Uncorrected Scribe v2 machine transcript. Neutral speaker labels are specific to this recording. Review flags: 4.
TXT · SRT · VTT · Untouched JSON · Review flags
Source: interviews/1300-empowering-minds-with-ai-steven-schrembeck-on-technology-self-control-and-mental-clarity.mp3
SHA-256: 28661444f85f0c64fd233548f0597043a622656de87707bc9a1f6ed69b785a94
00:00:00.460 Speaker 1 College football is back, so Hilton called in me, the superstition concierge, to
00:00:04.660 Speaker 1 make your fan rituals a reality. Need a room to match your lucky number? We got you.
00:00:09.100 Speaker 1 Wanna make sure our team doesn't wash your lucky jersey? [sniffs] Ooh, that smells
00:00:13.480 Speaker 1 lucky. [whistle blows] Hilton's unmatched hospitality can keep up with any
00:00:17.360 Speaker 1 superstition, even a marching band wake-up call at 5:55 and 55 seconds. Hit it.
00:00:22.220 Speaker 1 [marching band music] When you need a team that will do whatever it takes on game
00:00:25.300 Speaker 1 day, it matters where you stay. Hilton, for the stay.
00:00:30.620 Speaker 2 This episode is brought to you by PayPal. [upbeat music] Holiday rush? Skip the mall
00:00:35.020 Speaker 2 traffic and shop online with PayPal. Shop your favorite online stores like eBay,
00:00:39.700 Speaker 2 Fanatics, Walmart, Sephora, Target, and TikTok Shop, and choose to pay now or pay
00:00:44.520 Speaker 2 later at checkout, plus pay in four with no interest and no late fees.
00:00:50.460 Speaker 2 Shop now with PayPal. Pay in four subject to approval. Learn more at
00:00:54.260 Speaker 2 paypal.com/payin4. PayPal Incorporated, NMLS 910457.
00:01:02.340 Speaker 3 Good morning, guys, and welcome to the podcast today. I've got a fantastic guest for
00:01:07.200 Speaker 3 you coming out of the AI field, Steven Shrembeck from
00:01:13.020 Speaker 3 Georgia, rights out of, right outside of Atlanta. It's gonna be a really fascinating
00:01:17.400 Speaker 3 conversation because he's a startup founder working on AI for self-control and
00:01:23.380 Speaker 3 stability. And I'm always trying to get a pulse of the emerging technologies,
00:01:29.100 Speaker 3 trends, and fields, and I'm really in- interested and curious about, or in this
00:01:33.960 Speaker 3 area. So I'm really happy to have Steven on. So welcome, Steven.
00:01:38.100 Speaker 4 Good to be here. Nice to meet you, Chris.
00:01:40.180 Speaker 3 Yeah. You have a very interesting background and talk about your early experiences
00:01:45.580 Speaker 3 and what you're doing, and we'll dive right into the conversation.
00:01:49.400 Speaker 4 Yeah. I'm gonna sound like the opposite of impressive at first because I started out
00:01:53.540 Speaker 4 in, in college thinking that computer programming was something that was too hard,
00:01:58.760 Speaker 4 and then ended up switching into a computer science related major. From there, I did
00:02:03.760 Speaker 4 a pretty routine software development, and then I got a job at Amazon, not AI
00:02:09.419 Speaker 4 related, just regular software. About 2016, I started realizing that deep learning
00:02:15.260 Speaker 4 is having a second renaissance. This is incredible stuff. This is now like entering
00:02:19.460 Speaker 4 the level of complexity where as a regular software developer you can begin to
00:02:23.800 Speaker 4 understand it with a little extra work. So I started training deep learning models
00:02:27.240 Speaker 4 and, uh, mostly vision models, some audio models, stuff like that. And then of
00:02:32.700 Speaker 4 course there is a third wave of AI revolution here where natural language
00:02:36.920 Speaker 4 processing, the advances there combined with transformers and other deep learning
00:02:41.760 Speaker 4 models to give you, yeah, large language models that you're all familiar with. And
00:02:45.600 Speaker 4 all right, great, so now I'm gonna learn this. It's at an even higher level of
00:02:49.020 Speaker 4 abstraction. And so I've been building my own stuff for years, some models as well
00:02:54.700 Speaker 4 through my day job, and now I'm applying everything I've learned to just make tools
00:02:59.860 Speaker 4 that help people. Not like sensational AI, just like really practical. I'm using it
00:03:05.480 Speaker 4 where it's the right thing to do.
