0:00
Every day, every day, you should be in ChatGPT.
Search this video
0:03
I don't care what your job is, right?
0:05
You could be a sommelier at a restaurant, and you should be using ChatGPT every day to make yourself better um whatever it is you do.
0:17
Can I ask you about the story really quick?
0:19
And you have like you have like a list of stuff here that's like all amazing.
0:21
It's actually a lot of it's very actionable, but the reason I want to ask you about the story is for the listener.
0:25
Dharmesh founded HubSpot, $30 company.
0:28
You're the CTO, so you and you're you're an OG for web 1. 0 web 2. 0.
0:33
And your first uh round, or one of your first rounds, was funded by Sequoia.
0:37
Your partner Brian is an investor at Sequoia.
0:40
So, you are in the insider.
0:40
You you're an insider, I believe.
0:43
Uh you may not acknowledge it.
0:44
I don't know if you do or do not. You are an insider.
0:46
The cool part is that you're accessible to us.
0:49
When did you first see what Sam was working on?
0:53
And how long have you felt that this is going to change everything?
0:57
So, I actually have um known Sam before he started Open AI, and I got access to um the GPT API.
1:03
It was a toolkit for developers to be able to kind of build AI applications, right?
1:09
Effectively um And so, I built uh this little chat application that used the API, and so I could have a conversation with him.
1:16
So, I actually built that thing uh that night.
1:19
Uh it was it was a Sunday.
1:19
I have the full transcript 2 years before ChatGPT came out. So, that's 4 years ago?
1:25
Uh it's 2020, so 5 years ago. Wow, okay. >> Uh this summer.
1:29
And so, even then, it's like and like as soon as I like you sort of have that moment, the same that all of us have with ChatGPT, I just had it 2 2 years earlier.
1:38
And then I'm showing everyone, like Brian, you are not going to believe like I have this thing, you know, through this company called Open AI, and watch me like type stuff into it and see like see what happens.
1:47
It's And we would ask it like strategic questions about HubSpot.
1:50
It's like how should it like who are the top competitors like uh And they were shut even then, 2 years before chat, it was shockingly good, right?
1:58
But the thing you sort of have to understand uh but the constraints of how a large language model actually works is that you type, and you have a limited Just imagine this if we're going to just use the uh the physical analog uh there a sheet of paper can only fit a certain number of words on it.
2:14
And that certain number of words includes both what you write on it.
2:15
This is I want you to do this, and the response has to fit on that sheet of paper.
2:22
And that sheet of paper is what uh in technical terms would be called a context window.
2:26
And you'll hear this tossed around.
2:28
It's like, "Oh, this, you know, ChatGPT has a context window of whatever, or this model has a context window of whatever."
2:32
That's what they're talking about. All right.
2:34
So, why is that Why does anybody care about the context window?
2:37
It's like, "Well, sometimes you want to provide um a large piece of text and say, "Summarize this for me."
2:44
Well, in order for you to do that, it has to fit in the context window.
2:46
So, if you want to take two books' worth of information and say, "I want you to summarize this in 50 words," those two books' worth of information have to fit inside the context window in order for the LLM to process it.
2:56
Most the the frontier models are roughly 100,000 to 200,000.
3:00
They measure it in tokens, which is like 0.
3:02
75 of a word, but That's like a book.
3:04
So, yeah, is that a book?
3:07
So, I think it's an average I think the average book is like 240,000 words I think, but I'm not sure. That's not a lot.
3:12
So, what I the way that I use ChatGPT is I'll uh like let's say a fun way is I'll I'll put a historical book that I loved reading, and I'll be like, "Summarize this so I remember the details."
3:21
So, you're telling me that if it's a 1,000-page book, it's not even going to uh accurately summarize that book? It won't fit.
3:27
You like if you paste something large enough into ChatGPT or whatever um AI application you're using, it will come back and say, "Sorry, that doesn't fit."
3:36
Effectively, what they're saying is that does not fit in the context window.
3:39
So, you're going to have to do something different. All right.
3:43
A few episodes ago, I talked about something, and I got thousands of messages asking me to go deeper into explain, and that's what I'm about to do.
3:51
So, I told you guys how I use ChatGPT as a life coach or a thought partner.
3:55
And what I did was I uploaded all types of amazing information.
3:58
So, I uploaded my personal finances, my net worth, my goals, different books that I like, issues going on in my personal life and businesses.
4:07
I uploaded so much information, and so the output is that I have this GPT that I can ask questions that I'm having issues with in my life, like how should I respond to this email?
4:19
What's the right decision knowing that you know my goals for the future, things like that.
4:23
And so, I worked with HubSpot to put together a step-by-step process showing the audience, showing you the software that I used to make this, the information that I had ChatGPT ask me, all this stuff.
4:34
So, it's super easy for you to use, and like I said, I use this like 10 or 20 times a day.
4:37
It's literally changed my life.
4:39
And so, if you want that, it's free. There's a link below.
4:42
Just click it, enter your email, and we will send you everything you need to know to set this up in just about 20 minutes, and I'll show you you how I use it again, 10 to 20 times a day. Um all right. So, check it out.
4:52
The link is below in the description. Back to the episode.
4:57
I usually use projects, and I have like let's say a health project, and I'll upload tons and tons of books or tons of blood blood work, and I hope I'm hoping that it's going to pull from all those books in my project. Is that true? That that is true.
5:09
So, here and and this is a perfect segue, right?
5:11
Because this is the next big unlock.
5:12
So, the number one thing to like understand in our heads is there's this thing called a context window. Here's why it matters.
5:17
Um So, let's we're going to take a um we're going to pop that on the stack, and we're going to uh push down the stack, and we're going to come back to it.
5:24
But the thing you have to remember is two things.
5:26
Um number one, it doesn't know what it's never been trained on.
5:30
That's one of the limitations, right?
5:32
So, if you ask it something that only you, Sam, have in your in your files, in your email, whatever that the train the training model was I mean, the LLM was never trained on, it's not going to know those things.
5:41
Doesn't matter how smart it is, it's just information it's never seen.
5:44
So, it's not going to know that.
5:46
That's kind of problem number one. Problem number two.
5:48
So, let's say your website for Hampton was actually on um uh in the training set, right?
5:54
Because it's on the public internet or whatever.
5:57
But the training happened at a particular point in time.
5:59
Like they ran the training, ran the training, ran the training, and said, "Okay, we're done with the training now. The machine is done.
6:05
Let's let the customers in, right?"
6:07
Now, if the website changes, it's not going to know about those new updates that you made to your website because the training was done at a particular date if it completed its kind of training course, right?
6:18
So, those are the two things we sort of have to remember is that it doesn't know what it doesn't know.
6:21
And number two, that the things it did know were frozen at that particular point in time, right?
6:26
So, it hasn't seen new information.
6:28
And those are relatively large limitations, right?
6:30
So, especially if you're going to use it for business use or personal use, it's like, "Well, I've got a bunch of stuff that I will want it to be able to kind of answer questions about or whatever I inside my company or inside uh in my own personal life, how do I get it to do that?
6:42
Uh and so, here's the hack, and this is this was a brilliant uh brilliant discovery.
