AI is dangerous because it agrees with you

00:00:09,720 --> 00:00:34,360 [Eric] Welcome back to the Token Intelligence Show. AI is changing the way that we work, and on the show we will talk to you about the state of the art in using AI. We'll also cut through all the noise out there and hopefully give you wisdom to become a great leader and a great worker in this new era that we live in. And today, we wanna talk about AI's worst failure mode. 00:00:34,360 --> 00:00:34,650 [John] Ooh. 00:00:34,650 --> 00:00:47,660 [Eric] Uh, where do things go really bad when you're using AI to do work? And this is a subject that's near and dear to both of us, John. Uh, I want to, 00:00:48,840 --> 00:00:58,440 [Eric] I wanna start out with sort of going back to first principles. And there's a quote, it's probably one of my favorite quotes about writing- 00:00:58,440 --> 00:00:58,500 [John] Okay 00:00:58,500 --> 00:01:00,560 [Eric] ... from one of my favorite authors. 00:01:00,560 --> 00:01:01,080 [John] Who? 00:01:01,080 --> 00:01:07,570 [Eric] Uh, and people will ask about writing, um, you know, what is it like to be a writer- 00:01:07,570 --> 00:01:08,040 [John] Right 00:01:08,040 --> 00:01:09,740 [Eric] ... to, you know, to do these things, um, 00:01:10,760 --> 00:01:14,380 [Eric] the process and all of that. And so David McCullough, he's a renowned historian. 00:01:14,380 --> 00:01:14,530 [John] Yeah. 00:01:14,530 --> 00:01:16,030 [Eric] He died not too long ago. 00:01:16,030 --> 00:01:16,440 [John] Mm-hmm. 00:01:16,440 --> 00:01:23,590 [Eric] Um, but unbelievable writer. If you don't like history, I would actually recommend starting by reading some David McCullough. 00:01:23,590 --> 00:01:25,860 [John] Okay. What, what's like one of his top, uh, books 00:01:27,100 --> 00:01:27,330 [John] that you'd recommend? 00:01:27,330 --> 00:01:39,720 [Eric] Uh, the 1776, um, the John Adams se- uh, John Adams and, like, the 1776 stuff, which they made a, a s- uh, movie, you know, uh, series out of. 00:01:39,720 --> 00:01:39,940 [John] Okay. 00:01:39,940 --> 00:01:40,899 [Eric] People are very familiar with that- 00:01:40,900 --> 00:01:40,910 [John] Mm-hmm 00:01:40,910 --> 00:01:51,080 [Eric] ... it was sort of based on his work. But he's written biographies on, um, Harry Truman. Uh, he wrote one on the Panama Canal that's amazing. Uh, he wrote one on the Brooklyn Bridge. 00:01:51,080 --> 00:01:51,460 [John] Nice. 00:01:51,460 --> 00:01:59,440 [Eric] Which is a... The Brooklyn Bridge is incredible. Um, and he's just... Yeah, he's, uh, you know, a lot of really big, thick books. 00:01:59,440 --> 00:01:59,480 [John] Right. 00:01:59,480 --> 00:02:04,880 [Eric] He wrote a shorter one on the Wright brothers, which is a great starting point. Um, I'm obviously a huge fanboy. 00:02:04,890 --> 00:02:05,240 [John] Nice. 00:02:05,240 --> 00:02:06,810 [Eric] I got to see him speak in person live- 00:02:06,810 --> 00:02:07,070 [John] Oh, cool 00:02:07,070 --> 00:02:08,100 [Eric] ... before he died, which was- 00:02:08,100 --> 00:02:08,530 [John] That's amazing 00:02:08,530 --> 00:02:23,080 [Eric] ... which was amazing. He was very interesting. He came out on stage, and he didn't move. He didn't walk on the stage. He just stood in the same place with his feet planted, no notes, and talked for, like, an hour and a half, and it was unbelievably engaging. 