Transcript

Speaker 1 00:00.000
You ever wonder how Sam Altman takes notes, thinks about annual planning, thinks about sabbaticals, what he's going to actually work on, how he chose to focus on AGI, with those are the things that we talk about in this episode and we get answers. Let's get into the conversation
Speaker 1 00:13.920
with Sam Altman. All right, Sam. I want to begin with how is knowledge of LLMs changed how you think about writing and communication? I mean, I think we are going to all not all all
Speaker 1 00:30.000
I think many of us are going to write in a different way in the future. I don't mean like people are just going to use LLMs to like write stuff for them because one of the strangest things that I think happens is when people put a few bullet points into an LLMs, have it generate
Speaker 1 00:43.320
a nice email, send it to somebody else, and then they summarize it on the other end because we can't we we just can't agree that, you know, we just want the bullet points back and forth and there's still this societal nicety. But someone is going to build, probably somebody
Speaker 1 00:57.840
already has, built a first version of this like
Speaker 1 01:00.360
A great tool to write in a new way where you have this thing that is not, you know, expanding your bullet points, but is helping you discover new things in the idea space. And that's awesome. Like that's what computers do at their best, right? Is they they help they are a tool
Speaker 1 01:19.800
that help you do things you otherwise couldn't do. I've always thought it was strange how we've had this tools for thought, idea for decades and yet the vast majority of the way people
Speaker 1 01:30.040
right? Is they open up Microsoft Word? And they have no aid from a computer really. It's just like a typewriter. Yeah. I mean, it turns out that like writing is pretty good. I don't We can for sure make it better, but I understand why that's where we are. Tell me if this is
Speaker 1 01:49.680
baseless or accurate or where on the spectrum it is, but I find it interesting that there's a juxtaposition between words being more important on the input and then moving away from words with the output. So
Speaker 1 02:00.040
or a dolly. I think words are going to be a huge part of how we communicate with computers, how we program computers. And natural language is kind of the interface to computers that people want, I think. I think that it's been, you know, sci-fi predicted that for a long time.
Speaker 1 02:22.320
But I think a big part of the revolution of ChatGPT was you could just talk to a computer in plain English and get it to do
Speaker 1 02:30.040
all these things. Um, it won't be the only way we want to interact with computers, of course. And you'll have multimodal input as well as output, but we are very finely evolved to use language. There's also something special about text. Yeah, for sure. Searchable, malleable.
Speaker 1 02:50.720
There is a reason that this has been such a part of like to imagine humanity and human culture without language, it's like
Speaker 1 03:00.000
It's seems impossible. I can't do it. And even text itself, the there's a rigor to text. There's a rigor to thinking in text for sure. Yeah, I get it. I get it. Because you can point to specific words and sentences that you disagree with rather than just the overall vibe. So if
Speaker 1 03:18.640
we're having a conversation, I can't remember the exact word that you said, but if there's a transcription, I can say, "Ah, it was this that I really liked, this that I think we can make some minor changes Yeah. to How should chat
Speaker 1 03:30.040
I don't think we know yet what the writing of the future the process is going to look like. I would bet it's just like a safe baseline that it's not going to change all that much. I think we will have new tools that let people write in different ways and hopefully get more sort
Speaker 1 03:50.320
of idea refinement and generation out of the process. But uh you know this thing that people say of like well no one's ever going to learn to write anymore because now it's just like that
Speaker 1 04:00.640
That's not why people really write in the first place, like the kind of writing that you can just the kind of thing you can do by having ChatGPT go write your your kind of you know essay for English class. That's not real. That's not what this is about anyway. And if ChatGPT can
Speaker 1 04:15.360
help people do do a writing like activity and get higher quality thinking out of it, that's wonderful. Tell me about that. Literally if if we believe that part of the value a big part of the value of writing is to clarify
Speaker 1 04:30.040
your own thinking. And we can have new tools that help you do that better than before. That'll be a big win. What I think of ChatGPT is raising the returns to is the initial seed, the big bang moment of an idea. And this is a way that I like using ChatGPT is I know that I have a
Speaker 1 04:50.680
distinct idea if ChatGPT disagrees with me. And then once I have that idea, if I can clarify in some sort of way, then ChatGPT can help me find examples and stories, things that
Speaker 1 05:04.600
Totally. I think I I I I you know I try to like watch people in like very different walks of life use ChatGPT.
Speaker 1 05:12.500
and it's always eliminating. So I watched two students use it to kind of like help with their homework, do their homework to be honest recently. And one of them um basically just like put in their thing and wrote their whole essay and I was like appalled because I kind of knew
Speaker 1 05:32.380
that that was a theoretical thing that people were doing at you know significant volume or whatever but you hear about it but like to like watch someone just like do that and then get an
Speaker 1 05:42.500
essay that was, you know, bad, but like passable out of it was like that was like a real like what have we done moment. I was like this all in a way that, you know, I just hadn't I never seen someone do it before. And then I watched someone else use it in a in a very different
Speaker 1 06:02.620
more interactive way to try to do something more like what you're talking about, which is like I have this idea. I can't quite articulate it. I'm kind of stuck. Let me get unblocked and let me generate a bunch
Speaker 1 06:12.540
more ideas. And the thing that came out of that was far better than I think anybody would have done on their own. And I was like reflecting a lot on that and the first question was like a bad question. Like if you can just put something in and get a super interesting or that I
Speaker 1 06:30.540
thought not like I I a super passable response. I I I think we're just like asking people to do the wrong thing. Whereas if it's something that like gets them to want to think about a
Speaker 1 06:42.500
question differently and use the tool to help them get somewhere they wouldn't have gotten on their own. That's really interesting. How do you use ChatGPT everyday? I used to only use it for a few things and both ChatGPT has gotten better and I figured out how to use it more.
