Exclusive Analysis of ‘How Do You Use ChatGPT?’ With Anne-Laure Le Cunff

My take on the latest episode of How Do You Use ChatGPT? with Anne-Laure Le Cunff and the episode transcript is below for paying subscribers.

I’ll admit I was a bit surprised that Anne-Laure told me that she could only accomplish so much with ChatGPT. It seemed too good to be true, but I swear I didn’t pay her to say so—she just came right out with it.

We covered so much good stuff in this interview, but my favorite part of it was the end. Anne-Laure has been establishing a new meditation practice, and we used ChatGPT to coach her through how to improve it. We fed it a detailed log of her practice and how she’s been feeling that she’s been keeping. Then we did a trick I’ve been meaning to do on the show for a while: We asked it what kind of expert could help her improve the most.

It’s commonly known that ChatGPT gets better when you tell it to simulate being a particular kind of expert when it responds to you. But what people don’t tend to know is that you can ask ChatGPT which expert it thinks would be best—and then use the expert it suggests.

In this case, ChatGPT suggested that Anne-Laure chat with a physiologist (a healthcare professional focused on the functioning of the human body) to help her improve her practice. As a physiologist, it noted several ways she could make herself more physically comfortable as she meditates, like using a cushion and getting a white noise machine—things she wouldn’t have explored on her own.

The advice it gave her was fairly simple, but straightforward advice from a perspective we might not have considered is often the most valuable kind. If you’re interested in improving your use of ChatGPT, try asking it what kind of expert could solve your problem. It’ll make it significantly more helpful.

Timestamps:

Introduction 01:10
How to use ChatGPT to save time 02:11
Tips to breakdown research papers with ChatGPT 05:17
How to use ChatGPT to generate explanations tailored for you 09:38
Leveraging ChatGPT to find hidden gems on the internet (like recipes for obscure cheese) 19:51
How to create awesome YouTube thumbnails with ChatGPT 33:47
Incorporating ChatGPT into your writing process 51:13
Rapid-fire questions from X 56:52
Surfacing useful insights from Anne-Laure’s meditation journal 1:13:01
The case for journaling in the age of AI 1:29:04

Transcript

Anne-Laure Le Cunff (00:00:00)

I’m doing a Ph.D., writing a book, running a business, I have a team to manage, having to recruit someone and writing the job description, admin stuff, even grant proposals.

Honestly, I don't think I would be able to do everything I'm doing at the moment if ChatGPT didn't exist.

It's saving me so much time and several times a week, I'm thanking the creators of ChatGPT for making my current life possible.

Dan Shipper (00:00:35)

Hey, I want to just take a quick break from this episode to tell you about a podcast you should know about. It’s hosted by Nathan Labenz, who was actually on this show and did one of the most popular episodes we’ve ever had about how to use AI as a copilot for the mind.

Nathan hosts a podcast called The Cognitive Revolution, where he interviews builders at the edge of AI and discusses the profound shifts that it will unlock over the next decade. He talks to the researchers and entrepreneurs who are shaping this moment, like Logan Kilpatrick of OpenAI, who I interviewed for this show as well, Eugenia Kuyda of Replika, and Suhail Doshi of Playground.

To listen to the show, search Cognitive Revolution in your podcast player. Alright, now let’s get back to How Do You Use ChatGPT?

Dan Shipper (00:01:10)

Anne-Laure, welcome to the show.

Anne-Laure Le Cunff (00:01:12)

Thanks so much for having me.

Dan Shipper (00:01:14)

Of course. Yeah. It's so good to have you. We've been friends for a while. We've been following each other on our internet newsletter productivity journeys for a while, and it's really great to get to sit down.

For people who don't know you, you are a neuroscientist, a writer and an entrepreneur. You run an excellent newsletter and community, of which I am a subscriber, called Ness Labs. You're also pursuing a Ph.D. in neuroscience. And you're writing a book, so you are officially—I think I'm busy. You're officially the busiest person that I've ever met, and I'm always so impressed with everything you do. It's so amazing.

Anne-Laure Le Cunff (00:01:50)

Thank you. Yeah. I have lots of different interests, so I am a little bit busy at the moment.

Dan Shipper (00:01:56)

Maybe we should start there. How do you get so much stuff done in a day?