6:46
So, what they figured out is to say, "Okay, let's say you have a 100,000 documents that were never on the internet, that's in your company, it's all your uh employee hiring practices, your model of here's how we do compensation, all of it, right?
7:01
So, like all of you have 100,000 documents, and obviously you can't ask questions about those 100,000 documents straight to ChatGPT.
7:05
It doesn't know anything about those, never seen those documents.
7:09
So, this is And we talked about this two episodes ago, um this thing called vector embeddings and rag, retrieval augmented generation.
7:15
Um and I'll I recommend you folks go listen to that.
7:19
I think it's a it's a it's a fun episode, but uh I'll kind of summarize it, which is what you can do, and what we do is to say we're going to take those 100,000 documents, and we're going to put them in this special database called a vector store, a vector database.
7:29
And what we can do now is when someone asks a question, we can go to the vector store, not the LLM, go to the vector store, and say, "Give me the five documents out of the 100,000 that are most likely to answer this question based on the meaning of the question, not keywords, based on the actual meaning of the question."
7:46
So, it's called a semantic search is what the vector store is doing.
7:49
So, it comes back with five documents, let's just say.
7:52
Now, as it turns out, five documents do fit inside the context window.
7:57
So, effectively, we said, "Okay, well, yeah, it would have been nice had you trained on the 100,000 documents, but that was not practical because I didn't want to expose all of that.
8:03
I'm going to give you the five documents that you actually need.
8:06
I'm going to say give it to you in the context window, and now, as you can imagine, it does an exceptionally good job at answering the question when it knows the five documents that you should be looking You just gave it to them, right?
8:16
So, it's having this uh So, we'll kind of uh jump metaphors here.
8:20
It's like hiring a really, really good intern that has a PhD in everything. Right?
8:24
They went to school, they read all the things, read all the internet.
8:27
Uh the intern knows everything about everything that ever was publicly accessible. They're trained.
8:30
Show up for the first day of work. That's all they know.
8:34
They're not learning anything new, and they know nothing about your business.
8:36
Now, it's like, "Okay, well, I know you know everything about everything.
8:41
I have this question about my business.
8:43
Here five documents that you can read right now and answer my question."
8:48
It's like, "Oh, I can do that."
8:50
I like that analogy, the intern with the PhD and everything.
8:52
This is that's so much how it is, right?
8:54
It's It's as helpful and available as an intern, uh but it's as knowledgeable as somebody with a PhD and everything. Yes.
9:01
Um and then like you said, you know, my the another analogy for that is like it's a store.
9:06
You have shelf space, which is kind of limited, but they do have a back, and you can always get send the employee to the back and see if they can find it in the back for you, right?
9:14
That's kind of like what you're saying.
9:15
Put it in the database, they can go fetch the specific thing that you're asking for uh because, you know, you gave it access to the back.
9:21
You gave it a badge that lets it go in there.
9:22
Have you uploaded all of like Have you First of all, I want to know what your ChatGPT looks like.
9:28
I want to know how you use it on a like I just want you to just screen share.
9:31
Just like show me exactly what you do.
9:33
But also, have you uploaded your entire life?
9:35
Like have you uploaded all of HubSpot to ChatGPT where you can just ask it any question?
9:40
Yeah, multiple times, right? Um so, In what format?
9:44
Tell me how you did that.
9:44
Uh so, I did uh so, Open AI has um called this uh it's called an embeddings algorithm that takes any piece of text, a document, an email, whatever it happens to be, and creates this kind of point in the high-dimensional space, um called you know, called a vector embedding.
9:58
And you know, a point in high-dimensional space So, in three-dimensional space, physical space that we know of, we think of points being in three dimensions, X, Y, and Z axis.
10:07
Like, "Oh, here's where this point is in space."
10:09
High-dimensional space, you could have 100 dimensions, you could have 1,000 dimensions, and you describe each document as this kind of point in space.
10:15
So, what I've done So, it used to be in the early kind of GPT world, the number of dimensions you had access to was roughly like 100 to 200 dimensions.
10:24
And so, we would lose a lot of the meaning of a document, right?
10:26
They would sort of get it right, it would sort of capture the meaning.
10:27
Uh and then uh then we went to like 1,000 dimensions.
10:32
It's like, "Oh, well, now it can much more accurately sort of represent um and and capture a document of kind of arbitrary length and and be able to find it, uh give it a prompt or given some sort of search query."
10:43
Uh and then recently, within the last year, we've gone the the latest algorithm from OpenAI, embedding's algorithm is like 3,072, I think, uh dimensions.
10:53
But where where do you do this?
10:55
Do you just literally upload it as a project?
10:57
Are you you had to do it API connection?
10:58
What what did you How do you actually do this?
11:01
>> an API connection, right?
11:01
In fact, I'm running the most Let me see where it is now.
11:05
And anyone can do this, or you have special access cuz you're friends with No, anyone can do this.
11:09
The the API for the embedding's model, they have two versions.
11:13
They have the 3,000-dimension version, they have a 1,000-dimension version.
11:16
And is the results of this like are you driving a NASCAR and I'm driving like a scooter?
11:21
Like, is that the difference?
11:21
Like, if I just like, for example, what I would do is I'll just like download my company's financials and I'll upload it.
11:27
And then I'll like explain what my company does.
11:31
But the way that you do it is a lot different.
11:33
Now, are we talking a massive gap in results that you get versus what I get? Um yes. I don't know. The short answer is yes.
11:43
Uh and and the reason is like So, I do I do that as well in terms I'll describe a company or whatever.
11:47
I try to provide it context, and that's why it's called a context window.
11:50
You try to provide the LLM uh context for what it you're asking it to do.
11:54
Um you know, the difference is that, you know, because I can go through Like, and by the way, the richest and I'm working on a kind of nights and weekends project right now um that takes uh email, which, you know, So, you would be amazed.
12:07
Like, if you had to write No other words right now, if you did nothing but say, "I'm going to take all of my emails I've ever written uh that are still stored and give it uh to a vector store, use an embedding's algorithm, and then use ChatGPT to let me kind of answer questions."
12:22
So, if I want to say, "Oh, I want you to give me a timeline for when we start first started using Hub to main products or whatever.
12:29
How did that come about?"
12:30
Or what were the winning arguments against doing that versus whatever?
12:34
Like, it's shocking how good the responses are when you give it access to that kind of rich data, right?
12:41
>> Somebody needs to create just like a $10 a month a single website that's like, "Hey, make your ChatGPT smarter."
12:45
And it's a website where it's like, "Connect your Gmail, connect your Slack, connect your everything."
12:51
And I would pay them happily 20 bucks a month to just set this up for me so that my chat to to give my chat GPT like the extra pill that says, "You now have access to my data."
13:02
Is this a cuz cuz you're talking about like, "I have the API to the vector embeddings."
13:05
And like, "Well, I have the flux capacitor, too, but I don't know what to do with it, right?
13:09
Like, I need a button on a website with a stripe payment button that I could just connect the stuff.
13:13
Is it not Is there is there a caveman version of this?
13:16
Uh there's I mean, there are uh tools out there to do and there's startups working on it, right?