Speaker 1 06:57.540
And so the cool thing now is I really do use it as a general purpose tool. And I hope that a few years from now when you ask that I'll say I use it for most things that I do. Like every few months I find new ways to use it new ways to incorporate it's
Speaker 1 07:12.540
It's obviously still terribly integrated into most people's workflows, but that's just going to get better and better. When you're talking to friends, you're like, you should use ChatGPT for this. What are the themes that you're telling them to do? Uh, I mean, the thing that I
Speaker 1 07:30.820
hear about from my friends that they love it for the most is like computer programming, help in some way or other. And the number of people who say that's like transform my life.
Speaker 1 07:42.500
Yeah, I mean like, it's very gratifying here. It's a lot of fun. Like, there are other things where you know people say it's like change the way my kids learn or teachers say I change what their teachers that's great too and but I and then there's like incredible examples with
Speaker 1 07:57.700
healthcare the way people are using this for creative work but the programming one is like near and dear to my heart. Many of my friends are programmers so I hear about that a lot. Email. Yeah. You do a lot of writing by email and you've Uh I do a lot of like very short
Speaker 1 08:12.500
email. Like, I do a lot of like seven-word emails. And how does ChatGPT help you with that? Um, it's super good at summarizing long emails that like most long emails honestly I just stop. I don't even read, but if I have to read one, it's super good at like ChatGPT's ability to
Speaker 1 08:34.020
effectively summarize long pieces of content. I like a really long thread or whatever, very impressive. Yeah, it was just I got a tour of the
Speaker 1 08:42.500
library here. Yeah, that's a cool space. By the way, nice job. I like that space a lot. It's beautiful. Thank you. And the I saw the Encherto on the wall by Nassim Tolib and he says that basically the definition of a good book is one that can't be summarized and maybe there's an
Speaker 1 08:58.260
equivalent for GPT. There's a really interesting um There's a really interesting thing there which is that at some sense uh it took me like years to really understand this but Ilya would always
Speaker 1 09:12.540
say that what these models are really about is compression and we're going to go figure out how to compress as much knowledge as possible and that's what we're going to make AI. Compression is like the secret to intelligence and that was like I had to meditate on that for a long
Speaker 1 09:29.860
time and I'm sure I still don't fully understand it but there's something deep there. I was talking to your assistant she said that you think very clearly you're like a man of few
Speaker 1 09:42.500
words, but when you say something, it's it's really you're clear in what you want and you've really crystallized your message. I guess the part of that that resonates is I do try to like get at the essence of a problem and I I definitely don't like when other people communicate
Speaker 1 10:02.060
unambiguously. I thought it was really interesting in your conversation with Joe Hudson, how you spoke about the way that you've released
Speaker 1 10:12.540
anxiety from your life. How has that change in your internal state shown up in your thinking? I don't remember who said this, but someone I don't even remember if this is a friend. This is like a famous quote, but someone said like most
Speaker 1 10:25.040
People can't even let themselves think the interesting thoughts much less say the interesting ideas. And I think there is something about the world that has gone horribly wrong there. And I'm sure having like background anxiety running as a process makes it harder to think new
Speaker 1 10:43.800
thoughts and to focus for sure if you're like a bundle of anxiety and you have like a inner monologue spinning you in all sorts of different directions it's hard to really sit down and focus. But if you're like constantly
Speaker 1 10:55.040
self-critical if you're constantly saying, "Well, well, other people think about this." If I even, you know, I think a lot of people have I've heard I've heard people say things like, "Well, that might be an interesting idea, but I would like feel embarrassed or foolish to even
Speaker 1 11:10.920
like tell people that I was thinking about it or working on it." Like, if if you can't even let yourself like go pretty far down the path of an exploring that idea before you worry about what are the people going to think about it, that that seems bad.
