How this former NYT columnist uses ChatGPT to brainstorm ideas, do research, and find the perfect metaphor | Farhad Manjoo

[00:00] How do you walk through the process of brainstorming an idea instead of using Google? Right off the bat, it tells me, you know, about the main people in the administration who are talking about this. It gives me links to articles that I can read. This is the stuff that when I was writing a column every week, it would take me probably half a day or so to just find all the stuff and kind of figure out what I was going to write about. I'm presuming in the past you would have done this with colleagues in a newsroom and you could have these conversations live. [00:30] but in many ways it sort of has that same function because the interface is similar. Probably it's not as smart as that person, but it's maybe 80% and it's great and instant and available all the time. - I think there's a lot of fear that chat GPT or AI generated writing is slop and it's all generic. I love seeing this idea of you making the writing more specific and more impactful. - Quickly, I just discovered that it was so useful that now when I write, I have like two windows open [01:00] is the document I'm working on. [01:06] Hey everyone, welcome to How I AI. I'm Claire, product leader and AI obsessive on a mission to help you build better with this new technology. [01:15] Today we're talking about how AI is transforming the writing experience with none other than Farhad Manjou, former columnist for the New York Times and one of the most interesting voices in tech writing out there. Farhad's going to give us practical tips and tricks on how to make our own writing better using AI,

[01:30] And you're definitely not going to want to miss his special word-finding technique to discover that perfect idiom or metaphor. [01:37] Let's get to it. [01:38] This episode is brought to you by Enterprit. Enterprit is a customer intelligence platform used by leading CX and product orgs like Canva, Notion, Strava, Hinge, and Linear to leverage the voice of the customer and build best-in-class products. Enterprit unifies all customer conversations in real time, from gong recordings to Zendesk tickets to Twitter threads, and makes it available for your team for analysis. [02:08] Knowledge Graph that provides the most granular and accurate categorization of all customer feedback [02:14] and connects that feedback to critical metrics like revenue and CSAT. [02:18] If modernizing your Voice of the Customer program to a generational upgrade is a 2025 priority, [02:24] like customer-centric industry leaders Canva, Notion, and Linear, reach out to the team at interpret.com/howiai. That's E-N-T-E-R-P-R-E-T dot com slash howiai. [02:41] Hi Farhad, it's amazing to have you here. I'm super excited to see some of the workflows you use in your writing. But before we get in it, I have to ask, [02:50] As someone who writes for a living, [02:52] What made you curious about these tools versus skeptical? [02:56] I was a columnist at the New York Times for a while, and I was a columnist when ChatGPT came out in 2023, I think it was. And, you know, I just looked at it because everyone was looking at it. And then the first versions were not good enough to kind of help with writing. It was very poor writing. But it quickly got better. And...

[03:18] There was a lot of, just from creative people generally, there's this sense that like AI is a replacement, but I've always been sort of like a. [03:27] earliest adopter of things. And I really noticed, even when it was in its [03:32] you know, infancy and like not great. It could be [03:36] helpful for like [03:39] circumstances that in the past would take me a long time to do in Google. Like, for like, the most basic is just like finding another word like, this is better, the best thesaurus I've ever used, because you can like talk to it about, you know, what things that have meaning. [03:55] soon sort of started to become like a little bit of like a companion. No, it was sort of like, first I would just like consult with it [04:04] I don't know, maybe once or twice when I was writing an article. But quickly I just discovered that it was so useful that now when I write, I have like [04:13] two windows open on my screen one is chat tpt and one is [04:19] the document I'm working on. So let's just get into into the writing and I'd love to go through sort of your step-by-step flow about how you use this [04:27] companion through the whole process. So let's start with brainstorming. How do you [04:32] walk through the process of brainstorming an idea instead of using Google using some of these AI tools. It's become sort of like crucial in the brainstorming stage, especially after they added web search to it. So sort of it now knows like what's on the web. So... [04:47] A very sort of easy thing that I start with is like...

