We audited our AI use. We are now making a case for 'intelligence independence'

0:01 Quick question. How many AI models can you name? I'm not asking for how many you use. How many can you name? Okay, hold that number. We put two people who use this stuff every single day for a living. What they could not answer is the whole reason this episode exists. Welcome to Zerot. I'm Rohan and today is a little different. So, let me tell you the rules before we start. Brady and Pravin, my co-hosts, have not seen a single one of my questions. There is no script. In fact, let me read out to you a message from Pravin at 6:5 a.m. today. We're recording this around 10:00 a.m. So, just about a few hours ago. And the message reads, it's on Slack. what is today's ZS podcast about?

0:56 Right. So, I'm not kidding. They don't know. They don't know where this is going. And they certainly don't know what they're about to admit. We call this a cold read. A cold read usually happens. Pravin ready. Do you know what a cold read is? >> You walk in, you walk in without any knowledge, any preparation and kind of see how the chemistry carries it. >> I have a slightly different definition. by different definitions actually comes from magic and mentalism where a cold read is remember the time where somebody tries to like guess you know what you're thinking or what your mother's name is or something like that. It's called a cold read where they don't ask you any questions but they just sort of like see you and they're able to like guess it.

1:43 It's called a cold read. My my definition of a cold read is usually what happens in movies and scripts where you ask someone to come in and you give them a script to read as opposed to a warm read where the script is, they've read through it, they prepared in a cold read, they just given something to read. So this is a cold read episode. I hope we can do more cold reads in the future. And the reason why I hope it'll work is because Brady and Preine, you're not AI beginners.

2:10 You use AI every day. You're sharp about it. What makes the two of them perfect standins for the listener is because you two are users and of course you're my host as well. But if I can prove that there are gaps in the way you think about AI, then I can certainly prove that there are gaps in how all of us me included think about AI by the end of this are hopefully you'll know exactly where yours are. And now I'm referring to you the listener. Now a quick confession about time. If you're hearing this, you've probably already read the column I wrote based on the same topic about the intelligence that we rent. And you've probably seen that we're also running a live event on the same topic on the 1st of August about precisely this tickets are on sale. And fair warning, they'll get more expensive every week. So the links in the show notes is cheaper today than it'll be on the day when you finally click it. Here's a strange part.

3:12 When we recorded this though, which is today, none of that that I'm referring to existed. There was no column. There was no event. We hadn't worked out any of it. So, what you're about to hear is the conversation that talked us all into doing it. You'll be listening to three people convince themselves that the questions matter. Life without a net, it's a cold read. So one question for the next star are should you care about intelligence independence?

3:44 Let's find out by finding out what three people who really should know turn out not to. Brady Pravin, no prep, no mercy. Let's go.

4:22 All right, let's start with the questions. My first question is to Brady. Brady, what did you use AI for in the last 24 hours? Please be specific. Give me an actual task. Don't give me a category. I batched a bunch of decks that all needed to be updated. There were 10 of them and I just asked Claude to do this in one go based on my rules and restrictions. Basically let it run for I don't know it was probably a good 5 or 10 minutes and then it gave me everything I needed. I did some research for one of our upcoming podcast episodes on Zeroot.

5:07 what else? What else? I ran some tests on basically evaluating some HR matters with you know dummy data I fed something that I built just to see if it would simplify some of the things that would take me maybe two or three hours a week and that's it. >> Same question to you PGK. What do you use AI for in the last 24 hours?

5:37 >> Two things. First I do another podcast called 2x2 and we did an episode on cred and I had to get a transcript of that and for that I used claude to basically get generate a transcript. I did a fantastic job of like actually not claude I use assembly AI and claude to stitch it. So I have a bunch of tasks running together. That's one. The second is very frivolous but I really enjoy it. so one of the things that I've done is I have to make like protein and chicken every night for so what I do is I take a photo of my entire pantry with all the spice rubs that I have and I just tell Claude every day just tell me what do I need to mix in and it just asks me like two three questions. Do do you feel Indian? Do you feel Mexican? Do you feel Asian? And I just say whatever and it just says okay do the following.

