0:00
AI and creativity.
Search this video
0:00
I've asked dozens of writers for their take.
0:03
Some think it's a godsend.
0:05
Others think it's the literal scum of the earth.
0:08
So, here are the 12 best responses you're going to hear from literary award winners like Robert McFarland, builders like Sam Alman, and journalists like Ezra Klein.
0:17
This is what the people at the cutting edge are saying about how AI is going to impact writing, creativity, and also the culture.
0:29
I'm not completely against anything and I have not and not for lack of trying and I think not for lack of being informed or interested in the issue found a way that I consistently use AI in my work.
0:43
I'll sometimes use it right now as a replacement for Google searches and and is valuable for that and deep research has been good on some kinds of preparatory work for me.
0:54
It's just it's very hard to not be outsourcing the part of the work I need to do the most.
0:58
AI is better at the things I need to do than the things I don't need to do.
1:03
And so having AI step in which I will sometimes do in a crunch if I don't have time to like make the spreadsheet myself and like gather the data in the way I'd like to gather it or but having AI summarize a book or a paper for me is a disaster.
1:22
It has no idea what I really wanted to know.
1:24
It would not have made the connections I would have made.
1:25
It's the time I spent researching that was the most important time.
1:28
Having AI do research for me is a bad idea.
1:29
But, you know, if you had other people on the show, they'd give you something different.
1:33
I think somebody like Tyler Cowan has made AI much more fundamental to his workflow um than I have.
1:37
But I'm interested for the thing I will see that other people would not have seen.
1:43
And I think AI typically sees what everybody else would see.
1:47
So on one hand I hear what you're saying with that is the most important part of my work and I I agree.
1:52
I think that the years have proven that out.
1:55
On the other hand I wonder how much of it is like prompt issue of can you prompt it in distinct ways and and and all that.
2:04
I think of uh AI almost like a looking at the world through a kaleidoscope.
2:07
And I agree that the default prompt gives you the way that everybody else is looking.
2:11
But but that's that's assuming that I know what my prompt should be.
2:15
There's no prompt that would have told me just go read more books, man. You're not there.
2:22
You haven't gone far enough.
2:26
I'm not saying AI can't be useful.
2:26
Not saying it won't be useful.
2:28
But I think I'm pretty against shortcuts, and obviously you have to limit the amount of work you're doing.
2:38
You can't read literally everything.
2:42
But in some ways, I think it's more dangerous to think you've read something that you haven't than to not read it at all.
2:48
As somebody who knows the people at these labs and has like talked to them about how to use it and has access to the good models, it has not become an honored partner in my creative workflow. Right?
3:02
So then that seems like a pretty important fault line of there is your writing and you're like, ah, it's not that useful.
3:08
Then there is the I'm doing research, I'm trying to figure out what's going on.
3:11
I'll use a super concrete example.
3:12
We're talking about homelessness as a housing issue.
3:14
You know, the kind of thing that is really great for AI, like I haven't really thought about that.
3:21
I grew up in San Francisco, so it's like very near and dear to my heart.
3:23
Kind of watched the city get destroyed by homelessness.
3:26
I would use AI to basically say, "Hey, 03, can you give me a list of um how housing has changed over the years?
3:32
List of papers that have been written."
3:33
But what I'm hearing you say is, "Yeah, that's like a better Google."
3:37
That's not then there's something else of like I want you to write this for me and that you wouldn't use. >> Yeah.
3:43
Also, I wrote a piece about this a couple years back.
3:46
Um, it was called against the matrix theory of the mind.
3:51
And I think I used to conceptualize knowledge the way you see it in the movie The Matrix where it's like I wanted the port in the back of my mind that the little needle would go into and then I had read John RS's political liberalism, >> right?
4:11
I thought that what you were doing was like downloading information into your brain.
4:16
And now I think that what you were doing is spending time grappling with the text, making connections, it will only happen through that process of grappling.
4:26
And so the idea that you could speedrun that, the idea that it could just be summarized for you, part of what is happening when you spend seven hours reading a book is you spend seven hours with your mind on this topic.
4:38
The idea that 03 can summarize it for you.
4:40
In addition, there's all this stuff you just will not have read is that you didn't have the engagement.
4:44
It doesn't impress itself upon you. It doesn't change you.
4:47
What knowledge is supposed to do is change you.
4:50
And it changes you because you make connections to it.
4:53
What you're saying is knowledge is more embodied. >> Yes.
4:58
>> Than just looking up something on AI.
5:00
And then that because knowledge is embodied, it has a way of becoming a part of you and flowing through you into the writing. >> Yeah.
5:06
into the writing, into your thinking.
5:10
You know, when I sat down, I read a Rand report recently on housing construction costs in different states in California and Texas and and um Colorado.
5:15
If I just had 03 spit that out, it would have been some useful factoids.
5:21
it actually mattered that I spent more than an hour on the report marking it up like looking at the I just think there's something about your process of grappling with something that is what makes it that integrates it into the way your mind actually works.
5:38
I think shortcuts often operate off of a misguided metaphor of how our minds work, >> right?
5:48
>> That again we're we sort of have this download model.
5:50
download model. I'm downloading the information and Sam Bankman Freed is one of many tech people I've seen who would be like what's the point of all these books like shouldn't every book be a magazine article and ideally shouldn't every magazine article be a tweet and
6:04
the answer is no it shouldn't be in part because that actually doesn't contain the information I can tell that for sure having written books and then always thinking about how much fidelity is lost to them when I do the excerpts when I go talk about them on podcasts but also because it cannot change you. A magazine
6:18
A magazine article will change you less than a book will.
6:24
A tweet will change you less than a magazine article will.
6:25
You may not realize that at the time, but I think for most people, most of the time it's true.
6:33
It's a reason, by the way, podcasting is powerful because when they spend two hours with you over and over and over again, it's like what changes you isn't the wave, it's the erosion.
6:44
When you think of school, what comes to mind?
6:46
For me, it was lectures, homework, sitting at a desk.
6:48
And the worst part is it doesn't even work.
6:51
54% of Americans read below a sixth grade level.
6:56
That's over half of Americans. We need something new.
