Why Most ‘AI Experts’ Suck, And Ways You Can Blow Right Past Them, with Mindstone CEO, Joshua Wohle

Two years ago, I would have said generative AI and AI Assistants are the most democratized piece of technology ever built. Like, there hasn't been a time and place ever where for $20 a month, everyone would have access to the same capability models Everywhere. Now with O1 or Zero1 Pro, we're actually moving towards a world where you're paying for inference, right? So you're basically paying for thinking time. And the more it thinks, the more you get quality answers. And it looks like O3 is confirming that. So what that means is you're getting back world where the more money you have, the more intelligence you get. And that is something that if you then don't do the training to begin with, so you're still stuck on your crete here somewhere. You haven't even gotten yourself to the use cases that would allow you to merit 20amonth. You will definitely not engage with the $200 a month. Even if you somehow had the money, you would actually not feel comfortable to then pay $2,000 a month, when the $2,000 a month tier will inevitably at some point exist. Because if you're not experimenting enough to develop your confidence that you know which use cases and how to leverage the $200 into 2000, you definitely won't get the $2,000 use case to leverage into 20,000.

Joshua, welcome to the augmented life.

Thank you very much. Thank you very much for having me.

Oh, it's a pleasure having you. I've been looking forward to this conversation because we are, we're two entrepreneurs who, when we first met, worked in completely different fields. And, and now we've started, you know, companies that are, are. Are squarely in the heart of the, you know, incipient AI. And, and it's, it's in some ways the same, I think, and, and in some ways totally different. Are you feeling the same?

Yes. And the longer it goes on, the more it feels different. It's like the things that I thought I had learned from the first I realized in some cases were coincidence and in other cases, just the world is different. And so what I thought was working before no longer works now.

Right. I'm so relieved to hear you say that because I'm having the same experience. It's exactly right. Yeah. So, yeah, some of my lessons learned, I think, just had the wrong priors and just turned out to be contingent truths, not universal truths. Yeah. Well, let me put you on the spot for. Tell me a surprise, like, what's something you thought mattered a lot that turns out to not matter much at all or the opposite.

Two things. One, maybe I'll go a little bit controversial. The first thing, the first thing was I'm a product person, first and foremost engineer, product person. And I. I'm a big believer in general around. I mean it's almost hard not to, but like the kind of general agile methodology of going through things. I always thought that in the previous company, I didn't leave myself or we didn't leave ourselves enough time to be able to properly follow the process that would lead to the right things. We were constantly chasing the next deal and it was, it felt really accelerated. We were making a bunch of mistakes. And so the. Now that I got to do this another time around with a different team and with, well, actually in some cases the same team, but with much more funding to start, which is the benefit of a kind of serial entrepreneur having done it multiple times. Right. So you create yourself the space. I actually fell into the opposite trap where we ended up iterating through things so fast before. Before we even had any particular market traction. And so it is interesting how. How often we ended up circling back to the same problems because we just had badly experimented our way out of the problem space. So it's like it was almost as if we were too theoretically engaging and trying to get everything to be theoretically perfect in terms of the product development process and then ran into just wait, you know what? At some point reality is different and sticking with the same problem was a really, really important one. And then there's a similar mistake that I made, which was in team composition. It's that I almost very specifically hired a team around me that was very different to me in almost every aspect because it was almost like on paper, I wanted the perfect team. Like everyone was going to clash with each other and the clashes were going to lead to the best possible ideas. And you know what? When everyone clashes with each other, no one really likes the. That you create. It's like it was. When it was too much that became a real problem.

Yeah. Wow. That, that mirrors a transition that I've gone through where with, with my last company, with Human, I was absolutely obsessed from the earliest days with testing our ideas against the crucible of sales. Like, I was like, I think this is valuable. We all think this is valuable. Let's see if other people not only tell us it's valuable, but are actually willing to part with money because we do not want to waste the precious time of our lives on something that only we think is important. So, so we sold. We had paying customers before we had a functional dashboard. We had to like do one off reports. Right. Whereas quite possibly because of this time I can afford it. I, and you know, I put a lot of money into fulcra. We therefore put more money into upfront development. Before we certainly had testers, but we were not making sure we did not have that commercial discipline of asking for the money and using that as a signal of building. On the other hand, the advantage of that is that we are trying to build something that is like long term relevant for the future. And if you're doing category creation like the, you are building something that people do not yet know they need.

