Dear reader,
All my friends know that I’m obsessed with Readwise, which combines a web highlighter, a powerful read-it-later app, and every feature under the sun to organize and remember everything that I read. I’ve collected over 10,000 highlights there over the last 7 years.1
So I was honored when my friend
invited me to be the first guest on Readwise’s podcast. It’s a bit odd being on the other side of the interview, and more tactical/advice-y than the usual stuff I publish, but I’m resharing it in case there are useful tidbits—especially for any aspiring Substack writers here. Among other things, we discuss:Why Substack built Notes & how the algorithm works
Growth & confidence tips for beginning writers
My research/revision stack (ft. Readwise, Claude, ChatGPT)
Why I’m no longer as worried that AI will steal my job
Watch/listen by clicking above; or on Spotify, YouTube, Apple Podcasts, or Pocket Casts. A full edited transcript is below.
Episode transcript
This transcript has been lightly edited for length & clarity.
Erin Moore (01:31)
I want to start off by talking about your career at Substack, since a lot of Readwise users subscribe to a lot of Substacks, and some of them even have their own.
You joined Substack back in 2021. The pandemic was a really brutal time, especially if you were in a blue collar job, but for a smaller subset of people, myself included, it felt like a slower pace of life with this feeling of creative opportunity and abundance. What were those early days like at Substack for you?
Jasmine Sun (02:14)
The pandemic was interesting because it tore down all the old institutions and forced everybody to live in a world where nothing was normal anymore.
When Covid hit, I was studying abroad at Oxford and they sent everyone home. I did one quarter of Zoom school, then realized that this is not what education was meant to be. So I started looking for other things to do. That was when I started
, the technology publication I work on, with my friend . About six months in, I decided I probably wouldn't return back to school.I then got a job at Substack where I worked on writer development. The company was pretty small at the time—less than 30 people. I spent a lot of time coaching emerging writers on publication strategy and developing our first playbooks for how to succeed. But I was also doing product copywriting and user interviews and all sorts of other stuff—at a company that small, everyone does a bit of everything.
I was really sad to see how the pandemic and other forces had affected journalism's business model, and Substack felt like a glimmer of hope. It was one of the few places that was not just avoiding destruction, but making something new out of the chaos of the old world. It provided an outlet where a lot of people were able to be creative and make a living, including many people who would not have succeeded in traditional models. It was very inspiring to work on that.
Erin Moore (04:16)
Do you have a background in journalism? What led you to applying at Substack as a place to work?
Jasmine Sun (04:22)
I've never been a proper journalist; I only started doing my first reported stories in the last few months.
But I've always been a big reader and writer—since I was little, I had blogs on Medium and WordPress and all these sites. I would write long Reddit posts. I was more of an anonymous internet blogger than a proper writer. It's what I love about the internet. You don't need credentials, you don't need an institution; you can be a random kid in a suburb saying your opinions, and if they were compelling and you did your research, maybe somebody would actually take a look.
During college, I wrote an opinion column for the school paper. I started meeting journalists through Twitter because I would post the articles there. Sometimes a real journalist would actually read my writing and engage with me. I thought that was the most special experience in the world.
The evolution of Substack Notes
Erin Moore (05:33)
In the five years since Substack really took off, the platform has gone through a lot of changes. I'm mostly thinking about the powerful recommendations feature and the launch of Substack Notes.
spoke recently on the podcast about this idea of “Substack fatigue.” He explained how so many users now on the Substack platform are subscribed to more newsletters than they can possibly get through, and one of the most common moves a tech company at this stage will make is to implement an algorithm that can sift the signal from the noise.I've also seen conversations on indie writing platforms where people feel disenchanted with Substack for the move they've made with Notes. They fear that Substack is just going to become the next social media feed, like Twitter, and wonder how the algorithm will change the way in which people engage with and publish on Substack.
Jasmine Sun (06:47)
You're right, Substack has changed a lot. When I joined, we didn't have video. We didn't have Notes. We didn't have an app. We didn't have recommendations. It was like email, Stripe, and a website duct-taped together. And that was the service.