00:02:23,080 --> 00:02:24,400 [John] That's really cool. 00:02:24,400 --> 00:02:25,220 [Eric] It, it was, it was unbelievable. 00:02:25,220 --> 00:02:25,480 [John] Yeah. 00:02:25,480 --> 00:02:27,519 [Eric] It was counterintuitive. 00:02:27,520 --> 00:02:27,840 [John] Yeah. Right. 00:02:27,840 --> 00:02:28,040 [Eric] Uh, 00:02:29,420 --> 00:02:34,380 [Eric] anyways, in one of the prefaces to his, to one of his books, 00:02:36,580 --> 00:02:47,480 [Eric] he, he makes almost a side comment, and the, the, the comment is, "Writing is just a great deal of hard thinking." 00:02:49,240 --> 00:02:49,420 [Eric] You know? 00:02:49,420 --> 00:02:49,960 [John] Okay. 00:02:49,960 --> 00:02:54,970 [Eric] And, uh, what he's getting at, which is very true, 00:02:56,020 --> 00:03:15,190 [Eric] uh, and I think this would... People would also say that this is true of, like, writing code or, you know, doing a data analysis. But once you have something locked down in your mind, putting words on paper or writing the SQL query isn't... Uh, that's not the hard part, right? 00:03:15,190 --> 00:03:15,200 [John] Right. 00:03:15,200 --> 00:03:17,780 [Eric] And, and in the best state, like 00:03:18,820 --> 00:03:38,840 [Eric] what everyone loves, is when that feels almost effortless, right? It's just, "Okay, I, I have this idea of what I need to do or this concept or whatever really locked down in my mind," and so actually sort of making that material in whatever form that is, you know, for me that's words, for you it's data, whatever, it feels almost effortless- 00:03:38,840 --> 00:03:38,850 [John] Right 00:03:38,850 --> 00:03:49,579 [Eric] ... because you're just sort of translating these, um, well-critiqued arguments in your mind into some sort of, like, physical form, right? 00:03:49,580 --> 00:03:50,720 [John] Yeah. 00:03:50,720 --> 00:03:50,940 [Eric] Um, 00:03:52,180 --> 00:03:52,540 [Eric] and 00:03:54,220 --> 00:04:07,400 [Eric] we talk about that a lot on my team when it comes to writing because we review a lot of content from people around the company, and the stuff that's the most difficult is the stuff where it's like, "I don't think that you really thought about this well," right? 00:04:07,400 --> 00:04:07,790 [John] Right. Right. 00:04:07,790 --> 00:04:10,090 [Eric] And so if you start with a bad argument, 00:04:11,340 --> 00:04:15,470 [Eric] everything downstream from that is problematic, right? 00:04:15,470 --> 00:04:15,500 [John] Right. 00:04:15,500 --> 00:04:21,100 [Eric] And so again, with the example of writing, like the examples you use don't make as much sense, right? 00:04:22,680 --> 00:04:27,790 [Eric] Um, the narrative is harder to follow, right? There, there are just all these sorts of problems that flow downstream from it, right? 00:04:27,790 --> 00:04:28,860 [John] Right. 00:04:28,860 --> 00:04:41,370 [Eric] But I was also thinking about data and even business decisions, right, where if you have bad thinking up front, the downstream effects can be pretty significant. 00:04:41,370 --> 00:04:41,400 [John] Right. 00:04:41,400 --> 00:04:47,740 [Eric] So you deal with this a lot in data, right? Um, so tell us about that, and then- 00:04:47,740 --> 00:04:47,750 [John] Yeah 00:04:47,750 --> 00:04:51,120 [Eric] ... you had a really good example from Amazon, actually. 