6:23 I say okay and I mix it. So it just takes one thing off me. It's nice. >> All right. Now let me turn back to you again and ask you the question. Now which tool you already sort of revealed it. >> Yes. >> But you said that you use cloud. >> I use cloud almost exclusively and assembly AI sometimes but yes. >> Why? >> Why that one? >> Why that one? Okay. I think I've been conditioned to I' Okay. Two reasons.

6:48 I've been using it for a significant period of time. So I think that Claude knows me better than everybody else. it has the maximum context about me and well is also just universally seen as the best tool out there. >> Okay. All right. Same question to you Brady. >> With work stuff I use claude because it comes with work. so might as well. Personal stuff I still use perplexity because it's there and I can. >> Okay. Now I'm I'll flip back the question order back to Brady. On a scale of 1 to 10, how AI proficient are you?

7:26 And who in your head might be a 10? Oh, I think I'm probably like a six or a seven. who might be a 10? That's a good question. I do actually have a friend who is constantly running probably a few dozen agents on eight laptops. and he orchestrate or he he he runs them through his phone. So you know he could be traveling working remotely and and they just kind of you know are constantly there. I think he's a 10.

8:08 >> Pravin question for you as well. On a scale of 1 to 10, where are you and who's a 10? >> I'm a solid seven. I have some way to go. I don't use things like very complex workflows in agents yet. So I'm a solid seven. who's a 10? You know what I actually think? I think a 10 is employees at places like Anthropic or OpenAI. They are a 10. I think everybody else is just one step below them no matter who they are. And even I have friends like Brady who do all of those things. Did you see the spaghetti code that is clawed when it was accidentally leaked?

8:41 >> But but let's let's hold that. Let's come back to it. Okay. more questions. this time back to you PGK. What's the most impressive thing you've gotten AI to do for you? >> Your proudest moment with it. I think the the most impressive thing that it did for me was it was able to very successfully diagnose a rare heart condition that my cat had last year that I was very impressed by because I it's something that the vets missed a lot of people. And I just fed all the medical reports, all the blood tests and I was not very hopeful and this was like early day or actually when AI was not as powerful and was and it just came back with a diagnosis which was surprised me and it >> and what's the most impressing thing impressive thing that you've gotten AI to do for you?

9:30 That's a really good question because I haven't really felt impressed in a long time just because so much of it is talked up into being able to do so much. So I think, you know, if I really had to name a time I was really impressed, it had to be, you know, shortly after Chat GPT came out. And I was seeing how so many writers and poets were really scared because it could ape their style so well. And now we know, you know, since it's, you know, literally powered by language, that's a very natural thing it should do. So it seems less impressive now, but at the time, you know, it was very novel and it did impress me.

10:15 >> All right. Takes a lot to impress Brady. Now, if you're listening and if you probably answered all of that in your head, good. You're not just a spectator or a listener today. You're our third guest. All right. The next set of questions this time to you, BGK. >> Sure. >> Do you pay for AI yourself or does your employer? >> My employer plays for AI. I don't pay for AI myself. >> All right. >> And what about you, Brady? Both. my employer pays for Claude. I pay for Perplexity, but I've been thinking about getting a personal paid claude account as well.

10:58 >> Okay. Which one has a real history between your personal account? Since Brady, you you said you have a personal account and a work account. Which one has your real history? The version that actually knows you. >> I would say if you took the union between two sets, then that would be me. But you know each one kind of just knows my personal life or my work life. So in which one do I express more of myself?

11:31 I would say half half. Yeah. >> Okay. And what about you Pravin? You have just one cl. >> I have two clouds. I have one older one that I'd got as an individual account and a enterprise one that we have. I mean very similar to what Brady said which is the union of two of it sort of tells my life >> but that's not my question which one has I mean sure if there were three we could say the union contains more than the individuals but relative to the two which one has your real history >> the first one the older one that I have >> which is the your personal one >> the personal one yes you could say that >> okay now back to you again Pravin if you lost access tomorrow you know the cloud login the subscription the whole thing.