I personally feel really good about Substack's evolution over time. I'll also mention that these changes were planned from the beginning. In my very first conversation with CEO
in 2020, when I was interviewing for the job and asking about his vision, it was already the case that he wanted to have social media as part of the experience. I personally would not have joined Substack if it was going to be an email tool forever. I've always been very interested in how to design social media platforms in a way that's healthy and pro-social.Erin Moore (07:43)
When you say pro-social, what does that mean?
Jasmine Sun (07:45)
When people look at Twitter, for example, it feels like an anti-social environment. Saying something outrageous and being polarizing is what's going to go viral.
But I don't think there's anything wrong with the idea that people should be able to go on the internet and say things and connect with others. That has been one of the greatest gifts of my life. That's how you and I met, right? The question really is: How do you create a space where people can connect with others in a positive, healthy, and thoughtful way? How can you debate and disagree, but in a civil way instead of a trolly way?
There's nothing wrong with social media, but there is something wrong with the incentives that undergird a platform like Twitter. Twitter makes more money when you get more views; you get more views when you say something more extreme.
Erin Moore (08:50)
I think a lot of people had such a strong reaction to the launch of Notes because they saw it as Substack cargo-culting onto this model that was already very well known by Twitter. What is the Substack Notes algorithm actually optimizing for in terms of what gets into your feed and what gets prioritized?
Jasmine Sun (08:58)
Substack's business model is designed in a very simple way where the company only makes money when writers are paid subscriptions. There are no advertisers. There are no other stakeholders. Which means that Substack would never make a product decision that gains short-term views, but actually decreases the amount of money that writers are making.
I don't work there anymore, so I don't know how the algorithm works now, but I can give one example from when I did.2 Let's say that I restack a post; then someone sees my restack, clicks on the post, reads it, and subscribes to the writer. That's a really strong signal—both for the restacker of the post, because they're a great curator, but also for the person who wrote the post. That's the kind of thing we would prioritize—this note made someone click and read a longform post.
It's the opposite of how every other social media platform works right now, where if you share a link you're going to get crushed. That's the other reason that Substack built Notes. Long term, it's not sustainable for writers or creators to be dependent on networks like Twitter, LinkedIn, and Instagram that are trying to crush external links. We had to build our own discovery system because it was no longer reliable that writers could distribute somewhere else. Discovery is something that every writer needs.
Erin Moore (10:28)
That's something we are grappling with right now at Readwise. Over the last six months, the drop in engagement on anything with a link, even our full beta updates, has been quite disappointing. I can imagine it's even more crushing for someone who's a creator.
Jasmine Sun (10:44)
Right, exactly. I can understand why people feel jarred by the shift that Substack made. But the question I always ask is: What's the counterfactual in a world where Substack had not created Notes? The counterfactual is that writers and creators would see decreasing traffic over time as every other platform crushes external links. That means less of an ability to find your people. That means less of an ability to make an income.
Right now, I am a full-time writer and my income relies on writing. I'm really grateful to have a platform to promote my work that actually wants me to win.
Growth tips for Substack writers
Erin Moore (11:26)
I know that we have quite a few Readwise users who have ambitions of starting a newsletter or, maybe they wish they had started something way back in 2020. Have the product changes at Substack fundamentally changed how someone who wants to start a Substack should begin?
Jasmine Sun (11:49)
I think so. Cross-promotion within the Substack network has become really powerful.
Before, when Substack started and people asked how to grow their newsletter, I would give them tips on how to be good at Twitter, or how to syndicate with mainstream publications. Those are still good ideas. But that was also hard to tell someone who didn't have a Twitter account or a base of editor relationships at publications.
Now, when people ask me how to get started, I say to leverage the existing Substack network. Readers who use the Substack app—readers already on Substack—are far more likely to subscribe to more Substacks. It's the opposite of the “subscription fatigue” effect, which we've tried to study internally. What we've actually found is if you read one thing, you're more likely to read other things. If you pay for one thing, you're more likely to pay for more Substacks. It's the opposite of what people might expect.