00:04:52,410 --> 00:04:54,940 [John] Yeah. So, so I think there's two things. There's faulty assumptions- 00:04:54,940 --> 00:04:55,000 [Eric] Mm-hmm 00:04:55,000 --> 00:04:56,600 [John] ... and that's just the facts level. 00:04:56,600 --> 00:04:56,900 [Eric] Mm-hmm. 00:04:56,900 --> 00:04:58,980 [John] Like we assume this to be true, it's not true. 00:04:58,980 --> 00:04:59,260 [Eric] Yep. 00:04:59,260 --> 00:05:06,240 [John] Then the second is the thinking, is we're, we're, we, we might have the assumptions or facts, like, roughly correct- 00:05:06,240 --> 00:05:06,660 [Eric] Mm-hmm 00:05:06,660 --> 00:05:08,220 [John] ... but we're applying them poorly. 00:05:08,220 --> 00:05:08,460 [Eric] Mm-hmm. 00:05:08,460 --> 00:05:09,979 [John] I think I would differentiate between those two things. 00:05:09,980 --> 00:05:10,740 [Eric] Mm. 00:05:10,740 --> 00:05:13,700 [John] Um, and I think an interesting thing about AI, 00:05:14,840 --> 00:05:20,280 [John] the, the people... Most people have had the everyday e- experience of the AI agreeing with you, right? 00:05:20,280 --> 00:05:21,560 [Eric] You're absolutely right. 00:05:22,660 --> 00:05:23,500 [John] Exactly. 00:05:24,520 --> 00:05:33,680 [John] Um, and we've seen some, like, really bad effects from this, like people using it as kind of a, a personal counselor, coach, and then it really encouraging them to do, like, something terrible. Um, for example- 00:05:33,680 --> 00:05:35,630 [Eric] Sure, yeah. In extreme cases, cases, it's been- 00:05:35,630 --> 00:05:35,680 [John] Like- 00:05:35,680 --> 00:05:36,810 [Eric] It's been very bad, right? 00:05:36,810 --> 00:05:36,820 [John] Yeah. 00:05:36,820 --> 00:05:38,560 [Eric] People harming themselves or other people. 00:05:38,560 --> 00:05:49,240 [John] Yeah. And, and that's... Uh, yeah, and that's terrible. Um, it also happens, like you, like you, like you were just saying, like it leads you down a path that you kind of started it down and it'll just keep going. 00:05:49,240 --> 00:05:49,780 [Eric] Yep. 00:05:49,780 --> 00:06:03,540 [John] And it does it in code, which I think is more subtle. Um, and I'll, I'll share an example here in a minute, but I think it is the thing... It is a very human thing that it's mimicking, and it's the desire for, like, internal consistency. 00:06:04,072 --> 00:06:04,652 [Eric] Mm. 00:06:04,652 --> 00:06:10,152 [John] Like, you would not have good AI experiences, like sycophancy aside- 00:06:10,152 --> 00:06:10,252 [Eric] Yep 00:06:10,252 --> 00:06:14,272 [John] ... if it wasn't some internal consistency to the conversation. 00:06:14,272 --> 00:06:14,472 [Eric] Mm. 00:06:14,472 --> 00:06:17,352 [John] Like, if it was real scattered, like people wouldn't like using it. 00:06:17,352 --> 00:06:18,632 [Eric] Yes, for sure. 00:06:18,632 --> 00:06:18,692 [John] Um, but- 00:06:18,692 --> 00:06:32,062 [Eric] For sure. And so you sort of default to being agreeable because constantly, it- it's, it's exhausting to think really well and to get all of your arguments in order, right? 00:06:32,062 --> 00:06:32,072 [John] Right. 00:06:32,072 --> 00:06:36,221 [Eric] When we talk about this, like the map is not the territory, we distill reality down into- 00:06:36,221 --> 00:06:36,332 [John] Mm-hmm 00:06:36,332 --> 00:06:39,332 [Eric] ... things that are easier for us to understand and grapple with. 00:06:39,332 --> 00:06:39,352 [John] Right. 