[04:51] Say I was writing an article about Trump's tariffs and I wanted to know, [04:58] Just sort of generally, let's say I was arguing that the terrorists were great. And so I wanted to know, like, [05:05] What's the kind of general consensus in the news about the tariffs? And is there anyone saying they're great because everything I've read is that they're going to cause lots of trouble? So that's kind of a difficult question that in the past I would have just Googled, like spent a lot of time Googling, getting together, you know, [05:27] articles and kind of [05:29] synthesizing after reading a bunch of things. And now, I mean, maybe we could just ask it right now. I just made that example. Could you share your screen? [05:37] Okay, so I've been using the latest version, which is 4.5, which is just really great at writing. Like, it's sort of like the biggest writing improvement I've seen. But it is slow. Like, the earlier ones were just sort of much faster. So I'm going to just switch to 4. If it doesn't work very well, then we could just go to the old one. We'll switch back. The new one. But so then, like, if you turn on Web Search down here and you ask it something like, [06:06] Tell me about like [06:08] All the commentary. [06:10] on Trump's tariffs and. [06:13] Especially any that [06:16] Say the tariffs are good. [06:19] Okay, so right off the bat, it tells me about the kind of main people in the administration who are talking about this. It gives me links to...

[06:29] articles that I can read. This is the stuff that when I was writing a column every week, [06:35] It would take me... [06:36] you know, probably half a day or so to just find all the stuff and kind of figure out what I was going to write about. [06:42] But then now here, I could just kind of interrogate it and ask it for like, is there anyone in the automotive industry? [06:51] who has commented on the tariffs or. And I have a question while this is returning the results, which is. Yeah. Do you find the sources are of equal quality of what you would find if you were doing a Google search? [07:04] good, bad. How are you assessing the quality of the sources here? Since they added web search, they [07:11] put a little link next to all the things that they, next to the source of whatever statement they're making. So for example, like I just asked it, [07:20] Is there anyone who's in the automotive industry who has commented? So it showed me a Business Insider article, a Detroit Free Press article. [07:28] Reuters... [07:29] I generally... [07:31] If you ask it about news stuff, it generally will show you sources that are, you know, kind of well-known news sources. But it also just shows you everything. Like, at the bottom here, you can kind of click, and it shows you all the things it consulted. And, you know, if there's something that seems off, like, you can just check the sources. Yeah. [07:54] You know, initially, when it wasn't sort of [07:58] giving you links, who's telling you how it got this information. It was kind of really dodgy to use it for that kind of thing for brainstorming because you didn't know. And it was also like,

[08:07] There was this [08:08] real problem with like hallucinating where it would just make up stuff and then you wouldn't know where it found that. But now you can really like ask it for sources and then click and find those. And it makes it much faster and not only faster, like you can get kind of deeper into the subjects because you're asking kind of real questions and you're not spending your time kind of just like reading the articles and trying to figure out what's happening. [08:34] Yeah, that was my question, which is it seems like a really effective research tool. [08:39] But it also seems like it could take you on a path where you could actually identify new interesting things to explore or write about. So are you getting that? [08:48] effect by doing this sort of open-ended research? [08:52] The better that it's gotten, the more... [08:55] like deeply, it becomes kind of integrated in my workflow. So before ChatGPT, [09:01] The hardest part about writing an article was kind of figuring out where to start. And now I can just ask it sort of like, what is the most kind of compelling argument or sort of the main points or things that I should kind of highlight? I mean, I would have ideas of what to do that. But then I can ask it and it can suggest some things that I may not have thought of. And then we could talk about those things. And then [09:25] You know, it's not... [09:27] It's not as good as or it's like, it's not as good a writer as like an, an editor, a professional editor that I would. [09:33] work for. But it's [09:36] like as good as like a research assistant who understands uh you know who understands like the material and so you can you can get kind of deeper into it and like it can suggest new ways or new just new things you might not have thought of and um the other thing is it's like

[09:54] doesn't have... It doesn't have, like... [09:58] You don't have to worry about hurting its feelings. You could say, that's dumb. Like, that's a dumb idea or whatever. And, like, you could just have, like, this very... [10:08] kind of free and honest conversation. It doesn't care about like you... [10:12] misspelling stuff so like i type very quickly and like there's lots of misspellings but like it gets the gist of what i want and so it feels very much like you know like chatting with someone um like texting someone rather than kind of talking to like a computer so it's like it's like [10:29] very close to like how I used to talk to like [10:32] my research assistant at the New York Times, like, probably it's not as smart as that person, but, like... [10:38] It's maybe 80% and it's like [10:42] you know. [10:43] great and instant and available all the time. So there are those advantages. I like to say always on eager to please. Like that's one of its competitive advantages. Yep. Yeah. Well, let's actually get into the writing piece. I think this is the most fascinating part, which is how you use these tools to find the [11:01] the right words and the right phrases for when you're working on an article. So can you walk us through a couple examples of that process? [11:08] Okay, so this is based on something real that I was writing and it involved [11:13] Let me piece it in here. So it involves this phrase... [11:17] pay the piper, which is like [11:20] You know, it's an idiom that has a definite meaning, but I didn't want to use that.