The specific three tactics that I tell people are to use Recommendations, Notes, and guest posting. First, identify the other writers on Substack whose audiences you think would be interested in your work, whose audiences you want for yourself, or who are the people you admire in your field.
Then, turn on the Recommendations feature. It's totally passive; you don't have to do any work. You just pick 10 newsletters you love. You click recommend. It'll start asking your subscribers if they're interested in those other publications. And it also encourages those publications to recommend you back automatically. Then you have this positive sum ecosystem where people are sharing subscriptions with each other at no cost to themselves.
The Notes feed now generates as many new subscriptions as recommendations does. It might be more now, because that was when I was there, and the trend was that Notes was taking a bigger share of the pie. When you share your post to Notes, or when you engage with people who restack your posts, that drives a ton of readers to discover your work and subscribe to you. For example, when any of your readers likes your post, that will show up in the Notes feed for their followers and subscribers too. It's not constricted to your immediate network. For example, if you liked that Jackson Dahl podcast, I follow you, so I would see that in my feed: “Erin liked so-and-so.” My feed is full of what my friends are reading and liking now, which makes me want to read the same things.
Finally, asking bigger publications if you can guest post for them is really powerful. Oftentimes, once a writer has built an audience, they're starting to think, “I don't know if I can generate as much fresh content”—yet they have this huge audience that wants to read more. If you write a guest post using Substack's byline feature, they get to publish something for their reader, you get to reach their entire audience, and at the bottom of the post, it'll say “This is a guest post from Erin. Do you want to subscribe?”
Erin Moore (15:48)
I feel like I've discovered quite a few writers specifically through that byline feature.
Jasmine Sun (15:54)
It works really well. I've done it a few times and gotten a bunch of subscribers. You don't even need to write something brand new for that publication: if you work out the arrangement, you can take a post you've already written for your own Substack and have that person republish it with minimal edits. The
newsletter does a good job of this—it helps them promote smaller Substackers and drive them subscribers, but they also get a bunch of great content.Erin Moore (16:28)
You wrote this really beautiful essay that you titled “Exit Interview” on your personal Substack, talking about your reflections on your four-year career there. There's this one line that I love so much. You say:
I joke constantly about the uselessness of product managers, both before and after I became one. I harbored a default suspicion of grinning CEOs who described the workplace as a family or preach making the world a better place. I saw myself entering Substack as an undercover sociologist, not a loyal worker bee.
What has been your favorite sociological observation that you made at Substack?
Jasmine Sun (17:08)
There's so many things. One example that has to do with Notes is that for a while, there was a theory that Substack boosted Notes about how to grow on Substack. People would say that Substack is trying to self-promote our platform, so the algorithm boosts notes about getting your first 1000 subscribers or whatever.
The crazy thing about this is that not only is it false, but I think the opposite is true. We have tried to de-boost those because we find it annoying. Growth hacker discourse is not what we want Notes to be. But writers like that content. Everyone says they hate it, but keep clicking in and reading about how to get your first 1000 subscribers.
Erin Moore (18:03)
I have a confession to make. Last year when Notes first came out, I actually signed up for a course about how to hack it on Substack Notes, because I was thinking about starting a Substack for so long. But on Notes, I would see people saying “I'm afraid I'm just going to come on here and see a bunch of threadbois posting quick fixes to grow your audience to 10K overnight.” It's a very reasonable concern to have. Now when I go onto Twitter, that's the only content I see, and 99% of it is from people I don't follow. It's interesting how much you guys have anticipated that and tried to add in some barriers.
Jasmine Sun (19:02)
But that's the thing—there's only so much the platform can do, because that is what people want. It's a very human desire: everyone is anxious about their growth and success. Like I feel anxious about my own growth, so when I see writing advice, I sometimes click on it too.
But because people don't want to recognize that desire in themselves, they blame the platform. They say that the Substack algorithm is boosting this content. People externalize and project their impulses and make the company the boogeyman: It's their fault, and they're doing it because they're trying to make more money. And I knew from the inside that we weren't doing that. We were doing the opposite. That was a very interesting thing to notice about the way that people perceive technology companies.