13:20
Uh there's two pieces of good news.
13:21
One is there are startups working on it.
13:23
The challenge here is uh not that they're doing a you know, bad job.
13:28
The challenge actually comes down to uh if it were a startup and a startup came to you, it's like, "Oh, we just started last last week, but we've got this thing it it really works."
13:36
Uh in fact, I'm actually maybe an investor.
13:40
How willing would you be to hand over literally your entire life and everything that's in your email over to the startup?
13:46
That's the part of the challenge we have is that the access control that let's say you're using Gmail, which a lot of us use, when you provide the keys to your Gmail account to a third party, uh there's no degree of real granularity.
13:58
You can say, "Oh, I want it to read the metadata." That's like level one.
14:02
Level two access, I want it to read my full email.
14:03
And level three is I want to be able to write and delete emails on my behalf.
14:08
But if you wanted to like read like the actual body of the email, you can't say, "I only want it to read messages that are from hubspot. com."
14:14
Or I only I want to ignore all messages from my wife and my family or whatever in the thing.
14:19
There's no way to control that, right?
14:20
So, you sort of have to have a trust.
14:20
Is there any product that you would trust right now or that you can recommend that guys like Sean and I should use as ChatGPT add-ons or accelerators?
14:32
No, not I not that I don't trust them, but it's like I wouldn't trust really anyone right now with that.
14:36
And it's one of the reasons I sort of run it locally, even though I know these things are out there.
14:41
I predict what's going to happen is we're going to have uh any of the major players, and you can see this happening already, right?
14:46
We see this with um you have the ability to create custom GPTs in OpenAI and do projects and Claude, you have Google Gemini, which are essentially like a small baby version of this, right?
14:56
It says, "Oh, you can upload 10 documents, 100 documents, and it'll let you ask questions against it."
15:01
What it's really doing behind the scenes is creating a vector store.
15:04
That's effectively what it's what's happening.
15:07
Um my expectation is all the major companies um will actually have a variation of this.
15:14
Starting with Google should be the first one because they already have the data.
15:18
There is absolutely zero reason why Google Gemini does not let you have a Q&A with your own email account.
15:22
That's just like insanely stupid, right?
15:25
Like, I'll just go ahead and say it.
15:27
It's just it's just there's something not right with the world uh when they already have the data.
15:33
And it's like And they have the algorithm. They have Gemini 2.
15:34
5 Pro, which is an exceptionally good model, right?
15:37
So, they have all the pieces.
15:37
Uh but have not yet delivered.
15:40
What I hope is they'll listen.
15:42
Did tell me and Sean we're early adopters, but neither of us are technical.
15:46
What can we do to I want to get it on this, baby. >> All right.
15:50
So, give me give me two weeks.
15:51
Here's Here's what we Uh So, there's one thing I do trust that I trust myself.
15:54
Uh I'm I'm an honest guy.
15:57
Uh I'll give you like this internal app that I'm building.
16:01
Let you point your Gmail to it.
16:01
It'll go and it'll run for a day or two days or something like that.
16:04
And then you will be amazed.
16:06
You'll be able to ask questions.
16:08
Um and by the way, like and the thing I'm like working on now is once you have this this capability, right?
16:13
Like, step one is just being able to do Q&A, right?
16:17
It's like, "Oh, I'm just I'm going to step two, like imagine kind of fast-forwarding.
16:20
Like, it has access to all of your kind of history."
16:21
So, imagine you're able to say, "You know what?
16:24
I'm not doing this, by the way, but if I were, it's like, "I want to write a book about HubSpot and all the lessons learned and all like everything. It's all in my email.
16:32
Do the best possible job you can writing a book.
16:33
If you have questions along the way, ask me.
16:36
Other than that, write the book."
16:36
I think it would be able to write the book. Wow.
16:40
What what else are you doing with AI?
16:42
So, give me your day-to-day.
16:43
I like, for example, the CEO of Microsoft had this great thing where he goes, "I think with AI, then I work with my coworkers."
16:50
And that really shifted the way I worked cuz I used to brainstorm, have a meeting to talk about stuff with my coworkers, which was honestly always like a little disappointing.
16:58
I felt like I'm the one bringing the energy and the ideas and the questions, and I'm hoping that they're going to But dude, just sparring with AI first and then taking the kind of like distilled thoughts to my team of like, "Here's how we're going to execute." has been way better.
17:11
Like, that little one sentence he said shifted the way I was doing it.
17:15
How are you kind of using this stuff?
17:17
Yeah, so a a couple of things.
17:19
So, I'll Let's start the high level and we'll drill in a little bit.
17:21
So, uh what we're used to with uh ChatGPT, this is sort of your kind of early evolution of most people's use, is because it's called generative AI, you use it to generate things, right?
17:31
Generate a blog post, generate an image, generate a video, generate audio, all those things.
17:35
That's kind of the generation kind of aspect.
17:37
Uh and that's part of what it's good at.
17:39
Then you sort of get into the, "Oh, but it can also kind of summarize and synthesize things for me."
17:44
It's like, "Oh, take this large body of text, take this blog post, take this academic paper, and summarize it in this way or like so a 7-year-old would understand it" kind of thing, right?
17:51
So, that's the kind of step number step number two.
17:54
Step number three, and we're going to get into how this is now possible, um is you can do um effectively, you can take action um have the LLM actually do things for you.
18:06
Uh and then I always kind of put it broadly in the kind of automation bucket.
18:08
Like, I can automate things that I was doing manually before.
18:12
And then the fourth thing is around orchestration.
18:13
It's like, "Can I just have it manage a set of AI agents?"
18:16
And we'll talk about agents in a little bit.
18:19
And just do it all for me.
18:20
I just want to give it a super high-order goal.
18:21
It has access to an army of agents that are good at varying different things.
18:25
I don't want to know about any of that.
18:26
I just want it to go do this thing for me, right?"
18:30
And And that's sort of where we are on the so so the curve.
18:31
Uh the first three things are possible today.
18:36
And work well today, right?
18:36
So, you can generate As we know, it can generate blog posts, it can write really well.
18:39
Uh it can generate great images now, including images with text.
18:43
It can do great video now with uh you know, higher fidelity, higher character cohesion, all these things.
18:46
Uh Sean, so the thing the the vision you had 3 years ago when I was on was around creating the next Disney, the next kind of media company.
18:53
You have the tools now, my friend, to finally start to approach that, right?
18:57
But then you should sort of move into and this is what we were just talking about, this kind of synthesis and analysis thing.
19:02
This is Okay, this is where deep research kinds of features come in.
19:03
It's like, "Okay, well, I want you to take the entirety of the internet or entirety of what Sean has written about copywriting, and I want you to write a book just for me that summarizes all of that in ways that I enjoy because I like I like analogies and I like jokes and I like this and I like that.
19:17
Write a custom version of Sean Puri's book on copywriting, right?
19:20
That kind of synthesis um I think would be super interesting.
19:24
And then automation is now possible. So, agent.
19:25
ai is one of those things.
19:27
There's other, you know, tools out there that says, "Hey, I want to take this workflow, this thing that I do, and I want you to just do it for me." Give us a specific.