Absolutely. And to that same point, like had we not had that leeway at the time, we would not have been ready the moment ChatGPT came out, which then meant that we would not be here today. So like we would have run out of money basically in year one and it would have been a very different, very different type of situation. So ultimately I'm not entirely sure and this maybe it goes back full circle to the whole thing. It's interesting to figure out. What lesson do you take from it? If I were to do it again, exactly what would I have done differently? I think the one thing I would have done differently is stick with the same problem and not iterate your way out of the problem too quickly. But that was the biggest lesson I took.

Right. Well now with mindzone you do have engaged customers that are learning how to make themselves more capable using cutting edge AI tools. Now that itself is an interesting problem of let's say, meeting people where they are. Right. What are you learning about that end of innovation delivery? Did you have a customer base that, that is sort of similar in, in, in where you meet them or you know the. Or, or is it, or is it like super variegated? How does that work?

So it is, it is very varied and the same at the same time. So the, and I'll explain what I mean with that. The. I think learning how to use AI right now is very different to any other skill development that has happened before. And the reason for that is that it feels like magic in some cases. And so actually some of the secrets, the thing that we deploy is by making the magic seem approachable, you are getting people to a point where they get really excited about being able to wield these tools themselves. And that is it. That is the type of learning experience that actually is extremely rare. You have people that get passionate about leadership and management and whatever their craft is for sure, but it's not like hey, I'm going to teach you how to use this magic tool type of exciting. And so and that's the. So from a learning perspective like our actually this, this you can only do really with AI I don't know if there's a comparable but basically when you get someone okay, well I did not know that this possible to achieve with technology. They first get the excitement and then you show oh no, no, but you can do this yourself. And then they get really excited and then you break it down into okay, well it's actually very simple. You do this, this and this and here you go, now you can do it yourself. And the, the extent to which is both different for everyone as well as the same is that the biggest thing that we see is there is a big difference between what people think their respective levels of usage of AI is it their job. But there is actually most people don't know how to use it whatsoever. So the biggest problem we run into is that over 100 people, you might have 10 people that say they're experts, that say they're advanced and 50 that say they haven't even started yet. The reality is you're better off assuming that everyone is at a level of I haven't even started yet because the people that say they're experts are nowhere near expert. They're missing so many fundamentals, but they might be using it on a regular basis or they might even be talking about it to others on how they should be using it. But we're still in that phase where there's just a lot of fundamentals lacking. And so our approach tends to be baseline it because nine times out of 10 an expert is actually a beginner.

Yeah. Oh that that makes so much sense and maps to what I see actually across my own social graph. Like I would say that the I, I can almost use O1 Pro adoption as a determinant of this phenomenon. So I across my social graph the most of the people that I know are still kind of AI skeptics. They'll complain about hallucinations or whatever and I might can afford it. I and you know, I put a lot of money into Fulcra. We therefore put more money into upfront development. Before we certainly had testers, but we were not, you were not making sure you did not have that commercial discipline of asking for the money and using that as a signal of building. On the other hand, the advantage of that is that we are trying to build something that is long term relevant for the future. And if you're doing category Creation, you are building something that people do not yet know they need.

Absolutely. And to that same point, had we not had that leeway at the time, we would not have been ready the moment ChatGPT came out, which then meant that we would not be here today. So like we would have run out of money basically in year one and it would have been a very different, very different types of type of situation. So ultimately it's, I'm not entirely sure and this maybe it goes back full circle to the whole thing. It's interesting to figure out. What lesson do you take from it? If I were to do it again, exactly what would I have done differently? I think the one thing I would have done differently is stick with the same problem and not iterate your way out of the problem too quickly. But that was the biggest lesson I took.