A lot of technology is a mirror. Tech companies are just the market, right? If there's a demand, they'll meet it; if there's no demand, they won't. I don't think that's good for society in every instance, but I think about the news stories about politicians who say we have to ban TikTok because it's all teen girls dancing. But like, my TikTok feed is not teen girls dancing because I don't want to see that content. But if you do want to see that and linger just a second longer on those videos, that's what it's going to show you.
People don't like the way that technology holds a mirror up—not only to their desires, but to our most base, impulsive, short-term desires. Then the question is: How do you build products that give people not just what they want in the moment—how to succeed on Substack, or a very dopamine-triggering video—but that aligns people with their long-term desires to be healthier, smarter, and all of these deeper things?
How I started writing
Erin Moore (20:56)
A few months ago, you made the announcement that you were leaving Substack to pursue your own career as an independent writer. How did you know it was the right time to go out on your own?
Jasmine Sun (21:10)
I don't think it was a sudden realization. I always saw myself as a writer in addition to a technologist. In 2024, I started taking writing just a tiny bit more seriously. I tried to publish something every month. I joined a couple writing groups in San Francisco for accountability and community-building. I wanted to pitch and land a piece in a “real” publication. I was already trying to exercise that creative muscle more.
But I ran up against the barrier of time. I loved my job at Substack, but it was pretty demanding. I was PMing our Core Product team by the end; it was the app and chat and podcasts and video and publishing tools and so many different things. It was meaningful because of the mission, but after work every day, the idea of reading a dense book or writing an essay was impossible to me. I tried really hard to block out time for writing, but there can only be one top thing in your mind at any point. In prioritizing my job and letting that be the thing that occupies my brain space when I'm idling, that made it hard to be creative. I realized that I could only write when I was on vacation. I took a two-week trip to China, and I was able to write there because I had finally undone the part of my brain that was so attached to the day-to-day chaos of work.
It became pretty clear that if I wanted to take my writing more seriously, if I wanted to move up one more step, I was going to have to put aside the demanding job. And after spending about four years at Substack, that was a run I was proud of. It didn't feel like I was leaving too soon. I had learned a ton and built a bunch of really amazing relationships. So I was like, okay, the next phase is here. I know I'm going to be unhappy if I don't bet on myself for a little bit.
Erin Moore (23:44)
I'm really glad you did. I've noticed you've been publishing more frequently on your own Substack, and it's always a delight to read. I learn so much whenever I read your work.
On that note, you've been writing on Substack since 2020. Where did you find the confidence as someone who is so young to go online and start sharing your thoughts?
Jasmine Sun (24:12)
In the beginning, I wasn't trying to be read. I often wrote anonymously on blogs and wouldn't tell anyone I knew in real life. That helped right because it didn't require that much confidence—I was like, well, this will never come back to me. It was only after things I wrote were getting any traction at all, without attaching my name, that made me think, actually maybe people do want to read this. That's when I started writing more under my own name.
Erin Moore (25:05)
So many people talk about all the horrors that come with being able to start a random profile that's not connected to your face or real-life persona. People usually talk about that dynamic from a trolling perspective. I know you've shared some tweets of people trolling articles you've shared. I've definitely experienced it as a woman online with videos. I never really stopped to think about the flip side of that coin. Sometimes, anonymity can actually help strengthen your real voice and inspire this sense of creative freedom in a positive and beautiful way.
Jasmine Sun (25:49)
It's a double-edged sword, but if the trolls are gonna do it, we might as well too.
Also, a lot of folks tell me, “I really want to write more, but I'm scared. I don't know enough. I'm not an expert. I need to do more research. What if I say something and it's wrong?” But blogging is different from a publication because it's much more like starting a conversation. When I put an essay on my Substack, I'm not saying this is the definitive thing I'll believe forever, or that this is the capital-T Truth. It's much more like, here's how I'm thinking about a topic right now, and I'm inviting my readers to engage and help me develop those beliefs. If I get critiqued or I change my mind, that doesn't feel like a big deal. It's a big deal to get facts wrong in the New York Times, of course, but blogging is a dialogue.