19:33
What's a specific specific automation that you you've used that's like, you know, useful, helpful, saves you time?
19:38
I'll I'll tell you a couple.
19:40
One is around domain names, which is, okay, so I have an idea for a domain name.
19:46
Um and I'm going to type words in. And these things exist.
19:48
And I'll tell you the the manual flow that I used to go to is like, okay, first of all I can brainstorm myself and come up with possible words and various simple words whatever here's the things.
19:56
Then I'll say okay, which domains are available?
19:58
Absolutely zero of them that are good that will pop into my mind or like freely available to kind of just register that no one's registered before. Okay, fine.
20:04
Then I'll say okay, well which ones are available for sale?
20:08
Okay, what's the price tag?
20:08
Is that a fair approximation of the value?
20:10
Is it like below market, above market?
20:12
We don't know because there's no Zillow for domain names yet. Uh so I create that.
20:17
So I have something that automates all of that and says oh, so you have this particular idea for this concept for this business business whatever it is.
20:25
Uh here names, here the actual price points, here's the ones that I think are below market value, above market value.
20:30
Tell me which ones you want to register. >> That's in ChatGPT? No, it's in agent.
20:33
ai is where it lives right now.
20:35
But now there's a connector between agent.
20:37
ai and ChatGPT through this thing called MCP which you'll hear about uh a bunch of if you haven't already.
20:43
Um One thing I I want to kind of get out there just so we keep connecting the dots.
20:47
Um like what I want everyone to have this framework in their head.
20:50
Uh so we talked about large language models that can generate things.
20:53
We talked about the context window.
20:54
We talked about faking out the context window by saying oh, we can do this vector database and bring in the right five documents, stuff them into the context window.
21:03
Uh here's the other big breakthrough that's happened uh I'll say recently within the last year year and a half is what's called tool calling.
21:10
And what tool calling is is a really brilliant idea.
21:12
And the tool calling says okay, well the LLM was trained on a certain number of things, but if we had this intern that came in, it would be like saying okay, well whatever you know, you know, but we're not going to give you access to the internet.
21:24
Like that would be stupid, right?
21:24
We would give the intern access to the internet.
21:27
It's like if I ask you something that you weren't trained on, go look it up, right?
21:29
That that would be like like thing number one on the first day of work.
21:32
And as it turns out the LLM world the intern couldn't didn't have access to the internet.
21:38
All it had was whatever notes it happened to take during its PhD training in all things, right?
21:41
And so what tool calling allows, and this is a weird um weird approach to it, but this is because the way LLMs work.
21:46
So remember the the LLM it's architect is such that you give it the context window in, it spits things out. That's it.
21:52
It doesn't have and you can't reprogram the architecture to say now all of a sudden we're going to give you access to tool calling.
22:00
So here's the hack that they came up with.
22:01
They said okay, in the instructions that we give it in the context window we're going to say you have access to these four tools.
22:09
And it doesn't actually have access to the four tools.
22:12
It's that I want you to pretend like you hacks have access to these four tools.
22:16
The first tool is this thing called the internet.
22:18
And the way the internet works is you type in the query and it will give you some things back.
22:21
You have this other thing called a calculator.
22:23
And you can give it a mathematical expression and it gives you an answer back.
22:27
And you have this other tool that lets you do this.
22:29
And you can have n number of tools.
22:32
And so here's what happens.
22:34
In the context window happening behind the scenes ChatGPT, which is the the interface right now that is interacting with the LLM.
22:41
You're not talking with the LLM directly, right?
22:44
It gets a prompt and it says okay, by the way LLM, I want you to pretend like you have access to these four tools.
22:50
And anytime you need them, when you pass the note back to me the results, the output, just tell me when you want to use one of those tools. All right. So we give it a query.
22:57
It's like okay, well I want to look up like the historical stock valuation for HubSpot and when it changed as a result of is there any correlation to the weather?
23:06
Is it seasonal or whatever it is, right?
23:07
Uh in terms of market cap of HubSpot versus um um seasonal changes.
23:13
All right, well that's not something we have access to, but here's what actually happens. This is so cool, right?
23:15
So the LLM gets it and the LLM's in the context window that we gave it, we gave it instructions to pretend like you have these four tools.
23:24
One of which is stock price lookup, let's say historical stock price lookup.
23:27
It'll pass the output back to the application, not us and say and it in the in the output it says oh, please invoke that tool you told me I had access to and look up this result.
23:35
I want you to search the internet for X, what was the weather?
23:39
I want you to do this for the stock price.
23:42
And then we do that, we the ChatGPT application fill the context window with whatever it is the LLM asked for and then pass it back in.
23:52
So the LLM effectively has access to those tools even though it never accessed the internet, it never accessed the stock market.
23:57
But it pretended like it had access to it. And we never see this.
24:01
This is happening behind the scenes.
24:03
Now here is the big massive unlock, right?
24:06
Which is well, everything can be a tool, right?
24:09
Now you don't have to build this kind of vector store or whatever because you would never build a vector store of all possible stock prices from the dawn of time.
24:15
I guess you could, but then it's outdated immediately.
24:16
Now it's like what if we just gave it 20 really powerful tools including browser access to the internet?
24:24
Well, that's like a 10,000 100,000 times increase in that intern's capability, right?
24:29
And so that's where our brains should be headed now, which is exactly where the world is headed that says what tools can we give the LLM access to that will amplify its ability and cause zero change to the actual architecture.
24:42
Literally, it doesn't have to know anything about anything.
24:43
It's like I just want you to pretend that you have access to these tools.
24:46
Uh it doesn't need to know how to talk to those tools.
24:48
Doesn't need to know about APIs.
24:48
Doesn't need any of that stuff.
24:50
Do you think that I mean this is all about mind-blowing and you have an interesting perspective because you know, I think three episodes ago that you're on, you created this thing called Wordle. Was it Wordle? Word Wordplay. Wordplay.
25:02
That does like 80 grand a month.
25:04
It was just like a puzzle that you do with your son. It was amazing.
25:07
Um but now you have new projects.
25:10
You have agent AI, you have your other things.
25:12
But you still run a $30 company.
25:17
Do you think that the majority of value creation like am I going to is my stock portfolio going to go up because I own a basket of tech stocks or is the best way to capitalize as an outsider?
25:28
Obviously you start a company.
25:32
Or is it investing in new startups that are using AI or AI first startups?
25:39
It's a yeah, it's a big question.
25:39
I'm neither an economist nor a stock analyst.
25:42
But I will say this, the thing I'm most excited about with uh with AI and I actually said exactly this in a talk I gave um well before GPT on the inbound stage and I said you know, as AI is starting to kind of come up, it's not a you versus AI.
25:57
That's not the mental model you should have in your head.
25:59
It's like oh well, AI is going to take my job because it's me trying to do things that the AI is then eventually going to be able to do.
26:03
Uh the right mental frame of reference you should have is the you to the power of AI.
26:08
AI is an amplifier of your capability.
26:11
It will unlock things and let you do things that you were never able to do before.
26:14
As a result of which it's going to increase your value, not decrease it, right?