12:16 What would you actually lose? >> I think the thing I would lose the most. I mean, you're asking what is it that I would miss or Okay. The thing that I would miss the most and I would struggle the most with is actually doing the 2x2 podcast because that is the one that has all the transcripts, the various flows, the ideas, etc. You know, the rest of it I can sort of like I think I can figure it out.

12:40 >> What does that mean that you'd miss? What what what exactly would you lose with respect to the 2x2 code? >> I would lose the ability for like for instance it generates Python code for me in order to like get transcripts to basically write edit notes for my producer Rajie. Those are all things that I've written inside plot and it stays there. >> Okay. >> But outside that >> that's the only thing that you >> that's the biggest thing that I would lose. The rest of it I can still live.

13:04 I'd have to like think of ingredients and sorry mixing ingredients myself in the evening but I think I can live with that. >> And what about you Brady? What would you lose if you lost access to your subscriptions? >> I would lose some of the time I've been saving, right? it's not about actually what what exists in my history with either of these tools or scripts they've created or skills I've made or anything like that. It's it's more like the way I've I I work has changed. And if you took that away instantly, I would need to revert back to, you know, how I did things 2 or 3 years ago basically, which would need a lot of reorganization of my time and how how I even interact with some of my colleagues, I think.

13:59 Okay, back to you again, Grey. Have you ever thought about where your conversations and data actually go? I mean, have you ever checked? >> I have thought about I have wondered about this. I have not checked though. I think I'm kind of scared of the answer. >> All right. Question to you Provin. Have you ever wondered? >> I have wondered. I have checked. I I think just like I'm also a little scared of the answer. I remember that when one of my reasons why I wanted to switch to the enterprise cloud was I knew that in the enterprise cloud they don't train your data. They don't >> they claim not to train. They claim not to use your data for training, but I have not seen that note anywhere of late. So, I'm also wondering and my wife started using Claude recently and she told me that remember to go to your settings and turn something off and I said sure and I didn't I don't think I went back and checked but yeah I am scared of the answer too.

14:53 >> and and question to you first Pravin has your company told you which tools you are allowed to use? Has it banned any? Do you follow that? My company has not explicitly banned certain tools, but I think we do know what tools are allowed to use and I stick to them. >> Okay. What about you, Brady? >> I mean, same company. but I think it's more about how we use the tools, right? So, it's more like, you know, with paywalled material, we wouldn't just upload it willy-nilly.

15:32 so yeah, >> right. back to you listener. Are you noticing the pattern already? We're all sort of deep users of something that we actually don't own. In many cases, we didn't choose and we've never inspected. Now let's move on to the next set of questions. let me start with PGK and no googling. How many AI models can you name out loud right now? Go.

16:03 >> Oh, okay. Fine. let's see. there is Claude, there is Gemini, there is >> Claude is not an AI model. >> Oh, sorry. Claude has Oh, well, there are five AI models that I can name. There is Haiku, there is Sonnet, there is Fable, there is Opus, that's four. Gemini, Codex, Deep Seek, Kimmy, Serv. I guess >> those are companies. >> Those are companies, right? Yeah. Yeah.

16:36 yeah, I think I have around nine. Eight or nine right now, give or take. >> Heavily clustered around claude. Five. >> Heavily clustered around claude. That's right. >> Okay. All right. Same question to you, Brady. >> Oh, sorry. It's not Codex. It's GPD4, GPD 5. That's easy, right? You have GPT 3, 4, 5, 4. I can go there. So, yeah, more or less 9 10 I guess, >> right? So, you're thinking like an LLM now. Three, after three comes four, after four comes, right?

17:01 >> Sure. >> Okay. All right. Same question to you, Brady. I'd say no Googling, but then you're a perplexity user as well. So, no perplexing. How many AI models can you name out loud right now? >> Let me see. To add on top of what Pravin said, mythos. I think you missed that one. No, I said fable, but okay, fine. Mythos. Fair. >> Okay. what else? What else? Mini Max. Sakana.