Anne-Laure Le Cunff (00:02:02)

Honestly, this is very relevant to this show, and I swear I'm not making it up just to make it relevant. But I don't think I would be able to do everything I'm doing at the moment if ChatGPT didn't exist. As you know, you mentioned I'm doing a Ph.D., writing a book, running a business. I have a team to manage, a bunch of things to do. And there are so many things that would have taken me hours before, and that now take me maybe 10 minutes. That adds up, right? Just all of this time that I'm saving—boring stuff sometimes like having to recruit someone and writing the job description, it gets 80 percent of the way there by using ChatGPT, admin stuff, even grant proposals. Sometimes I use it for that. So it's saving me so much time and several times a week, I'm thanking the creators of ChatGPT for making my current life possible.

Dan Shipper (00:03:03)

That's so funny. I swear this was not discussed beforehand. I didn't pay you to say that, but it makes me excited to hear that. I think it's just great that it is furthering your creative life. Cause A, I think that's a key thing that I feel about it that I want to bring to the people who listen to this. And so it seems like you feel that too. And then B, I'm just excited to see how you use it because maybe I'll pick up some things that I can use—maybe I'll get a Ph.D. or something, you know?

Anne-Laure Le Cunff (00:03:42)

Why not?

Dan Shipper (00:03:43)

So maybe we can dive into that. Maybe we can dive into a little bit of some of your use cases for it. Do you want to start with—I know you said one of the things that you use it for is research, like talking to research papers. And where are you getting the papers from? Do you have some sort of Google alerts set up or some search somewhere that feeds you the papers?

Anne-Laure Le Cunff (00:04:06)

Yeah. So I have an alert that feeds the papers into my inbox. So I receive it over email and I can have a quick look very often just from the title. You can know that maybe that has nothing to do with the research that you're doing. And if it's relevant, then I'll open it, download the PDF, which there are various ways to do. I have access to all of them because I have an academic email address, but if you don't there are other ways to get access to these PDFs, and you can upload it to ChatGPT and ask your questions.

Dan Shipper (00:04:36)

Got it. That's really cool. Okay, so basically you're getting these papers in your email. You're scanning through them. You're doing like a first little triage yourself just on the title. And then if you're interested, you're like, okay, I'm going to upload. So it looks like you have a paper here that you're going to show us. What is this paper?

Anne-Laure Le Cunff (00:04:51)

So this paper—I'm just going to demo it with one of my papers that I have published because I didn't ask for permission from another researcher to upload their papers, and people can be a little bit worried when it comes to research and rights and redistribution. So, just for the sake of explaining how it works, I'm going to upload one of my papers. I am giving myself authorization to use this paper for this show.

And so the first thing you want to do is obviously to upload the paper. So here I have uploaded the paper as a PDF, and then you can ask it a question. Something I have found, which is a bit silly with ChatGPT, but if you don't tell them to answer the question based on the uploaded PDF, sometimes they will just ignore what you've uploaded.

So, when I type the question, I always say, “Based on the PDF attached, please answer the following question.” And in this case, we can ask something like, “What are the key considerations to take into account when implementing a”—this is a long word—”neurophysiological measure of cognitive load in ADHD.” You don't need to necessarily understand the question—I'll still explain because it's not that complicated. But when you want to measure cognitive load, which is the mental effort that you put into doing something, there are different measures that you could use: You could use eye tracking, you could use electroencephalography, you could use fMRI. There's lots of different ways of doing it. and you have some considerations that you need to take into account. if the participants have ADHD versus if it's just neurotypical participants. So that's a question we can ask to this paper, which was a review of all of the different measures used in ADHD. So—

Dan Shipper (00:06:50)

Before you ask the question, it sounds like one of the things you do first is you just review what is in the paper? Like, is it relevant to you? Do you have a thing that you usually say for that?

Anne-Laure Le Cunff (00:07:06)

So, I always read the abstract before I upload a paper here. There's no point in uploading something that has nothing to do with my research. So, I always have a sense that it is at least relevant to my field of research. And then the questions I asked allow me to know if it's relevant specifically to my research without having to read the entire paper yet.

So I press enter and send it. ChatGPT is reading the document and then we get an answer. And it gives you a bullet point list of all of the different considerations to take into account based on the paper.

Dan Shipper (00:07:53)

And what do you think of these considerations?

Anne-Laure Le Cunff (00:05:4)

They're actually relevant and they are the ones that are in my paper, which I can tell you now. I wouldn't necessarily be able to tell you if it wasn't mine, but because I wrote it, I can confirm that this is exactly what's in the paper. And it's very helpful because, as you can see, I get four bullet points instead of, I think, this paper is 30 pages. So four bullet points with this answer. And, you can keep going and you can ask more questions saying—am I the only one who's weird, but I would say, “thank you.” I always thank ChatGPT.