[11:24] It's, you know, kind of cliche and people say that all the time. And so I would just... [11:30] Take that. [11:31] And... [11:33] Paste it in here. [11:35] and [11:36] This is something that, like, Google couldn't give me before. Like, you know, you could get... [11:41] a thesaurus, but that's not going to give you sort of like this... [11:46] Like, it's not going to help you search idioms. So, you know, like, these are kind of easy... [11:52] foot the bill, pick up the tab, settle the bill. But it could get, like, a lot deeper than that. Like, I often have these extended conversations with it about, like, just, like, [12:02] weird things in English that we [12:04] we think we know the origin of, but we don't really, or sort of what it means exactly and how those differ and like nuances. So like I had this sentence or something like it, which is involving pay the piper again, which is... [12:17] So for months, the mayor ignored public outrage, [12:20] Let me paste it in here too, so... [12:23] over the polluted lake, eventually he realized he had to pay the piper. And that's just basically like not the correct usage of pay the piper. But it was like the closest I could think of, you know, like, you know, any of those others foot the bill or something. But... [12:38] I wanted something like... [12:40] I want to say this... [12:41] in a catchier way. [12:43] But also... [12:44] with some kind of metaphor that describes like paying for something or [12:51] that like your... [12:53] previous actions

[12:55] Or coming home to roost or something. [12:58] it's just like a very vague idea of the word you want. Um, [13:04] And so then it suggests... [13:06] They gave you the chickens came home to roost. They gave me the chicken one, yeah. Like, the devil came to collect... [13:11] I... [13:12] That's not bad. I might use that. And it's not like something I've heard a lot before. So... [13:18] So this one, the storm he'd been whistling past finally broke. I would be like, that doesn't make any sense because you don't really whistle past a storm. And then we could just have like a conversation about it. This episode is brought to you by Vanta. Building a business? Achieving ISO 42001 compliance shows your customers that you're taking the necessary steps to ensure responsible usage and development of AI. But the process can be time consuming, tedious, and very expensive. [13:48] achieving compliance can be done in a fraction of the time and at a fraction of the cost. 95% of the required document templates are pre-built for you, accelerating the process, helping you demonstrate trustworthy AI practices and scale your business. [14:03] Start with Vanta's free ISO 42001 checklist, which gives you a breakdown of the compliance process and the road ahead. [14:10] Download it at vanta.com slash howiai. That's v-a-n-t-a dot com slash howiai for the free compliance for AI checklist. [14:23] One thing this makes me reflect on that I'm curious your point of view is, you know, you and I are dialing into this from our respective homes. And I'm presuming in the past you would have done this with.

[14:32] colleagues in a newsroom and you could have these conversations live. And I'm imagining it's [14:36] very beneficial to just have this partner [14:39] for you to bounce ideas off and get this cycle on. Is that something that kind of in this [14:44] post-remote world has also been a benefit for you? [14:48] If I was working with colleagues, basically I would talk to them through Slack. So it's essentially like... [14:54] a very similar interface. You know you're not talking to a colleague, you know you're not talking to a human, but [15:00] In many ways, it sort of has that same function because the interface is similar. Like instead of a Slack chat, this is a chat GPT chat. But otherwise, like we could sort of still have that conversation. So then I asked it. [15:12] you know, like, can we fix that storm imagery to make it the [15:16] make it more coherent and it says, you know, it suggests some others like the storm he'd been pretending wasn't coming finally broke. Those are like much better. And [15:26] Like I had this thought [15:28] Like, I think a fear that people have, you hear it from like professors and you hear it from like professional writers and just creative people generally, is that like, [15:37] AI is gonna like replace you? [15:40] And [15:41] That could easily happen. [15:42] But like, [15:44] I find that it speeds up [15:46] A lot of the things that you used to spend a lot of time thinking about, I used to be [15:52] perfectionistic or persnickety about like the specific words I use in a paragraph before I could start writing the next one. And... [15:59] This allows me to just sort of like get to a point where I'm comfortable enough with it. And then I can really spend a lot of time like working on edits to like fix this particular word or sentence. It feels much more like you create a rough draft and like because of this tool, it kind of you form it into like something that you like more often. And it's really like my work, like even if it suggests some of these things, like.