Right. Well now with, with Mind Sewn, you, you, you do have, you know, engaged customers that are learning how to make themselves more capable using cutting edge AI tools. You know, the, that itself is an interesting problem of let's say meeting people where they are. Right? The what, what are you learning about that like end of innovation delivery. Right. So did you have a customer base that is sort of similar in where you meet them or the, or is it like super variegated? How does that work?

It is very varied and the same at the same time. And I'll explain what I mean with that. The. I think learning how to use AI right now is very different to any other skill development that has happened before. And the reason for that is that it feels like magic in some cases. And so actually some of the secrets the thing that we deploy is by making the magic seem approachable, you are getting people to a point where they get really excited about being able to wield these tools themselves. And that is the type of learning experience that actually is extremely rare. You have people that get passionate about leadership and management and whatever their craft is for sure, but it's not like, hey, I'm going to teach you how to use this magic tool type of exciting and, and so there. So. And that's the. So from a learning perspective, like our, our actually this, this you can only do really with AI. I don't know if there's a comparable, but basically when you get someone to okay, shit, well I did not know that this was possible to achieve with technology. They first get the excitement and then you show, oh no, no, but you can do this yourself. And then they get really excited and then you break it down into okay, well it's actually very simple. You do this, this and this, and here you go, now you can do it yourself. And the, the extent to which is both different for everyone as well as the same is that the biggest thing that we see is there is a big difference between what people think their respective levels of usage of AI. Is it their job? But there is actually most people don't know how to use it whatsoever. So the biggest problem we run into is that over a hundred people, you might have 10 people that say they're experts, 20 that say they're advanced, and 50 that say they haven't even started yet. The reality is you're better off assuming that everyone is at a level of I haven't even started yet. Because the people that say they're experts are nowhere near expert. They're missing so many fundamentals, but they might be using it on a regular basis or they might even be talking about it to others on how they should be using it. But we're still in that phase where there's just a lot of fundamentals lacking. And so our approach tends to be baseline it because nine times out of 10 an expert is actually a beginner.

Yeah. Oh, that, that makes so much sense. And maps to what I see actually across my own social graph. Like I would say that the I, I can almost use O1 Pro adoption as a determinant of this phenomenon. So I, across my social graph, the, the most of the people that I know are still kind of AI skeptics. They'll complain about hallucinations or whatever.

And.

Then a subset, a subset has already subscribed to one Pro and they're having the most fun. And, and, and all of us in, in that like small camp, we're like highly confident that we are unprofitable customers. Like so yeah, there's like this huge divide, like $200 a month. How could that possibly be worth it? And then you know, if we're using it, we're like.

I'm sure my GPU and power costs are far higher than 200amonth. Well, and also just if you're using, depending in how you're using it and if you're, if you're professionally active in any type of knowledge intensive activity, the return is dramatically higher than those $200. And the, this is actually one of the bits that I'm a bit scared about at the moment, which is that two years ago I would have said generative AI and AI systems are the most democratized piece of technology ever built. Like there hasn't been A time and place ever where for $20 a month, everyone would have access to the same capability models Everywhere. Now, with O1 or Zero1 Pro, we're actually moving towards a world where you're paying for inference, right? So you're basically paying for thinking time and for the more it thinks, the more you get quality answers. And it looks like O3 is confirming that. So what that means is you're getting back to a world where the more money you have, the more intelligence you get. And that is something that if you then don't do the training to begin with, so you're still stuck on your crete here somewhere. You haven't even gotten yourself to the use cases that would allow you to marry $20 a month. You will definitely not engage with the $200 a month. Even if you somehow had the money, you would actually not feel comfortable to then pay $2,000 a month when the $2,000 a month tier will inevitably at some point exist. Because if you're not experimenting enough to develop your confidence that you know which use cases and how to leverage the $200 into 2000, you definitely won't get the $2,000 use case to leverage into 20,000.