So I say, “Thank you.” And I say, “Still based on this PDF,” because sometimes it forgets. “How has eye tracking been used to measure cognitive load in ADHD?” And so, again, it's telling that pupil dilation, fixation patterns, psychedelic movements, definitely things that are included in this paper. So it all makes sense. So this is great and if—obviously I know the answer already, but if I didn't with this, it would be enough for me to at least know that this belongs in the to-read pile and not in the I don't care pile.

Dan Shipper (00:09:32)

And do you use it like once you're in the middle of reading something, do you use it to look things up or sort of do a deeper dive or do you find that you don't need it for that?

Anne-Laure Le Cunff (00:09:38)

Oh, no, that happens all the time when I'm in the middle of reading a paper. And sometimes I generally don't understand what the author is talking about. It's especially true if it's from an adjacent field of research where we're kind of talking about the same things, but they don't necessarily use the same words that I do when it comes to neuroscience. And so I know we're talking about the same thing, but I don't understand what they're saying. And so what I can do, same, is upload it here and ask questions.

I really like explain me like I'm five, where it’s like, explain me like I'm five. What does this author say? Or tell me what were the findings? What were the limitations? and I can even use it as a translator asking, can you translate this in a language that I can understand from my field in cognitive neuroscience? So it can really be used as a kind of translator for research papers as well.

Dan Shipper (00:10:37)

You're so right. You're hitting on so many points that I've been talking a lot about on this podcast and writing about a lot. And it's so great to see that you feel it too. ‘Cause the key underlying thing that I think is going on here that people might miss is I think ChatGPT is an incredible tool for human coordination, allowing humans to work with each other because we have basically we have all this knowledge that's all out there and there's so much knowledge that like a single human cannot even know even a small percentage of it, let's say. And what ChatGPT is able to do is find the right little bit of knowledge, compress it, and reformat it, and translate it so that you can understand it and use it in your context.

Thinking of ChatGPT as a summarizer is something that we're maybe a little bit familiar with, but one of the really interesting things about what you said that is a nice twist too is that it translates for you. But the thing that people miss is that it translates from English to English. And what I think ChatGPT lets us see is that there are many different versions of English that are spoken by many different subcommunities who all have their own ways of talking about things and the fact that we have different versions of English makes it harder for us all to work together. And ChatGPT as this universal translator is able to allow someone like you who's working in cognitive neuroscience to understand the words and conventions of a different community. And that's just going to be net beneficial to progress for humans.

Anne-Laure Le Cunff (00:12:24)

Absolutely. And I love that kind of lateral translation from one field to another. But also you can use it in a more vertical manner where, for example, you know when you're like a second-year Ph.D. student and you're reading a paper that has been written by a professor in neuroscience that has been in the field for 15 years, they are, without even realizing it, maybe going to use more jargon and it's going to be a little bit more difficult for you to understand, or maybe they're even going to use concepts that you haven't even studied yet at school. And so again, you can have a little bit like the explain me like I'm five, but you can go like, explain me like I'm an undergraduate psychology student, or explain me like I'm a product manager at a tech startup, or explain me like I'm insert whatever you are, what your role is and what your current knowledge is. And so that can also be this vertical translation that's filling the gap between the knowledge that you have today and that additional knowledge that is in this paper, this article, et cetera, but that's a little bit out of reach at the moment, right?

Dan Shipper (00:13:30)

Right. That makes perfect sense. I think it's so valuable because previously. You might read something and not understand the terms and then end up having to spend a day looking up each individual term or trying to find something that could fill in the gaps because there's probably something that does fill in the gaps, but it's really hard to find and ChatGPT just does that without really even blinking. And, yeah, I think that's amazing.

Anne-Laure Le Cunff (00:14:02)

Yeah. I have another example of this where when I submitted a paper for peer review, one of the peer reviewers came back and the comment was, but what does it mean based on this theory of dyslexia, basically, and I had never heard of that theory. And if you don't know the theory, that comment makes no sense, because you're like, what do you mean? What do I mean in regard to that theory? What part of the theory? And I actually copy-pasted that comment from the reviewer in ChatGPT and ChatGPT was like, oh, the reviewer probably means this because that's the part that a lot of researchers are still debating and you're making a claim here. And maybe you should not sound so sure what you're saying, because this is something that is still a topic of debate. And see, this is an example where I don't even know what I would have typed in Google to get an answer to this.