[16:22] It's suggesting ideas, and then I'm... [16:25] I'm thinking about them and integrating them, and I don't feel like [16:30] it's writing for me, which is something that I'm worried about. Like, right. You know, this is like, [16:34] Is this really my work if I [16:37] I tossed it off to like an AI, but it really feels like it's integrated into my writing rather than kind of replacing it. [16:45] Yeah. And what I love about what you're showing us here is I think there's a lot of fear that [16:50] chat GBT or AI generated writing is slop and it's all generic. And I love seeing this idea of you making the writing more specific and more impactful by using these tools instead of less. [17:02] Could we I love this, you know, idiom metaphor seeking [17:05] And you mentioned a thesaurus. Do you do this at the word level, too? Oh, yeah, I do it at the word level all the time. And basically, that's how I started. So, like, what are... [17:15] alternatives. [17:16] to [17:18] Outrage. [17:19] So it gives me a whole bunch. [17:21] And then... [17:22] You know, and this I could have found in a thesaurus, but probably not all of these because they're not like exactly... [17:29] you know, um, linked, like it would be hard to find all of these in a thesaurus. So outrage, like. [17:36] Fuhrer, condemnation, like fuhrer. [17:40] is a good word and i feel like i would have found that in a thesaurus but it's just so much faster and easier to ask this than like [17:47] go on Google, type in the word, find the kind of correct link to like the good thesaurus or whatever. And then if it wasn't quite right,

[17:56] Like, you couldn't get any... You couldn't ask it about, like, other words kind of like it. So it's basically like... [18:02] I... [18:02] You know what? [18:03] super thesaurus, just at the word level. And, um, [18:09] And you can also ask it if, like, your word is, like... [18:13] If you're using a word and you're not quite sure that that word is correct, you can ask it [18:17] if it's correct. And like, ask it sort of the shades of meaning about it. And then, you know, find an alternative if it's not. So in that way, it functions as like, [18:28] I've never, not even a human editor, [18:31] You talk about like, [18:32] the specific words you're writing for, like you would talk, you know, for like a, [18:37] a specific part of the article or when you're editing. But, like, as you're writing, like, getting the right word is, like, something that used to all happen in my head. And now you get a chance to, like, talk it out and then, you know, get a real result at the end, which is, like, a tiny thing. Like, you changed outrage to furor. But, like... [18:56] It used to take me like three minutes or something to figure out like some other word. And now I could do it in... [19:03] you know, 10 seconds. [19:04] What I like about this super thesaurus that I see here is it actually categorizes the words depending on the intent you want to... [19:12] drive forward. So I'm seeing here there's kind of the straight up, you know, synonyms for these words. And then there's what's more dramatic, what's more colloquial, what's more ironic. And that's a really interesting, I'm guessing, surface area to explore against.

[19:26] What if I chose, like, grief? [19:28] from this list, which is [19:30] totally does not work um and i could just ask it does this work [19:35] For months, the mayor ignored public grief over the polluted lake. It should tell me that that's not quite right. [19:41] I think if 4.0 doesn't tell you it's not quite right, 4.5 will definitely tell you. I found it's a slightly more critical reader. [19:50] Yeah, so it says, that's close. Public grief has a mournful, sorrowful tone. More about sorrow than anger. [19:58] So it gives me a way to keep grief in there while changing the sentence correctly. [20:03] slightly. But then it also gives me examples that don't involve grief that like, [20:09] It tells me essentially that that's not quite the right word. So you're able to, you know, [20:15] go over the surface area, find the right words or phrases, [20:20] Use those, integrate them to your own writing. [20:23] And then you're working with an editor, but you also have used... [20:27] tools as a first reader. So what does that process look like? Or what are the things that you want out of [20:33] AI as a first reader that you find really helpful as a writer. [20:37] So the way that I've been using it recently is like... [20:39] I will start writing an article and I'll write maybe like [20:45] I don't know, like six or seven paragraphs, like just like the start of it or so. And I want to know if I'm like... [20:52] heading in the right direction. And that [20:55] You know, I wouldn't have called an editor to ask about that in the past because I'm not done with the article. So I can pass it off like just those six paragraphs or whatever and say, you know, does this get.