That's right. That's right. If I gave everyone that I know, if I'm like, use my credit card sign up, it's fine. The I. I don't know how helpful that would be because it seems to me like the little tricks, it's not even tricks that I have tacit procedural knowledge of how to get better results out of Claude and chatgpt and fascinatingly did not see this coming. Like, increasingly the Gemini like family of products that are similar to like. The problem is the reason why I use tacit is that I don't even know where I would begin to try to catch someone up to. To what I am doing. Because it's a lot of little things, right? Like, like, it's. It's how my grandmother thought about cooking, right? She, she, like, she did it extraordinarily well and extraordinarily quickly, but could not write a cookbook. She'd be like, well, you add enough salt until it's like salty enough, like you, you know.

Yeah, I mean, I, I both agree, but also it's. It's obvious it's my job literally to try and break that down. So I'm in a slightly different position. Like, I spend my days trying to communicate this in a way that works for others. But I do agree that. So, for example, O1 is not yet a model that we really teach because the reality is so few people use it. Like we're we're focusing on on general models that are available to to company users and and 01 so 01 Pro I now over especially over the holidays actually that's where I really started to use it because the the when you have to think through really thorough strategic questions that have many different parameters it's extremely good. And so you need to find the types of use cases where you where you can leverage that type of information into real advantage and so on. But I totally agree with you that now I have very specific ways of using it and the more I use it the more I find myself already about only about three weeks into properly using it or about a month or so when I go back to GPT4O I'm like oh, the only reason I'm doing that is because I don't want to wait five minutes for an answer.

Yeah right, right.

But it's qualitatively not in the same league and that's the thing like so you end up having to compensate like 3 or 54 iterations with GPT4 might get me to the result of the OF01 Pro in a particular context. But yeah it's an important thing.

I wonder now that we are thinking specifically about this I'm reminded of my pre Google Internet search abilities. Like I felt like I put a lot of time into how to get a good result out of Yahoo and out of altavista and it really mattered. So so, so the difference between my fingers on keyboard using AltaVista to find something on the Internet versus others like was was measurably better but that stopped being true. Like Google with PageRank was so much better that made your skills with using it matter less. I wonder if something like that will happen here.

So I think I think the same will happen but actually I think the effect is slightly different which is that I agree that it pulled the bottom up but it didn't get the top down. And so basically so the even in Google today like very simple small things but being able to time bound a query on Google or looking at similar websites to a very simple commands that are not hard to do. They will give you a dramatic edge on how to how to do a Google search and get you to the result much faster. But there are small few techniques but only very few people actually use them. So you're right that most of the stuff has been abstracted away but it has basically created kind of two tiers where you have the ones that do the minimum work to at least have those few things and then you have everyone else that just takes the default. And this is exactly the world that I think that we will end up walking into, which is people that will take the effort to go one or two steps above and beyond what the model does. And those that are just going to default use whatever the model has been programmed to try and compensate for, which is going to be. It has to be tailored to the average individual, which means going to be tailored to the average level. Intelligence is going to be tailored to the average whatever they define as average type of thing. But it's going to be an interesting experience there.

One particular way that I'm actually looking forward to seeing where this plays out is in changes to software. In light of the fact that power users will be armed with the ability to write one off code essentially so cheaply that I think there will be new ways for us to interact with our software with a little bit of a coding assistant to make the. The user interface, the pre built user interface. No longer the limiting factor of how I can interact with my software. And again I highly doubt that literally everyone will use that capability. However, I strongly suspect this will transform what it means to be a power user.

Totally agree. And I actually do think that everyone will end up there in this case. It's just a time question kind of when do they get there? This is something that I thought Google got really right in their demo. Never actually did it yet. But I don't know if you remember about a year and a half ago they did this UI on demand demo at Google I O which was. So it was. You ask a question and the ui, the, the. The AI builds a UI in order so it decides what is the right level of visualization that I need the interface that I need to build in order to present the answer to this query to you. And if that is a multiple choice question that it's asking, well, you get a multiple choice interface in front of you. If that is a visual based Alcro because you asked for dishes that you want to evaluate then it will give you like a gallery type of thing. That. That is something that I definitely can see moving into. And as personal experience I'm already the amount of times that you can now ask for an app to be built and basically it just gets built without you having to do anything. And the. If you're close to this space and you're observing how quickly the bar moves up where things that I would have asked a month ago wouldn't have worked and today they suddenly, they, they, they do and just. It's absolutely crazy how fast this is moving.