00:06:39,352 --> 00:06:44,572 [Eric] And so if AI were constantly on every turn challenge- you know, challenging your assumptions- 00:06:44,612 --> 00:06:44,702 [John] Yeah, exactly 00:06:44,702 --> 00:06:46,312 [Eric] ... to make sure they were 100% correct- 00:06:46,312 --> 00:06:46,672 [John] Right 00:06:46,672 --> 00:06:48,192 [Eric] ... it would be exhausting- 00:06:48,192 --> 00:06:48,512 [John] Right 00:06:48,512 --> 00:06:51,032 [Eric] ... you know, as that form factor over and over again. 00:06:51,032 --> 00:07:00,092 [John] Yeah. So that initial starting point, whatever you're doing, writing code or, or whatever, however you start the steering, it will naturally go that direction- 00:07:00,092 --> 00:07:00,162 [Eric] Yeah 00:07:00,162 --> 00:07:01,332 [John] ... I think is part of what we're saying here. 00:07:01,332 --> 00:07:11,732 [Eric] Mm-hmm. Yep, absolutely. So let's... I wanna dig into the AI stuff, but these are all... You know, one thing that we consistently see with AI is that none of these problems are new, right? 00:07:11,732 --> 00:07:12,392 [John] Yeah. 00:07:12,392 --> 00:07:16,652 [Eric] And so what is... You me- we were talking about, before the show- 00:07:16,652 --> 00:07:16,662 [John] Right 00:07:16,662 --> 00:07:20,272 [Eric] ... we were talking about, okay, what are instances where, like a bad decision or bad thinking- 00:07:20,272 --> 00:07:20,442 [John] Right 00:07:20,442 --> 00:07:21,242 [Eric] ... or bad assumptions 00:07:22,552 --> 00:07:25,082 [Eric] caused a really big downstream- 00:07:25,082 --> 00:07:25,082 [John] Right 00:07:25,082 --> 00:07:26,052 [Eric] ... problem? 00:07:26,052 --> 00:07:27,222 [John] Right. This is one of my favorite ones. 00:07:27,222 --> 00:07:27,232 [Eric] Okay. 00:07:27,232 --> 00:07:31,012 [John] So it's Amazon. Yeah, the competitors of Amazon. Um- 00:07:31,012 --> 00:07:32,172 [Eric] Okay, the competitors of Amazon. Yeah. 00:07:32,172 --> 00:07:34,192 [John] So if you remember, Amazon was a book, an online bookstore. 00:07:34,192 --> 00:07:34,632 [Eric] Mm-hmm. 00:07:34,632 --> 00:07:39,012 [John] That's what they did. Um, and, and they always had a bigger vision than that. Like, if you go back or you listen to- 00:07:39,012 --> 00:07:39,772 [Eric] Yep 00:07:39,772 --> 00:07:42,552 [John] ... read books or listen to interviews with Jeff Bezos, it was always bigger than that. 00:07:42,552 --> 00:07:42,692 [Eric] Mm-hmm. 00:07:42,692 --> 00:07:44,092 [John] But it was a online bookstore. 00:07:44,092 --> 00:07:44,632 [Eric] Yep. 00:07:44,632 --> 00:08:04,172 [John] And then, and then they started getting into product. So when they start getting into product outside of books, the fundamental underlying belief that apparently all of their competitors had is really interesting. And, and I... And I'll, I'll state it here in a minute, but the interesting part before I get, get into it is, um, one, did they know they had this belief? 00:08:04,172 --> 00:08:05,052 [Eric] Mm. 00:08:05,052 --> 00:08:09,772 [John] And then two, if they did, why, why did it take so long for it to get invalidated? 