7:01
And that's why I'm excited to have Alpha School as the flagship sponsor for How I Write.
7:04
At Alpha, you go to school and do all your academics in only 2 hours per day.
7:09
So that then you can spend the rest of the day learning life skills like launching a business, writing a musical, or giving a TED talk.
7:17
You know, the kind of things that make you a capable and well-rounded person. And it's working.
7:21
They have academics figured out.
7:24
Alpha classes test in the top 1% nationally, and more than 90% of alpha students say they love school.
7:31
Now, I used to live in Austin where Alpha School started, and I was consistently blown away with what was happening there.
7:38
I personally mentored a bunch of Alpha students, and what stood out to me is how much the team believes in the potential of kids, way more than what I ever got in school.
7:46
Alpha's approach and the success it's already having is why it's the most talked about K through2 school in the world right now.
7:54
And it's also where I want to send my kids to school one day.
7:56
So, if you want to learn more, go to go. alpha. school/howite.
8:05
That one thing that only the artist has access to is the feeling out of which something was created.
8:11
>> AI will never have feeling.
8:11
AI will have the simulacum of feeling.
8:14
AI will never write the great American poem, the great French poem, the because it hasn't suffered.
8:21
I mean AI has not the capacity to suffer because even if you try to make it suffer, meaning write a command that is to execute failure, it'll already be succeeding at executing failure.
8:34
It will never know what it's like to collide with its own impossibility.
8:37
So AI can only ever succeed.
8:42
And without suffering, what kind of true art can there be?
8:46
>> That that is what the essence of art is.
8:49
>> I'm not I I don't subscribe to the tortured genius myth.
8:51
I don't think it's necessary to suffer in order to create.
8:56
But I do think that out of what we have suffered and do suffer comes that restlessness to find meaning to find beauty to find wonder to give voice and shape to what we feel that can be so lonely.
9:09
I mean I guess it goes back to the question why do we read?
9:14
>> And if we read to be moved and to be changed I think there are ways to change intellectually which is with information and AI can do that.
9:23
it it AI can tell me about the eye of the scallop without me spending many days in many you know scientific journals and papers where it can give me a pretty good enough you know >> but had I not spent all those days I wouldn't have written about it with feeling >> right >> and I think all the writing for me that moves me is made of feeling and time >> of which AI has neither >> it's an instantaneous unfeilling delivery of pure for information.
9:55
>> I don't directly use AI for writing typically.
9:58
Now I sometimes I do in the following sense if I'm writing on a legal issue and I'm not a lawyer.
10:05
>> I will ask 01 pro for the relevant legal background to something I'm writing on.
10:10
>> So I just wrote a column about declassifying classified documents.
10:12
I don't know that law very well.
10:15
I asked the AI for a lot of background on the topic.
10:19
Uh, I didn't use what it gave me, but now I feel like I'm not an idiot on the topic and what I wanted to say.
10:26
Whether or not it's correct, you can debate, but it's not what you would call flatout wrong.
10:29
Uh, but I don't let it write for me.
10:32
I want the writing to be my own.
10:34
It's like my little baby, so to speak. I don't care.
10:37
Whenever it's better than I am, I'm still not going to let it write for me.
10:40
Also, a lot of the sources I write for wouldn't let me.
10:43
Uh, I agree with that decision on their part, but even if they would let me, I wouldn't do it.
10:49
Most of all, I use AI when I read things.
10:52
So, I use it as the secondary literature.
10:54
So, I'm preparing a podcast for a British historian.
10:56
She does history of Richard II and Henry V.
11:02
Now, in the old days, I would have ordered and paid for 20 to 30 books on those kings.
11:06
Now, maybe I've ordered and paid for two or three books on those kings, but I'll keep on interrogating the best LLMs about the topics of her books.
11:16
Helen Caster is the historian.
11:18
And just keep on going >> and I acquire the context much more quickly.
11:22
Uh, it's pretty accurate.
11:22
Keep in mind, I'm the questioner.
11:25
So, if there's a modest degree of hallucination, it doesn't matter for me.
11:30
I'm not giving the answers.
11:30
And uh I can now do many more podcasts than I used to because I'm using AI for my prep. >> Okay.
11:38
So do you think that this is about saving time or improving the quality of your prep?
11:43
>> It's improving the quality of my reading which I use for my pleasure reading.
11:45
So I reread Shakespeare's Richard II which is a [clears throat] wonderful play.
11:52
Again in the old days I would have piled up a lot of secondary literature.
11:56
>> I'm also rereading Weathering Heights.
11:56
I just keep on asking the AI what do you think of chapter 2?
12:00
like what happened there? What are some puzzles?
12:02
What should I ponder more?
12:04
How does that connect to something else later in the book?
12:07
And it just gets me thinking.
12:07
It's the new secondary literature for me. So, it's more fun.
12:12
I learn more just by being more actively involved asking a question.
12:17
I think that improves my epistemics of the reading.
12:19
Uh, and then I think I'm smarter about the thing in the final analysis.
12:24
But mostly I'm doing it because for me it's fun and a pleasure to learn.
12:28
When you're looking at how other people use AI and you're like, "Ah, you're using it wrong."
12:32
What is the thing that they're doing wrong?
12:37
>> They're asking it questions that are too general.
12:39
They're not willing to put in enough of their own time generating context.
12:43
Now, maybe they don't have the time.
12:45
If that's what's sufficient for them, I mean, fine.
12:47
Uh, but I think they end up not sufficiently impressed by the AI because they're using it as a substitute for putting in their own time, which again for them might be fine, but it's not what I want to do.
12:57
I want to put in more and more of my time to learn and have it complement that learning.
13:03
And if you do that and keep on whacking it with queries and facts and questions and interpretations, you'll come away much more impressed than if you just ask it, oh, what what does the rate of price inflation mean?
13:16
Or, I'm interviewing Tyler tomorrow.
13:18
What questions should I ask him?
13:18
Then it's pretty, you know, mid.
13:20
Is that the term people use now?
13:22
[laughter] Like mid is fine.
13:24
Mid is called mid for a reason.
13:28
But at the end of the day, you will be asleep on the revolution occurring before our eyes, which is that it's getting smarter than we are.