Erin Moore (26:56)
I often get caught up on trying to think of the unifying theme of a Substack if I start one. There's so many good ones out there. For someone who has a lot of interests, how did you decide what your Substack was going to be about?
Jasmine Sun (27:35)
I still don't think I have a theme. When I told everyone I was quitting my job to start a Substack, people would ask, “What's your Substack called?” And I'd explain that it doesn't have a name; it's just under my own name. The reason is because I don't want to commit to a beat, to only writing about one thing.
Here's what I always told writers when working with them: people are not subscribing to download floating units of content and information. You can get that from the New York Times or something. People go to Substack for a new lens on the world. Every writer has a distinct way that they see and process the world—the things that grab their attention, the ways they analyze or synthesize—newsletters are great for conveying that ineffable stuff. It's taste, subjective judgment, curation, right? There are writers whose work I would read about anything; I don't really care if it's parenting or AI or philosophy3—these are all interesting because of the person who's writing it. A lot of the best Substacks are about a lot of things.
For me, I wanted a space to be creative and expressive more than I needed to build a concrete, revenue-generating business right this moment. For anyone who is prioritizing creative work, literally write whatever you want in whatever form you want, and your people will find you. I wouldn't worry so much about sticking to one topic because a great writer can make anything interesting.
Erin Moore (29:32)
I think a lot of people will find that comforting. I see a lot of people, and in myself too, trying to figure out “What is my thing that I want to speak about?”
There's some very consistent stylistic things that happen in your writing. You make a lot of references to things happening online—trend observations, and subcultural memes that could only really be picked up by someone who is chronically online—but you're also writing consistently. How do you balance being so online, especially writing about technology, with finding time to actually sit down, reflect, and write?
Jasmine Sun (30:15)
I don't know that I'd give anyone advice because I am really online. I do spend too much time on the internet. I do need to touch grass more.
But this has been interesting to figure out. Before this year, I was not writing consistently. Part of that was having a job, but part of it was figuring out where to make space in my schedule to be creative. When I was a PM, I could write marketing copy or prep for a meeting in 15 or 30 minute blocks. But I can't write an essay in a 30 minute block. I've met people who can—I think they're insane and built different and I deeply respect that—but I could never, right?
In my writing process, there's collecting mode and then there's writing mode. What I've realized is that collecting takes a really long time. It's really important, but it can happen kind of in bits and pieces over weeks or months. I will literally have an Apple Note where I drop bullet point thoughts of ideas that feel vaguely related—a book I read, a quote I heard, something I saw on the internet. I'll have like 20 bullet points that don't make sense in a note that I'm collecting for two months. But I think it's important to always be passively collecting and curating. Meanwhile, I'm assuming that somewhere in the back of my mind, the ideas are connecting together.
One day, I'll decide that it's time to write. That's when I go into lockdown—I block out like two eight-hour blocks in my schedule. And I'll force myself to have a draft by the end of the day. It doesn't have to be like a good draft, but I know you can be in collecting mode forever. So sometimes you just have to put a deadline on it.
Since I mostly publish on Friday mornings, I don't do anything on Thursdays right now.4 I can float around the rest of the week—do coffee chats and see my friends and sit around reading in the park—but every Thursday, I make zero plans. I won't get coffee, I won't get dinner. My only task for the day is to take all my notes and highlights and turn them into a post.
Erin Moore (33:03)
It's that fast of a turnaround? Like you sit down and start the thing, knowing that the next day you're going to have to hit publish.
Jasmine Sun (33:10)
This year, because I'm trying to go for a weekly cadence, I do a lot of Thursday grinds. That only works if I have been thinking about the idea for a long time, if I have that research already collected. If I start thinking about it on Thursday, it's not going to be a good post.
