of American philanthropy Hundreds of billions of dollars in new philanthropic capital will soon become liquid. The OpenAI Foundation holds 26% of OpenAI, worth about $220B at today’s valuation. Anthropic’s seven co-founders have pledged to give away 80% of their wealth and have instituted the most aggressive donor matching program for employees in tech history. How much does this all add up to? And how meaningful is that in the context of philanthropy today? I was doing some simple napkin math to wrap my head around the scale of what’s coming, and radicalized myself in the process. I had dramatically underappreciated the scale of the philanthropic capital that’s about to become available and the corresponding gap in talent and organizations that will be needed to make the most of it. This piece aims to directionally sketch the scale of what’s coming, the gap in operational capacity needed to absorb it, and what we can do to fill it. (Link to full post in reply)

When you're deciding what to study in college, don't try to predict what will be valuable in the future, because that's so hard that you'll probably get it wrong. Instead focus on what you personally find most exciting. You can't get that wrong.

Why You Should Write Your Own Essays Writing is thinking. That sentence is so short and plain that it barely seems worth saying, but it may be the most important claim in this essay. When you sit down to write—actually write, from the blank page, with no assistance beyond your own mind—you are not merely transcribing thoughts you already have. You are generating them. The resistance of the page, the demand that one sentence follow another in some defensible order, the necessity of making vague intuitions precise enough to survive contact with words: this is cognition of a kind that cannot be replaced. We are now several years into a period in which artificial intelligence can produce, on command, parsing and polished prose on virtually any topic. The output looks good. It is fluent, grammatical, and organized. It arrives with headers and transitions and topic sentences in all the right places. It has the aesthetic properties of competent writing. And this is precisely what makes it dangerous—not because AI-generated text is always wrong, but because the gap between looking right and being right is where intellectual life actually happens, and AI closes that gap from the wrong direction. Consider what a large language model actually does. It produces sequences of words that are, in a statistical sense, plausible given the words that preceded them. It is an extraordinarily sophisticated pattern-matching engine. When it writes an essay about, say, the causes of the First World War, it generates sentences that resemble what a knowledgeable person might write on the topic—but it arrives at those sentences through a process that has nothing to do with understanding the First World War. It has no model of the world. It has a model of text about the world, which is a very different thing. The result is prose that is often correct, sometimes subtly wrong, and occasionally complete nonsense delivered in the confident register of an encyclopedia entry. This creates an asymmetry that should trouble anyone who cares about the quality of their own thought. Generating plausible-sounding prose is cheap. Verifying that plausible-sounding prose is actually correct is expensive. If I hand you a well-formatted five-paragraph essay on monetary policy, you cannot assess its accuracy without already knowing a fair amount about monetary policy—at which point you scarcely needed the essay. The whole exercise collapses into a peculiar loop: the people best positioned to verify AI output are the people who least need it, and the people who most need it are least equipped to catch its errors. This is not a minor inconvenience. It is a structural problem that degrades the epistemic environment for everyone. And people do catch on. Not immediately, and not consciously, but they catch on. When a colleague, a student, or a writer begins producing text that is always fluent and never quite has a point—when every paragraph reads like a reasonable thing to say but the whole never accumulates into an argument that surprises or challenges or teaches—readers develop a kind of immune response. They begin to skim. They stop expecting to learn anything. The text begins to function as bureaucratic filler, something that exists to satisfy a formal requirement rather than to communicate. We have all experienced this with corporate communications and boilerplate legalese, and AI-generated prose is rapidly joining that category. It is writing that is optimized for the appearance of competence, and the more of it that floods the world, the less anyone trusts that any given piece of writing is worth their sustained attention. You should care about this even if you are not a professional writer. The ability to construct an argument on paper is the ability to construct an argument in your mind. When you outsource the writing, you outsource the thinking, and what you get back is not your thinking—it is a statistical average of how people have discussed similar topics in the past. It will not be wrong in interesting ways. It will not lead you somewhere you did not expect to go. It will not force you to confront the part of your argument that does not actually hold together. All the productive discomfort of real intellectual work is smoothed away, and you are left with a product that teaches you nothing about what you actually believe. There is a particular danger for students, but the problem extends well beyond education. Anyone who uses writing as a tool for reasoning—which should be everyone—faces the same temptation and the same cost. The executive who lets AI draft a strategic memo never discovers the contradiction between two of her stated priorities. The researcher who lets AI write the literature review never notices the gap in the existing work that could have become a paper. The citizen who lets AI compose a letter to a representative sends something that sounds like every other letter and vanishes into the noise. Writing badly, writing slowly, writing with effort and frustration and deleted paragraphs—this is not a flaw in the process. It is the process. The difficulty is where the value lives. A first draft that comes easily and reads smoothly has almost certainly skipped the hard part, which is the part where you figure out what you actually think and whether it survives scrutiny. If you are not struggling, you are probably not learning, and if you are not learning, then what, exactly, is the essay for? I am not making a Luddite argument. Tools are good. Spell-checkers are good. Access to information is good. But there is a difference between a tool that augments your capacity to think and a tool that substitutes for it. A calculator helps a mathematician work faster. A machine that generates proofs the mathematician cannot follow does not make the mathematician better—it makes the mathematician irrelevant. The question you should ask about any writing tool is: does this make my thinking sharper, or does it make my thinking unnecessary? If the answer is the latter, you should be alarmed, not grateful. Write your own essays. Write them poorly if you must. Write them slowly. Let them be difficult and frustrating and full of sentences you will later cut. What you produce will be yours—not in the sentimental sense, but in the epistemically meaningful sense that it will represent what you actually believe, tested against the rigorous and unforgiving demands of the written word. In a world filling rapidly with fluent, confident, empty text, that is worth more than you might think.