00:03:09.100 Speaker 3 Really interesting foray, and I was actually have this kind of curiosity 'cause I
00:03:13.440 Speaker 3 was watching this YouTube with a podcast with, it was Marc Andreessen and Ben
00:03:18.120 Speaker 3 Horowitz and like Peter Diamandis and a lot of these AI innovators, and they were
00:03:22.920 Speaker 3 talking, and also Sam Altman, and he was talking about large language models as the
00:03:27.300 Speaker 3 baseline and was very akin to the browser wars, like which, which company is gonna
00:03:32.780 Speaker 3 come out with the, the most dominant one. And right now it looks like OpenAI, but
00:03:37.400 Speaker 3 does doing a lot of things with their LLaMA. But the question I have for you, 'cause
00:03:41.180 Speaker 3 it really sparked my curiosity, is like these AI models are based on language, but
00:03:47.620 Speaker 3 they, he was talking about AI models being based on video and pictures. And what
00:03:53.620 Speaker 3 are your thoughts on models, AI models based on things that are not based on
00:03:58.560 Speaker 3 language, but like videos and pictures?
00:04:02.820 Speaker 4 Yeah. This is slightly less explored, but this is coming to the foray in
00:04:08.580 Speaker 4 multimodal models. There's been a lot of advances in the encoding. At the lowest
00:04:12.720 Speaker 4 level, your large language models have taken little chunks of words. You can imagine
00:04:16.600 Speaker 4 the word, I don't know, waffle, broken up into just a few pieces. So sound wa and
00:04:20.980 Speaker 4 then ful. It's not necessarily how it gets split, but that's about right. That's
00:04:24.260 Speaker 4 like the lowest, that's the atoms of large language models. And then those little
00:04:28.440 Speaker 4 pieces get rearranged in a sort of autocomplete way, and that's how it thinks,
00:04:33.680 Speaker 4 right? It thinks at that low level. It's, those are how it represents its
00:04:37.340 Speaker 4 information. But how do you go from the word waffle to a picture of waffle?
00:04:41.680 Speaker 4 Something only knows text. It has no conception. This is like describing color to
00:04:46.380 Speaker 4 somebody who has never seen it in their entire life. So this is a dimension that
00:04:50.660 Speaker 4 they've never experienced. So how do you get them to understand it, a picture of a
00:04:55.220 Speaker 4 waffle intuitively like we do? And there's a lot of techniques for converting from
00:05:00.020 Speaker 4 one to the other, but what they've started doing is actually training these things
00:05:04.980 Speaker 4 at the same time, and so that the lowest level encodings are shared representations
00:05:10.380 Speaker 4 between all of them, so that you can actually seamlessly move from one to the other.
00:05:14.800 Speaker 4 It's really crazy. You input like the text waffle, and then it has an internal
00:05:20.260 Speaker 4 representation of the different meanings of the word waffle in context, like what a
00:05:24.040 Speaker 4 waffle is, what it looks like. It can actually generate output out the other side of
00:05:29.480 Speaker 4 a picture of waffle, and that doesn't necessarily look good 'cause that's not what
00:05:32.780 Speaker 4 it's for. But it, it shares, like it co-locates these representations together so
00:05:37.940 Speaker 4 that... And you can do that with sound, and then you can train it again with other
00:05:40.900 Speaker 4 low-level stuff. And now it begins to be able to intuit like how these concepts are
00:05:46.080 Speaker 4 related in ways other than just text. And I think that OpenAI's model can, what is
00:05:50.880 Speaker 4 it, Sora, its low-level encodings are basically pixels and sound.
00:05:56.980 Speaker 4 But it als- it does this technique. It also starts to marry these things together
00:06:00.720 Speaker 4 with words, and it's pretty cool. It's hard to train. [laughs] These are not easy to
00:06:06.300 Speaker 4 build. No one's gonna train them anyway. Just use theirs.
00:06:11.250 Speaker 3 It's really fascinating 'cause I'm just always trying to just keep grasp 'cause, you
00:06:14.830 Speaker 3 know, when OpenAI's, like their ChatGPT came out in November of, uh, '23.
00:06:20.750 Speaker 3 Was it '23 or what was... It s- it seems so long ago. It was very clunky. It was
00:06:24.310 Speaker 3 just like the texting, it wasn't really... But now it's like the, the large language
00:06:28.410 Speaker 3 models have been trained really well that you put in text or even images now, the
00:06:33.730 Speaker 3 output is actually pretty decent and it just reminds me of when I was using Google,
00:06:37.590 Speaker 3 like Google for search. It was just all text, right? But then you could use like
00:06:41.230 Speaker 3 Google Lens, like pictures, and you could use your voice as well, and it's like what
00:06:46.310 Speaker 3 you were describing, this multimodal. Which brings me to my next question is, 'cause
00:06:50.650 Speaker 3 you talk about everyday usage of AI and I had another guest who was also, he was,
00:06:56.250 Speaker 3 his focus was to educate those who don't have access to AI so that they're not left
00:07:01.010 Speaker 3 behind. So when you talk about these, they're these tools to help everyday people.
00:07:05.870 Speaker 3 You talk about developing AI tools to help individuals regain control over their
00:07:10.790 Speaker 3 attention and choices in a increasingly distracted world. Elaborate on that and, and
00:07:16.670 Speaker 3 what that means.