Right, right. I'm seeing this in the Fulker user base. I don't know how many other startups are doing this, but we just decided to absolutely embrace this future of what Scott Belsky is calling DIY software and what I've seen you call personal software, which is this user driven, maybe a one off app. Like I have a need right now and I want it to look this way and maybe next time I'll write the code differently. So in the context of Folker, what we've done is we're helping everyone connect all of the smart devices in their lives to the, you know, single Fulcrur data store and we give them out of the box visualization so you can just explore all of your own data. You, you, you have to be able to see it to have ideas about how to use it. But anything you can see is also accessible via an API with the, you know, with us with a machine readable spec. So, so you do not even need to read the API documentation. You can just ask, you know, your favorite model to read it. And, and, and if you want to do some custom one off like data science on yourself, you know, we got you, you just go nuts. You, you want to, you want to take all of your daily data and, and, and set it to music, right? Like, like daily song generation, whatever, right? And right now I would say that most of our users are actually adopting because they just want to stop visiting 12 dashboards. They just want to, you know, see everything in one place. So, so, so it, it is very much. They're, they're consuming the interface that we made for them and a minority of users are using the API to do, to do one offs. But I, I think the, I think the ratio is going to switch where what's going to happen is, you know, people do tune in to the Super View but in the total number of custom views that have been built and what people are doing that's just very situational will come to constitute the majority of usage. That's my prediction. So I have no idea how fast that'll happen.

Yeah, I think it's going to go faster than what most people realize. The speed of this is breathtaking at the moment. It's, I mean as a, an engineer by background like the I, I used to take a lot of pride at being at the cutting edge of whatever technology was able to, to achieve and in the Last two years, I, I've definitely transitioned to a place where I can no longer confidently say if something is possible or not. But also I cannot, I can't keep up. And that's a really weird situation for me to be in because I always was able to keep up in, in terms of technology. Like, that was a thing that I. That was part of my value is I am up to date and I'll be able to have that conversation. And now it's kind of like if you're, if you're out for a week, you're behind. It's like, what happened.

Yeah, but you've been countering that with some, I would say, conscious disciplines starting, I think more than a year ago with almost daring yourself to write custom GPTs just all the time. But now that transitioned to like DIY software. I, of course, was particularly interested in something that you shared. I think now it's been a few months that was like, you know, personal. That I'll just call it personal tracking. How has that gone? What's your favorite creation so far? Let's. Let's go there.

I think I call the personal productivity platform the ppp. The now, the. This is something that I firmly believe in, which is that as we're moving, and we already did, to a degree like we've had you as an individual are no longer. Or the ability for you to be productive is no longer just determined by you as a person, but also by you and your setup, which it to a small degree could have been. The laptop or the PC setup that you have, the phone that you use, the types of applications you use to manage your life, those do have a direct impact in how you're able to manage workload coming in and workload going out. And what is true with AI now is it's that, but on steroids, it's basically you. Now, in the definition of that, that personal productivity platform, comes the ability to outsource to external agents whatever work comes in. And if you're set up in a way to effectively outsource and effectively manage the parallelization of various agent workflows as the work comes in, you as an individual will be able to do dramatically more work than the same other individual with the same intellect, the same situation, but without that type of platform. And so before that might have been, say a 2x, maybe 3x increase if you're really, if you were really well set up, but we're now looking at 5 to 10x and if not very soon, you're looking at unbounded Depending on your level of parallelization, a single individual might be able to do the work of 100 that are not set up with AI in a similar type of way. And so that is the bit that I'm excited by and a bit scared about. When I rationally think about this, it's like I know I'm on the cutting edge, so I'll be part of the people that have at least the higher multiple of that productivity increase, I'd say. But it does take a lot of work and it is extremely different to anything we've seen in the past because it does increase to a degree, very strongly increases the divide between people and what they're able to manage. Because if before the multiplier was two to three, that's, and our entire society was, was governed around the idea or set up that, that that type of multiple was possible. If now you're looking at one to a hundred, that is not our social mechanisms, our, our performance benchmarks, they're not built to try and take those things into account at the moment and that there's going to be some cracks showing everywhere because of that.