My research and revision stack
Erin Moore (33:36)
How do you use your Readwise highlights in your writing workflow? Do you typically have an idea for a post, and then hunt for highlights connected to that? Or do you typically aggregate highlights together, and the core idea for a piece emerges more naturally from those disparate highlights?
Jasmine Sun (33:58)
I'm always highlighting as I read. On Kindle, on Twitter, on articles. I'm highlighting all the time. Highlighting is my way of taking notes. I know that some people take notes while they read. I admire that, but it's so much work and it's so slow. For me, highlighting helps me create outlines of everything I read.
I'll often have an idea for an essay and know that it's related to a book I read, but I won't remember what was in the book. Instead of rereading the whole book, I have 40 Readwise highlights from that book, that when I read in order, form an outline of the book. They show exactly what I found interesting.
More recently, I started using Readwise Chat. Sometimes I don't remember what books I've read, especially now that I've been reading and using Readwise for a long time. During the early brainstorming exploratory phase, I put the general gist of an idea into Readwise Chat and ask if I've read any books about like this topic before, and it will surface highlights from something I read years ago that I totally forgot about. The chat feature helps me make connections that I would have forgotten.
Erin Moore (36:09)
When we first launched Readwise Chat, you emailed one of our co-founders and me feedback on the chat. This chat that you had actually ended up becoming this article that got published, and then you ended up going onto NPR Marketplace to talk about it. Would you be open to sharing the chat for the article?
Jasmine Sun (36:32)
Yeah, so I was starting a piece about the defense-tech revival at Stanford. When I was in college at Stanford, I read a lot of history books about the history of Silicon Valley and how the DoD played a big role in funding the early labs and technology breakthroughs at the university. But I read these books a long time ago and forgot what was in them.
So I queried my highlights to say, I'm writing an essay about this topic, is there any historical context that I should bring in? And I got this good summary of just the influence of the Cold War on industry and academia, et cetera. Then I would go into the relevant highlights and scroll through titles like Palo Alto and What Tech Calls Thinking. Then I went to my actual Readwise database and reread my highlights to refresh my mind about the context in those books.
Erin Moore (37:48)
It was really fun to see the genesis of this essay, and then for it to show up in the real world, get picked up, and inspire so many other conversations.
Jasmine Sun (37:57)
It was an interesting piece to write because it ended up being viral on campus and quite controversial in a way that I didn't expect. But it was definitely meaningful for one's work to actually do something in the world. My impression was that the article really did reinvigorate a campus conversation over what kinds of careers students wanted to go into, and the ethics of working with the US military. I was happy to see that conversation happen.
Erin Moore (38:34)
In addition to Readwise Chat, are there any other AI tools that you use for creating and publishing?
Jasmine Sun (38:41)
I use AI a lot now, which feels taboo to say as a writer. But I feel like people should be more open and know the capabilities of the tools. I use AI in every part of my process now.
In the very early phases, it's similar to Readwise Chat. I'm using AI to generate essentially Wikipedia pages or overviews on topics I'm interested in. For example, I was interested in the idea of political culture because I was in Taiwan and thinking a lot about democracy there. But it's been a hot second since I've taken a college political philosophy class. So I told ChatGPT Deep Research something like: “I'm really interested in political culture, political philosophy, and democratic theory because I want to write an essay about Taiwan. Can you build me a syllabus of 10-15 core foundational texts that I should read to build my basis in this discipline?” It did that, and it was really helpful.
Erin Moore (39:46)
And you actually ask it to compile links for you?
Jasmine Sun (39:48)
Yeah, like, here's my essay, please give me a bunch of background reading to do. Give me some homework!
And sometimes there are nuances that I don't understand when I'm reading. Let's say I don't remember what deliberative democracy is, or that I notice Jürgen Habermas's name keeps coming up. I will have a ChatGPT window open where I can ask: What's deliberative democracy and who came up with it? Can you explain who Habermas is, but also how he differs from Chantal Mouffe's theory of democracy?
I could be doing these things with Google, but ChatGPT is faster, so I do it more. The counterfactual is that I see that phrase and don't look it up at all. AI makes it easy to learn as I'm reading.