00:07:18.250 Speaker 4 Just like the general concept, and a lot of these general concepts you can start to
00:07:21.970 Speaker 4 apply in your own life. You don't need to use anything that I'm making or anything
00:07:25.050 Speaker 4 else. But think of it this way, like large language models specifically, and all the
00:07:29.490 Speaker 4 stuff that gets bolted on top of them, right? The... I'll just shorthand it to large
00:07:32.830 Speaker 4 language models. There's a lot of magic under the covers. It's the same with Google.
00:07:35.790 Speaker 4 You type in a search box, but you just know that there's an ocean of code under
00:07:39.390 Speaker 4 there. So there's a lot of stuff that goes into this, but for now assume it's large
00:07:43.890 Speaker 4 language models. It's basically commodity intelligence. You can think of it as not
00:07:49.370 Speaker 4 like an especially bright person, just like your average person, like a, a, an
00:07:54.270 Speaker 4 average level of reasoning in a box. A general reasoning, and that is really
00:08:00.290 Speaker 4 powerful. One thing you can use that for, yes, you can use it to replace people.
00:08:05.770 Speaker 4 Yeah, however you feel about that, good or bad, is neutral. But you can use it to
00:08:10.430 Speaker 4 replace commodity intelligence. That's... Everyone tends to focus on that. But you
00:08:15.570 Speaker 4 could also use it in places where it would be unethical, unreasonable, or just not
00:08:21.070 Speaker 4 preferable to have a human do the same thing. So think of it like a hazmat suit, or
00:08:26.730 Speaker 4 if you've seen the show Cher- Chernobyl or whatever, where they have, they try to
00:08:30.130 Speaker 4 get the robots to go in and clean it up. [laughs] So like y- you put it in an
00:08:34.950 Speaker 4 environments where you wouldn't want a human. So you can imagine there's a lot of
00:08:38.990 Speaker 4 environments like that on the internet. It's a lot more dangerous, predatory.
00:08:45.270 Speaker 4 There's a lot of influences. So some simple ways to use it are just send an AI ahead
00:08:50.730 Speaker 4 of you just to go look for things that it's triggering or is it good, and just
00:08:56.250 Speaker 4 pacify it or sanitize some of the things, or even just censor some of the things you
00:09:01.030 Speaker 4 don't want to experience. Get rid of the noise, the intentional bias and inf-
00:09:05.830 Speaker 4 influences, all that sort of stuff, right? So you can think of this as having a
00:09:10.110 Speaker 4 hazmat suit. So that is a simplistic way of looking at that, but there's a lot of
00:09:14.410 Speaker 4 other environments and situations in which you would like to have human-like
00:09:18.630 Speaker 4 intelligence evaluate a situation without a human so that you don't, a human
00:09:24.610 Speaker 4 doesn't have to. So not just saving work, it's also sparing a conscious [laughs]
00:09:30.530 Speaker 4 creature from having to experience or go through this.
00:09:35.170 Speaker 3 Interesting. It's, and I actually have some follow-up questions around just
00:09:38.570 Speaker 3 developing tools for mental clarity and focus, and also self-control and ADHD. I
00:09:44.210 Speaker 3 just wanna understand it more. If somebody... 'Cause I'm hearing a lot, 'cause I use
00:09:48.090 Speaker 3 the AI for a lot of content creation and, but I hear a lot of people are starting to
00:09:53.090 Speaker 3 use like AI for chatbots and assistants. If somebody want- wanted to learn how to
00:09:57.990 Speaker 3 use AI for this area, where would they go? Where would they start? How would you
00:10:02.730 Speaker 3 recommend them play around and tinker?
00:10:06.510 Speaker 4 Most awesome part about this like third wave of AI, assuming you don't count the
00:10:10.810 Speaker 4 early work before the early 2000s, is that it's mostly at, it's mostly
00:10:16.830 Speaker 4 these big foundational models that are trained on millions if not more dollars.