13:34
You'll just think it's a cheap way to achieve a lot of mid tasks, which it is.
13:37
Also, some humans will become masters of the tools.
13:43
How much writing they still will be doing.
13:44
At what pace, I'm not sure.
13:44
But it's a a major psychological adjustment and a lot of people who thought they would be writers, I will predict they won't be.
13:54
Just like the number of jobs for certain types of computer programmers is plummeting. >> Yeah.
13:59
And that will come to other areas.
14:02
By no means all areas, but a lot of kinds of writing are one of them.
14:06
Something like generic corporate writing is the first to go from what I can see.
14:11
Uh it could be writing a biography of a person the AI cannot really do.
14:14
It may help you a lot but it can't go out and interview you know the the high school teacher and so on.
14:22
Writing memoirs of course the AI cannot do writing that is more subjective more personal I think the AI already can do it very well especially deepseek program we can get to that soon but uh I'm not sure readers want the better product from the AI.
14:40
They may want it from a human being. I feel that I do.
14:42
So, if I read a brilliant memoir written by an AI, but it corresponded to no actual life, I would read a few of those, but at some point, I'd get bored, and I don't think I would keep on reading them, even if they were better than human memoirs on average.
14:57
>> So, how does all this influence what you're choosing to to write?
15:00
Like, should you write books, articles?
15:02
How is all this shaping that?
15:04
it's affected my writing uh very significantly already.
15:10
So there's two quite different effects at work.
15:12
One is simply that AI is progressing quite rapidly and that changes the world quite rapidly.
15:16
So if you're writing a book that takes say two years to write and a year and a half to come out.
15:23
Maybe there's some other delays.
15:25
We're talking four years, right?
15:27
>> There's a lot of topics you just can't write on.
15:29
>> Like you can't write a book on AI. It's crazy.
15:32
You could write a very good book, The Early History of AI, which is frozen in historical time.
15:38
>> Now, maybe the AI will write that book better than you could, but at least you could consider doing it.
15:44
>> So, what I call predictive books, uh, books about the near future, they don't make sense anymore.
15:49
You've got to cover those by writing on this ultra high time frequency every day, every week, something like Substack, Blogging, Twitter. Uh, that's a big change.
15:56
So some of the recent stuff I've written is about the more distant past that is frozen in history.
16:06
But the other question is what can the AI soon enough write better than you can.
16:11
And it may not be that the AI writes a book. Don't fixate on book.
16:16
Maybe the AI know better than to write books.
16:18
They're just like a box and you can ask the box.
16:20
Any question you might read in the book.
16:22
That's what I suspect is the case.
16:24
Not that there'll be all these books written by AIs. It's inefficient.
16:28
like why this single package for everyone?
16:29
Just give people the box.
16:32
So, it's a question box >> and uh what you're writing had better be more interesting than the question box.
16:40
So, the book I've started writing recently, we were discussing it before filming started.
16:44
It will be called mentors.
16:46
It's about mentoring and also being a mentee.
16:51
And first, I think the extent literature is weak enough the AI maybe can't do a great job.
16:56
But even if the AI could do as good a job or better than I can, I don't think people want to read that book from an AI.
17:04
>> I think they want to read it from a human who has been a mentor and a mentee.
17:08
Just like I don't want to read all these phony memoirs from AIs, even if some of them are good.
17:12
And that's a human book that only a human can truly write sincerely and credibly.
17:20
So I'm going to write fewer books in the future because of this.
17:22
I may not write any more books after this book on mentors.
17:25
A lot of books I would have written, they're now obsolete.
17:27
I I feel I'm wise enough to recognize that.
17:30
And I'm not going to write less because I'm going to do more of this super high frequency writing. Much of it about AI.
17:39
To me, the idea that, you know, AI can can write is sort of founded on the notion that there is no soul.
17:46
To me, you write with with something of a soul.
17:49
And you know, and the fact that the fact that science says there is no soul and that the you know, the idea of the soul really doesn't make a lot of sense like the sort of like inner Tom or inner David, you know, it's kind of it's kind of kind of odd.
18:06
>> Um, but I I can I can tell you so in in Shelter Island, you know, I have there's a shed out in the backyard.
18:16
>> This at your house out there? >> Yeah.
18:17
And then what's one of the reasons that we bought the house is that there's the shed where the guy the previous owner used to have his business and do his business.
18:24
And it was a that shed was a dump for a really a really really long time.
18:30
You know, people used to come to the house.
18:34
Oh, I can't wait to see your writer shed.
18:35
Then they'd kind of look at and go >> this way. >> Yeah.
18:39
Let's go back to the house.
18:39
You [laughter] know, so it was kind of >> get a little thirsty. A good drink of water.
18:44
Um, but we so we but we finally got it fixed a little a little fixed a little while ago.
18:50
And one of the things that I had, you know, I had these rows of bookshelves installed, you know, in the wall and there, you know, you can fit a lot of books in there and and I had a lot of books and so and so the big the big question I had to I had to decide was like how am I going to organize the books? Alphabetical >> type? I don't know.
19:12
I had this like idea like I was going to do it chronologically like the books were going to be in there chronologically and so it begins with the Bible and ends with Finnegan's Wake, you know. >> That's cool. >> Yeah.
19:27
Or maybe even pre there maybe maybe even pre-Bible, you know, there's some like some stoic >> an epic of Gilgamesh. >> Yeah. Exactly. How did you know? Yeah. But yeah, exactly.
19:37
And um and the the cool the really interesting thing about doing that is that the organization of those books tells its own story and the story is of is of humanity.
19:55
Like everything it does is wrong essentially.
20:00
I mean we [ __ ] up again and again and again.
20:03
Any piece of technology that we develop, we put we either got from or put to use in war.
20:08
Any sort of religion that we talk about is, you know, is a wrong and b a tool of tyranny.
20:15
I mean, we [ __ ] up again and again and again.
20:22
And yet for every [ __ ] there is a writer, somebody who is grappling with the [ __ ] and bearing witness to the pain and the suffering and the joy and the hope and the whole freaking shitow. And you see it.