Speaker 1 11:25.000
this idea that you have around people spend so much time trying to think about how to be more productive, but you're like hold on, hold on, hold on. Let's talk about how to really think about what we're going to work on in the first place. Yeah. How does writing help you do
Speaker 1 11:37.360
that? So first of all, I I I strongly agree that if you have a choice between spending some effort, thinking about what's work on versus how to like be a little bit more productive in this new method or that new method uh with a very you should have a very high bar uh for doing
Speaker 1 11:52.440
anything but thinking about what what to work on. Um
Speaker 1 11:55.360
I think that's just sort of a higher higher impact thing most of the time. Of course that doesn't work all the time at some point you actually have to go execute but I I I often see people who I think are really talented um work super hard are super productive. Just not spend
Speaker 1 12:14.200
much time or surprisingly often not really spend any time at all in a meaningful way thinking about what they're going to work on. And I think that's like the high order bit. Uh so that's that's part one. In terms of variety
Speaker 1 12:25.040
writing is a way to do that. I I think of writing is sort of a like externalized thinking. Um I I I still if I have like a very hard problem or if I feel a little bit confused about something, have not found anything better to do than to like sit down and make myself write it
Speaker 1 12:38.640
out. Um write out like what I'm you know, how I'm thinking about it, what I think somebody should be trying to like figure out how to explain it to myself or to somebody else. So I I think it's just like it is a super powerful thinking tool. Um I write for my write things down
Speaker 1 12:52.360
for myself uh or for the
Speaker 1 12:55.040
most and for like private groups the second most and public at this point very rarely. What are the different parameters of clear communication? They're sort of the sloganeering, there is a good tagline, there's also the depth, the idea ma's. Yeah, actually I think clear
Speaker 1 13:07.840
communication is very much less important and very much downstream of actually clear thinking. So if you know what you're going to do, if you've and if you've like figured out how to like reduce that to the essence of why it's a good idea and what the plan is going to be, what
Speaker 1 13:22.160
the priorities are going to be, then communicating clearly about
Speaker 1 13:25.000
that is not so hard. But getting clear about the actual ideas is really hard. And so I think unclear communication is is a symptom of unfocused thinking for the most part. Napoleon, he has a line about the importance of clear directives, clear communication because when you're
Speaker 1 13:44.880
in the battlefield, you need to be able to articulate things simply and have alignment for the team. Lots of similarities with what you're saying. I mean, I don't think that's just Napoleon. I think
Speaker 1 13:55.000
I think that, as I understand it, I haven't studied a lot of military history, but that's like a pretty common refrain. Like that seems to have been born out by history. Um, but I also think that's like born out in business. Uh, that clarity, speed, quality of execution uh all
Speaker 1 14:10.200
linked. Of all the things that you've written, what are you most proud of? This is not false modesty, truly none of it. Writing's not my gift. And I'm okay with that. Like, writing is super valuable to valuable to me as a tool
Speaker 1 14:25.520
for thinking for communicating with internally with the org, but there's nothing I am I hope I will do things that like stand the test of time and matter to the world. It's not going to be my writing. But that doesn't mean I don't get a lot of value out of it. I think that to
Speaker 1 14:45.680
give you a little bit more credit, maybe the purple prose isn't your gift, but a piece like how to be successful really influenced me.
Speaker 1 14:55.000
Thank you. I appreciate that. To make every next thing that you do be a footnote to what you've done before, that's a profound idea. Yeah, I I mean I think I I hope that like I will contribute some ideas to the world that matter. I again I hope all of those matter much less than
Speaker 1 15:11.040
opening eye does. Um but that's nice of you to say so. I genuinely appreciate it. What got you to start writing the personal blog? I wanted to like practice writing. I had this like sense I had watched Paul Graham write and he's an amazing writer. I never had any
Speaker 1 15:25.000
aspirations that I was going to be anything like that, but I I had seen how powerful it was for helping start up founders and for getting to invest in good start up founders. Um so I wanted to get I wanted to like
Speaker 1 15:37.500
try to get good at it. I I'm like I'm not a naturally gifted writer. But I believe like you know with practice anybody people can get good at a lot of things. I wanted to like kind of continue doing the thing that seemed to work so well for Yeezy getting good founders. But
Speaker 1 15:59.580
honestly it wasn't It's not my calling in life. I don't really do it anymore. You wanted to be a novelist. That astounded me. Uh
Speaker 1 16:07.540
I did, but only for the like romantic life of it. Not that I thought I was ever going to be a good writer. It just seemed like this like very cool friend who was sit you know smoking in a cafe in Paris and yeah. You can still do that. I could. I could. Probably not the path my
Speaker 1 16:23.100
life is going to go down but I could. So it turned out I'm like not a very good writer and I'm not going to be a blogger and that's okay. But I am still very happy with the experiment because I learned that I can like
Speaker 1 16:37.500
right for myself to clarify my own thinking and that has been super powerful. Even the ability to like write a message to like explain to a team what a plan is and why we're going to do it. I think doing that in writing versus doing that in a meeting is often very powerful. Have
Speaker 1 16:54.260
you done that recently? It's like if we're starting a new project or if we're putting together some sort of like plan that we're going to execute on forcing myself to write it down rather than just like sit in a meeting and let it spit all around. It's been very good. Do you
Speaker 1 17:06.900
have a format?
