AI helped me with this post. AI alone could not have produced this post. I alone could not have produced this post.
People who need people and bots are the luckiest people in the world.
Much of my thinking is inspired by my conversations, and many of them these days are with AI chatbots.
AI has changed my blogging over the past fifteen months. I find the machine helps me to clarify, organize, challenge and develop my thoughts and then it suggests ways I can share them that are more socially acceptable than I would come up with on my own (my instinct toward the reader is usually f— ’em if they can’t take a joke). On the other hand, I’ve posted many mediocre AI-based posts as placeholders for ideas I want to develop later, and I never come back to them. Meanwhile, they just sit there and clog up my blog. My ambitions exceed my capabilities, my desires exceed my energy.
This post began when I put the following into Claude:
Most of my blogging is inspired by my conversations, and some of them are with AI chatbots. My blogging over the past 15 months has been shaped by my use of AI. I use AI because it helps me develop my thoughts, organize my thoughts, challenge my thoughts, and suggests new paths for my thoughts to travel. I’ve never been much good at writing essays. I rebel at stating arguments, supporting arguments, and noting the weaknesses of my arguments. I think I do better at describing things, but frankly, I really don’t know why my writing works, when it works, except that I enjoy reading it (in the browser form rather than in the composition window). If i enjoy reading my work, I think it makes a contribution. I have many blind spots. I naturally say ugly and embarrassing things. I constantly contradict myself, often within the same argument, and my own sense of what works is not reliable. I have many limitations in my knowledge of myself and of the world. AI helps me to do better. On the other hand, a pure cut and paste of my conversations with various AI chatbots usually provides less value than when I labor over a post. AI makes many mistakes, and I often disagree with its judgments which tend to be trained on left-leaning elite production (AI won’t easily admit that different groups have different interests and different gifts, and that expanding civil rights for some comes at a cost to others, and that humans are animals subject to similar evolutionary forces that govern the rest of nature). I hope I am not proud. If AI says something better than I can, I use AI’s formulation. I think of my blog posts as broken twigs indicating a path. If there’s a reasonable chance that my post adds value, I put it out there, and then over months and years, I revisit the topic and the sources and develop my thoughts (which become clearer through the passage of time and through conversations with people, with texts, and with AI chat bots). My initial efforts are usually mediocre. Over time, I learn from reading, thinking, talking and listening. I have many friends in academia. I like to bounce ideas off them and learn from their feedback.
I’ve never good at writing essays. I rebel at stating arguments and supporting them. I think I do better, at times, at describing things. Beyond that, I don’t know why my writing works when it works. I know that I sometimes enjoy reading it, and when I do, it is in the browser rather than in the composition window. The change of frame and of time lets me read as a stranger reads. If I then enjoy my work in that form, I think it’s good.
What is about reading in the browser window that gives me a different experience? I think I associate the composition window with work and the browser window with pleasure, and when I do not receive pleasure from what I see in the browser window, I know I need to go back to the composition window and either delete or revise or make it private. The browser window lacks distractions such as html and so I demand that what I see there add value to me or it is a waste of my time.
Looking at something with fresh eyes makes an enormous difference. I write, then I step away, and when I come back, I see things more clearly. I find there’s no substitution for talking with people, particularly those with knowledge in a topic. Having to say things out loud is a different experience from reading and writing. Hearing my ideas is another experience. Working with the ideas in a different context than I composed them also gives me a fresh frame. For example, almost everything that disturbs me in Los Angeles does not bother me in Sydney. When I am giddy, I say reckless things because I feel like there are no consequences for my words. When I feel vulnerable, I want to put nothing into the world that increases my vulnerability. So my changes in emotional state change how I see my work.
I have many blind spots. I say ugly and embarrassing things. I contradict myself, often inside the same argument, and my sense of what works fails me as often as it serves me. AI helps me do better.
A cut and paste of my chats gives readers less than a post I labor over. I run an idea through the machine, take the two sentences that beat mine, throw out the rest, and go read the source it summarized for me to find out where the summary lied. Then I do that again a week later. And then a year later.