I have a very simple personal goal with this respect, which is I'm trying to build the foundation of my life systems in such a way that I would be capable of running multiple successful companies without physically, mentally and emotionally destroying myself. Like patently. There are people who manage to successfully run multiple companies, but they also seem to not be having enjoyable lives. So is it in fact possible to not have to pay that trade off with human? The length of time we took to go from zero in revenue to ARR in the 20s was pretty fast. That was about a five year sprint. And then we were Ebitda positive in the 20s and it felt really good especially to crossover because it just meant that the company was sustainable. But I was emotionally devastatingly draining. And it's hard to talk about these sorts of things because in some respects if you're running a fast growing startup, you're living the dream. So you're not, you know, supposed to be complaining but, but it was very.

Very hard running the fast growing startup. Yeah, yeah, right, right, right.

I do think that that there, it just has to be the case that the, the very mechanisms that you're describing right now should make you capable of so much more. And, and as a result, can we not consciously choose to like buy with that increased capability, actually let's call it emotional safety margin. So, so it's not like oh well, since I could do one successful company and destroy myself I will do 10 and destroy myself. How about eight? But I get good sleep.

I mean, I, I, I wish, I wish I'd agree with you there. The I, my mental state agrees with you, but I also know that somehow I'm building minds on again and I'm still going through similar things even though I don'. Anymore. And I, I have had to come to terms with, I guess at some point the realization that part of the pain is it's not that I seek it out, but it's, that is part of the differentiator. If you're trying to build things that are exceptional, then that requires exceptional effort. Otherwise the thing is you have a lot of other people around you. And so if you're trying to do something better than them, you have to put in something if they have the same tools that you might end up competing. And this is again the thing which is if it's capital related, like okay, you can throw a lot of capital at more intelligence and do that, but you just end up competing with other people that have the same, if not more capital. And if they have more capital, you're going to have to find a way to somehow outcompete them. And so you end up back in the same loop. And the question, at least for me, where I ended up is that everyone has a bar at which they stop being okay to, to give more to try and win. And, and so, but that bar doesn't move. So you're, I think the way you're thinking about it works really well so long as you keep winning. The moment you feel that there is somebody else doing something slightly more, you're like, oh, well, but I have a little bit more to give, so I'll just go back into that and then go a little bit further. And I don't know if that's just me, but definitely.

Oh my God. No, you're still right. Yo, dude, you just blew up my fantasy.

Oh my God.

That is, that is absolutely right. Because the, if you are doing something innovative, then even if you are winning, or I should say even if you're going to win, it's actually ambiguous at the time whether what you are trying, like can't work or just hasn't worked yet. And while that ambiguity obtains, what all the hardy entrepreneurs that I know do is they act as though it's option B, not option A.

That's exactly it. And this is the, so I mean last year for me was a direct result of exactly that. Like if you we were in the right space doing the Right thing. We were going from zero to, to revenue making. We got to profitability at the end, which was really good. Everyone outside will looked at this. It's like, holy. This whole thing just took off like crazy. And it was the most gruesome year of my life. Like, I have not gone through that before. And it, it honestly makes me feel stupid sometimes because it's like, wait, why the hell did I do this? Because the. I've done my exit. I've gone through it. And exactly what you said was the thing I set out. It's like, hey, I want to be able to do this, but do this in a way that is also mentally healthy. What do I do? I get myself back into exactly the thing and it's even worse.

No, I'm sad to say that I feel you. I'm in the same place. I'm holding on to this fantasy that I'm going to build the foundational system that is going to unlock the true sustainability, but I have not yet achieved. Well, okay. Speaking of making yourself more capable.

Yeah.