Speaker 1 17:07.500
of sorts? No. No. I mean I try to like keep it under I don't think long is good. Yeah. So I try to keep it short but beyond that no real constraints. Tell me about your just communication lessons that you've learned from Peter Thiel. He is so distinct in the way that he
Speaker 1 17:25.260
communicates and know you've spent a lot of time with him especially early in your career. He's an amazing communicator uh and one thing that he does super well is he comes up with these uh like very evocative very short
Speaker 1 17:37.500
statements that really stick in your brain. And I don't know I don't know how to do that. I don't really know anybody else who does that like he does, but it's uh he has like very interesting things to say and very interesting ways to say them. And most people you're lucky to
Speaker 1 17:53.580
get one or the other. He is like a very rare combination of both. It's super impressive. What do you think contributes to that? He thinks about the world in this sort of like deeply unconstrained way. He has you know I mean the
Speaker 1 18:07.540
First thing anybody would say say about him is he's a truly brilliant original thinker. And that's just rare. There's a boundlessness about your thinking that really stands out. Like I feel like you have that same sort of lack of constraint. I think he's he's more of a like here
Speaker 1 18:27.580
is this totally here is a totally different view on something that no one else has ever expressed and
Speaker 1 18:37.500
now sounds like obviously at least interesting and often obviously correct. And I think my view of the world is often more like can we just do more? Like we have this like vector. Can we push on it harder? Is that like the David George sense of like everything is possible that's
Speaker 1 18:56.100
not limited by Yeah. the constraints of physics Yeah. And also that there's not enough people don't tie back to Peter. Um I remember
Speaker 1 19:07.540
for sometime someone asked like a long time ago someone asked him what was your biggest investment mistake ever. And everybody expected him to say something like well I invested in this company but all the money and it blew up and he said the biggest mistake I don't know if it
Speaker 1 19:18.580
was B or C but the biggest mistake ever let's say was not investing in the series B of Facebook. And that is the kind of mistake I try not to make. So I'm like a big believer in find what is working and like go aggressively after it. Ideas are such a power law and it's about
Speaker 1 19:35.740
finding that core thing and just
Speaker 1 19:37.540
doubling, tripling down on that. Yeah, I think that the really good ideas are rare and when you find one, you should quadruple down on it and should be the only thing you push on. You know, you should probably push on a few of these things. In writing and business, whatever, I I
Speaker 1 19:55.940
I really I really really believe in this principle. And I mean, I think this is why like all business almost all business books are terrible, right? There's like three good ideas in 300 pages. And what a reader wants is
Speaker 1 20:07.540
is three good ideas in one page. Yeah. Did Paul Graham teach you anything specifically about writing? Yeah, mostly just by reading his essays. I think like many other people, my introduction to the startup world and excitement about it came from reading PG's essays. He's like an
Speaker 1 20:23.620
unbelievable writer and that was a topic of like great interest to me and many other people. Um, I think a whole generation of us like copied PG in all of these ways. Uh,
Speaker 1 20:37.980
And so, although he was never like, "Let me teach you a class on how to write." I and others clearly took a lot of inspiration because I think he just does it in a style that resonates so much.
Speaker 1 20:50.160
clarity, precision, density. Like if you go read average business book versus PG essay, it's like they're both business writing, but other than that, they're like different species. There's no posture. He says interesting stuff. He says it clearly. He doesn't waste your time.
Speaker 1 21:07.720
Nothing feels fake. Pitching, coming up with the story. How does writing factor into that? Uh Again, I think of like writing
Speaker 1 21:20.040
as a tool to think more clearly or to get to the essence of something. And then hopefully when you're in a pitch meeting for your startup or whatever, you've already figured out how to get that down to the clear essence of it. Um and if you can It's really dramatically different
Speaker 1 21:40.400
to be on the other side of the pitch if the person has like gotten their thinking clear ahead of time or not. It's also a bonus if they're a clear communicator and and they I I
Speaker 1 21:50.040
I can like think of a few examples of people who I think are exceptionally clear thinkers and horrible communicators, but it's rare. Like I had to sit here earlier as you were talking about that and think. Um, and so if someone can get their thinking clear before a pitch, then
Speaker 1 22:07.920
they can get across to you what they're trying to do. And there are a lot of people who can do this without writing, but I often find their writing is is really is really helpful. And I often find that there are these ideas that I think I'm super clear on.
Speaker 1 22:20.280
And then I try to make myself write it down write down like a one-page summary and I was like, "Oh, I didn't really understand that in the first place." Do you do a lot of Google Docs exchanges with friends? I used to. I used to like all of life it's just been in this like weird
Speaker 1 22:37.440
through the looking glass past year and a half or whatever it's been but not even that much. Um since ChatGPT launched all of like the normal hobbies of life pretty much have gotten attenuated.
Speaker 1 22:51.040
When you were doing that, how did it help? What did you ask for? Be like, I'm thinking about this I'm thinking about doing this thing or I'm thinking about this idea just cuz it's interesting um what's the next step or tell me where I'm wrong. And you can do like a lot of that
Speaker 1 23:07.560
over dinner parties and make a lot of progress. You know, you can like host friends for a weekend and talk about something a lot and make a lot of progress. But there is something about the process of trying to crystalize it onto a sheet of favor that has to be like
Speaker 1 23:20.000
internally consistent. That doesn't let you like hide from the weak points. The constraint I like to give people is it needs to be short enough that you can send it to me in a screenshot. Like a mobile phone screenshot? Mm. I I Not for everything Yeah. but I like that I
Speaker 1 23:39.600
personally think that's like maybe too constraint for some important ideas all even though I directionally super agree with you that like short is short is critical. How much of your own writing
Speaker 1 23:50.040
The inspiration is born from conversation. A lot. But but but it it kind of like comes in as this jumble of ideas and then writing is helpful because it you know I I I think of like conversation as this very generative process and then you've got to like grind it down to the
Speaker 1 24:07.480
essence and that is best done like sitting in front of a big monitor with no one else around. The image of tangled headphones came into my mind. Interesting. You know. From
Speaker 1 24:20.040
For me, the image is much more like grinding down rocks than than untangling something. Cuz it's more like a process of like removing than untangling. And when you have all these like slightly different ideas banging against each other, you kind of end up with the right core.