00:10:21.710 Speaker 4 There's, nobody can run these things on their computers, not really. Nobody c-
00:10:25.290 Speaker 4 definitely can't train them. It's highly sophisticated, but they're so
00:10:28.650 Speaker 4 generalizable. It turns out that basically everything I learned [laughs] about how
00:10:33.210 Speaker 4 to construct these neural networks and the software on top of it is more or less
00:10:37.030 Speaker 4 useless. Okay, it helps conceptually understand it. Like the playing field got real
00:10:41.190 Speaker 4 level because it's mostly prompt crafting and some light software engineering, but
00:10:46.230 Speaker 4 even the AIs can write the software. So it has leveled the playing field immensely
00:10:51.410 Speaker 4 to, to the point where, yes, it's a skill to learn how to interface with these
00:10:55.070 Speaker 4 things, but once you get good at using ChatGPT, like stitching together some
00:11:00.190 Speaker 4 software to do that in an automated way is basically just logic, like common sense,
00:11:04.450 Speaker 4 like just trial and error. And then you can use no code tools, Bubble, and there's
00:11:08.730 Speaker 4 LLM specific ones like, uh, Flowise, I think is one. There's a couple like them,
00:11:12.970 Speaker 4 where it's just like a little branching structure. If you've ever seen like a f-
00:11:16.590 Speaker 4 flow diagram. It's okay if this, here's the prompt, then do this. Like you've got
00:11:21.930 Speaker 4 customer intake or you've got, uh, new patients or something. Did they fill out this
00:11:26.030 Speaker 4 form? Okay, was this empty? Check for errors, like talk to them about it. If that's
00:11:30.130 Speaker 4 all good, then send them to this sign-up page. Okay, then send them to this thing.
00:11:34.910 Speaker 4 It's easy. If you're a human being with a little bit of patience and you're willing
00:11:38.430 Speaker 4 to l- learn a little skills, then you don't really need software
00:11:44.210 Speaker 4 developers like me. Uh, so that's the good news, is that 80% of the use cases you're
00:11:49.810 Speaker 4 enabled to do by yourself, and it's much more approachable. So those are the tools
00:11:54.390 Speaker 4 I'd recommend trying out, but there is no replacement for getting in there and
00:11:58.330 Speaker 4 learning the nuances of how it works, like intuitively. And this is something I'd
00:12:02.270 Speaker 4 recommend actually learning how to intuit if you don't already use the AI tools.
00:12:08.630 Speaker 3 Yeah. I, I love this process, how you're describing just tinkering and just playing
00:12:12.550 Speaker 3 around. Um, go in and just like just see what, you know, that's how you best learn.
00:12:17.510 Speaker 3 Um- Well, the other question I have for you is, so you talk about, um, the AI tools
00:12:23.242 Speaker 3 assisting those struggling with ADHD or addiction and, you know, what are... how is
00:12:28.622 Speaker 3 this possible? What are specific features designed to help those struggling with
00:12:32.982 Speaker 3 ADHD and addiction?
00:12:34.702 Speaker 4 There's lots of different hows. It's like a constellation of different tools, and a
00:12:38.562 Speaker 4 lot of it's just regular software. But one of the fundamental problems is you have
00:12:42.202 Speaker 4 an, the untrustworthy operator problem. Let's say you've got a bunch of system that
00:12:46.842 Speaker 4 keeps you healthy, in check. I meditate at this time. I take my medication at this
00:12:50.281 Speaker 4 time. I don't go buy drugs or alcohol 'cause [chuckles] I'm not doing that. I
00:12:54.462 Speaker 4 don't... I, I make sure that I check into work, or I, I leave by this time. You can
00:12:58.942 Speaker 4 imagine a whole bunch of systems that people with ADHD and autism rely on. However,
00:13:04.142 Speaker 4 the operator themselves is not trustworthy, both in the sense that they forget, uh,
00:13:09.442 Speaker 4 they have low object permanence, they just don't even realize that they should be
00:13:12.562 Speaker 4 doing this, or they have emotional irregularity or motivational instability. So you
00:13:18.042 Speaker 4 simply can't operate in a way that's... Okay, I have a personal rule not to go buy
00:13:23.022 Speaker 4 alcohol on the way home. Okay that's not gonna stop you. [chuckles] Like, y- you-
00:13:27.742 Speaker 4 the operator themselves can't be trusted. So now what do you do when you- the system
00:13:32.622 Speaker 4 you rely on, you can't trust the person who's operating it? And you can have a
00:13:37.082 Speaker 4 nanny, or you can have [laughs] I guess a, a babysitter. You could have a sponsor.
00:13:42.002 Speaker 4 There are piecemeal solutions to all this stuff. But what would be really nice is if
00:13:46.702 Speaker 4 you had a system of accountability that understood the nuance. There's that
00:13:51.882 Speaker 4 commodified intelligence. This is where the AI part comes in. What state are you in?