17:32 >> he's just making stuff. Okay. No, he's not. He's not. That's a company though. I think Fugu is the >> Oh, yes, it's Fugu. Yes. Yes. >> let me come back to you, Brady. Qen. Qen. >> Quinn. That's right. Quinn. Okay. Brady, have you ever used DeepSk? Have you ever used GLM2? Have you ever used Quen? Have you ever built anything? I mean, have you ever used anything which was built outside of the US? Anything which is open-source/openwaits?

18:07 >> Deep Seek, yes. Not GLM. QN, yes. >> My answer is just simply no. outside of anything that is built by Google, Anthropic, a little bit open in the beginning but not not for a long time. So yeah, those are the three. >> Okay, you haven't. Okay, great. Now let me ask you Pravin, have you ever run a model on your own laptop fully offline and do you know that you can? >> I know I can. I have been reading a lot about local models and I've posted it as well, but I haven't gotten around to doing it yet.

18:39 >> Okay, great. Brady, have you ever run a model on your own laptop? Not yet, but I have a deprecated gaming laptop that is ready for that somewhere sometime later this year. >> Okay. All right. A followup question to you, Brady. Do you know the difference between an open weight model and a closed one? And does the difference matter to you? >> H that's a I cannot articulate that right now. Just answer the second question then. Does the difference matter to you?

19:19 >> It should. >> Okay, we'll take that. What about you, Pravin? >> I can only explain it in this terms that an openweight model is the closest that a model can get to being open source in the sense that you can take it and you can you can tune it to a very specific use case that you want to use it for and build something on top of it. That's all I can say at this point in time intelligently. I'm sure there are some AI researchers who are listening want to write in with angry email saying I'm wrong but that's broadly where I'm at.

19:49 What's the second question? >> The second question was does it matter to you? >> Does it matter to me right now? not really especially >> and back to you the listener. Now there are dozens of AI models but most of us including the two professionals in this room seriously use just one or two. Let's move on to the next set of questions. Now I'll go to you Pravin.

20:22 Where do the useful things that you build with AI actually end up? Do they go somewhere or do they just evaporate when you close the tab or close the software that you want? >> No. The most useful things that I build with AI more or less not all of it but maybe 80% of it stays in my laptop and on the cloud because I use claude co-work so which ends up as I've started creating MD files for everything. So it ends up >> markdown files.

20:53 >> Those are markdown files. So those are markdown files. The work that you create ends up in files which sit on your laptop >> and the cloud it's synced. So it's not like it's if my laptop disappears still in the cloud so I can take it from there. >> Okay. All right. What about you Brady? >> exactly the same answer. >> H Okay. So let me ask you slightly more I think a deeper question. Have you taught the AIS that you use about yourself, your style, your projects? And if you have, could you take them to a different tool tomorrow?

21:30 For me, the answer to both is yes and yes. But those are not both neither of them are strong yeses. They are like >> seven on 10, eight on 10 yeses. Which means that I've not taught them very deeply about everything that I do and how I do it. I've not taught them for every single project. I've taught them for a few projects. Can I take them to another tool? I think I can. In fact, that's one of the reasons why I started using markdown files where I wanted that optionality if I wanted at some point in time.

21:58 >> How do you teach them? >> I wrote instructions, I wrote very basic skills, context files, all of these things. >> And and the reason why you're saying you can take those to another AI model is because those sit as files on your machine. Correct. What about you, Brady? >> So, have I taught my tools about myself? Not in a very deeply involved way. In fact, I actually recently asked both Claude and Perplexity, "What do you know about me?" Because I was just curious about like how how that history has translated into I guess like context that it has of me. and it was able to say that, you know, I'm involved with something called zero shot, but it was very confused about whether that was a podcast or a written column, which it's both actually.

22:57 Claude knows I'm part of the Ken. Claude knows I, have been helping with training some of our new reporters, for example. But you know these are very I guess like superficial reads of who I am, what I do. it's very task based. I don't think it would be able to construct like you know a set of responsibilities that I'm involved with like on a day-to-day basis.