Speaker 1 24:38.080
Mhm. If you were to read a book, what would it be about? I mean, a lot of times people say like, hey, this AI thinks things really important. Can you recommend me a book to read? And I kind of think about it and say, no, not
Speaker 1 24:50.040
really. So, I think I would try to like write the book for the people that ask what they should read about AI. And I think I would start with like here is the historical context of other technological revolutions, why this one will be similar, why it'll be different. Um, here's
Speaker 1 25:07.440
how the technology actually works. Here's what is possible right now. Here's so this is going to impact your life this year. Here's the range of things that might be possible in five years and how it might impact your life then. And then if we really kind of let ourselves
Speaker 1 25:20.000
dream out 100 years. Here's like what this means for all of us. And if I was your editor and I was like, "Sam, what is the biggest thing that people are missing right now?" What would your answer be? Well, that's why I'm not going to write the book. Uh, I I I I haven't had time
Speaker 1 25:34.240
to like think about that and I don't think I will need time soon. Mhm. Where did the all lowercase thing come from? Um, I mean I was like I lived online as a kid and that was just I don't know I stopped using the shift key. I
Speaker 1 25:50.040
Do it if I'm still if I'm writing something that feels like a school paper. I just I I actually wrote something that I may do as a blog post, but it's like super long. It's like 20 pages. It's way too long. Um and I may just not have time to edit it down, but it was still
Speaker 1 26:01.600
interesting to write.
Speaker 1 26:02.500
Um, but like for something like that, I still, you know, capitalize it perfectly. So it's like still in there somewhere. I like that. I may not have time to edit it down. There's something about that. That it's really the editing that takes work. Yeah, for sure. I heard a nice
Speaker 1 26:16.380
line from David Ogilvy. He said I'm a terrible writer, but I'm a great editor. That's a real skill. That's very tough to do, especially on your own stuff. Do you get help with editing? Like is that is is that something that happens like in Google Docs here or how do you think
Speaker 1 26:31.380
about it? The you know
Speaker 1 26:32.540
The things that are like written just for like an internal document, those those don't really get edit. I mean that's I I kind of write it once, maybe I read it once if if I have extra time and I'll just send it out. But for like internal coordination why I think writing is
Speaker 1 26:45.700
super valuable. So that's not like getting edited for publication. Internal coordination, why do you use those words? Oh, if like if there's like a bunch of teams that have to agree on what we're doing. I think like having a written We are like a document heavy culture in that
Speaker 1 26:59.580
sense. Um I think that's a good thing.
Speaker 1 27:03.260
Is that document heavy culture something that you got from Matt Machari? No, that predated him. Predated him. Did I see that? No, actually that's interesting. I think it's probably something about like the academic culture of researchers that started to hear. In what ways did
Speaker 1 27:21.860
people's thinking reveal themselves through the writing of YC apps? The biggest thing that you that I took away most of the time is
Speaker 1 27:33.020
how rare clarity of expression in a YC application is. And it's rare even though we say like this is really important and it seems obvious that that's what you should try to do. But I found on the whole that people who did not express themselves clearly in a YC application did
Speaker 1 27:53.500
not run the company in a clear way, did not explain to the team what they were doing, then explain to investors, to customers, everything else what they were doing in a clear way. And
Speaker 1 28:02.500
That is a very hard way to have a chance of success for a company. Um so much of your job as a founder or anyone leading any kind of company is is like evangelist in chief. Mhm. And it's hard to be an effective evangelist without clear communication. When you were at White
Speaker 1 28:27.140
Combinator, you had a big initiative of open sourcing knowledge around a course and you wrote
Speaker 1 28:32.540
a book called the startup the startup playbook. I would say I wrote like a pamphlet. Okay. Well, it's okay. You wrote a 50-page book, but tell me about why you did that and the process of writing the book. Um, I think getting the knowledge out about how to do startups is just
Speaker 1 28:48.460
like a clear net win for the world. It's not the most important part of what YC does, like the the one-on-one mentoring support the network, that's all more important, but putting the knowledge out there is is I think a good and easy thing to do.