00:13:55.882 Speaker 4 Can I trust you? Yes or no? Like, normally you can change your rules. It's just a
00:13:59.182 Speaker 4 rules engine. Like, I do this, I don't do this, make it hard to do this, hide this
00:14:04.282 Speaker 4 from me. It's more or less it, it knows you can't be trusted, so there's like this
00:14:10.342 Speaker 4 balance between this is where you need the intelligence. Do I let you just turn all
00:14:13.762 Speaker 4 this stuff off? Do I try to block a transaction at a liquor store? Okay I'm just
00:14:18.842 Speaker 4 buying this in order to make a, a flambé or [laughs] later. There's a lot of nuance
00:14:23.502 Speaker 4 here that brute force, like simplistic tools don't work. So in... you need something
00:14:29.182 Speaker 4 that understands. Also, you don't want it, it interrupting you or stopping you or
00:14:33.282 Speaker 4 turn- blocking your websites when you're trying to do stuff for work. There's an
00:14:36.842 Speaker 4 enormous amount of nuance in when to apply, like enforcement, and the net result of
00:14:41.962 Speaker 4 all this is that it just makes it easy to do what you wanna do, harder to do what
00:14:45.762 Speaker 4 you don't wanna do, and it makes it more or less impossible to not realize what you
00:14:49.462 Speaker 4 should be doing. And then the rest is more or less just set up your own rules. So
00:14:55.002 Speaker 4 that's the general system. But yeah, it's for sleep, it's for electronic addictions,
00:14:59.582 Speaker 4 avoiding substance abuse, spending control. These are the most common problems that
00:15:04.682 Speaker 4 people have, and so that's what we're focused on. But the principles are general,
00:15:09.642 Speaker 4 right? It's whatever rules you got for life. But it's just imagine a little, you
00:15:14.102 Speaker 4 know, guardian angel that is constantly paying attention to what you're doing.
00:15:17.742 Speaker 4 There's some AI there as well. And then whatever information you choose to give it,
00:15:21.242 Speaker 4 it's using that to evaluate whether it should step in to enforce your own rules or
00:15:25.882 Speaker 4 not if you don't.
00:15:27.562 Speaker 3 Really interesting. And then when you talk about ADHD and addiction, the other
00:15:31.762 Speaker 3 question is how do the A- AI tools you're developing help users main- attain mental
00:15:37.342 Speaker 3 clarity, and can you share examples of how these tools are used?
00:15:44.222 Speaker 4 Moving things. I'm operating from the state that your standard human at rest is
00:15:50.022 Speaker 4 already healthy, so if that's not true, that's a different target audience. So
00:15:54.302 Speaker 4 assuming that at rest when healthy you're functional, then it is mostly about
00:15:59.422 Speaker 4 removing influences. It's about decreasing your level of stimulation to improve your
00:16:04.022 Speaker 4 baseline dopamine. So a lot of this is like the hazmat suit I mentioned, like it's
00:16:08.182 Speaker 4 creating different interfaces to influences. Okay, you wanna listen to podcasts? You
00:16:12.362 Speaker 4 go through this other tool that desanitizes it. You wanna browse the web, you want
00:16:17.242 Speaker 4 to look at social media, you're gonna do it through this alternate interface that
00:16:22.062 Speaker 4 doesn't allow these things to pull your attention away. You can also tell it what
00:16:27.022 Speaker 4 your intentions are. This is what the first version of the software was. Tell it
00:16:29.922 Speaker 4 what your intentions are, and then it's paying attention to what you're doing, and
00:16:33.222 Speaker 4 if there's a discrepancy, it can just make you aware. But if you choose to... You're
00:16:37.101 Speaker 4 doing this to yourself. If you choose to follow up by adding more restrictions,
00:16:40.002 Speaker 4 "Hey, I said I was doing this thing for work and I'm not doing it after 45 minutes,"
00:16:43.882 Speaker 4 go through this pretty well-known psychotherapy routine, mirroring. There's... None
00:16:48.602 Speaker 4 of this is new, right? I'm just putting together existing Legos. Follow this step,
00:16:52.282 Speaker 4 then this step. If I completely ignore it, then I want you to turn off my internet.
00:16:56.702 Speaker 4 Not everyone's gonna do this kind of thing, but for certain problems it makes sense.
00:17:01.722 Speaker 4 Like substance abuse, for instance. You need a hard guardrail which says, "If a
00:17:06.042 Speaker 4 transaction comes through or is trying to be processed from these merchants or in
00:17:10.522 Speaker 4 this way, deny it." And that's just a hard enforcement. There are other rules, of
00:17:15.902 Speaker 4 course, to... It's never gonna be able to stop you. The purpose is just to add
00:17:19.202 Speaker 4 friction or to remove the presence of things that are making your life worse so that
00:17:24.722 Speaker 4 you don't have to use willpower. It's the, the whole purpose is you cannot rely on
00:17:29.442 Speaker 4 willpower. These people specifically, I'm definitely in the target audience, cannot
00:17:34.382 Speaker 4 rely on willpower reliably. So this removes, it adds a backstop, and ultimately
00:17:39.582 Speaker 4 allows you to retrain your brain for what you want and what you pay attention to
00:17:43.722 Speaker 4 because there's something, there is a digital adult in the room that will eventually
00:17:48.662 Speaker 4 step in, and that is something that we lose when neurodiverse people come from
00:17:54.182 Speaker 4 childhood to adulthood. There's nobody to step in anymore if you live alone or
00:17:58.502 Speaker 4 you're supporting yourself or others are dependent on you. You need reliable ways to
00:18:04.382 Speaker 4 not crash and burn. There's no one to step in. And so now this sort of preserves
00:18:09.662 Speaker 4 autonomy and independence and allows some software to step in by your own design.