Dan Shipper (00:15:02)

That makes total sense. Yeah. There are so many of these like ungooglable Google blind spot things that it sort of fills in. I love that. Did you just copy paste any of its responses back to the reviewer? Did you spend some time thinking about it?

Anne-Laure Le Cunff (00:15:14)

No, because, first, ChatGPT does sound like ChatGPT. So I think anyone would know that, even though it is getting a little bit better. And second, I didn't upload the entire paper at the time. So it was lacking context, but at least I understood the question. And so I could address the feedback. Whereas before that, I was literally staring at that question from the reviewer. And had no idea where to start.

Dan Shipper (00:15:41)

That’s great. That's great. I love this use case. I know you have some more stuff to show us. So, if you're cool with it, I see one of the things you said that you use it for is deep research on online ghosts, and I'm very curious to learn what that means.

Anne-Laure Le Cunff (00:15:59)

Yes, it is actually a little bit adjacent to what we just talked about, where sometimes you don't even know what to Google. In this case, it's different. You know what to Google, but it's not returning any results. And you know that knowledge has to be somewhere on the internet. But somehow it is not part of the circle of your map. It's a little bit like—you know, when you play map-based games like Age of Empires or games like these, and you can only see the places that you have explored already. So you have no idea what's beyond this. So it's a bit like that. And so those online ghosts, they are traces of their existence, you know that somewhere this information is available, but you keep on googling and you find links, but they go nowhere or someone mentions some sort of information that they read somewhere, but they didn't link back to the source where they found it. And so you can't find the answer to your question and the example that I want to share has nothing to do with neuroscience. I'm French-Algerian, and I've been trying to reconnect a little bit with the Algerian side of my culture. And one of the ways of doing this for me is to learn a little bit more about Algerian cuisine.

I've been eating a lot of Algerian food when I was at home as a kid, but it was always the same basic stuff. And I've discovered as an adult that Algeria actually has a very rich culinary culture with lots of different dishes that I have never heard about. And so I'm trying to learn more about them and I'm even trying to make some of them. And something I discovered recently is that Algeria consumes a lot of cheese. They make and eat a lot of cheese, which is funny for me being half-French, because I thought that was a very French thing. But apparently, Algerian people also do this. And there was this specific cheese that I found that seemed very interesting and not that complicated to make, but it was impossible to find a recipe for it. I tried it in English. I tried it in French. I could not find anything. And I'm going to show you a little bit about how I managed to find it and then I published a recipe in English because I need to contribute back to the internet, even though I don't even know if anyone is going to ever look for it.

Dan Shipper (00:18:27)

I love that. Wait, I'm sort of curious. How did you become informed of this cheese so as to know that you would want to find a recipe for it?

Anne-Laure Le Cunff (00:18:41)

I was reading an article about Algerian cuisine in general. And there was just one paragraph saying, and Algeria is one of the biggest producers of cheese in Africa. They love cheese. They have so many different ones and I had never heard of that. I was never served cheese at home as a kid. So it was completely new. And that's when I fell into a rabbit hole of figuring out the entire cheese industry in Algeria.

Dan Shipper (00:19:08)

Okay. And what is this specific cheese called?

Anne-Laure Le Cunff (00:19:10)

It's called kemariya. And it's made with goat milk and it's very, very simple. You can make it in a couple of hours and the way you serve it is with honey and some pistachios and tea and no bread. You just eat the cheese like that.

Dan Shipper (00:19:30)

And is it sort of like a chèvre? Is it very soft or is it harder or what's—?

Anne-Laure Le Cunff (00:19:39)

It's very soft because you don't age it, so yeah, it's very soft. I'm going to show you and I can describe it because I know some people listen to the audio only. So I initially found this cheese with a different name: Takemmarit. It has two names because you have Arabic and Berber people in Algeria. So you have two languages. And ‘cause it's a cheese that was initially made by Berber people from the north of Algeria and they call it takemmarit.

And so, as you see, “What's the recipe to make takemmarit at home?” “I'm sorry, but I wasn't able to find a specific recipe for takemmarit.” It's very rare that you ask a question to ChatGPT and literally it says it doesn't know. And so then I give it a bit more information. I'm like, “Here's what I found on Wikipedia about the cheese.” Right? And so I'm like, “This is the cheese.” It was on the French Wikipedia—there's nothing on the English Wikipedia. The French Wikipedia explains a little bit more and then ChatGPT goes, “Thank you for the information. Now that I have this, I can tell you, these are the ingredients and here are the instructions.” But I'm reading the instructions and I'm like, that doesn't really make sense. And so I asked, “Are you sure you'd make the cheese first, then mix it with the leben, which is one of the ingredients. ChatGPT says, “You're correct to question that. My previous description may not accurately reflect the traditional method of making takemmarit.”