26:18
But in order for that to be true, you actually have to use it. You have to learn it.
26:23
You have to experiment with it.
26:25
And the only real way to get a feel for what it can and can't do is you have to do it.
26:30
So I'll give you the very very simple, everyone should do this.
26:33
I do this uh personally is that anytime you're going to sit down at a computer and do something research, whatever it is you're going to do you should give ChatGPT or your AI tool of choice a shot at it.
26:44
Try to describe and pretend like you have access to this intern that has a PhD in everything.
26:52
It's like okay, well maybe it doesn't know anything about me or whatever. Fine.
26:54
So then tell it a few things about you.
26:56
But imagine you have access to this all-knowing intern that has a PhD in everything.
27:00
Give it a crack at solving the problem that you're about to sit down and spend some time on.
27:02
And what you will invariably find, number one is you'll be surprised by the number of times it actually comes up with a helpful response that you would never have expected it would be even remotely able to do. Like how can it do that?
27:15
It's because it has a PhD in everything, right?
27:17
And so now actually we'll talk about reasoning and whether models are actually doing that or not if we have time.
27:21
But um so that's my advice is every day, every day um you should be in ChatGPT.
27:30
If you're a knowledge worker at all, it doesn't matter actually.
27:31
You don't even have to be a knowledge worker.
27:32
I don't care what your job is, right?
27:33
You could be a sommelier at a restaurant and you should be using ChatGPT every day um for to make yourself better um at whatever it is you do.
27:41
And that might be the introduction of that orthogonal skill.
27:44
So uh orthogonal means that line that's 90 degree intersection to another line.
27:48
Uh and the most common use is when we have an X and Y axis. Right?
27:50
It's like oh, the X axis and the Y axis are orthogonal to each other because they have 90 degrees separating them.
27:55
The common usage when you say oh, that's an orthogonal concept, it means it's unrelated.
28:00
It's completely different.
28:02
That's like the Y and X axis are completely independent of each other.
28:05
You can say oh, you can be here on the X axis but here on the Y axis and they're not related to each other.
28:08
So that's what I mean when I say orthogonal concepts or skills or ideas. Um yeah, anyway.
28:13
Is there anything you disagree with that's kind of the consensus?
28:17
Because a lot of things you're talking about like hey, AI is going to change everything. It's super smart. Agents are coming.
28:22
They can do some stuff now, more stuff later.
28:23
These are all probably right, but they're also consensus.
28:26
I'm just curious like is there anything you disagree with that you hear out there that drives you nuts where you're just like oh, people keep saying this.
28:32
I think that's either wrong, it's overrated, it's the wrong timeline, it's the wrong frame, it's uh whatever.
28:39
Is there anything that you disagree with that you've heard out there?
28:43
I've heard variations of two variations I disagree with.
28:46
One that I've I think spent so much time uh hopefully kind of talking folks out of, which is it's just autocorrect, it's not really thinking.
28:54
Um and and that's a matter of like what do you think thinking is, right?
28:58
It's like okay, well if it produces the right output to which we think would require thought.
29:02
Um so I think that is is is flawed reasoning to say oh well, and this often comes from the the smartest people, the most experts in their field because oh, it's really like a stochastic parrot.
29:12
You'll hear this phrase, which is it's like a probability driven pattern matching base.
29:17
It just so happens it's been trained on the internet, but it's not really like human intelligence.
29:22
And I agree with that phrasing, which is it's not like human intelligence, but that does not mean that all it's doing is sort of mimicking stochastically uh you know, all the things I've read before because in order to do what it does it is a form of creativity different from what we normally experience.
29:35
That's kind of thing number one that I kind of disagree with.
29:39
Thing number two is people are thinking I both disagree with the oh, the scaling laws are going to continue forever indefinitely that the more and more compute we throw at the more knobs we put on the machine the smarter smarter it's going to get.
29:50
I think there's going to be a limit to that at some point is like nothing goes on forever.
29:55
It's going to asymptotically move towards we're going to come up with new algorithms.
29:57
That's GPT can't be the do all end all things, right?
30:00
There will be a new way um um you discovers.
30:03
I think that's going to happen, but I I think the smarter I I did I see this other people have said it the best way to kind of think about AI right now is uh as you use it is to kind of truly find find the frontier of what it's incapable of.
30:20
It's like okay, it can sort of do this thing but not very well.
30:23
If if that's the way you describe its response, you are exactly where you need to be, which is if we can sort of do it right now it's sort of if you have to squint a little bit it's like ah well it's kind of something, but wait 6 months or a year, right?
30:36
Like it's that's the beauty of an exponential curve it gets so better so much better so fast that if it can sort of do it now it will be able to do it and then it'll be able to do it really well.
30:45
That's the inevitable sequence of events that's going to happen.
30:49
>> Sam, have you heard this about startups?
30:50
There's like a kind of the smart money in startups believes that the the right startup to build is basically the thing that AI kind of can't do right now.
30:57
That's the company to start today because you just have to stay alive long enough give it the the 12 to 18 month runway that it needs for the thing to go from didn't really work very well to like oh my God this is amazing, but you've built your brand your company your mission you've your customer but you've been building that all along the way and you're basically just betting you're going to be able to surf the improvement of the model.
31:20
>> that's that's how I feel about my company.
31:22
My company is not even related to this but at all, but in terms of like our operations were like things are very manual and I'm like oh my God once I'm able to finally implement AI when it can work for this purpose my profit margins are going to go through the roof.
31:34
I mean that's how I that's how I feel about it, but it which isn't entirely related to that Sean, but a little bit.
31:39
I one one one thing I'll plant out there since this is my first million week we like talking about ideas.
31:45
At a macro level here's the entirely new pool of ideas that I think are now available and a trend that I think is inevitable, which is as agents get better and better, right?
31:56
Right now most of us when we use use AI use chat GPT we use them as tools, which is great. Perfect fine.
32:05
Over time you need to shift your thinking and think of them as teammates.
32:08
Think of them as that intern that just got hired, right?
32:10
And and as a result of that so let's let's assume for like let's stipulate that I I'm all we don't know is how long is it going to take for me to be right is that we're going to have effectively digital teammates that are part of all of our teams.
32:25
Every company is going to someday have a hybrid team consisting of carbon based life forms and these kind of digital AI AI agents.
32:31
Okay, so if you accept that the way that's going to happen is not going to be like all of a sudden we one day wake up and every organization now starts kind of mixing them.
32:39
What's going to happen is it's going to slowly introduce this way it's like oh I have this one task whatever that an agent is better at is well it's reliable enough for the thing and the risk is low enough I'm going to have it do that.
32:48
We already see elements of that.
32:52
But here's what's going to happen as a result of that kind of gradual kind of infusion and adoption of that technology the way to win and the opportunities that get created is like how do I help the world accomplish this end state that I know is going to come.
33:04
So here I'll give you some examples.
33:07
If we were to hire if you Sam were to hire a new employee tomorrow here's what you would do.
33:12
You would say oh well I'm going to onboard that employee spend a couple of days I'm going to tell them about the business whoever's managing that employee let's say it was a direct report of yours maybe you'll have a weekly one-on-one or every other week or whatever.