20:43
I can I can I can tell you the whole story.
20:45
I mean like modernism comes from World War I. Beat America.
20:52
Jack Carowak comes out of World War II.
20:56
Don [ __ ] Dilo comes out of Vietnam and and and the you know the the the rise of the technology state from that.
21:06
It you can just see like you know writers are there to address this flow of history.
21:11
And I would I would say that it's like the only thing maybe that humanity has ever really done right is bear witness to its own thing.
21:20
And you know I you know I think anybody who writes I'm not going to say oh I'm bearing witness I don't want to be pretentious but I think that anybody who is writing seriously from the heart is bearing witness to this onward stamp of this of this blind genius being that is humanity you know and that's the soul is the the part of us that is willing to and in fact needs to bear witness to the truth, to ourselves, to God, all of it. That's the soul.
22:02
And you see it you see the soul do its work over the course of of human history.
22:09
And I I think what I'm winding up into is like a long anti- AI speech because like why the the [ __ ] would we ever want to give that up?
22:21
It's like the one thing that we can do and the one thing that we're we're g, you know, we've been given we've been given that gift.
22:27
I love using ChachiBT to help me write something.
22:35
Um, especially like as I've been trying to write this thing, if I get like stuck, it's a sort of like super theosaurus if I just can't figure out how to phrase something or I'm like struggling with something or like just can't get something to flow.
22:47
But it's definitely not like going to replace coming up with the ideas anytime soon.
22:54
It's an incredible tool for writers.
22:56
Like incredible tool for writers, but definitely not a writer.
22:58
Like a sparring partner, like a collaborator, like someone you can like give like a subtask to.
23:05
>> 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."
23:14
And then I'll take the sentence, quote it, and then it'll give me the output.
23:18
It's really good at that. >> Yeah.
23:21
>> How do you think that we should be training writers differently in a chat GPT world?
23:25
>> I heard this story once.
23:25
I I don't know if it's true or not, but it was like some creative writing teacher.
23:29
They would have these students come and, you know, first day or whatever, she'd like give an assignment, which is write the first paragraph of your novel.
23:37
And people would come in with all of the standard like freshman and college mistakes, like you know, way too many like stretched metaphors, way too much like flowery language.
23:48
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 that, cut that.
24:02
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 torturously overwritten >> and then the class would read them and they would say like okay what happened here like what's the and the answer was there was like no story at all.
24:18
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.
24:31
And the thing from this like teacher is that we might teach people to write beautifully but uh there's there's no interesting story.
24:39
On the other hand, you have these like sort of massly massively 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.
24:53
And and the question is like can we make it easier to get both and can we teach people how to use these tools >> 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?
25:09
>> I don't see any evidence whatsoever that AI seems to be killing writing.
25:11
I mean there's like a lot of bad AI writing like plastered over the internet.
25:15
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.
25:23
I don't think Paul Graham is sitting around being like, "AI is gonna, you know, kill my writing here."
25:28
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.
25:36
And we have much bigger issues to worry about at that point.
25:40
Even if that happened, um, let's say we have a system that can write better than a human.
25:44
Uh, do you think that the most popular novel of 2027 has a human name on it or not?
25:54
Like a human writer on it or not?
25:58
>> I think yes, >> I think it does, too.
25:58
When I finish a great book, the first thing I go do is like I want to know about the writer.
26:02
I want to know their life story.
26:04
And I don't think I'll ever have that feeling to like AI writing.
26:08
Um there there's there is something about you read an incredible book and you kind of you get to connect to a person even though you don't literally know them.
26:19
You feel like you do and you feel like you have this important shared human experience and that is like some significant percentage of the enjoyment of a great book to me and I bet we keep doing that.
26:29
The output of LLMs is is peculiar.
26:32
In one sense, it's well written in the sense that it tends not to be in academies, in jargon, in the sentence structure tends to be pretty >> um plain and sound.
26:44
Uh the even the progression of ideas tends to be orderly.
26:50
There tends to be an introductory sentence and concluding sentence.
26:54
>> Um so in that sense, it's good writing.
26:57
It's bad in the sense that it is so generic and prosaic and you could almost recognize the output of a large language model. It's so so benile.
27:06
>> Uh now perhaps it be trained and perhaps if you prompted it, you know, don't don't be prosaic. Don't be uh plain.
27:11
Uh it'll be interesting to see whether we come up with of uh with any kind of style or or freshness.
27:19
But it's not the way it's designed.
27:22
It's designed as a mashup, as a pastiche of, you know, literally, you know, billions of of examples out there.
27:30
And it's an interesting question why it should be so uh at least why it's pro style should be sound.
27:38
I mean, I think that's the best you could say about it.
27:39
Whereas the pro style of most, you know, academics, you know, most lawyers, most bureaucrats is not sound. Why is it better?
27:47
not sound. Why is it better? One possibility is just been hammered into shape through the fine-tuning and the the feedback that is instead of just regurgitating uh an amalgam of >> right the reinforcement learning >> the reinforcement learning and the uh the stuff they don't really talk about
28:06
much but that might be essential that is they're real human beings who force it to uh into you know a five paragraph essay >> right >> the other is that uh and this is completely speculative but we know in visual beauty often a composite is more attractive than the uh elements that went into the composite. So that if you
28:25
So that if you take a bunch of faces and you morph them together, the um non-existent human being that comes out of the morphing is more handsome or prettier than all of the men's faces and women's faces that went into it.
28:39
>> Really >> take a high school yearbook and you mash together the several hundred faces and the the result is is is is pretty attractive. >> Is that right? >> Yeah.
28:47
Now, whether that can be true of pros style, that is that if you were to eliminate all of the um god-awful convoluted constructions and just came up with a kind of the generic sentence structure, it wouldn't be beautiful, but it would be clear. >> That's a hypothesis.
29:12
I think of it as the most unbelievable tool I've ever seen to improve every single part of the work I do.
29:18
Like I am I have been blown away.
29:21
I've been a power user now for a year and a half, right?
29:24
Like just aggressively using every tool that comes out, trying to understand the intricacies of the tools, like stay trying to stay ahead of a field that's rapid rapidly developing.