00:08:09,772 --> 00:08:10,272 [Eric] Mm. 00:08:10,272 --> 00:08:10,452 [John] Um, 00:08:11,592 --> 00:08:21,252 [John] and I don't know the answer to that, but, but the underlying w- belief for shopping, say, things other than books especially, say, "Oh, well, people need to see it in person." 00:08:21,252 --> 00:08:21,502 [Eric] Right. 00:08:21,502 --> 00:08:21,522 [John] Um- 00:08:21,522 --> 00:08:23,742 [Eric] "I need to touch it, feel it, try it on," whatever. 00:08:23,742 --> 00:08:24,552 [John] Yeah, try it on. Yeah. 00:08:24,552 --> 00:08:24,852 [Eric] Yep. 00:08:24,852 --> 00:08:28,672 [John] Especially like, like Amazon starts doing clothes. People are like, "Ah, that's never gonna work." 00:08:28,672 --> 00:08:28,932 [Eric] Mm-hmm. 00:08:28,932 --> 00:08:31,832 [John] Or, or even, um, yeah, oth- other things that- 00:08:31,832 --> 00:08:33,572 [Eric] I think about headphones, right? Like- 00:08:33,572 --> 00:08:34,392 [John] Yeah, yeah 00:08:34,392 --> 00:08:34,532 [Eric] ... like- 00:08:34,532 --> 00:08:35,082 [John] Like what if people- 00:08:35,082 --> 00:08:35,802 [Eric] ... people are, why would someone- 00:08:35,802 --> 00:08:35,802 [John] Yeah 00:08:35,802 --> 00:08:37,072 [Eric] ... buy it if they can't try it on, right? 00:08:37,072 --> 00:08:57,652 [John] Yeah, exactly. Um, and the... And, and, and Amazon understood that, too, but they did two, like crazy cool strategic things. One is they figured out that, um, there's a proxy for this. Someone else tried it out, tried it on, whatever. Left a review- 00:08:57,652 --> 00:08:58,312 [Eric] Mm 00:08:58,312 --> 00:09:01,901 [John] ... said that they loved it, and that was enough for people to buy. 00:09:01,901 --> 00:09:02,292 [Eric] Yes. 00:09:02,292 --> 00:09:09,602 [John] Number two, so it was a two-part thing. That was, that's really... The reviews were, like absolutely critical, and they had to be trustworthy reviews. They had to be not fake. 00:09:09,602 --> 00:09:09,612 [Eric] Yep. 00:09:09,612 --> 00:09:11,202 [John] But I'll, I'll... There's a lot that went into that. 00:09:11,202 --> 00:09:11,792 [Eric] Mm-hmm. 00:09:11,792 --> 00:09:15,932 [John] It's like, all right, all right, cool. This person loves it. They, you know, whatever. 00:09:15,932 --> 00:09:15,952 [Eric] Right. 00:09:15,952 --> 00:09:17,552 [John] Great. But the second part 00:09:18,672 --> 00:09:23,632 [John] is, um, that might not be enough. The second part was, oh, you can return it, and it's- 00:09:23,632 --> 00:09:23,662 [Eric] Yeah 00:09:23,662 --> 00:09:26,732 [John] ... easy and it's free, and like the, the most generous- 00:09:26,732 --> 00:09:27,002 [Eric] Lo- yeah, low consequence 00:09:27,002 --> 00:09:28,772 [John] ... return policies, especially back in the day. 00:09:28,772 --> 00:09:29,312 [Eric] Totally. 00:09:29,312 --> 00:09:30,662 [John] The most generous return policies. 00:09:30,662 --> 00:09:30,672 [Eric] Yep. 00:09:30,672 --> 00:09:33,092 [John] People had never even heard of it being that generous. 00:09:33,092 --> 00:09:33,322 [Eric] Yep. 00:09:33,322 --> 00:09:33,352 [John] And- 00:09:33,352 --> 00:09:35,272 [Eric] I mean, I think Zappos was, like the other- 00:09:35,272 --> 00:09:35,652 [John] Yeah 00:09:35,652 --> 00:09:36,452 [Eric] ... the other one that was like- 00:09:36,452 --> 00:09:36,592 [John] Right 00:09:36,592 --> 00:09:37,312 [Eric] ... crazy, right? 