Speaker 1 29:03.540
And what is something that you learned while running YC that you feel like really influences the way that you run open AI? A big part of YC was just like encouraging founders to be more ambitious and to like kind of go after what they believe in and I think there's a lot of that
Speaker 1 29:20.940
in the company too. What is something that you're excited to do with your writing with GPT that you can't do now? The thing that I have been thinking about
Speaker 1 29:32.500
is uh how can I use ChatGPT to just like make writing feel higher volume and lower stakes? Like how I still like if I have to go write like a 10-page thing that still feels like a huge thing to have to go do. And there's like a lot of activation energy I have to like write wait
Speaker 1 29:52.940
time and like the right mood and then I have like hours of uninterrupted focus. And if if if using ChatGPT and I haven't figured this out yet but I've been thinking about it
Speaker 1 30:02.780
Can somehow mean like it's the kind of thing I do when I'm like in an Uber for 15 minutes because it just makes the activation energy that low. That would be very cool. How can GPT amplify different personalities? You know, one of the things I like to use it for is hey rewrite
Speaker 1 30:17.940
this in the style of Amir tolls or Tyler Cowen. How can GPT continue to do that? Well, future future versions of GPT will be very capable of that. What
Speaker 1 30:32.540
the fair thing to Tyler Cole is in that case, we're trying to figure out. Um, so it's like not an obvious question. Um, but for sure what everybody agrees on is there there can be many personalities that are not based on real people. And that's a cool thing to have. And the fact
Speaker 1 30:55.580
that you can have call let's call them like personas. You can have like Chachi BT
Speaker 1 31:02.540
We mix things in different personas. I think that'll be helpful in the creative process. Um, the thing that I hope for more than anything else out of ChatGPT and
Speaker 1 31:15.040
transformations is that it will be a tool that lets us do things we just couldn't do before, think of ideas we just couldn't have before. Be more creative than we could be before. And this is kind of the archive technology, but I think this is a going to be a particularly great
Speaker 1 31:34.680
example of it. Creativity not limited by skills but by the ability to think of the idea in the first And place not even that like if these tools like can help you think of the idea but
Speaker 1 31:45.040
you have got to have you've got to be a great curator like I don't know exactly what it's going to be like. Um but I do know people are going to get very good at using the tool like they do with any new tool and that will expand the realm of human possibility. Hey, I want to
Speaker 1 32:02.560
tell you about a new site that I built called writing examples. We take writers like Steinbeck, Orwell, Seinfeld and break down what makes their writing so good. If that sounds like
Speaker 1 32:15.040
it's kind of your thing. We'll go to writingexamples.com. And if you go there, you enter your email, I'll send you my three favorite editions right away. All right, back to the episode. One of the things that I really admire about you is how deliberate you are about thinking,
Speaker 1 32:32.360
about what to work on. And I'm curious how you thought about your choice to work on AGI. And what that process of envisioning that one thing that you're going to focus on?
Speaker 1 32:45.000
was all about. Your process is the right word for it, right? Like it all of these things sort of start as these like almost jokes. Not quite a joke, but like a a sort of like somewhat ridiculous idea. Um at the now working on AGI seems like the obvious only decision for me at
Speaker 1 33:01.360
least. Um but at the time it seemed like a pipe drain. But I think ideas in general are very fragile. Good ideas, the best ideas are extremely fragile. And
Speaker 1 33:15.280
There is an unbelievable amount of value in figuring out a setup, a method, whatever you want to call it for not killing very fragile but potentially very great ideas. This comes down to like how you think about it, what your process to make a decision is. It comes down to like
Speaker 1 33:30.680
who you surround yourself with. I think a particular kind of toxicity to avoid are the people who are like so smart they understand why every great idea is bad. But I think in the the very early days
Speaker 1 33:45.000
the main thing is not to accidentally kill good ideas. So tell me about fragility and how writing factors into this. The thing that is most important to me personally about writing is like externalized thinking and organization, magnification, whatever you want to call it of
Speaker 1 34:06.960
vague ideas. I find it astonishing how much writing just for yourself uh sometimes
Speaker 1 34:15.040
for a small group of other people you're exploring an idea with, but mostly writing just for yourself helps clarify what you actually think helps like sharpen stuff in a way that for me and I think for a lot of other people is somehow impossible to do just like thinking
Speaker 1 34:31.320
carefully on a long hike. Like in your head. Yeah, it's harder to hide really messy thinking when you have to actually write it down and look at like stare at it. So tell me more about the process. As you thought about your plan in the early days of Open AI
Speaker 1 34:45.040
I in terms of focusing on this, what was the sort of final output of that process where you said, "Let's do it." I do remember intermediate stages where it was like talk to like a bunch of people, uh have all these ideas write out like okay, here's what we're going to do. Like
Speaker 1 35:00.240
here's our here's our plan. You would write some of those down and it would be like very obvious to you immediately. Like okay, this actually makes you feel it. You feel you feel it or you think it through and when you when you stare at it like it's one thing to like
Speaker 1 35:15.040
like have a couple of beers with some friends and say, "We're going to build AGI." Um and it's another to say like, "Okay, here's like here's like a full cohesive plan for what it's going to look like." And that makes some of the bullshit fall away. Um so many of those we'd
Speaker 1 35:31.560
write out as we were thinking through the different things we could do and how we would It's going to be an organization, you know, we're all going to go join some university research lab like that helped get rid of some of the silliness. And again, now it all seems so obvious
Speaker 1 35:43.560
that this feels weird to even say.
Speaker 1 35:45.040
Because like of course this is what we're going to do. Right. But at the time it was deeply non-obvious or a lot of other people would have been doing it. That would be sort of my like evidence point for it. And then eventually if you write something down that looks like
Speaker 1 35:54.880
credible enough you send it around to other people. Uh they have the same experience. They might rewrite it, they might edit it, but they also kind of say like all right when I have to like stare at this in black and white it's a little little different. I'm a big believer in
Speaker 1 36:05.760
getting like input from lots and lots of people um especially on like hard questions of what to go do in the broader sense.