29:34
And by the way, I'm not a I'm not a technical person.
29:35
So I'm not an engineer and I'm not like a founder of a tech company.
29:40
So my interest in this is purely at the level of like tooling like what tools are being developed for people like me to do our work better.
29:48
And honestly there there's been one other moment when I've been this blown away and it was the first time I connected to the internet.
29:54
Like the first time I connected to the internet and I was like wait wait wait wait I'm connected to literally everything and everyone if I want to be.
30:01
The the great challenge for any kind of writer is one of the great challenges is like simply like getting stuff on the page that you think reflects the thing in your head.
30:11
You have an image in your head >> and in your head it's perfect.
30:14
It's like every pixel is exactly where it needs to be.
30:18
Now you need to describe that image to somebody else and that's where the challenge is.
30:23
>> AI helps you do that better and faster. How does it do that?
30:26
Well, I'll give you an example actually from just a couple days ago.
30:30
I am I'm working to edit in my role as a as a publisher.
30:33
I'm working to edit a book about grief and the woman who wrote the book lost her daughter uh two years ago.
30:39
This book is extraordinary.
30:41
I actually haven't been able to edit it without weeping. Like it's that good.
30:44
Part of that is I'm a parent and when you read a book about somebody's child dying, you just it it hits too close to home.
30:48
There's a section in the book where she talks about how Judaism emphasizes the need for the body to be buried and in the ground very very quickly. Yes.
30:59
You're supposed to wait a maximum of 3 days.
31:02
That's the outer limit or something. >> Yeah.
31:04
Basically, I think the the soul or the spirit leaves the body after 3 days. >> Yeah.
31:08
And she talks about how there's actually a decent amount of writing in like Jewish liturgy about the importance of the speed.
31:15
Like why the speed matters?
31:17
Where does that come from?
31:17
And there are sort of two like abiding explanations, right?
31:21
And and one is that the spirit needs to to be back to God as quickly as possible or back to this back to, you know, the heavens as quickly as possible.
31:29
The other is that bodies if they were left out could become decayed.
31:35
They would start to like the the the flesh itself would would start to smell and rot and all these other things.
31:39
And so there was a there was a reason that was spirit and there was a reason that was flesh.
31:44
I wanted to understand that better and to understand what the rabbitical sources said about that better so that I could confirm that one she was telling me like she was speaking the truth. I'm not Jewish.
31:53
I don't come from this tradition.
31:55
But also if in case I missed anything like did we miss any quotes or ideas or something from the Talmud that should be in here.
32:00
Could I like go and spend a month like really getting you know understanding Judaism? Sure.
32:06
But instead, I could use AI to like take the paragraph and and ask AI to really give me sources and quotes and texture and nuance and please explain, you know, what this word means and what this word means, like is this the actual explanation?
32:20
And it gave me really incredible thoughts so that I could verify the information.
32:24
I could share some things with her that I learned that might make for better editing, right?
32:30
like it actually enhanced that paragraph taking it and putting it in AI and just asking it a series of questions.
32:34
There are people who could say that's unfair and I would say well all I did was just Google on steroids, right?
32:41
Like all I did was just like Google what I what would have taken me 10 Google searches took me one query with Claude.
32:47
The the point is that I think what it does is it one it enhances your ability to research and it allows you to pull sources together that you wouldn't have otherwise thought about. Right? There's a second thing.
32:59
Psychologically, having an editor that's always on call is new and super powerful.
33:05
I It used to be, and it still is, that you have editors and they help you punch up your work, right? No writer works alone.
33:11
They have people who work with them.
33:13
I now have the world's greatest editor, the world's smartest editor who never complains, who never gets tired, who never turns off.
33:21
So, if I'm working at 4:00 in the morning and I need my editor, I've got it. It's right there.
33:27
So if I have a paragraph that feels flabby or if I have a word that doesn't quite sit right, I can put it in and say what are some other options for this or what's wrong about this? Right?
33:36
So that's like the editorial function.
33:40
There's another important function that I use it for.
33:41
I think and this is something that t that one of my another one of my kind of people in the hall of heroes is Tyler C.
33:47
I know you're close to him as well. >> Love Tyler.
33:50
>> One of the things that Tyler Tyler talks a lot about is like the ability to argue the opposite position is insanely important.
33:55
And so sometimes, particularly with opeds, when I have an op-ed I've finished or something I've written where there's an argument, like a real argument, I will have Claude argue the opposite position for me, I will have it write take write a takedown of this oped.
34:09
Please criticize in 700 words this oped.
34:13
And I tell it, I said, "And do not hold back."
34:15
I'll like add personality to it.
34:18
Once I see how it's criticized the piece, I can then go back to my piece and strengthen it against those criticisms because everybody's got blind spots.
34:25
And so I have to actually be able to like unearth those blind spots so that I can address them.
34:29
And I would say nine times out of 10, the arguments that it makes really force me to stop and think.
34:36
Everything I've just described obviously has hazards.
34:38
There are hallucinations. There are problems.
34:42
There are times when it gets source material wrong.
34:43
But that's the whole point of human operators running these these, you know, these LLMs and using them judiciously.
34:49
I'm not substituting for the work of getting up every day and doing writing.
34:53
I'm using the tools in the places where it matters.
34:55
And I cannot emphasize enough how much better and faster my work is as a result.
34:59
Bji Shinvas has a line where he says AI is not end to end, it's middle to middle.
35:07
>> And what he means by that is when people criticize AI, they say, "Well, you're gonna need to figure out what is the thing you want to do in the first place.
35:16
Hey, there's going to be hallucinations.
35:17
You have to check it at the end.
35:18
That's the end to end stuff.
35:20
Humans have to work at the very beginning to set the vision.
35:22
Humans have to work at the very end to to verify, to fact check, to make sure everything's good.
35:28
But a lot of that middle to middle work can now be done by AI.
35:32
And and also I I think look, I think a lot of the criticism about AI is being leveled by people who don't use it.
35:37
I'll just be honest like I think a lot of the people who are critiquing it are people who have an understandable anxiety about how it might threaten their work, their income or just their like position in life, right?
35:49
But a lot of them don't actually use AI.