33:23
That one-on-one will consist of looking at the work they did whatever it's like oh over here you did this or whatever and it could be copy editing it could be anything whatever the role happens to be you're going to give them feedback, right?
33:33
That's what you would do for a human worker.
33:37
All of those things have a direct literally a direct analog in the agent world, right?
33:41
And what we're doing right now is we're hiring these agents and expecting them to do magic just like if we hired an exceptionally smart has a PhD in everything employee and expected them to do magic with no training no onboarding no feedback no one-on-one no nothing.
33:56
Well, your results are not going to vary they're going to be crap because you just not make the investment in getting that agent.
34:02
Now the the big unlock here so whether you're an HR person or whatever is like figure out like well, what does employee training look like for digital workers?
34:09
What do performance reviews look like for digital workers?
34:12
How do we do that how do we do recruiting for digital workers?
34:15
How do we like what what are all the mechanisms that need to exist?
34:17
What is a manager of the future?
34:18
What are the new roles that will be created as a result of having these hybrid teams?
34:24
It's like okay, well now maybe we're going to need someone that's like the agentic manager.
34:29
That knows all the agents that that are on their team or whatever and has kind of built the skill set to how to do recruiting for their team how to do performance reviews how to do all of that but for agents or hybrid teams you know versus just purely human ones that
34:41
that's just a whole other and we're going to need software we're going to need the onboarding we're going to need training we're going to need books written we're going to need all of it to kind of adopt and it's going to take it's going to take years, right? It's It's not overnight.
34:51
Two years ago I asked you is it going to be horrible or is this going to be amazing and you said I saw this with the internet nothing is as extreme as the most extreme predictions.
35:04
I listened to you and I trusted you then.
35:06
I actually think knowing what I know now I'm actually more fearful than I was a couple years ago where I'm like oh this is actually going to put a lot of people out of work and it's maybe not good or bad, but things are going to change drastically more than I thought.
35:20
And my so I I don't remember how I phrased the question, but is this going to change the future more than you thought 2 years ago or less than you thought 2 years ago?
35:32
Has your opinion on that changed?
35:34
I still think they're going to be unrecognizable my my macro level sense and this is maybe just my inherent optimism about things is that it's going to be kind of a net positive for humanity.
35:45
And this is the other thing that you know lots of people would disagree with me on this like oh well is this an existential crisis here to the species um and I've not said this before, but I'm going to see how it sounds as the words leave my mouth I'm probably going to regret it, but in a way we are actually and and Sam Sean you said this earlier we're sort of producing a new species, right?
36:07
So that's like saying okay, well Homo sapiens as they exist absent AI is likely not going to exist.
36:11
So the way we know the species as it exists today with where we have a single brain and and natural form you know four appendages whatever maybe that's going to be different um but I think of that as an extension of humanity not the obliteration of humanity, right?
36:25
That's the that's you know human 2. 0 or n.
36:26
0 of the way we kind of think of the species right now.
36:29
So I'm I think things are still moving very very fast and this is the this is why I think humans have issues with exponential curves.
36:37
We're just not used to them when something is kind of doubling or every n months it's hard to wrap our brains around how fast the stuff uh you know can move.
36:47
Things that we thought were like the things we have today Sam um if we had just described them to someone a year and a half ago there's like ah well chat GPT is cool whatever, but it's never going to be able to do that.
36:59
And now we're like those are like par for the course, right?
37:03
Like we we can do like things that were literally like ah there's no way no way.
37:06
It's like yeah it's good at like text and stuff like that, but that's because it's been trained on text. Now it can do images.
37:10
Well, it can do images but like videos at 30 frames a second that's like third generating images per frame of like a per second of videos like all of that.
37:20
It's like yeah, but you know diffusion models the way they work is because you're not going to get you you're going to get a different image every time.
37:25
So how are you going to create a video because it requires the same character the same setting in subsequent frames.
37:29
That's not how the thing is arc that's not how image models work.
37:33
And we solved all of those things, right?
37:34
Now we have character cohesion setting cohesion video generation anyway.
37:37
So my answer is it's exactly not exactly, but it's close to like yep this is what exponential advancement looks like.
37:46
I'm still of the belief that we're going to have more net positive.
37:49
That is not to say that in the interim there's not going to be pain and there's two things I will put out there as cautionary cautionary words.
37:55
One is in the interim um anyone that tells you that there's not going to be job dislocations or not going to be roles that get completely obliterated is lying to you.
38:04
That is going to happen it's already happening, right?
38:07
It's that there is no world in which that does not occur.
38:10
That's kind of thing number one.
38:11
Thing number two and we didn't talk about this but we should have um is that because of the architecture of how LLMs currently work maybe they will figure out a way to do that they produce hallucinations.
38:21
That's just a fancy way of saying it makes things up. Right?
38:27
And that's sort of okay but not okay because it doesn't know it's making it up.
38:32
Because of the way the architecture works it's like the intern that thinks it's been exposed to all there is to know in the world it's like I know all the things you ask me a question I know I know all the things so I'm going to tell you the thing that I know.
38:43
It's like well, yeah, but you didn't know this and you what you said is actually factually like probably demonstrably wrong and it has has absolutely zero lack of confidence in its output which is fine for some things if you're writing a short you know fiction story or something like that it's not great at all for other things like health care related where you need kind of predictable accurate responses.
39:02
So I think we need to be aware of the limitations around it when we're doing research and things like that and the problem is when we have relatively I'll say naive I don't mean this in a disparaging way uh folks that are naive to a subject area asking chat GPT for things where it can't judge the response, right?
39:20
We're sort of taking it on faith that it's chat GPT and our mesh set has got a PhD in everything.
39:26
So of course it's going to be right.
39:27
Well, no it's often not right.
39:29
And it's kind of up to us to figure out what our kind of risk tolerance is.
39:33
It's like when is it okay for it to be wrong?
39:35
How would I test it for my domain for my particular use cases?
39:39
Yeah, so What do you think about this situation where Zuck is throwing the bag at every researcher, $100 million signing bonuses, even more than that in in comp and he's poaching basically his own dream team.
39:54
He's like, "Okay, you're not going to I can't acquire the company.
39:56
Well, why don't I go get all the players?
39:58
If you can keep the team, I'll keep the players."
40:00
And he's going after them with these crazy nine-figure offers A $100 million signing bonus and $300 million over 4 years, I think is what I saw. Is that true?
40:09
I think that was like the higher Yeah, so the higher end.
40:11
And some people have said there's even like billion-dollar offers to certain people that are out there. This is like job offers.
40:17
So, Dharmesh, like, were you shocked by this?
40:19
Cuz I mean my reaction to this was that's First time I heard it.
40:23
Then I was like, "Wait, the source is Sam Altman. Why would he say that?"
40:26
And then I was like, "Okay, that's insane."
40:27
And then I an hour later I was like, "Wait, that's actually genius."
40:31
Cuz for a total of 3 billion or something, he can acquire the equivalent of one of these labs that's valued at 30, 40, 50, or 200 billion dollars. What a power play.
40:40
I know obviously you're investor in OpenAI, so you know, maybe you don't like this.