The, the I, I, I think you, you, you're ahead of most of us in, in your experimentation in making tools with AI models that are, are making your life better. Of the experiments that you've tried thus far, what are you most excited about? You know, adding to or growing from here? And like, let's narrow this to especially, you know, outside of work. Right. So the, the, it's unquestionable that Mindstone is helping a whole bunch of people literally get more done at work. But I'm, I myself am personally preoccupied in, in adding tools of like, personal fulfillment and capabilities outside of work. Like, the. What, what did you learn from ppp? What's, what do you want to keep and what do you want to build on from here?

Yeah, so it's, it's partially success and partially like just iteration. Like part of this whole thing for me is learning on, on the go. And so I, I must have rebuilt my own flows so many times last year from the ground up. Again, like, it was an interesting learning process. The, the biggest thing that I found is I no longer make a decision without running it through a, an AI assistant anymore. Anything that is of importance at the cost of getting a second opinion on things or to get someone to play devil's advocate or whatever I'm thinking about both personally and professionally. Is there, well, at a personal level, a little hack for anyone? That, that one said these models are dramatically better at choosing the what TV show or movie you should be watching than Netflix. Amazon, Apple TV or any of them will ever be. It's.

I haven't tried that. Oh my God, I gotta try that.

Literally night and day. You feed it a list of like the 10 TV shows you really liked and I kid you not, when I asked it it was going through the first 10 that it gave me, I had to go through and say yep, absolutely love that. Already saw it. Yeah, absolutely love that. Already saw it. Like it's so spot on. And the same for books by the way. And it's interesting because it makes for, for those of you that understand what how the models are built it makes total sense why these models very specifically are great at that type of recommendation. But that's been a very strong quality of life improvement for me. So the other, the books, very direct examples I'm now reading was it Final Empire from I forgot the name of the author now But I, I notoriously have a hard time finding great books. I love a specific type of fiction and, and I, I think nine times out of ten I start a book and I just, it's not really what I want and so I have to try another one and it got this one right. Not only was it one book, it's a full series and now I've been whatever nine months into the series. Yes, it's really cool. Fantastic, right?

Oh man. I had a similar like I can't think AI for this but I'm eternally grateful for the, the friend that introduced me to E M Banks and the culture series. Like you know, I got one and then vacuumed them all up but of course you're right like ask the giant multi parameter models to like explore literally the latent space and find like the cluster that includes these examples.

Exactly.

Yes, obviously we should all be doing this.

It makes a ton of sense.

I use something called the unschedule technique to add into my productivity system like doing non productive things like like why don't you what so I'll like schedule my TV watching otherwise it does not happen.

Oh I'm exactly the same. The yeah, I have, I have to schedule in my lunchtime because otherwise lunch doesn't happen.

Right, right. This had not occurred to me before in my own personal operating system documentation. I actually wrote a line many years ago that is the answer to have you thought of everything? Is of course not right and the.

All of our lives are.

Are far more novel than most human lives in the evolutionary history of humanity as a species and the just developing a a simple convention of running decisions by your favorite LLM strikes me as just lowering the, the cognitive load in a way that probably has some sort of accretive effect over time. Because the answer to have you thought of everything? Is of course not. But at least you can have a private thought partner.

So if you think about it, the more important the decision, the more people you are likely to get involved to get opinions. The problem is there's for that. So like before you get married you might have talked to a whole bunch of different people before. Like you do that before you have kids. You will have. I mean I definitely talked to a bunch of different people before having our child. And, and there are some decisions where you know what? It's just not worth getting taking some unless it's an enjoyable conversation. It's just not worth spending somebody else's time on. But that doesn't mean it wouldn't benefit from their time. It's basically you're protecting their time because otherwise you'd constantly be talking to everyone about every single thing that you're doing in your life. Now when the cost of intelligence drops to zero, it means that all of those decisions that otherwise would not have hit the bar to merit wasting in a way somebody else's time now benefit from a double checking from a quick back and forth which not only makes them likely better as an outcome where you get to, but also actually accelerates your own thinking. Because how many times do you have the blank page child problem on a particular situation? You don't quite know how to approach it. And now you always have an approach. The blank page problem is no longer a thing because AI will give you the ignition in a way.