Speaker 1 36:15.640
And now, as you do annual planning and you think in one, maybe three year time frames, is that process the same, different? It don't do this with like as much rigor as I should, it hasn't been annual, but maybe like every
Speaker 1 36:27.500
two years. I've written a document for OpenEye, I called it literally our plan. Nice. And the first one was like 25 pages. And that was like lots of hours of talking to people getting feedback, but it was like a sharpening process the whole thing. There was then one later that
Speaker 1 36:42.060
was like 15. There was then one that was like four. I believe we could do like a half page version now. And I think that's like a good that's a that's a great sign of progress. Yeah. How much write are you doing day-to-day now? Every weekend, I mean every weekend I'll like write
Speaker 1 36:56.820
something and
Speaker 1 36:58.020
Usually share it with like 10 people internally or something just like here's a thing I've been thinking about that we should do. I have been working on something I actually plan to publish which is rare for me now about just sort of what the world looks like if we get AI driven
Speaker 1 37:12.460
abundance and like why that's important. But it's like it's a long way to go. As you think about how AI is going to change writing, you know, what are comparatively what skills are going to be more valuable versus less valuable?
Speaker 1 37:28.060
In a world where like AI can do lots of things for you, having great ideas, knowing what you want the AI to do and AI can do anything is really important. Taste, creation, like expert level, you know, like whatever it is that PG does, Yeah. that's going to be super valuable. I
Speaker 1 37:43.420
love using ChatGPT to help me write something, um especially like as I've been trying to write this thing. If I get like stuck, it's a sort of like super Thesaurus. If I just can't figure out how to phrase something
Speaker 1 37:57.540
I'm like struggling with something that like just can't get something to flow. But it's definitely not like going to replace coming up with the ideas anytime soon. It's an incredible tool for writers, like incredible tool for writers, but definitely not a writer. Like a sparring
Speaker 1 38:13.420
partner. Like a collaborator, like someone you can like give like a subtask to. Yeah. That's a lot of how I use it is a lot of times I have a word that I'm struggling with and I'll say, "Give me 10 words that would work in this sentence." And then
Speaker 1 38:28.260
take the sentence, quote it, and then it'll give me the output. It's really good at that. Yeah. Yeah. How do you think that we should be training writers differently in the ChatGPT world? I heard this story once. I don't I don't know if it's true or not, but it was like some
Speaker 1 38:42.940
creative writing teacher. They would have these students come and you know the first day or whatever she'd like give an assignment which is write the first paragraph of your novel and people would come in with all of the standard like freshman and college mistakes.
Speaker 1 38:57.540
like, you know, way too many like stretched metaphors, way too much like flowery language. Um, and then she'd go through this like exercise of I think a standard one first which is cut one metaphor from every page. Cut one unnecessary word from every sentence. Cut this, cut
Speaker 1 39:16.060
that, cut that. You take this like 10-page thing down and you cut it down to one page and it would like it would not be so tortuously over written. And then the class would read them and they would say like, "Okay, what happened?"
Speaker 1 39:27.540
happened here? I like, what's that? And the answer was there was like no story at all. There was the the instinct was try to like write this like beautifully whatever kind of satisfying to write thing, but it's no fun to read. Like the readers want a story. Yeah. And the thing
Speaker 1 39:44.900
from this like teacher is that we might teach people to write beautifully, but uh there's there's no interesting story. On the other hand, you have these like sort of mass
Speaker 1 39:57.540
massive mass market successful. I don't even know what. Like I'll pick on like the Twilight books or something. Quite interesting story. Horrible writing. Sure. And the question is like, can we make it easier to get both? And can we teach people how to use these tools? Do you
Speaker 1 40:15.100
have a sense for how good ChatGPT storytelling is? Like if I turn on voice mode and read it to a kid, how much better is that versus mom? I think the storytelling is not yet very good, but I would expect it to get better.