So we have a little bit more back and forth. And then I realized that this cheese has another name. So I tell ChatGPT, “I found another name for that cheese: kamariya. Does that help with finding a recipe? Please have a look in French and in English.” And so then it looks like it's finally understanding what I'm talking about. So the recipe is starting to make a lot more sense. And it also links to the source. But then I noticed that it contradicts itself with something it said before. So there's one recipe where it is one ingredient that's mentioned. And then there's another recipe where that ingredient doesn't exist. So I'm asking, ChildGPT is so nice with me, saying, “You're absolutely right. I apologize for the confusion and update the recipe again.” At this point, I'm a bit tired, so I just check before I go further with this. “So there's no cooking involved,” I ask, and it says, yes, it says, “It's just heating the milk but there's no actual proper cooking involved.” So I'm just going to spare you the entire, very long conversation that I'm having, but the breakthrough moment was when I asked it to look it up in Arabic, which didn't occur to me before, because I don't speak Arabic. My mom didn't teach it to me. And so I was sticking to the two languages that I know.

Dan Shipper (00:22:39)

Wait, can I ask you a question? At this point—’cause it had given you a recipe, it had given you a couple of different recipes before. It seemed like the recipe that it had just given you was pretty much right, but were were you still sort of suspicious that it wasn't quite right?

Anne-Laure Le Cunff (00:23:01)

Yeah, I could tell that it was hallucinating part of it. And so the ingredients sounded right, but the order, the steps felt a bit wrong, and this is why—I spare you that entire part—but I asked a bunch of questions about how to make other types of cheese and that didn't match. It would have been a really weird cheese if you did it with these steps. Not that I know anything about cheese—well, now I know quite a bit after this conversation, but at the time I didn't. And so I was definitely suspicious that something was wrong with it. I was getting closer, but that was not the actual recipe. And this is when I asked it to check in Arabic and it found a YouTube video, which I didn't even know ChatGPT could do this until this conversation. So, it found the video and I watched it. You can see a little bit of what it looks like. Sorry, this is so funny. The transitions. So that's the video.

Dan Shipper (00:24:19)

And this is something that you just never would have found. You couldn’t Google for this really, you know?

Anne-Laure Le Cunff (00:24:25)

No, I couldn't Google for it. That's in the language I don't even speak. I was already struggling with the name in French and English of that cheese. So, no idea how to write this in Arabic. And again, it's a cheese where there's very little information online. So there's no way I would have found that video without ChatGPT. And the part where I was incredibly lucky here is that the video actually has no voiceover and it's all subtitles explaining the steps. So what I did is that I used the camera of Google translate to watch the video. And translate the subtitles to English.

Dan Shipper (00:25:13)

Oh my God. That's amazing. That's so cool. Were you just taking pictures each time for each frame? Or did you download the video? I don't know how Google camera works to be honest with you.

Anne-Laure Le Cunff (00:25:25)

So it's live. So I was posing, translating with the camera. You don't need to take a photo. You just need to point your camera at whatever you're looking at. And I had ChatGPT open in another window and I was just typing things there in the text box, but in a very messy way, because the next step was to ask ChatGPT to clean it up and format it as normal recipe instructions. But yes, that's what I did with the Google translate camera.

Dan Shipper (00:25:54)

That is the best. That is so cool. I love it. Can we see what came out?

Anne-Laure Le Cunff (00:25:59)

Yes. So you can see here, I say, “Thank you, because I still thank ChatGPT—”

Dan Shipper (00:26:04)

I always do that too. I agree with that as a best practice.

Anne-Laure Le Cunff (00:26:10)

You never know what's going to happen in the future. So I say, “Thank you. Here's the transcript of the video.” And as you can see, it’s really just bullet points. It's very messy. I didn't change anything from the direct translation of the subtitles to what I copy-pasted into ChatGPT. And then I asked it, “Can you clean it up by combining steps, removing unnecessary information, fix the grammar where needed, and fix the spelling of kemariya throughout because I butchered it when I was typing it. And voilà, the recipe that I obtained after hours of collaborative work with ChatGPT.

Dan Shipper (00:26:55)