00:09:37,312 --> 00:09:37,652 [John] Right. 00:09:37,652 --> 00:09:37,832 [Eric] Um- 00:09:37,832 --> 00:09:40,932 [John] Right. So that with the proxy of you didn't- 00:09:40,932 --> 00:09:40,982 [Eric] Mm-hmm 00:09:40,982 --> 00:09:46,732 [John] ... try it in person, but somebody that you, um, trusted tried it in person. 00:09:46,732 --> 00:09:47,372 [Eric] Yep. 00:09:47,372 --> 00:09:52,472 [John] Uh, trusted by a proxy in that you believed it was a real person. You believed there was no incentive to, like write a fake review, whatever. 00:09:52,472 --> 00:09:53,292 [Eric] Yeah, yeah. Yep. 00:09:53,292 --> 00:09:53,772 [John] Um- 00:09:53,772 --> 00:09:54,312 [Eric] For sure. 00:09:54,312 --> 00:09:57,472 [John] And, and that bet obviously went really well for them. 00:09:57,472 --> 00:10:11,812 [Eric] Totally. Totally. I, I mean, this, this is very true. I bought something the other day. It was a z- uh, one of, um, those little, uh, extensions that you put on a zipper to make it easier to pull a tab on like- 00:10:11,812 --> 00:10:11,842 [John] Yeah 00:10:11,842 --> 00:10:12,492 [Eric] ... a little cooler- 00:10:12,492 --> 00:10:12,612 [John] Okay 00:10:12,612 --> 00:10:12,992 [Eric] ... that we have. 00:10:12,992 --> 00:10:13,212 [John] Yeah. 00:10:13,212 --> 00:10:14,362 [Eric] Like it broke, and I was like- 00:10:14,362 --> 00:10:14,362 [John] Yeah 00:10:14,362 --> 00:10:15,192 [Eric] ... "Oh, man, like- 00:10:15,192 --> 00:10:15,202 [John] Yeah 00:10:15,202 --> 00:10:16,852 [Eric] ... where do you get this?" So I went to Amazon. 00:10:16,852 --> 00:10:17,432 [John] Right. 00:10:17,432 --> 00:10:19,592 [Eric] You know, and of course there's a million options. 00:10:19,592 --> 00:10:19,732 [John] Right. 00:10:19,732 --> 00:10:22,881 [Eric] And like the images are AI generated, and you're like, "Okay, which-" 00:10:22,881 --> 00:10:22,881 [John] Right 00:10:22,881 --> 00:10:25,652 [Eric] ... you know, $6 versus $7. Is there- 00:10:25,652 --> 00:10:25,681 [John] Right 00:10:25,681 --> 00:10:26,472 [Eric] ... a better choice here? 00:10:26,472 --> 00:10:27,532 [John] Right. 00:10:27,532 --> 00:10:33,532 [Eric] But I just picked the one that had a high enough number of people who took a picture of it on their own zipper. 00:10:33,532 --> 00:10:33,912 [John] Uh-huh. Sure. 00:10:33,912 --> 00:10:35,271 [Eric] We're like, "This is what it looks like," and I'm like- 00:10:35,272 --> 00:10:35,282 [John] Right 00:10:35,282 --> 00:10:36,092 [Eric] ... "Oh, that looks fine." 00:10:36,092 --> 00:10:36,252 [John] Yeah, yeah, yeah. 00:10:36,252 --> 00:10:38,172 [Eric] You know? Like someone else did it. Uh- 00:10:39,272 --> 00:10:39,321 [John] Totally. 00:10:39,321 --> 00:10:40,432 [Eric] So okay. 00:10:41,452 --> 00:10:48,992 [Eric] What were, what were the negative downstream effects of not believing that that's enough for Amazon's competitors? 00:10:50,252 --> 00:10:56,192 [John] Yeah. I mean, I mean, I think they were huge. I mean, Walmart struggled like forever. 00:10:56,192 --> 00:10:56,412 [Eric] Yeah. 00:10:56,412 --> 00:11:03,492 [John] Like I think, I think in the last like couple years they kinda got it now. Like they, they they've kinda caught up. Um, you remember Jet? 00:11:03,492 --> 00:11:03,672 [Eric] Yeah. 00:11:03,672 --> 00:11:04,302 [John] Remember jet.com? 00:11:04,302 --> 00:11:05,352 [Eric] Oh, yeah. Huge. 00:11:05,352 --> 00:11:09,312 [John] I mean, Jet, Walmart bought them, and that still wasn't quite enough, um- 00:11:09,312 --> 00:11:09,452 [Eric] Yeah 00:11:09,452 --> 00:11:10,162 [John] ... for whatever reason. 