40:44
Maybe you have a different bias here, but I'm just from one kind of like leader of a tech company to an to another, like, what's your view of this move?
40:52
I think it's it's one of the crazier moves.
40:55
If I had to use one word, I would say diabolical.
40:58
Not stupid, not silly, but diabolical. And here's why, right?
41:01
This is the like in the grand scheme of things.
41:03
So, this is not just a Oh, can we use this technology and build a better product that will then drive X billion dollars of revenue through whatever business model we happen to have?
41:12
There's a meta thing at play here that says whoever gets to this first will be able to produce companies with billions of dollars of revenue or whatever, right?
41:20
Because that's it's like kind of finding the the the secrets of the universe, the mystery of life kind of thing.
41:25
It's like, "Okay, well, whoever wins that and gets there first will then be able to use the technology internally for a little while and be able to just kind of run the table for as long as they want.
41:34
So, there's it's got incalculable value, right?
41:36
The upside is just so high that no amount of like if you can increase your probability even by a marginal amount if you had the cash, why wouldn't you do it, right?
41:46
It's So do you think A, do you think it'll work?
41:50
Do you think this tactic will work for him?
41:51
Do you think he will be able to build a super team?
41:52
Is he just going to get a bunch of engineers who now have yachts and don't work?
41:56
Like, what's going to happen when you give somebody hundred million-dollar offers?
41:59
You you put together this smash together this team of I think he's got a hit list of 50 targets.
42:03
And I think like, you know, something like 19 or 20 of them have come on board already.
42:08
Yeah, what's your prediction of how this plays out?
42:10
It feels a little bit like a Hail Mary pass, right?
42:13
That's Okay, they're going to take this.
42:15
It's like, "Okay, well, there's not a whole lot of things we can do.
42:17
Yeah, the the chips are down.
42:17
I'm going to mix metaphors now, too.
42:19
Um But but but that that works sometimes. It works sometimes.
42:23
That's exactly why people do it.
42:25
It's like, "Okay, well, what other option do we have, right?
42:27
Like, everything else hasn't worked yet, so let's try this thing."
42:31
But the I think the challenge still I think it's a diabolically smart move.
42:36
We're not I'm not going to use the word ethics or anything like that.
42:37
But here's the challenge though, right?
42:39
If you if we were having this conversation we'll call it 2 years ago, give or take.
42:44
Um OpenAI was so far ahead in terms of the underlying algorithm and this is even before ChatGPT hit the kind of revenue curve that it's hit.
42:50
Just just raw the GP GPT algorithm was just so good and they were so far ahead.
42:54
It was actually inconceivable for folks, including me that others would catch up.
43:00
It's like, "Okay, well they'll make progress, they'll get closer, but then OpenAI is obviously going to still keep working on it and they're going to be far ahead for a long, long time."
43:08
That's proven not to be true. Right?
43:10
We've seen open-source models come out.
43:11
We've seen other commercial models come out. There's Anthropic.
43:12
I look and they have by most measures comparable large language models, right?
43:17
Within like one standard deviation, they're they're pretty good and sometimes they're better at some things, worse at others, but it's not this single-horse race anymore.
43:25
So, the thing that I'm a little bit dubious of is that even if you did this, you pull all these people together like it didn't really work for OpenAI in the truest sense of the word, right?
43:32
Like, they weren't able to create this kind of magical thing that It's like, "Okay, maybe they end up doing it you know, somewhere else, but I think there's more smart people out there.
43:42
The technology has caught up.
43:43
DeepSeek proved that you could actually I think they actually have an actual innovation in terms of reasoning models and things like that versus kind of the early generation large language models. So jury's still out.
43:55
How much better is a $300 million So, $100 million a year engineer over like a $20 million engineer?
44:02
Is it like I I followed some of these guys on Twitter and it was they're fantastic follows.
44:08
And do you think that their IQ is just so much better or is it because they've had experience Is it really because they just saw how OpenAI works and they want that experience?
44:18
Are they like is this like espionage?
44:20
What is How good could a $100 million or $300 million a year engineer be?
44:26
Well, that's the thing though. This is software, right?
44:28
So, this is a you know, a world of like 95% margins.
44:30
So, let's say yeah, I think part of the value is yes, they're super smart, but even high human IQ asymptomatically moves towards a a certain ceiling, right?
44:40
You take smartest people in the world, however you want to measure IQ.
44:43
And so, that doesn't explain away the value, right? That's not that.
44:45
It's not that they've seen the inside of OpenAI and they have some trade secrets in their head that they can then kind of carry over.
44:51
It's like, "Oh, here's how we did it over there and here's how we ran evals and here's how we did you know, the engineering process."
44:55
They'll have some of that because we always carry some amount of kind of experience in our heads.
45:00
I think the larger thing I think kind of primary kind of vector of value is they sort of have demonstrated the ability to kind of see around corners and see into the future, right?
45:11
They believed in this thing that almost no one believed in at the time.
45:14
They sort of saw where it was headed and they were working at it, chipping away at it, whatever.
45:19
And that's much rarer than you would think for really smart people to do this stupidly foolish, seemingly stupid foolish thing.
45:25
It's like, "You're going to do what now?" Right?
45:27
And we're still asking ourselves a variation of that question that we would have asked 3 years ago, except now we have ChatGPT and we have the things in it and we're still like "Well, you say that we're going to have like these kind of digital teammates that are going to be able to do all these things and it can't even do this simple thing right, right?"
45:41
Like, we sort of keep elevating our expectations in what we believe is or is not possible.
45:47
They sort of know what's possible and they almost think of what many of us would consider impossible as actually being inevitable.
45:52
Have you guys Has HubSpot Have you made any of these offers?
45:56
I don't think so, but that's that's not the game we're in, right?
45:57
So, we're not we're not in that league.
45:59
We're not trying to build a frontier model.
46:01
We're not trying to invent AGI.
46:02
We're at the application layer of the stack.
46:04
So, we want to benefit from it, right?
46:05
You know, we didn't in any layer of my entrepreneurial career I have not been the guy in the center of the universe sort of company sort of thing.
46:15
>> you're not like, "Oh man, I met this person.
46:17
Like, we need to offer like an NBA contract in order to secure this this guy."
46:20
No, and there's a reason for this, right?
46:23
It's like for the kinds of problems we're solving, what's the There's a sports term about the best all trades to the player or something like that that over replacement cost.
46:29
>> WAR, wins above replacement is the metric they use in sports.
46:33
So, yeah, it's just not it's not worth it given our business model and given what we do.
46:36
I have one last thing on the kind of AI front cuz this is one of the things um answering your question, Sean, in terms of things I disagree with folks on is that there's um yeah, a group of people um very smart that will say, "Oh, well, AI is going to lead to a reduction creativity broadly speaking, right?
46:55
Because you're just going to have AI do the thing.
46:56
Why do you need to learn to do the thing?"
46:57
And I have a 14-year-old, right?
46:58
So, it's like, "Okay, well, if he just uses AI to write his essays and do his homework or whatever, um it's going to it's going to reduce his creativity."