23:31 >> Okay. Thank you. Let me switch back. actually let's stay with you really now. Stop thinking about yourself and think about the team that you work with or teams or people or colleagues that you work with. Where does everyone's AI learning pile up? What do you mean by pile up? >> Where do they reside? Where does the AI learning reside or do individuals end up quaintly reinventing the same wheels again and again?

24:02 H >> I'm asking about my earlier question was about you as individuals. How do you use it? Does the tool know about you? Right now I'm zooming out and asking about your teams. Right? And I want you to answer where does the AI learnings of the teams that you work with reside? Is there >> so AI learning at the team level? So I wouldn't say there's an organized effort but I have had I guess like very impromptu sessions with specific team members who have been trying to use Claude or whatever other tools they've been using on their own and you know they felt like they should be getting more out of it. So I would show them what I do. I would open up Claude. I would, you know, step through all of my prompts from the past that led to an outcome that they were after. I would show them how, you know, how I use co-work I would show them, you know, how to build a skill. but this was very I guess like off the cuff. It wasn't like a formal learning plan, so to speak.

25:26 >> Okay. Same question to you, PG. I'm just going to be more blunt and say that I think most of my team members and I include the two of you here. I think our skills are mostly at an reside at an individual level rather than at a collective level. I mean a very simple thing is that all three of us are like power users of claude and we are on the Ken enterprise and we don't to the best of my knowledge have a folder or a project called zero shot that the three of us collaborate and use on. So that itself is one indication. So it's more or less a >> people do reinvent the wheel on their own.

25:59 >> People do reinvent the wheel on their own. I have reinvented the wheel multiple times on my own. So yes, definitely true. >> All right. The next question is a hypothetical question. Let's start with you Pravin. If you switch jobs next month, >> does your th is feeling like therapy. I like it. Go on. Huh. If you switch jobs next month, does your quote unquote AI self, everything it's learned about you, come with you, or does it stay behind somewhere else?

26:26 >> If I switch jobs next month, I don't see how my AI self can come with me. I mean, I can probably like download all of it and try to put it, but I'm sure that would not be ethical. So, I wouldn't do that. >> No, I don't think it would. I really don't think it would. would be >> I am spiritually detached from my AI self. I do not feel attached to it at all and I am okay with just being severed from it and rebuilding something from scratch in a new new setting.

26:59 >> I think that's broadly true for me as well. >> All right. question to the listener. Ask yourself the real question. Are you actually compounding as you use AI or are you also usually starting from zero every single morning? Let's move on to the next set of questions. Now let me start with you Pravin. Does your company have an AI strategy or does it just have AI users?

27:32 And does anyone actually own that question even? I don't think there is a very strongly be honest. >> No, I am being honest. Okay. I don't think there is a very strongly articulated or super long-term AI strategy, but I do think there are >> there's one one strongly articulated version from the editorial side. >> Well, yes, that's also true. not using it is also definitely a strategy where you basically say that we will not use chat GPT for or any kind of AI for writing any articles or anything that the subscriber would read or see and that's true we do that I think for instance we use AI to build certain kinds of products and we have an AWS instance it's there so there is certain use cases that we want to use it for so there is some form of it but I don't think that's a >> broadly the company doesn't have an AI strategy it has AI users >> it has AI users it has AI I instances, it has AI use cases that we want to build on, but I wouldn't call it very strongly a strategy yet.

28:34 >> And ideally, who should be owning that question? >> Probably me. I think yeah. >> All right. Brady to you as well. Does your company have an AI strategy or does it just have AI users? >> I if if this were a binary choice, we just have AI users. I think there are individuals within the company that have imagined what the AI strategy should be but we don't have a unified line out of that.

29:08 >> All right. Followup question. Who decides at the place that you work which models are allowed and where your data is permitted to go? Is it anyone? >> I think there are loose or I wouldn't say loose. I think there are like guidelines and guardrails that we have but I wouldn't say that there is like if you're asking me is there a document somewhere that lists out here is what is allowed and what's not probably not but if you talked to five or 10 people at the ken and said what can you use and what can't you use I think people would broadly give you the same answer >> okay what about you Brady >> the editorial team does have a document for all the writers and editors that spells out how or how not to use AI I should say that's really more important on their side. but orwide no we do not. All right I'm going to zoom out to a very different level now which is the country and let me start with you Pravin off the top of your head does India have its own AI models and does it matter to you personally?