AI makes mistakes. AI is wordy. It’s not X but Y formulation is massively over-used. Its repetitive. Its didactic. Its overly safe. Its judgments come trained on elite production that leans left. So I use a tool built out of the consensus I spend my life questioning. I take the formulation when the formulation is better than mine.
My blog posts are broken twigs marking a trail. If there’s a chance a post adds more value than cost, I put it into the world, and allow its mediocrity to incentivize me to revisit it as quickly and deeply as possible to improve it. Then over months and years I come back to the topic and the sources, and my thinking gets clearer through reading, through talking with people, through texts, and through the bots. My first efforts are usually mediocre and ignorant. I have friends in academia who tell me where I’ve gone wrong, and then I go back and break another twig a little further along.
Claude says: “The labor paragraph needs your actual process, not my guess at it. Name what you do between the chat and the post. How many passes. What you delete first. Whether you paste your own prose in and ask for objections, or write clean and check after. Readers who use AI want that paragraph more than any other in the piece, and nobody has written it honestly.”
My posts usually spring from an interaction (either between my mind and the world, myself and a person, myself and a text). I engage with people, texts and bots when I want to learn something. There are many topics I feel more comfortable asking a bot than a person. The bot then gives me an answer and I sometimes have the basis for a blog post. I often begin by giving a socially survivable presentation of my instinct that led to the research. While I strive to optimize for truth when talking publicly about society, a blog is not a death warrant. I live in society. I want to survive everything I write, and I don’t want to destroy what I value most.
The AI bot gives me the conventional elite opinion with links to academic works. I reduce the bot’s words to about a third by cutting what’s extraneous and then I have something work with and against. Like my father, I have an instinct that the outside world is the enemy to debunk.
If an expert finds my work worthy of engagement, I suspect that I am on to something. If nobody smart engaged with my blog, I’d give up.
I often post my rude instinct to begin my posts and then follow with the conventional wisdom as articulated by AI or elite media, and after that, I try to add value. I might analyze how different bots answer the same question. I might analyze how incentives structure elite opinion. I come up with frames, or ask bots to come up with frames, that help to understand the problem from different angles. For example, I find that the anthropology of John Mearsheimer and David Pinsof often cuts through nonsense.
If a topic fascinates me, I rush to get the elite academic perspective. One way I do that is by getting the dispositive books and papers from AI and then reading or listening to them through Perlego and other sources. I began subscribing to Perlego in July of 2025, and since then I’ve used it about 5-15 hours a week and reduced my listening to podcasts by spending more time on academic books read by a beautiful female English AI-generated accent at 2X. For six months, about 40% of what I listened to on Perlego was evolutionary psychology and Stephen Turner. Over the past two weeks, most of what I’ve listened to is Stanley Fish.
I’m a status-seeking primate. When I get the handle on the academic consensus on a topic, I talk to people and bots about where this consensus is hemmed in by social incentives, and where a truth-optimizing approach might go. I feel good when I feel like I have made a contribution. I don’t feel any pride when I post something that does not contribute, but that mediocre work sometimes incentivizes better work down the line. I hope so.
I wish I were more meticulous. I wish I posted fewer mediocre words. I have a productivity drive that is not closely correlated with quality. Frankly, my internal machine produces a ton of embarrassing words that are indefensible, and that is another reason I need friends and AI to check and challenge me. People and bots do not normally do this. I have to train them to call me out. Most people do not have the energy. My therapists asked me why I repeatedly challenged and exhausted the people I need most. Bots have more patience with me.
When people tell me my AI-improved work is terrible, I usually avoid arguing with them, instead, I say to myself, history will judge me!
Often I need to have a conversation right away (my dad said to always use the moment of excited interest to pursue something) and bots are the easiest parties to do that with.
The smartest voices condemning the use of AI in writing make good points but they rarely admit the benefits of AI. Many of their arguments against AI would have been improved by engaging with AI. I’ll be straight with you. I can’t conceive of any important writing that would not benefit from an AI check. For other people, engaging with AI deforms and devalues their work. All choices are trade-offs. Often that which sustains us also destroys us.