Erin Moore (41:05)
It's interesting that you're using ChatGPT to cull together what you called a syllabus rather than what a lot of people do, which is running a Deep Research search to get this super lengthy compiled study.
Jasmine Sun (41:29)
I use Deep Research, but there's just so much value in reading the original texts. It's like people who read summaries of books. You lose so much. There are all these decisions that you make when you're summarizing, choosing what details get included and what details don't, and the nuance of what makes a specific writer interesting gets lost when compressed into an AI summary. If I want to understand something super deeply, I need to read the original text. AI helps give pointers to those texts. But the act of poring through a text and noticing what examples call out to you and what you find interesting—that's really important.
For issues that I don't need to understand at that level of depth, I'll take the summary at its word. But in my Deep Research query I still say things like “Prioritize peer reviewed articles and then journalistic articles, but don't give me low-quality sources and random blogs.”
I also mentioned NotebookLM. One thing I like about Notebook is that it ties its answers to the exact quote in the original document you upload so that you can understand the context. It's really only good for that; in general, it's not the best product. But for example, one time I was reading policy reports about immigration and AI talent. I wanted to know what percentage of researchers at top labs were Chinese nationals. I had all these hundred-page reports, and I knew the answers were in there somewhere, but didn't want to read the whole thing. So I put the PDFs and questions in NotebookLM. It would answer, but more importantly, it would send me to the exact quoted text where the statistic was given. Then you can check what the AI says against the nuance of the original framing. Otherwise you may end up misstating a fact if you don't read the original.
Erin Moore (43:50)
I really like the idea of still reading the original but using AI to point you directly towards the area you need to reference, especially for these intense research and academic PDFs.
Jasmine Sun (44:01)
For people who are anti-AI in all cases, they don't realize the alternative is often me not reading the hundred page report at all.
Erin Moore (44:14)
How do you use Claude in your writing workflow?
Jasmine Sun (44:19)
I think Claude is the best with prose, so I use Claude for revision. It's not as good at research because it didn't have web search until recently. But I give Claude most of my final drafts of essays before I publish. I can pull one up.
This was probably written at like 2am, which is why I said I have one hour to edit this essay. I want feedback, but not too much. In the past, when I asked Claude for writing feedback, sometimes it'd tell me to rewrite my entire essay. And I'm just like, this is unhelpful and depressing because now I'm just thinking about how I don't have time.
I tell it to act as an experienced nonfiction editor. I want it to provide three strengths because I want to feel encouraged, and I can't have it only flaming my work. Then I ask for five focused improvements, with the comments placed in context of the essay. Then I paste in my essay.
Erin Moore (46:12)
The bolding format is very useful because I've experimented mostly with ChatGPT and it's just this long block of text. I like this format that you're using.
Jasmine Sun (46:28)
Yeah, I like having a separation between my work and the AI's feedback. I don't want it to get mixed together.
I don't have to take the AI suggestions either, obviously. I probably take around 30%. In this context, Claude said to consider adding a bridge sentence. So I said, “Can you give me one?” In most cases, I don't copy-paste the exact suggestion that they give me. But it helps unblock me, because it's easier to work from revising something else than a blank page. Even if I'm not using the exact thing Claude gives me, which I don't think I did in here, it gets my brain moving.
AI and the future of writing
Erin Moore (47:31)
You've written quite extensively on AI. In fact, you even co-authored a paper with Mozilla about how AI can be used to benefit the larger public and not just a handful of corporations.
It also seems like maybe this wasn't always the case. In one of your articles, you acknowledged that you used to be somewhat of an AI skeptic. You were worried about ChatGPT jeopardizing your career as a writer, and so many writers and artists do see AI as an existential threat to their craft. What changed your sentiment about AI?
Jasmine Sun (48:08)
Using it more, to be honest. It's easy to reflexively critique these technologies, but when you use them in your workflow, you realize where it can help you and also where it doesn't—where I still feel like my own taste and judgment and creative skills are valuable. It's easier to have an extreme view like “It's coming for all of our jobs” or “It's shit and can't do anything” when you don't use the technology.