Speaker 1 40:27.500
We're still at a place where the models are just generally improving so much. I mean there's areas that we could push on that'd be better for storytelling, but if the model just gets a lot smarter and also if we train it to be better at storytelling, um that will help. How do
Speaker 1 40:41.620
you do that? You show it a bunch of examples of what makes a good story and what makes a bad story, which I don't think it's like magic. I think we really understand that well now we just haven't tried to do that, yeah. When you're sitting down to write and you're thinking about
Speaker 1 40:56.020
creating a focus state
Speaker 1 40:57.500
What is it that you're doing in your process to really create that? I used to think like oh I got to get in the perfect place and I got to like set a time that I'm going to like go to this coffee shop and put on my noise cancelling headphones and I'm going to be in a right
Speaker 1 41:08.060
there. And now I will take any 11 minutes uninterrupted that I can get like sitting in the back of a car laying in bed like whatever it is. I mean if I do have like if I had like a perfect thing it would be like you know Saturday morning with a cup of coffee and nothing
Speaker 1 41:25.260
scheduled. And that is great like a
Speaker 1 41:27.500
I got to sit down and like if I have to write like a long thing, I will try to set that up. But most of it happens in like short chunks in the back of a car. You know what I use a lot is I use the voice
Speaker 1 41:40.000
feature. I take it and I ask it to just clean it up and I find ChatGPT to be so helpful with that because I'm much more generative with my mouth than I am with my fingertips. Interesting, for me it's the opposite. Really? Yeah. I'm convinced there's ideas I would never have
Speaker 1 41:56.040
sitting and talking with people that I just need to sit and type for. This is like obviously a very common observation but but figuring out like the right amount of being with people talking you know
Speaker 1 42:10.000
getting exposed to like a lot of ideas and then having some time alone to think, to write, to just sort of like do some deep work, whatever that is. I think obviously this is a super important pattern to a lot of people, definitely to me. My sort of like roughly rough rhythm is
Speaker 1 42:27.080
I'm like, you know, in the office kind of non-stop all week. Uh, I have no time to think. It's just like kind of crazy packed. And then on the weekends I have like long quiet blocks and I'm not really around people and uh that
Speaker 1 42:40.000
cycle is very important to me. And is that fractal? Like do you sometimes take a few weeks off or anything like that? I used to. I think that's like really good. Like when I've taken like long chunks of time off, I would do like a month of like non-stop hanging out with people
Speaker 1 42:55.160
and then like a month of, you know, being in the woods on the beach whatever. That doesn't really happen anymore. Yeah. Do you take notes during the week that you reflect on or is it just Yes. on your No, I'm a huge notebook. Oh, tell me about that. There's all these like fancy
Speaker 1 43:07.040
notebooks in the world. You don't want those. Um, you definitely want to
Speaker 1 43:10.000
viral notebook because one thing that's important is you can rip pages out frequently and you also want it to lie like flat and open on the table. And if you like open pages, you want them to like, you know, like be able to lay like this whatever. You definitely want to be able
Speaker 1 43:23.400
to like rip pages out. I'm a big believer of like I take a bunch of notes and then I like clearly like rip them out so I can look at multiple pages at the same time and I can like crumple them up and throw them on the floor and I'm done like when our house cleaner comes in on
Speaker 1 43:36.920
like a, you know, whatever. There's just just these pile of
Speaker 1 43:40.040
couple papers that I'm like type my notes in or whatever on the floor. You definitely want like a kind of paper that is like good to write on which is a feel thing but most paper is terrible to write on. Um you want hard front and back to the notepad so and you also want
Speaker 1 43:57.080
something that can fit in a pocket. I think the Uniball micro .5 pen is the best pen overall but the Muji .36
Speaker 1 44:10.000
for points 37 in dark blue ink is a very nice pen for other reasons. Uh so those are the two I would use, but I think this kind of notebook and one of those two pens is the right answer. And how many notes you're writing per day on that thing? Uh I go through one of these like
Speaker 1 44:24.600
every three two or three weeks. Oh wow, so you're taking a lot of Well, this you can see how much I've ripped out. Like this used to have like 100 pages in it or something. So that's how you think about it. So you're going to basically take the notebook and then you rip out the
Speaker 1 44:35.320
pages Pretty and you much don't have completed notebooks. I don't have completed notebooks. Wow. What inspired this? Where does this
Speaker 1 44:40.000
come from. Lots of trial and error, uh many kinds of notebooks, many pens, many different systems. This one's really good. Another thing I've been thinking about when it comes to the influence of AGI on creative mediums is just the competence with the written word is going up so
Speaker 1 44:55.080
much. And here's what I mean. There's now, you know, with Sora, you can create videos using text as the input. You can do that with music. You can do that with images. And that's a big change in terms of the influence on of writing on our world. Again, for me
Speaker 1 45:10.280
like writing is a tool for thinking most importantly and I don't think that's going anywhere and so I think it's like it's really important that people still learn to write for this reason in the same way that even if there's going to be like less traditional coding jobs coding
Speaker 1 45:24.040
is a great way to learn to think too. So you should still learn to code. So when you say it's important that people learn to write what does that mean? What it means to me is that I like figured out this tool to think more clearly. Now if there's a better way to think more
Speaker 1 45:37.480
clearly I great I would switch to that.
Speaker 1 45:40.040
definitely not found that yet. A final question that we can close with is there's just a lot of people out there who are saying that AI is going to kill writing and they're angry about it about it and what do you make of that? I don't see any evidence whatsoever that AI seems to
Speaker 1 45:54.520
be killing writing. I mean there's like a lot of bad AI writing like plastered over the internet. Um and there's like a lot of like bad student assignments that have probably been written by AI. But I don't think anyone's serious. I don't think Paul Graham is sitting around
Speaker 1 46:08.760
being like AI is going to
Speaker 1 46:10.000
you know, kill my writing here. I think it would have to be like full super intelligence before I was like, okay, this is going to replace human writing full stop and we have much bigger issues to worry about at that point. Even if that happened, um let's say we have a system
Speaker 1 46:26.800
that can write better than a human, uh do you think that the most popular novel of 2027 has a human name on it or not? Like a human writer on it or not?
Speaker 1 46:40.240
I think yes. I think it does too. When I finish a great book, the first thing I go do is like I want to know about the writer. I want to know their life story. And I don't think I'll ever have that feeling to like AI writing.
Speaker 1 46:52.500
um there there's there is something about you read an incredible book and you kind of you could connect to a person even though you don't literally know them. You feel like you do and you feel like you have this important shared human experience and that is like some significant
Speaker 1 47:08.060
percentage of the enjoyment of a great book to me. And I bet we'd keep doing that. All right, Sam. All right. Thank you very much. This was fun. This was fun.