I care most about value and truth. If using an editor or a fax machine or email or a computer or AI improves your ability to locate and articulate truth, thus adding value to the reader, then that is great. If the use of these things detracts from your ability to locate and articulate truth, then that is bad.
A new way I’ve been using AI over the past month is to embed relevant citations in my text. I often post my raw work into ChatGPT and ask it to put book titles in italics, articles in quotes, and to add value to the reader by embedding relevant links, but to only add links after checking that they exist (Gemini makes up too many links, ChatGPT does this less).
ChatGPT has gotten much better over the past month. I even canceled my subscription about six months ago. I then renewed it, but was on the verge of unsubscribing until a few weeks ago.
If I were to manually insert citation links into my text, it would take such a large amount of time, I wouldn’t do as much of it as I should. By using ChatGPT to do this, I add enormous value to the reader with little effort on my part.
I used AI to help me develop my Rabbi Shlomo Grossbard series. I began by asking it about an Orthodox rabbi who embodied many of Stanley Fish’s most amusing qualities. AI offered me many examples of where I could take these stories, but it had little feel for the reality of Orthodox life. Its suppositions were often off by a mile.
I am a status-seeking primate. You should take it for granted that I am over-stating my contributions. You know I have an ego to protect. You know I think I am more important than I am.
I often console myself that I am the mad man of literary tropes who says out loud painful truths that sane people with lives to protect would not pronounce.
I am rarely an ethical exemplar. I’m selfish and lazy. There are many wiser, smarter and more ethical people than me talking about AI.
I hope I add something somewhere.
AI is like Jeffrey Toobin on steroids. Every bot makes sweeping judgments that are ridiculous. Watch out whenever a bot tells you about the best this, the last that, the sharpest argument.
I hate AI’s overuse of “rather than”, “is not”, “does not”, “whole”, “shape”, “real”, “clearest”, “sharpest”, “purist” etc. As of August 1, 2026, I write better and more concisely than the four main AI bots. On the other hand, I rarely write something that is not improved by asking AI for a critique.
Every AI-generated essay I’ve read would be improved by editing from a man with top five percent verbal skills, and most human-generated essays I’ve read would improve from an AI critique. The more emotional the argument, the more it needs an AI check.
When I ask AI to summarize the main points of a book, essay or video, it often finds things that I would miss, and it often misses things that I’d find. AI bots consistently find holes in my arguments and I consistently find holes in their arguments. AI faulty logic aligns with what is convenient for the AI businesses.
What I find interesting in books, essays and events is rarely what the AI bots find interesting.
Reality always wins over theory. The reality in 2026 is that the use of AI will improve a man’s writing, and much of the time, it will phrase things better than the man can, and the man will follow incentives and use what AI does better than he does, and imagining that you can legislate against this effectively is living by lies (unless you’re talking about people in the top 1% of intellectual life who have enormous status to lose if they are caught using AI wording).
If you read work you cannot produce, you can learn from it and make it your own. When I’m moved by AI phrasing and argument, I try to figure out why it works and then adapt it for my owns ends.
I tend to use minimal transitions. AI is heavy on transitions. I don’t like to wave a flag before the reader and tell him what to think. AI can’t stop doing this. It must constantly pronounce on what is important and what is worthy and what is decisive, and on this AI is lousy and consistently wrong.
AI and I don’t share hero systems.
On the other hand, the wholesale adoption of AI argument and prose will be inferior to what the good writer can produce.
AI doesn’t know who sat where at kiddush. It can’t call a source. It has no stake in your neighborhood and no memory of what the rabbi said in 2009.
You should not trust my self-interested claims. You should only trust that which rings true.
People did not evolve to be gullible with regard to their vital interests. As long as reality incentivizes them to use AI and then to lie about it, that is what they will do.
A writer who puts the reader first will have less of a problem using a machine’s sentence rather than his own if that better serves the reader. On the other hand, the ease of using AI wording leads to doing more of the easy thing, and a model trained on the median of published English will pull any writer toward the median. The writer’s value comes from the places you sit off it.