00:18:13.782 Speaker 4 You're doing this to yourself. It's configurable. So that's the premise.
00:18:18.342 Speaker 3 Really the next question I have for you is, 'cause talking a- about just going away
00:18:22.822 Speaker 3 from the AI and just talking about more broader, talk about this what are your
00:18:27.022 Speaker 3 thoughts on society's current challenges with self-control, and how can technology,
00:18:32.002 Speaker 3 particularly AI, foster more mindful and intentional behaviors?
00:18:37.578 Speaker 4 Anyone relatively studied in like psychology or sociology recognizes just how we
00:18:42.398 Speaker 4 like to think we're, we have free will, but... And like we have a lot of agency. But
00:18:47.438 Speaker 4 most of our behaviors and the ways in which we can be influenced are very well
00:18:51.278 Speaker 4 documented. Like we are very well doc- they're super well-documented blueprints on
00:18:56.918 Speaker 4 how to get humans to do what you want them to do. And the sad part of this is, is
00:19:02.118 Speaker 4 that has been very well perfected and engineered upon by basically everything.
00:19:07.138 Speaker 4 Everything from business processes to the well-known attention and influence
00:19:12.538 Speaker 4 economies, ads, all this sort of stuff. Um, it's obvious people have figured out
00:19:18.098 Speaker 4 really how to perfectly manipulate others through technology. The nice part is that
00:19:22.538 Speaker 4 you can do that to yourself, and you can also detect that, and you can stop it. So
00:19:28.138 Speaker 4 that is the role of software in this, is that you can use software to harm yourself
00:19:32.598 Speaker 4 or to be harmed, and you can just accept that. Or at the individual level or an
00:19:37.178 Speaker 4 organizational level, you don't have to wait for society to change. You can just
00:19:41.658 Speaker 4 make your own tools or set your own rules, like none of this has to be software. But
00:19:45.918 Speaker 4 what to access and what not to access, um, and become aware of what's
00:19:51.318 Speaker 4 manipulating you or influencing you. And so that is really how I see it, is yes,
00:19:56.918 Speaker 4 collective action is great. The man is slow, frustrating, and often ineffective.
00:20:02.158 Speaker 4 Solutions are not great. So if you're waiting for US Congress to pass laws banning
00:20:07.278 Speaker 4 social media so that your 12-year-old isn't being unduly influenced, I think you're
00:20:13.018 Speaker 4 gonna be waiting a long time and you're gonna be disappointed. However, you can
00:20:16.258 Speaker 4 begin to take action now beyond just... A lot of it's simple, and a lot of it is the
00:20:20.598 Speaker 4 willingness to
00:20:23.058 Speaker 4 notice how things are making you feel. That is su- such a simple thing to say, but
00:20:27.938 Speaker 4 notice what is making you feel bad, become committed to that, and then be willing to
00:20:32.598 Speaker 4 live differently or to be creative on how you remove those, remove or blunt those
00:20:37.958 Speaker 4 influences on you. That's like the most important thing to do regardless of what ch-
00:20:43.298 Speaker 4 tools you use, 'cause nothing is ever perfect. You gotta fit it to your life. Remove
00:20:47.038 Speaker 4 the bad stuff and do more of the good stuff. I don't think that any of this is
00:20:50.858 Speaker 4 actually surprising. My insight for building a business in software was that people
00:20:56.458 Speaker 4 know what to do. The average person knows what is healthy for them and not healthy
00:20:59.918 Speaker 4 for them generally, but they have trouble actually doing it or not doing it. So
00:21:05.958 Speaker 4 that's where you should focus. Less on shame. Stop using effort and willpower, and
00:21:10.318 Speaker 4 just make it so that it's really hard, if not impossible, to do the things that are
00:21:14.338 Speaker 4 bad for you, and it's really easy, if not inevitable, to do the things that are good
00:21:18.898 Speaker 4 for you.
00:21:19.278 Speaker 1 [upbeat music] College football is back, so Hilton called in me, the superstition
00:21:23.118 Speaker 1 concierge, to make your fan rituals a reality. Need a room to match your lucky
00:21:27.098 Speaker 1 number? We got you. Wanna make sure our team doesn't wash your lucky jersey?
00:21:31.038 Speaker 1 [sniffs] Ooh, that smells lucky. [whistle blows] Hilton's unmatched hospitality can
00:21:35.758 Speaker 1 keep up with any superstition. Even a marching band wake-up call at 5:55 and 55
00:21:40.458 Speaker 1 seconds. Hit it. [marching band music] When you need a team that will do whatever it
00:21:43.938 Speaker 1 takes on game day, it matters where you stay. Hilton for the stay.