35:51
Like they don't actually sit down with the tools and sit down and like put things in and play around and learn how to use it.
35:57
Learning AI, learning how to use these tools effectively is a skill unto itself.
36:03
Learning how to ask better questions, create better prompts, use it for things like numerical analysis, use it to write code.
36:08
I had a friend who's an engineer at a very prominent technology company that is doing a lot with AI. Let's just say that.
36:13
And I asked him about AI because engineers in particular are also feeling the same pressure that AI could replace them, that it could be a threat to their livelihoods.
36:20
And I asked him what he thought. He has a family.
36:22
He has a stake in the future of his financial life, right?
36:26
He's not independently wealthy.
36:28
And he said, "AI is the sharpest knife that an engineer has ever been given."
36:32
He said, "But it's still a knife.
36:34
I've still got to cook the meal." Right?
36:36
And so I think of it the same way.
36:38
I'm still cooking the meal when it comes to the Kobe book.
36:41
I'm still cooking the meal when it comes to all of my other projects, but AI is an exceptionally sharp knife.
36:46
And it is unlike anything I've ever seen.
36:48
It there's one final piece that people don't necessarily think about that I talk about a lot when I encourage people to try using AI.
36:55
My my daughter doesn't have a hard time writing.
37:01
>> It's because she has no self-consciousness about the quality of her writing.
37:04
>> She doesn't care if it's good or bad. She just does it.
37:06
And it's it's an unadulterated childlike approach to creation.
37:10
And by the way, it's like marvelous.
37:12
Like all parents have experienced this at some point.
37:13
It's why, by the way, we put our crappy drawings of our kids stuff on the fridge just cuz like we're just like, "Well, look at this." You know?
37:18
Um but the the the bigger point is ego is not interfering, self-consciousness is not interfering between the impulse that my daughter has to write something and the act of writing it.
37:30
between my impulse to write and the act of writing is people who are going to hate me.
37:36
The people things are things people are going to say on Twitter like like all the interferences that get in the way.
37:41
all of the stuff that clo like clouds up cabbages up your brain that leads you to say, >> "Yeah, >> you know, I should shove this essay.
37:50
It's not going to be any good."
37:50
Like, and what if what if somebody doesn't like it?
37:52
Or what if they think less of me?
37:53
Or what if they think what if they think what if they think?
37:55
What AI does is actually allows you to short circuit a lot of that noise because it gets you to a finished product much more quickly and effectively.
38:04
So, I'll give you an example.
38:06
There's a there's a particular uh AI tool I like called Whisper.
38:08
I think it's called Whisper Flow. Yeah. >> WPR.
38:11
And it's really, really, really effective voice dictation.
38:14
And so sometimes if I'm feeling really stuck or I feel like I'm not like I'm not myself that day or something's off, I will use Whisper Flow to just dictate thoughts and it'll do a very good cleaned up transcription or translation of what I just said.
38:28
And then at least I've got a bad first draft that I can play around with.
38:32
And I didn't get in my head about it.
38:34
And I didn't wait and wait and wait and agonize and research and clean the kitchen and make sure my shoes were all lined up by the door and all the stuff that we use to excuse ourselves from actually finishing the work itself.
38:42
AI is an anxiety antidote.
38:45
It's the people people think AI causes anxiety and I'm for sure for some people it does.
38:52
For me, it's been the ultimate antidote to the thing that prevents me from getting my work done because I can get to a bad first draft very, very quickly and then I can bring to it polish and flare and attention and refinement and all of those things.
39:05
And I think that's seriously underestimated. Think of it like this.
39:09
Could I write my books by by by hand? Yeah, Robert Caro does.
39:11
So, you know, no shade to be literally like literally by hand. >> Yeah.
39:17
Could I write them by typewriter? Sure.
39:19
But I'm choosing to use a word processor.
39:20
That's how I think about AI.
39:23
AI is the next iteration of that.
39:23
I am choosing to use this tool.
39:25
The words are still my own.
39:27
The research is still my own.
39:28
The facts are still my own.
39:28
The interviews are still my own.
39:30
All those things are still, they still belong to me.
39:32
They're still coming from my brain.
39:34
AI simply accelerates all of that and gives me a far bigger just canvas of research and potential options for words and phrases and paragraphs.
39:41
And so I don't I don't see how any modern writer couldn't use AI.
39:45
And I think the idea that it is like stealing work or no, it's been trained on work, right? It's not stealing.
39:52
It's it's it's borrowing to learn.
39:55
So did Google books, but there was no outcry about Google books, right?
39:58
Like like there were there's plenty of information on Google that is the quote unquote like like cliffnotes version of a given topic or fact or book or novel or something.
40:07
There's not the same outcry.
40:09
I think of it very much in the same spirit.
40:11
And I think the courts are going to, by the way, going to find the same conclusions, too.
40:13
Think about the current narrative around AI.
40:16
It's like we people who have never thought about thinking, never thought about thinking. We're engineers.
40:22
We we thought about how to build things from the ground up using code.
40:25
There's been a lot of philosophical thought on what is like what it talks like justify a belief.
40:32
That's the field of epistemology right through to like what it means to like define agency.
40:35
And you have all this AI researchers now who are like we've built this code and it's we've anthropomorphized it in such a way that this language calculator is going to become conscious.
40:45
I'm like these are very loaded terms.
40:47
These are very loaded terms that you haven't introspected on a lot.
40:50
We don't even know what those definitions mean.
40:51
We don't have a collective definition even what the word intelligence means.
40:55
There's no single answer of what intelligence means.
40:58
There's something that a spider can do when it creates a web that's intelligent that's going towards a preference and a goal. Guys, bees and ants. Exactly.
41:05
There's like murmmoration in birds.
41:05
There's like murmmoration in birds. many different types of intelligence to have the arrogance to say if we can make it resemble our intelligence in this way like to build these kind of very smart very human sounding you know language calculators because we can see ourselves in it that is a primary form of intelligence like I think something that's way more scary from a existential
41:27
risk point of view is synthetic biology we'll actually try and build biological intelligence right like how do we build organisms from non-organic matter like that to me is a little bit more like we should be thinking about that as opposed to like if we just keep adding thousands more trading inputs to this giant language calculator data it's going to walk out the screen and kill us. [laughter]
41:46
[laughter] But it comes with this arrogance right of like everything is logical and we are so with such a great paradigm science that we figured out what consciousness thinking is like no we have like we don't have a definition of what the goal even is.