00:11:10,162 --> 00:11:10,192 [Eric] Totally. 00:11:10,192 --> 00:11:10,982 [John] I don't know the details- 00:11:10,982 --> 00:11:10,982 [Eric] Totally 00:11:10,982 --> 00:11:19,902 [John] ... of what happened there. But I think the, the, they missed out on this very new, um, uh, channel- 00:11:19,902 --> 00:11:20,332 [Eric] Mm-hmm 00:11:20,332 --> 00:11:28,132 [John] ... that, again, was insignificant for a while. And, and if Walmart had asked all their competitors like, "How many, uh, you know, how many online sales do you do?" You know- 00:11:28,132 --> 00:11:28,332 [Eric] Yep 00:11:28,332 --> 00:11:32,082 [John] ... Kmart, Kmart's like, "Ah, like, not many." It's like, "All right, we don't need it." You know? 00:11:32,082 --> 00:11:32,092 [Eric] Yeah. Yep. 00:11:32,092 --> 00:11:33,212 [John] Like you can imagine, like- 00:11:33,212 --> 00:11:33,412 [Eric] Absolutely 00:11:33,412 --> 00:11:36,092 [John] ... in whatever circle at the time, it was like, "All right, well, it's fine." 00:11:36,092 --> 00:11:36,692 [Eric] Yeah, yeah. 00:11:36,692 --> 00:11:37,832 [John] Um, and- 00:11:38,872 --> 00:11:38,891 [Eric] I- 00:11:38,892 --> 00:11:39,332 [John] Yeah. 00:11:39,332 --> 00:11:45,332 [Eric] No, I, I was gonna say, I think that the, um, if we go back to first principles, there's sort of, 00:11:46,672 --> 00:11:52,992 [Eric] I, I would say two things that jump out to me here. One is that am I... 00:11:54,632 --> 00:12:02,092 [Eric] Th- this is very interesting actually, now that I think about it. It, I think what Jeff B- what Jeff Bezos and, and the Amazon team 00:12:03,512 --> 00:12:07,492 [Eric] thought really well about was human nature. 00:12:07,492 --> 00:12:08,052 [John] Yeah. 00:12:08,052 --> 00:12:10,592 [Eric] Right? Um- 00:12:10,592 --> 00:12:15,121 [Eric] Because what you're describing with, like, it's enough for someone, the review is enough, 00:12:16,192 --> 00:12:18,002 [Eric] is a tale as old as time. 00:12:18,002 --> 00:12:18,052 [John] Yeah. 00:12:18,052 --> 00:12:19,592 [Eric] I actually remember, um, 00:12:21,752 --> 00:12:25,912 [Eric] I bought Apple, uh, AirPods Max, which are the over-ears. 00:12:25,912 --> 00:12:26,052 [John] Yeah. 00:12:26,052 --> 00:12:28,542 [Eric] And I debated the decision for a very long time. 00:12:28,542 --> 00:12:28,552 [John] Right. 00:12:28,552 --> 00:12:29,702 [Eric] I mean, I think over a year. 00:12:29,702 --> 00:12:29,702 [John] Right. 00:12:29,702 --> 00:12:31,732 [Eric] 'Cause they're super expensive, right? 00:12:31,732 --> 00:12:31,882 [John] They are. Still are. Yeah. 00:12:31,882 --> 00:12:33,531 [Eric] Um, they still are very expensive. 00:12:34,652 --> 00:12:37,552 [Eric] And, you know, but I travel a good bit for work- 00:12:37,552 --> 00:12:37,572 [John] Mm-hmm 00:12:37,572 --> 00:12:41,802 [Eric] ... and, you know, I, I, you know, and I really appreciate high-quality audio when- 00:12:41,802 --> 00:12:41,832 [John] Right 00:12:41,832 --> 00:12:46,012 [Eric] ... I'm listening to music. And I had a pair of headphones that I had for a very long time. 00:12:46,012 --> 00:12:46,132 [John] Right. 