[21:06] My point across quickly enough. [21:09] Is there a way you can, like, suggest a way to, like... [21:13] get to this argument like much quicker am i sort of like doing too much like um unnecessary commentary here you know it's just basically like questions about like [21:22] writing structure, it's not going to find like [21:26] logical inconsistencies or something in your argument. Like, I don't think it's that sophisticated, but it will find, like, you know, better ways to say something if you pass it, like, a first version. And basically, so that's what I do. Like, I write... [21:40] Several paragraphs, I pass it to it. I sort of flex... [21:44] get its... [21:45] input, kind of change the article, then I'll write more, then [21:50] I'll [21:51] You know, I'll basically like read the article, read the words by myself, and then sort of pass it off to ChatGPT and... [21:59] Just like work on like polishing after that. So in that sense, it's like a first reader, but it's also like reading while I'm writing it. So it's like even more kind of integrated than like the first person that you would like present the kind of roughest draft to. [22:15] Well, I appreciate you. Now I'm trying to think of the right... [22:19] right metaphor idiom here, you know, raising the curtain or showing us how you do this behind the scenes, because I think something like writing is really mysterious. High quality writing is really mysterious, folks. And I think you've shown us how [22:34] Technology can have a role in that that still allows someone like you with an amazing independent voice to

[22:40] write great stuff that has impact on the public. And I think that's pretty... [22:44] Pretty cool. So I'm going to wrap up with a couple... [22:47] lightning round questions. I have to ask the first one because I've been observing you copying and pasting a bunch. What is one thing if you had a magic wand and you could have a tool that would make this process easier for you? [22:59] you'd love to see. Is there something that you want? [23:02] One of them is it like doesn't have like very good persistent memory. So like if I talk to it [23:07] about something yesterday and then we get back to it and i'm maybe in a different chat and i [23:12] I sort of have to go back and look at that chat and kind of figure things out. And I can't say like, [23:18] Tell me all the things we talked about last week about this article. Another one is, like, I would love if they had... [23:25] like the ability to share the screen so that [23:28] I could just instead of copying and pasting, I could just ask it about like the sentence over here in a different app and. [23:34] It does for some apps, but I don't think it does for like all. So you can, so [23:40] Cursor is this programming app that you can [23:42] connect to it, but like it doesn't for most apps. And so kind of improving that feature would be great because then I wouldn't need to copy and paste and it would sort of know what's [23:51] on my screen at any time. [23:53] You are the first person that I've seen as a true writer of non-technical documents [23:59] show a little snippet into using cursor for writing. So I think that's a really exciting. [24:04] little tidbit you showed us there. Okay. [24:06] My last lightning round question, everyone has a different answer.

[24:10] When AI does not do what you want, it's getting the wrong answer or it's just not responding. [24:16] What is your strategy? Do you cajole? Do you bully? Do you yell? Do you compliment? How do you get AI? [24:23] To get over its own hurdle and do what you want it to. [24:26] Yeah, I find myself being very kind of brusque with it. [24:30] I have, there's this like freedom of saying like, you're totally, like, I would tell a person like you're on the wrong track. Like, let's. [24:37] think about something else, but I can just tell, like, this is a very stupid thing. Please, like, let's talk about something else. Like, and so you could be much more direct with it. And I feel like that really works, like being direct. But sometimes if it's, like, there are lots of things where it just can't help you. And I feel like... [24:55] I have to figure out a place at some point where like, [24:58] we're talking in circles and it's not really like helping me. And then I kind of have to do it without the AI. Well, this has been super interesting to watch. Thank you so much for giving us an honest look into how [25:09] AI is changing and improving the craft of writing. [25:13] Cool. Thank you so much. This is fun to talk about. [25:24] You can also find this podcast on Apple Podcasts, Spotify, or your favorite podcast app. Please consider leaving us a rating and review, which will help others find the show. You can see all our episodes and learn more about the show at howiaipod.com.

[25:41] See you next time.