00:21:49.858 Speaker 2 This episode is brought to you by PayPal. [upbeat music] Holiday rush? Skip the mall
00:21:54.278 Speaker 2 traffic and shop online with PayPal. Shop your favorite online stores like eBay,
00:21:58.938 Speaker 2 Fanatics, Walmart, Sephora, Target, and TikTok Shop, and choose to pay now or pay
00:22:03.758 Speaker 2 later at checkout. Plus, Pay in 4 with no interest and no late fees.
00:22:09.698 Speaker 2 Shop now with PayPal. Pay in 4 is subject to approval. Learn more at
00:22:13.518 Speaker 2 paypal.com/payin4. PayPal Incorporated, NMLS 910457.
00:22:22.318 Speaker 3 Very insightful, and I'm sure the audience is gonna have a lot of follow-up
00:22:25.798 Speaker 3 questions. I have two more questions before we have around three, four minutes left.
00:22:30.618 Speaker 3 The other question is, the... One is AI can feel impersonal, and how
00:22:36.598 Speaker 3 do you ensure your tools foster genuine human agency and are accessible? Because I
00:22:42.478 Speaker 3 was... There's this concern from people that have access to AI and that know how to
00:22:47.638 Speaker 3 use it. They're gonna use it to, of course, enrich themselves and progress, whereas
00:22:51.258 Speaker 3 the common average person that doesn't know or is gonna get left behind. How do
00:22:57.157 Speaker 3 you address this?
00:22:58.897 Speaker 4 That's a profound problem, and I think that it's accurate. But that's a question
00:23:03.658 Speaker 4 that is well beyond the scope of just AI. This is any technology, what happens when
00:23:07.738 Speaker 4 you have exponential, uh, abilities. Like some people are becoming superhuman.
00:23:12.558 Speaker 4 That's a major challenge that is really a philosophical one. So that's... I won't
00:23:17.438 Speaker 4 weigh into that too much other than I would advise you don't need to be on the
00:23:21.478 Speaker 4 cutting edge. But, uh, uh, usually being about 20, 30% behind the cutting edge,
00:23:25.478 Speaker 4 being an early-ish mainstream adopter is about good. Just pay attention [laughs] and
00:23:30.118 Speaker 4 try to use these things. There was a second part to that question that I forgot.
00:23:35.438 Speaker 3 Yeah. It was this, the kind of the, the next question is just with AI,
00:23:41.598 Speaker 3 how do you help people use technology to create more meaningful lives? Just 'cause I
00:23:47.238 Speaker 3 know in our day and age, it's really, it's just more efficient, instant
00:23:51.278 Speaker 3 gratification, just more effective and just bam. But ultimately, with technology,
00:23:55.938 Speaker 3 we're losing meaning. For example, we're very connected, but we're in the, we're
00:24:01.058 Speaker 3 what some of the most disconnected people in human history, even though the
00:24:05.498 Speaker 3 technology is supposed to bring us together. So how, how do you, uh, use AI to
00:24:10.078 Speaker 3 create more meaning for people?
00:24:12.618 Speaker 4 Okay. That's a great question, and I'm willing to be late for my next thing to
00:24:17.078 Speaker 4 answer that because that's an amazing question. And I feel very strongly about this.
00:24:22.258 Speaker 4 We should not be replacing the behaviors that we want to do.
00:24:28.478 Speaker 4 We... This is where capitalist profiteering, like efficiency mindset falls flat, is
00:24:33.178 Speaker 4 that it, it... If we keep going, we've automated away everything, right? We have AI
00:24:38.438 Speaker 4 raising our children. We have technology that does everything for us. Eventually, it
00:24:42.258 Speaker 4 can just live for us, and what are we? You can just be- Put on a, uh, a feeding
00:24:47.006 Speaker 4 tube, um, and just [laughs] go unconscious. Uh, what are you replacing? Uh, I think
00:24:52.906 Speaker 4 this is where we, we have to be quite mindful of the fact that we are...