30:16 India does have its own AI models. The word own has some quotes around it but still it does have its own AI models. I'm thinking of serv right now. I know that servum has a few AI models that is built. Yeah, it's built it started with open weights and they did some changes. So I know that there is some own AI model definitely. >> Does it matter to you personally? >> Does it matter to me personally? the question of whether India should have or does have or does not have >> it does not matter to me personally. It matters to me from as an Indian and I would like India to have it.

30:52 >> You separate your personal opinion. >> No, personal opinion as in like I think it doesn't matter to me personally today like whether it has or does not have but I know that in the future it will matter and it matters a lot. >> Great. So you do feel it matters. >> It does matter. Fine. It's a simpler answer. It does. >> It's a tricky question for you. you work with the Ken but you're not in India but still I think you know you're you're intelligent enough to kind of answer this question right off the top of your head should India have its own AI models in an ideal world yes can it I mean it does but you know how that looks in the future I don't H okay let me ask you a follow-up question now if the models that you use every day was suddenly built somewhere else and the rules of how you can use them changed overnight would it impact you yes absolutely it would impact me because it changes what I can do or can't do with those models.

32:09 >> Okay, Brady, is that the same with you as well? >> It impacts me every day right now, right? Like Claude is not offered in Hong Kong. To use it for work or to use my personal account there, I got to get on a VPN. So, this leads to some inconveniences. it also is kind of a precarious position because maybe you know someday in the future they will kind of block certain IP addresses that are known to be used by VPNs right so there is this question mark of like is this like a sustainable thing basically thank you for saying that out loud and to the listener back again this I mean what we're discussing right now is what most of us rarely end up thinking about we usually optimize just for the chatbot that's in front of us on our screens and we rarely think about what's the road that actually leads up to it and what if someone cuts access to that road.

33:16 Let me go to the next bunch of questions now. Let me switch to you PGK. What are you actually afraid of as your AI use deepens over the next few years? What I'm really afraid of is I start making mistakes with what is it that I am supposed to do by myself without AI and what is it that I'm supposed to do with AI. I think I'm learning that. I think I have a fair understanding of it now, but I'm afraid that I'll make those mistakes.

33:50 >> Okay. And Brady, what about you? What are you afraid of over the next few years? I think one of the most important I guess mindsets to have when we're using any AI tool is knowing when to say no, right? And this is like a very stark contrast between somebody who's fresh out of school and somebody with say 10 or 20 years of experience. Like people who are more experienced know when to say no when they see output that doesn't satisfy them. It might look right but it doesn't feel right in some sense. So my worry here is like if I keep using AI in an intensive way my intuition there kind of gets dulled or I don't keep sharpening it. like there's a there's there's a possibility where I kind of like stall in my progress as a professional where I stay at the same level where I feel more productive because I have more output but my intuition isn't being refined. So that's what I worry about. to the listener remember neither Pravin nor Brady actually know the questions that I'm going to ask them. They had no clue what this episode is. And the reason why I'm saying that because my literal follow-up question to Brady, the next one was, "Do you think this is the use of AI is making you sharper or duller?"

35:24 Right? And Brady just answered that question before I even got to it because he literally said the same thing that I suspect it's going to make me duller and not sharper over time. But let me still pose the same question to you, PGK. Do you think as you continue using AI and and you you admitted that you are a power user as a power user are you becoming sharper or are you possibly becoming duller? The real answer is I'm becoming a little more I'm becoming more productive but I don't think I think it's it takes a lot of work. It's very similar to what Brady said to decide.

36:00 >> Yeah, you can't deflect that. the the question is between sharpness and dullness because one could argue that there is a version of duller productivity. So that's a different sort of metric, right? Like it's it's like that what is that fable of the wood cutter with the sharp axe. >> So one sharp cut is worth 100 dull cuts. So it's not about productivity, but do you think you're becoming sharper or duller? >> I'm not becoming duller. I'm definitely becoming a little sharper, but I don't think it's I wouldn't say it's a lot.