47:05
I understand that particular kind of line of reasoning that says, "Yeah, if you just have it do the thing, um you're not going to uh but I think the the part those folks are missing is that uh you know, creativity is kind of in the literal sense of the word is like, "Okay, I have this kind of thing idea in my head and I'm going to express it in some creative form, be it music, be it art, be it whatever it happens to be."
47:27
Um and the problem right now is that um whatever creative ideas we have in our head are limited in terms of how we can manifest them based on our emerging skill set.
47:40
So, Sean could have a song in his head right now that like he may be composing things in his head, but until he learns the mechanics of how to actually play an instrument, whatever the instrument happens to be, there's no real way to manifest that, right?
47:51
We don't have we can't tap into his brain and do that.
47:52
Um so, in my mind, AI actually increases creativity because it will increase the percentage of ideas that people have in their heads that they will then be able to manifest regardless of what their skills are or are not. And I I love that.
48:06
So, my son he's a big Japanese culture fan, big manga fan, and Japanese comic books and and anime.
48:10
Um and so, he he's an aspiring uh you know, author someday.
48:15
And what he can do now, right?
48:17
And he's been able to do this for years, which is so, he's always had Again, he likes fantasy fiction as well.
48:22
So, he's had these ideas for writing things, but he lacked the writing skills.
48:26
He doesn't know about character development, doesn't know about any of these things.
48:27
So, what he uses ChatGPT for is he's got this like 2,000-word prompt that describes his fictional world. Here are the characters.
48:35
Here's the power structure.
48:37
Here are the powers people have.
48:37
Here's what you can and can't do.
48:38
And then the way he tests the world is he turns it into a role-playing game.
48:42
It's like, "Okay, I'm going to jump in the world.
48:44
Now, you ChatGPT, I'm going to do this. Tell me what happens." Oh, this happened.
48:47
Okay, now I'm going to do this.
48:49
Okay, well, now you got this power.
48:50
Like, and so, it will sort of kind of pressure test kind of his world.
48:54
And so, that's an expression of his creativity cuz the world was sitting in his head, but now he can actually share that with friends, maybe turn that into a book someday because it's going to take the ideas that he has and hopefully in the meantime, he will kind of develop some of those foundational skills, but he doesn't have to wait until like 12 years of writing education before he can take this idea as a child.
49:14
He has lots of creativity.
49:14
Uh but as a practitioner, most of those things that he would love to be able to manifest in the world, he has nothing uh close to the skills required, whether it's drawing or writing or anything.
49:23
So, I think that's what AI can help us kind of elevate.
49:27
Um what And once again, we have to use it responsibly, uh but it should be able to elevate our skills.
49:32
I want to show you guys a uh Uh, example of this real quick.
49:35
So, I had this idea not long ago, a couple weeks ago, of um, creating a game using only AI.
49:44
So, uh, I don't I don't know if you guys ever played the Monkey Island games from like when I was a kid, I played Monkey Island.
49:51
It was a incredible game.
49:53
It's got basically this guy wants to be a pirate.
49:54
It's like this very funny but like eight-bit art style game.
49:56
And so, I created a version of that called Escape from Silicon Valley.
50:00
I didn't create the whole game but I created like the art. But like check this out.
50:03
So, So, I go into I go into AI and I basically start creating the game art.
50:07
And so, it's like the story is basically like deep in San Francisco, the year is 2048.
50:13
The Rock is starting his third term in office.
50:17
Uh, you know, Nancy Pelosi passes away, the richest woman on Earth.
50:19
And then, you know, Elon is promising that self-driving cars are coming really, really soon for sure for real this time.
50:26
And here you are, you're this character and you're in um, the OpenAI office.
50:28
And basically the idea is like >> at that. What's that? Look at the Charlie bar. Yeah, yeah, exactly.
50:35
I was I was putting in some references to like, you know, stuff that I thought was um, would be cool. That is so cool.
50:40
What did you use to make those images?
50:43
So, that right there was just ChatGPT and Midjourney um, mix.
50:45
I tried using, you know, Scenario and a couple other like game-specific tools. Like check this out.
50:51
So, like I created all these like tech like characters.
50:53
So, it's like I created Zuck and Palmer Lucky and like Chamath and Elizabeth Holmes in jail. >> awesome.
50:58
And I had it basically write the scenes for the levels with me, like write the dialogue with me, create the character art. Dude, that looks sick. Why don't you do that?
51:06
Um, well, because I did the fun part in the first two weeks.
51:11
So, I was like, oh, the concept, the levels, the character art, the music, the seeing what AI could do.
51:15
But then to actually make the game, the AI can't do that.
51:20
And so, I was like, oh, now I need to like I mean, people have built games in years building it.
51:25
It's like, oh, this is like minimum 6 to 12 months doing this like very, very arbitrary project.
51:28
It's like Um, I still love the idea and I'm going to like package it up the whole the whole idea. Darmesh, last question.
51:35
Um, just really quick, where do you hang out on the internet that we and the listener can hang out to stay on top of some of this stuff?
51:45
Like are there like who's a reputable handful of people on Twitter to follow or reputable websites or places to to hang out at? That's interesting.
51:52
So, I spend most of my time um, on YouTube, as it turns out.
51:57
Um, and and I I've sort of given to the given to the vibe, so to speak, and let the algorithm sort of figure out what things I might enjoy.
52:06
And gets it right sometimes, gets it wrong sometimes.
52:09
So, it's a mix of things.
52:09
But um, the the person that I think uh, if you want to kind of get deeper into like understanding uh, AI, there's a guy named um, Andrej Karpathy.
52:21
I don't know if you've come Uh, just search for Karpathy.
52:25
Dude, you want to know how I know I I get so many ads that says like, Andrej Karpathy said this is the best product or Andrej Karpathy showed me how to do this.
52:33
Now, I'm going to show you.
52:33
Like I don't even know who Andrej is other than ads run his name to promote him.
52:37
Yeah, I mean, he's yeah, one of the the true OGs in in AI.
52:42
But he has this his orthogonal skill or one of them.
52:44
I think he's got like nine you probably like a nine-tool player of some sort.
52:49
But he's able to really simplify complicated things without making you feel stupid, right?
52:55
So, he's not talking down to you. He's like, okay.
52:57
Like here's how we're going to do this.
52:59
We're going to kind of build it brick by brick and and you're going to understand at the end of this hour and a half how X works, right?
53:03
Um, and it's and and he's amazing.
53:07
>> So, that would be awesome.
53:08
>> so him, any other YouTubers or Twitter people or blogs?
53:10
On the business side actually, um, like, you know, Aaron Levie from Box is actually very, very thoughtful on the uh, if you're in software and in business and the AI implications there, I think he's really good.
53:20
Uh, Heath Henshaw, who you both know now at Dropbox through the acquisition has been on fire lately on LinkedIn.
53:26
Uh, so he's one I would I would go back especially the last like three, four months uh, and read all the stuff he's written. I think he's on point.
53:33
Um, yeah, so Those are awesome.
53:35
Darmesh, thanks for coming on. Thanks for teaching us.
53:38
You're one of my favorite teachers uh, and entertainers.
53:39
So, thank you for for coming on, fun. Likewise. Thank you. That's it. That's the pod.