Over the last year I started forcing myself to use AI as much as possible because I knew I wanted to write about it. I didn't want to write about AI without having a personal sense of where it works and doesn't. That's my way of assessing the capabilities. Yes, there are evals that get published about the hard math problems it can do, but I know writing pretty well, so my way of measuring the state of AI progress is using it for that.
Also, I don't feel like I’m less creative when NotebookLM tells me which part of this hundred page report is relevant. I feel like I’m doing better research—I’m being more rigorous than I would be in a world without AI. But I also notice where Claude’s sentence suggestions are super generic, or when it takes something interesting and quirky I wrote, and says it’s kind of weird. And I'm like, you know what Claude? You're wrong about that because you don't understand art.
Erin Moore (49:45)
AI has no taste, right?
Jasmine Sun (49:50)
Exactly. Using AI also shows me where my human side is still valuable. You mentioned earlier that my writing often weaves in internet references, and Claude actually hates this. Every time I have a piece of casual language or slang, Claude will say “This isn't very befitting of your serious subject matter.” And I’m like, Claude, you don't understand anything about why people read my work! It’s a bit reassuring because I know the thing that I deliver as a writer isn’t something that AI does. I know that my specific audience wants my internet references. Hopefully that personal lens is at least what some people come to my work for.
I don't know. I think AI will change writing a lot. It will change creative work. Certain jobs will be eliminated, right? If one’s work was writing standard issue 500 word articles about the results of a sports game every day, that stuff will probably be automated—it's already being automated. But for the writing that I find interesting and want to do, I think AI makes my work faster and better.
Erin Moore (51:13)
What type of writing do you think will come to be most valuable in the future where AI is automating so many other tasks?
Jasmine Sun (51:20)
One is deep reporting and investigation. When I’m writing reported pieces, I interview people, and I have to build trust so that they trust me with their words. The task of reporters—building relationships with sources who might feel nervous talking to a journalist, getting them to believe that you’ll tell their story in a good faith way—I don’t think AI can do that. Reporting is a very human task.
The other thing is more quirky personal stuff, things that aren’t the Wikipedia-ized versions. For me, I use internet references or whatever, but for somebody else, they might write in a style that is grammatically incorrect, but is playful and interesting and engaging. Or personal essays that relate personal experiences to broad phenomena. Claude does not know your individual experience, but what often helps people understand topics is taking things like AI or politics and making them personal. That's something AI can't do, and people are always going to want that.
Relationships and connection and the creativity that comes from that will be pretty AI-proof.
Erin Moore (52:50)
Before you go, what's been your most highlighted book from this past year?
Jasmine Sun (52:54)
I read the book Liquidated by Karen Ho. She's an academic anthropologist who wrote an anthropology of Wall Street. She worked at an investment bank for like six months, and also did a bunch of traditional anthropological interviews with colleagues and people at other banks to understand the cultural practices of Wall Street. She related, for example, how banks would recommend layoffs at firms to how bankers themselves are constantly getting laid off and rehired within Wall Street. Bankers are not only used to precarity, but they like that it’s a survival of the fittest, and you have to be adaptive to change—then they impose that worldview onto everyone else. She shows this through deep interviews and personal experience, which I thought was such an interesting approach.
Erin Moore (54:11)
I'm gonna have to read that. I'm a huge fan of Barbara Ehrenreich's work, and she was well known for doing those extended studies of living alongside the people she was doing research with.
Thanks so much for joining me. This was a lot of fun!
Jasmine Sun (54:30)
Thank you for having me. And I love Readwise, and couldn't imagine reading or writing without it.
Thanks for reading or listening,
Jasmine
This is not sponsored by Readwise; I just really like the tool! (But if y’all do want to sponsor me… let me know.)
This got edited out of the video, but it's an important disclaimer, so I'm adding it back to the transcript.
I've moved my focus day to Wednesday and my publish day to Thursday due to schedule changes, not that it matters!
Share this post