00:24:58.706 Speaker 4 At an individual or even an organizational level, what are you omitting? What are
00:25:03.466 Speaker 4 you actually replacing? Did you get rid of the things that you wanted to do? The
00:25:07.346 Speaker 4 purpose of an AI tool is the obvious that most people talk about, to replace the
00:25:12.046 Speaker 4 jobs and tasks that people don't want to do, or that it can do it better. Sure,
00:25:16.286 Speaker 4 that's fine. Go use it for those things. That's obvious. Pay attention to where it's
00:25:20.946 Speaker 4 overrunning and recognize that there were... there is meaning in work, there's
00:25:24.966 Speaker 4 meaning in helping people, and that you're not remove... What are you doing with
00:25:28.926 Speaker 4 this extra time? Unfortunately, if the common denominator, like the status quo, is
00:25:34.486 Speaker 4 that you go back to the attention economy, you start grazing for information, or
00:25:38.806 Speaker 4 you're in a world where nobody else is around, in the physical world it's very
00:25:43.106 Speaker 4 lonely because they're all zombies, that's bad. What are you freeing up your time
00:25:48.706 Speaker 4 and energy for? So you want time, energy, and attention freedom. That's awesome. You
00:25:54.226 Speaker 4 can keep opening that up, but if the way you allocate it is just pointless leisure
00:26:00.386 Speaker 4 or it has no meaning for you, then you're just getting more of a terrible thing. So
00:26:06.786 Speaker 4 this is, uh, again, a neutral tool. All it does is give you more time, energy, and
00:26:11.866 Speaker 4 attention freedom, maybe some financial freedom as well. But what are you gonna do
00:26:15.766 Speaker 4 with that? You have an excess, you have a surplus of these things, but how you
00:26:19.706 Speaker 4 allocate that for meaning is critical. It doesn't... It creates more of a thing, but
00:26:25.006 Speaker 4 it doesn't mean that you're gonna use your f- newfound freedom and resources. That
00:26:28.326 Speaker 4 is an entirely different, uh, conversation and problem. I have a, a lot to, to say
00:26:33.626 Speaker 4 on those, but I think we'll have to leave it at that for now.
00:26:38.726 Speaker 3 Yeah, I really enjoyed this conversation, and very insightful just coming from a
00:26:42.866 Speaker 3 SaaS founder and just a leader in this area. And what if people wanted to follow you
00:26:47.646 Speaker 3 and reach out to you and connect with you, how could they do that?
00:26:51.106 Speaker 4 Uh, there's a website, impossiblelaboratories.com. If you're a software dev or a
00:26:57.006 Speaker 4 designer or project manager, I, I'm looking for collaborators. This is a lot of
00:27:00.446 Speaker 4 work.
00:27:00.686 Speaker 3 [laughs]
00:27:01.446 Speaker 4 But now that I've started, like, doing user interviews and all this stuff, running a
00:27:04.506 Speaker 4 business is, like, crazy hard. It's... I don't have time for all this.
00:27:07.386 Speaker 3 [laughs]
00:27:07.886 Speaker 4 The loose collection of contractors plus me is not cutting it anymore. So yeah, if
00:27:11.386 Speaker 4 you... It sounded cool and you wanna try it out or you wanna help, this is mission
00:27:15.006 Speaker 4 driven. Uh, it's got profit motive because this is America. You can't do anything
00:27:19.706 Speaker 4 without a profit motive. But really it's a... This is not the easiest way to make
00:27:23.606 Speaker 4 money. I could just-
00:27:24.386 Speaker 3 [laughs]
00:27:24.866 Speaker 4 This, this is about helping people. So if that sounds cool, give it a shot. That's
00:27:28.966 Speaker 4 impossiblelaboratories.com. I'll put some links on there.
00:27:32.586 Speaker 3 Excellent. Yeah. And, uh, keep up the great work, and I really enjoyed this
00:27:35.806 Speaker 3 conversation.
00:27:37.386 Speaker 4 Awesome. Well, it was nice to meet you, Chris. And, uh, cool. Yeah, maybe we'll do a
00:27:40.566 Speaker 4 follow-up in a while.
00:27:43.106 Speaker 5 Fall savings are happening now at The Home Depot. Save up to $500 on select
00:27:47.766 Speaker 5 appliances, plus enjoy free delivery when you spend $998 or more. From prepping, to
00:27:53.306 Speaker 5 cooking, to cleaning up, Frigidaire has you covered. The custom flex temp drawer
00:27:56.866 Speaker 5 switches from fridge to freezer. With more than 15 cooking modes, including
00:28:00.106 Speaker 5 stone-baked pizza mode, you can enjoy restaurant quality pizza in as little as two
00:28:03.706 Speaker 5 minutes. Then clean up with the Frigidaire dishwasher featuring the industry's
00:28:06.886 Speaker 5 fastest normal wash cycle. Shop fall savings now at The Home Depot. Offer valid
00:28:10.806 Speaker 5 through October 21st. U.S. only. See store online for details.
00:28:13.306 Speaker 6 Football is back and PrizePicks is bringing out their best offer yet. On PrizePicks,
00:28:18.546 Speaker 6 you can make all the picks you want, all in the same app. And now PrizePicks is
00:28:22.906 Speaker 6 available nationwide. Lock in for kickoff because everyone and everything is on
00:28:28.286 Speaker 6 PrizePicks. What you on?
00:28:30.146 Speaker 7 Download the PrizePicks app today and use code SPOTIFY to get $150 instantly in
00:28:35.366 Speaker 7 lineups if you win your first $5 lineup. PrizePicks, run your game.
00:28:40.366 Speaker 6 Must be 18 or older and in legal states. Visit prizepicks.com for restrictions and
00:28:42.566 Speaker 6 details.