36:35 >> All right, let me ask you a follow-up question then. If someone else offered to set up you know an AI system for you, what do you wish you could do today? No constraints. Like what do you wish you could do if the heavy lifting of actually setting that up someone offered to do it for you free of cost? >> H okay. things that I would like. First, I want a list. >> Oh, I have like a broad scaffolding. Ha, see, I'm using scaffolding. Anyway, no, the broad thing is number one, I need an AI system that is completely private. First, I want to be really convinced that whatever I don't want to start thinking what should I say, what should I not say, is it going to be used for training. I want to be really confident on what is it used for, what is it not used for first. Second, I wanted to run more things in my life. I wanted to handle my emails. I wanted to handle you know a lot of other tasks and which I don't want to give it to it right now. I want to be a little more unfettered with it. And finally, I want an AI system that when I try to ask it something, it tells me that I'm not the right person to ask this for. You should be the one to have the answer to this.

37:55 You should think this through. And I can tell you something, but I probably going to be a bad answer. I would love an AI system that just pushes back and says, "Can't answer it or I'm going to give you a bad answer." >> Okay, same question to you, Brady. What do you wish you could do with AI if someone else offered to just do the entire setup for you with no constraints? >> Okay. a few weeks ago I had a dream and that dream was about a game and I want a system that can help me build that game.

38:30 >> What would someone set up for you? >> That's a good question. something that could debate me on the narrative first. so it would need to be able to challenge any plot holes first of all. It would need to be able to offer suggestions to remedy those plot holes. It might need to be a I feel like I can outright cla maybe.

39:08 >> So you're saying if someone offered to set it up for you, you'd want a more intelligent AI. better than Opus 4.8 or you so you need Fable/Mythos. >> I do. Yes. >> All right. Okay. my follow-up question is to you Brady. Roughly a year from now, do you want to be more valuable because of AI or more replaceable because of it? And if you've ever thought about it, what would be the difference?

39:40 >> That's a good question actually. maybe like two or 3 weeks ago, I even posted on Slack where I feel like my role at the Ken is changing and someday I I hope I can actually automate it away because, you know, I feel like a lot of the things we do will eventually be automated away. And I actually want that because I'm one of those people who believes that it frees me up to do more creative things like more not dreamy things but stuff that is not midcurve basically like off to the edges.

40:24 >> Oh okay. Got it. All right. Same question to you PG. A year from now do you want to be more valuable or more replaceable? more valuable. Why would anyone want to be more replaceable? I think people say >> just say he wants to be replaceable. >> No, no, I hear Brady. I'm saying that I I don't understand that construct at all. I think a lot of people say that sometimes even I have said that I definitely don't want to be more replaceable. I want to be more valuable.

40:47 A lot of things that I do may become replaceable, but I'm going to I want it I want to be more valuable. Very clear. >> Let me ask you a question then. If you had to take charge of AI the way you take charge of say your money or your health and know you're really into working out and fitness and all of that what's the very first thing you do after taking charge of it like what would you actually do >> right now a lot of usage for you I suppose is just in the professional context it's expected out of you but if you would actually be conscious about it what would you do like you don't just go and work out kind of without a goal in mind, right?

41:27 There's a certain >> consciousness to it. >> I think there's I think there's generally two parts to this. One part of it is I think there is I want to basically be able to set up or very aggressively set up nonAI time and say that >> these are certain times of the day where I will not use AI at all. And then the second part is I want to know that the times that I use AI, it's going to be so easy for me to use it. I just have to say do this and it just does it and I don't run out of tokens. Not not that I run out of tokens much nowadays, but I just want it to just do it and not come back to me. So I think that separation is very very important for me.

42:16 >> Same question to you Brady. So when I talked about making a game, I think the deeper the deeper desire here is more more like using a very advanced thing and entity to make your visions come true when you do have a vision. you know and and part of it is like I feel I feel that a lot of people are given this tool but it's aimed at very menial things as it should be but then you know there's always like this this constant struggle to make like bigger things become reality and that's the thing that I really want AI to do I know this is like a really vague answer but That's kind of what I'm reaching for.