41:58
We don't even know what intelligence is that we're somehow going to get there even though we don't really know what the goal is.
42:02
know what the goal is. So yeah, the Arab the like current hubris in STEM to disregard the humanities have been thinking about these questions for a long period of time like it's just it would just save much more time if people were talking to each other like every I mean a lot of the machine learning
42:17
examples come from behaviorism they used to in the past like looking at like Pavlov's models and trying to understand like how you do reward functions that was some of the beginning basis of machine learning and then it kind of separated out and became much much more about code and not even looking at the behavior as an intelligence intelligent animals. It's like this divorce, like
42:32
It's like this divorce, like you say, like I mean, how much is that slowing down progress?
42:36
This extreme arrogance between both camps, right?
42:40
Like everyone needs to take a massive fility pill.
42:43
>> Have you written with AI at all? >> No. >> Why? No.
42:48
You're like, >> cuz why would I?
42:50
It'd be like hiring someone to have sex for you.
42:52
[laughter] You know, I love writing.
42:55
I love I love the materiality of it.
42:57
My partner has you My partner volunteers in a lot of things.
43:01
So one one point she had to do governance papers for a charitable organization.
43:06
You know how their board works like just legally ease small print kind of stuff.
43:11
So she used AI to generate that because AI could access the kinds of rules that a charity should have in terms of how it operates so it's legally protected.
43:19
So for completely non-creative stuff like that sure I'll go to AI.
43:24
You know my shoulder hurts.
43:27
I'll go to AI on the computer.
43:29
But for anything creative, I said, "Why would I want to use it when it's the very creative language that I like?" So, I've never used it.
43:35
And and I suspect we'll tire of that quickly.
43:37
I'd rather read a bad short story by a human than a good one by a computer.
43:41
Because after all, what you want in art is connection.
43:45
It's connection with another human being.
43:46
You know, a chatbot that's your boyfriend or girlfriend, that's short term.
43:51
Long term, you don't want that.
43:52
You do want that human connection.
43:52
So I find AI will probably be a useful tool in medicine.
43:56
It possibly a useful tool in in in in the law in the legal system.
44:01
But for the creative endeavors, even for music, I gather AI music can be quite good.
44:06
But it's a question of whether you know it or not. It's the deception.
44:08
If I'm told it's AI music, if it's really background stuff, like, you know, really ambient background stuff, then possibly I guess I'll listen to AI.
44:16
But if it's something that actually is meaningful, I want it to be human because it's precisely that imperfection that I want.
44:22
I think the question that's coming to mind here in terms of how much I want to read something written by AI is how much does it matter to me that there's a human on the other side of this.
44:33
So, I need to draft up a contract later this week.
44:35
I'll probably use AI to help me do it. Yeah.
44:40
>> Uh to think through the finances and stuff like that because when I send that out, the person isn't looking at me saying, "David, did you dig into the depths of your heart and soul to write this?"
44:51
But then if uh a friend is getting married, this happened to me a few a few years ago. So AI was early.
44:59
This was, you know, chat GP3 time and we're at the wedding and uh we're sitting around.
45:05
There's probably 100 people at the wedding.
45:06
And at every single table was a note that was written from the bride and the groom about why we meant so much to them.
45:17
>> And so I sit down and I open it up and you know, it's probably 80 words, 100 words.
45:21
I'm like, "Wow, that is that is so touching that they did that."
45:24
>> Don't tell me it was AI.
45:26
>> And um we have some drinks.
45:29
We're out partying that night and I go up to the groom.
45:31
I say, "Hey, I just want to let you know that was one of the nicest thing I've ever seen at a wedding."
45:38
And he gives me the old the old wink and he says, "Yeah, we got AI to write it."
45:43
And I was just like, "Dude, >> dude, I would have so much rather >> you written something that's half as sincere as that, >> filled and littered with typos >> than to have AI write that."
45:58
It felt like a violation and a betrayal.
46:02
>> We're in this weird moment where like >> the LLM's line, the generative, you know, generative AI like it just it just literally started working and it already works incredibly well.
46:09
So, it's like this step function thing from doesn't work at all to works incredibly well and it works incredibly well for helping on writing and it works incredibly well for generating art.
46:18
And we're sitting here in this moment where like in theory like within a year or two there is going to be a generative AI tool where you feed it an essay and it will give you the short film, right?
46:26
But it doesn't quite exist yet, but we're almost on the verge of that.
46:30
And then by the way, if you were going to do that today, you could hand do that.
46:32
You could use, you know, majour or stable diffusion or whatever and you could do what you're talking about like very quickly.
46:36
And I could probably literally do it myself, right, with by just feeding him the text.
46:39
And so like we're at this, but like you know, as you know, like most professional writers aren't doing this.
46:47
Most professional artists, you know, like we're all sitting here knowing that this latent ability is there and we're not actually doing it yet.
46:51
And so, and it's hard for me to judge this, but like it may be that five years from now, you know, we we'll be sitting here and it'll be like, "Oh my god, this has all changed because all of a sudden both verbal artists and, you know, visual artists are are able to express themselves in so much more rapidly and so much more vividly.
47:09
And now you've got a giant change happening in real time, which is the new coding.
47:14
The old coding interface was your code and then you run it and then you see where all the bugs are and you try to fix the bugs.
47:19
The new coding interface is you've got your code and then you've got a chatbot.
47:22
You got the co-pilot and and the co-pilot is like reading your code and like giving you like real-time comments on like, you know, you idiot, you got this wrong, right?
47:31
Um or the chatbot is actually writing the code for you and then you're giving it feedback.
47:35
You know, you idiot, you got this wrong, right?
47:37
Um and so now you got this con, we had this old concept in the old days called pair programming where you'd put two coders together in front of the same keyboard and they could talk to each other and kind of write code together.
