Lex Fridman: preference

Transcript context

…arrier for communication and a portal into understanding the spirit of a people connected by that language. It’s all a weird and beautiful puzzle, and I’m just excited to get the chance to explore it. All right, I got a question on how I prepare for podcasts. This has evolved and expanded more and more over time. There are some podcasts that I prepare hundreds of hours for. In AI terms, let’s say, first, I’m training a solid background model by consuming as much variety on the topic as possible. A lot of this comes down to picking high-signal sources, whether it’s blogs, books, podcasts, YouTube videos, X accounts, and so on. For this conversation with President Zelenskyy, for example, since February 2022 I’ve spoken with hundreds of people on the ground, I’ve read Kindle or audiobook, about 10 books fully, and then I skimmed about 20 more. And I don’t mean books about Zelenskyy, although he does appear in some of them, I mean books where this conversation was fully in the back of my mind as I’m reading the book. For example, I read Red Famine by Anne Applebaum. It’s about Holodomor. Does it directly relate to Zelenskyy? Not on the surface, no. But it continues to weave the fabric of my understanding of a people, of a history of the region. But it’s really important for me to read books from various perspectives. And I’m always trying to calculate the bias under which the author operates and adjusting for that in my brain as I integrate the information. For example, Anne Applebaum’s book, Gulag, is very different from Aleksandr Solzhenitsyn’s Gulag Archipelago. The former is a rigorous, comprehensive, historical account, the latter is a literary, psychological, and personal portrait of Soviet society. Both I think are extremely valuable. On the bias front, for example, The Rise and Fall of the Third Reich by William Shirer is a good example. It is full of bias, but he was there, and to me, he has written probably one of the greatest if not the greatest book on the Third Reich ever. But like I said, it has a lot of inaccuracies and biases. You can read about them online if you like. But my job in this case and in all cases is to adjust based on my understanding of the authors’ biases and take the wisdom from the text where it could be found and putting the inaccuracies aside into the proverbial dustbins of history. ases is to adjust based on my understanding of the authors’ biases and take the wisdom from the text where it could be found and putting the inaccuracies aside into the proverbial dustbins of history. As I’m reading, I’m writing down my thoughts as they come up, always digging for some deeper insight about human nature. If I’m at my computer, I’ll write it down in Google Doc, sometimes use Notion or Obsidian. If I’m not on my computer, I’ll use Google Keep. For example, if I’m listening to an audiobook and I’m running along the river, if a good idea comes to mind, I’ll stop, think for a few seconds, and then do a speech to text note in Google Keep. By the way, listening to audiobook at 1X speed. Old school. And eventually I get a gigantic pile of thoughts and notes that I look over to refresh my memory. But for the most part, I just throw them out. It’s a background model building process. By the way, LLMs are increasingly becoming useful here for organization purposes, but have not yet been useful, at least for me, and I do try a lot for insight extraction or insight generation purposes. I should mention that my memory for specific facts, names, days, quotes is terrible. What I remember well is high-level ideas. That’s just how my brain works, for better or for worse. I realize that sometimes forgetting all of the details and the words needed to express them makes me sound simplistic and even unprepared. I’m not, but that’s life. We have to accept our flaws and roll with them. Aside from books, I also listen to a lot of podcasts and YouTube videos where people are talking about the topic. For the President Zelenskyy episode, I listened probably to hundreds of hours of content from his supporters and from his critics; from all sides. Again, I choose who to listen to based not on their perspective, but based on SNR, signal-to-noise ratio. If I’m regularly getting insights from a person, I will continue listening to them whether I agree or disagree. In the end, this turns out to be a lot of hours of prep, but to say that it’s X hours per episode is not accurate because a lot of this preparation transfers from one guest to another even when there’s an insane level of variety in the guests. We’re all humans, after all. There is a thread that connects all of it together somehow if you look closely enough. For more technical guests in STEM fields, I’ll read papers, a lot of papers, and also technical blog posts and technical tweet threads. This is a very different process. For AI or CS-related topics, I will run other people’s code, I will write my own, implement stuff from scratch. If it’s a software company, I’ll use their tools and software relevant. But in the actual conversation, I constantly am searching for simple but profound insights at various levels of abstraction. Sometimes this means asking a trivial question in hopes of uncovering the non-trivial, counterintuitive but fundamental idea that opens the door to a whole new way of looking at the field. bstraction. Sometimes this means asking a trivial question in hopes of uncovering the non-trivial, counterintuitive but fundamental idea that opens the door to a whole new way of looking at the field. And actually, every guest is their own puzzle. Like preparing for Rick Rubin was me listening to hundreds of songs he produced and even learning some on guitar like Hurt by Johnny Cash. Preparing for the Cursor Team episode meant obviously I had to use Cursor fully for several weeks; all of its features. I switched completely from VS Code to Cursor. For Paul Rosolie, round two especially, I literally went deep into the jungle with Paul and almost died fully taking the leap toward adventure with him. When it gets close to the conversation, I’ll start working on the actual interview questions and notes. And there I’m asking myself what am I personally curious about? I love podcasts. I’m a big fan of many, many podcasts, and so I ask myself, “What would I want this person to explain on a podcast? And maybe what aspect of their thought process or their humanity would I want to be surfaced or have the chance to be surfaced?” In the actual conversation, I always try to put my ego aside completely and do whatever it takes to have a good conversation and serve the listener. This means asking questions, simply trying to define terms and give context if needed, being open-minded, vulnerable, curious, and challenging the guests when needed. Despite the claims on the internet, I do ask a lot of challenging questions, including follow-ups, but always with empathy. I don’t need to be right. I don’t need to signal my moral or intellectual superiority to anyone. I try to do the opposite actually, because I want the guest to open up. And I trust the intelligence of the listener to see for themselves if the guest is full of shit or not, to detect the flaws and the strengths of how the guest thinks or who they are deep down. A lot of times when interviewers grill the guest, it doesn’t reveal much except give a dopamine hit to the echo chambers who hate the guest. As I said in the intro, I believe the line between good and evil does run through the heart of every man. The resulting conversations are sometimes a failure, sometimes because they are too short, sometimes because the chemistry was just not working, sometimes because I fucked it up. I try to take risks, give it everything I got and enjoy the rollercoaster of it all no matter what. And as I said, I trust the listener to put it all together and I trust the critic to tear it apart. And I love you all for it.…

Stored transcript either side of the excerpt. The highlighted words are the published quote; the surrounding text is unedited source, never generated.