Well, you know, you're touching on the theory of the, of the moat for Fulcra, which is. I also have played it out in exactly the way that you just did. So imagine the cost of intelligence goes to zero. The unsolved problem that remains is what I call the observability problem, which is in order to get effective help from anyone, from a machine or another person, they need to understand your situation. And so, so like the fundamental theorem of Folker was actually there's a whole bunch of connected devices in your life that are all siloed which if brought together collectively solve for the observability problem. So then you reduce the effort of, of like giving someone the, the inputs into what's up. So, so that you don't have to explain all of that right? Like, like if the only way to do all of that is, is via your fingers, it's still a lot of it's still too much work.

Absolutely. Well, and let alone the fact that you probably don't even know how to properly explain it.

Totally that's exactly right. Yeah. Even my probably still untutored use of AI augmentation on my personal data shows me that I have had patterns of my behavior for years without ever realizing it.

Yeah, I can certainly imagine that.

It is funny that the human brain is often described as being this amazing pattern recognition machine. And, and obviously there has to be some truth to this. But on the other hand, I think we all have examples among our friends and family of like total lack of self awareness of their own patterns. And so if you can see that in other people, the next move is you go wait a minute, I must be doing that also.

Well, this is the other bit as well, which is that, I mean there's one the observability there which is different. I guess you're not aware of your own metrics in the way that somebody else in some case is because they observe it in a different way. But there's another effect here which is the what do you do when that intelligence is more intelligent than you? And so I must say that this was an interesting experience for myself over the holidays as I kind of went through what is like mind stone strategy for 2025 and I was using 01 Pro a lot. And there were definite moments where I was like wait, this thing is getting close to just like it's, it's be. It's beyond my thinking process here. I would not have thought about that. It kind of felt, not all the time, but on some moments it was really, should I be trusting this thing or should I be trusting my own thing? It was kind of like, wait, interesting. It tells you something about maybe too much that I don't want people to know about my own intelligence that we're already at that point with 01. But you, it is an interesting, it's an inevitability. I think in terms of where we get to the question is just when. But what, what do you do with that? What do you do when you end up with an AI that has more data than you? You know, that they reason better than you, you know, and they're coming to a different conclusion than you. Interesting. Now what?

Quite. Yes. Oh yes. And there's an interesting trust dimension to this where. Well, it's patently the case that sometimes people call this the Nobel effect. People who have achieved great wisdom and intelligence in one domain tend to perhaps overestimate themselves across all domains and imagine that they're an expert in places where they patently are not. The very uncanniness of these machines means they could earn your trust by appearing to outperform you in certain domains such that then they're not. Yeah. And so, so, so, so you over trust them in the wrong dimensions. Like, like, you know, are these things spiky in their capabilities or are they truly even is. Is just a really hard eval problem.

Yeah, for sure. It's. And it's, it's not one that I have a solution for you. It's not one that I had experienced until literally just a few weeks ago. So like I was kind of like, yeah, it was that getting close to that barrier.

That's awesome. I have not had that experience but it's because I haven't been trying the systematic thought partner engagement that you're describing, which I'm going to be doing now. So I can run into these same kinds of envelopes. That actually sounds super cool.

It is really cool.

On that note, let's wrap. We've, we've talked about what Mindstone is doing. How do listeners find you online and learn more about you and also about Mindstone.

So you can go to mindstone.com we have everything there. So if you're interested in learning how to use AI to be more productive or with your job or for your team, come and talk to us about it. You can find me on LinkedIn. So, Joshua Vola. I post quite a lot. I try to. This is part of how I'm trying to drive kind of awareness out there. I try to post a lot about practical applications of AI for everyone. So if you're interested in that, follow me mostly on LinkedIn. I'm decently active on X as well, so if you want to follow me there, you can also do that. Fabulous.

All right, well, thank you for this time. We covered a lot of ground. This is an absolutely fantastic conversation.