47:45
kind of write code together. Now you're going to have pair and that was always a rare thing because most people didn't do that but now you're just going to have pair programming happening much more broadly which is human plus machine and by the way I think the this hasn't happened yet but I think the same thing is going to happen for every other form of writing >> right so I think pros writing and
48:00
fiction writing and everything else in the future I think most professional authors are going to be working in that kind of format and and you'll kind of you'll have this continuous dialogue going with an AI at the same time that you're you're you know that you yourself are are in charge of the overall product >> the only AI I'd use although I do try and switch it off is is obviously like spellch check. You know, you get the the
48:18
You know, you get the the red and that is a kind of AI and I suppose it's been >> it's been scrutinizing us for a long time.
48:25
But actually I and because my writing often is highly violating of grammatical norms.
48:29
Um they're like the grammar check goes wild does not like it.
48:35
And I'm like good this is this is excellent.
48:38
I could give you an example of what gra as it were Grammarly does not like.
48:42
So this I I sometimes um like to remove verbs from from sentences.
48:48
I leaving them technically not sentences, but this is um seeing this huge Canadian river for the first time from the sky.
48:56
Um my first sight of the river itself catches my breath. New paragraph.
49:04
A world snake in the green. New paragraph.
49:07
Cliffs dropping near sheer to water. New paragraph.
49:12
House-sized boulders on the banks.
49:12
Time falls from the rock faces above. New paragraph.
49:17
Water blue, black, and glossy in the deeper calmer runs.
49:19
Peep brown where it stretched towards and away from rapids, churning green, gold, and cream in the falls. End paragraph.
49:27
So that's five paragraphs, five sentences.
49:30
Sounds doing weird stuff.
49:33
Cliffs dropping near sheer to water.
49:36
Um, and Grammarly Grammarly does does not approve.
49:40
Uh, which I take to be a badge of success.
49:44
[laughter] >> It's sort of the paradox of Grammarly that it makes the average piece of writing, I think, much better, but it hurts a kind of write.
49:54
It takes away from distinctiveness and individuality that underlies so much great writing.
50:01
I think that's so well put and I like the idea of it hurt hurting that.
50:06
I think that's a good way of putting it.
50:08
Sam Salmon Rashi yesterday I think said writers have nothing to fear from AI until it can do comedy.
50:15
And I thought that was an interesting idea.
50:17
Like the hardest thing of all is to make us is to make us laugh.
50:22
But um I I think I think writing broadly does have a lot to fear from AI and we should resist it.
50:29
Including the absolute piracy of of our books that we now know large language model composers have been undertaking for uh for for years now and feeding absolute feeding them without any kind of recompense or licensing or permission sort into the creation of LLMs.
50:48
>> Our copyright law does not protect things created by AI. Hm.
50:49
So if it's been created by AI, you do not own it at McMillan. >> No one can own it.
50:55
It's not it's not subject to to to copyright protection.
50:59
So now, as a businessman, am I going to invest in something that has zero barrier to entry?
51:03
If I publish that book tomorrow and it's deemed not to be covered by copyright, everyone else could publish a different version of that book.
51:10
And at what level is it not copied by copyright?
51:12
Like in page 297, I redid the third word in sentence 21 on that versus this was written by AI chat GPT.
51:21
>> Well, this is there's no there's no clear line for this and that's the problem.
51:24
You know, there's a there's a test and I I just honestly can't remember what the phrase is, but it's along the lines of, you know, substantially created by a human.
51:30
You know, the the uh copyright only protects works that are sub I think it's substantially created by human, but you you can't quote me on it.
51:38
I got to get the the exact information.
51:40
But there's a test and it's not a bright line.
51:43
>> So then if I come in and I'm going to work with you, do you have to basically look me in the eye at this point and say, "Did AI write this? How how did you use AI?"
51:52
>> Well, we we've had we've had um in our contracts forever, and I don't think this is different for most publishers or any publishing contract I've ever seen where it's you make the author makes a representation that the work is their sole creation and they and then they're they're they make a representation that they will identify any parts aren't.
52:07
So here's a perfect example.
52:09
There are plenty of books that might have a song lyric or might cite someone else's book.
52:14
You know, part of the editor's job is to, you know, identify those, but really it's the author's job to say, "Hey, I quoted so- and so here."
52:20
And for fiction as well, right? You've seen it. So, this hasn't changed.
52:23
And I mean, I've been in working for publishing companies for the last 25 years.
52:27
I've been in the entertainment business, you know, for the last, you know, 35 years.
52:30
This is no different than it's ever been.
52:33
So, then back to our thought experiment.
52:35
So, now I'm sitting down.
52:37
I'm going to write a book.
52:38
John, give me some advice. I'm going to use AI. I want to use AI.
52:41
Um, how should I be working with AI?
52:44
How should I be tracking?
52:45
Do I need to be tracking all the AI that I use?
52:47
Like, how do I do this in a ethical way that speaks to your creative side, the business side, the legal side?
52:55
>> I think it goes back a little bit to what we were just talking about a couple minutes ago.
52:58
If you use it as a tool, >> Yeah.
53:01
>> You know, a great example would be a research assistant, right? Sure.
53:03
If you used AI as a research assistant, you said, "Look, I've read these 10 books."
53:09
And you had permission to dump them into AI, you know, because that's a whole other thing.
53:12
And you then query, you say, "Okay, my theme or that I'm trying to get across is this or I'm trying to get documentation or, you know, trying to annotate this thought."
53:20
AI could help you find that way quicker than having to read through all those books again. That's a fine use of AI.
53:26
I think that's no different than being able to, you know, scan an ebook, right?
53:30
I would call that Google on steroids category.
53:33
I guess that's fair enough.
53:35
But so that to me is, you know, that's a that's a tool for fiction.
53:38
It might be maybe it's a thought partner.
53:40
Maybe it's like, hey, here's my manuscript.
53:43
And then you can start querying the the the chatbot to say, um, you know, give me your your your two cents on this character development or something.
53:53
Maybe that's a a a good example, but I don't know.
53:56
I haven't given this incredible amount of thought, but those are things that just come to mind about how you could use AI to maybe sort of as a tool