43:12 >> All right. Thank you. That was a great set of questions and answers. So, let me play back what we discussed. Now, we use AI every single day. We rate ourselves. Six, seven. And yet most of the times we don't have a strong stance about how exactly like you know we want to use it, where the data goes, what we want to do with it etc. Most of the times we can't name many other tools other than the ones that we are actually using. we keep a lot less of our usage than like Pravin you said that you have a lot of MD files but a lot of the questions and thinking that you've done with cloud over the years would have been in the form of interactions and prompts >> yes MD files is a very recent thing >> where are they inside cloud and if you delete my account it goes away >> that's right and so there are a lot of choices that we don't make or we sleepwalk around. We are typically deep in just one product or habit and mostly a blank other than just surface level awareness of yes it exists and by the way we do this for a living right we are professional journalists we run a business we are quote unquote you know power users of AI not in the crazy agentic dev mode like you know feeding them 24 by7 sort of way which is a little made up but otherwise Now the reason I'm saying this is because I don't think it was ever I mean well it was but it's not in question any longer whether AI and by that I mean LLMs which is the current iteration of AI is powerful or not. Of course it is. We know that right? I don't think that's up for debate any longer. The question really is, are we using it in a way that makes us more valuable over time or are we just becoming dependent on it rented?

45:26 We are in some senses like you know most of the times our employers are paying for usage and we are essentially renting that intelligence right. and I think it's fair to say that very few people almost nobody no person no company and barely any country bar a few has actually sat down and figured out what they need to do and certainly not India because when I asked you how many models are there you're like I think it's serv model I don't think we know when was the last time we heard any of the server leadership say anything about we keep talking about server but the funny thing is the server never talked talks about itself, right? Which is a very weird inversion.

46:10 And that's really it. That's what I'm calling the need for intelligence independence, right? I mean, this is we're using it. I mean, I mean, I'm with Brady on this. I'm not sure it makes me more sharper over time. I fully agree with you that it's making me more productive in the medium to short term, but am I becoming sharper as a result of this? No. I mean, I'm probably using a lot of knowledge and experiences that I had pre AI but amplified and magnified because of AI. So, my core sharpness or my prei sharpness is being used and if it's being used it means it's being dulled but I'm not sharpening myself which was probably the point that Brady was making as well.

46:56 Right. >> Interesting. and I think that's the difference between drifting into AI, which all of us have, me included, versus taking charge of it. So, here's us three regular daily users. And just between the three of us, we really had no stance. we were mostly in one corner of the tool sets, largely clawed. almost no memory which is persistent and nobody really setting or defining strategy as well, right? all of us quote unquote rent our intelligence by the month and we don't own the upside, right? We get the productivity but the upside isn't ours. And if that sounded even a little like you, the 1st of August is built just for you. We're actually putting out the models, the setups, the methods on stage in Bangalore and you'll be able to walk out able to do something about all three of them at three levels and not just nod at them. The link is in the show notes.

48:06 Like I said earlier, it's cheaper today than it'll be next week. And that isn't a trick. It's the whole philosophy. We'd like to reward people who move first, our subscribers. And if you've listened this far, that's you. There's also a short form in the show notes. use it to tell us which models you want to discuss. You could nominate someone or you could just tell us how dependent you secretly are. We're trying to build the event with you, not at you. Brady and Pravin, thanks for walking in blind and doing the cold read. And dear listener, please go find out how many models you can name. Then come and tell us in person on the 1st of August. This was zero shot and this is me Rohind Dharmmakumar.

48:54 That was zero shot the Ken's weekly podcast on the biggest developments in artificial intelligence. Our hosts and commentators are Pravin Gopel Krishnan, Rohan Dharmakumar and me Brady. Our sound engineer is Rajie Cenne who makes everything sound spectacular. Don't miss our zero shot columns which are published every Saturday. We'll be back with more for this podcast next week.