00:12:46,132 --> 00:12:52,352 [Eric] You know, and they were getting, you know, sort of... I think one of the things, like, broke. They were still- 00:12:52,352 --> 00:12:52,972 [John] Yeah 00:12:52,972 --> 00:12:53,012 [Eric] ... usable. 00:12:53,012 --> 00:12:53,912 [John] Right. 00:12:53,912 --> 00:12:54,341 [Eric] Anyways, 00:12:55,432 --> 00:12:57,952 [Eric] I had a friend who had AirPods Max- 00:12:57,952 --> 00:12:58,112 [John] Mm-hmm 00:12:58,112 --> 00:13:02,922 [Eric] ... you know, over-ear headphones. And I said... He was wearing them on a call or something. 00:13:02,922 --> 00:13:03,012 [John] Okay. Yeah. 00:13:03,012 --> 00:13:08,702 [Eric] And I was like, "I am... I wanna buy them, but, like, I, it's, is it?" 00:13:08,702 --> 00:13:08,702 [John] Yeah. 00:13:08,702 --> 00:13:09,272 [Eric] "I just-" 00:13:09,272 --> 00:13:09,702 [John] So, yeah. 00:13:09,702 --> 00:13:11,312 [Eric] "It's so expensive. Like, I don't know." 00:13:12,372 --> 00:13:23,832 [Eric] And he just said, "I..." He's It was kind of funny. He's like, "I debated for, like, a year to buy these." He's like, "Someone convinced me that it was worth it." And he's like, "They're just worth it." He's like- 00:13:23,832 --> 00:13:23,842 [John] Right 00:13:23,842 --> 00:13:26,552 [Eric] ... "They are that good. The audio quality's good." 00:13:26,552 --> 00:13:26,702 [John] Right. 00:13:26,702 --> 00:13:28,931 [Eric] "The long-term comfort, you can wear them on a-" 00:13:28,932 --> 00:13:28,992 [John] Right 00:13:28,992 --> 00:13:30,152 [Eric] ... you know, six-hour flight." 00:13:30,152 --> 00:13:30,612 [John] Right. 00:13:30,612 --> 00:13:35,232 [Eric] He's like, "Everything." He's like, "It really is, like..." He's like, "It's one of those products that just- 00:13:35,232 --> 00:13:35,242 [John] Right 00:13:35,242 --> 00:13:36,872 [Eric] ... it's so expensive, but then when- 00:13:36,872 --> 00:13:36,882 [John] Mm-hmm 00:13:36,882 --> 00:13:38,902 [Eric] ... you use it, you're like, 'Yeah, it's probably worth it.'" 00:13:38,902 --> 00:13:38,912 [John] Right. 00:13:38,912 --> 00:13:41,092 [Eric] And he's like, "And they're so sturdy." He's like, "I mean-" 00:13:41,092 --> 00:13:41,431 [John] Yeah. 00:13:41,432 --> 00:13:42,752 [Eric] And I've, I've had the same experience. 00:13:43,932 --> 00:13:51,192 [Eric] That is this, that is, you know, a very, very similar thing to, like, mul- reading multiple reviews- 00:13:51,192 --> 00:13:51,712 [John] Mm-hmm 00:13:51,712 --> 00:13:57,182 [Eric] ... and getting validation from another person. So I think the Amazon team thought really well about human nature. 00:13:57,182 --> 00:13:57,212 [John] Right. 00:13:57,212 --> 00:13:58,792 [Eric] We know that this is enough, right? 00:13:58,792 --> 00:13:58,802 [John] Right. 00:13:58,802 --> 00:14:00,772 [Eric] And maybe they didn't know how much of it was enough- 00:14:00,772 --> 00:14:00,902 [John] Yeah 00:14:00,902 --> 00:14:03,042 [Eric] ... or what the thresholds were on- 00:14:03,042 --> 00:14:03,042 [John] Right 00:14:03,042 --> 00:14:04,892 [Eric] ... the price that you would pay for a certain product- 00:14:04,892 --> 00:14:04,992 [John] Yeah 00:14:04,992 --> 00:14:05,712 [Eric] ... or whatever, right? 00:14:05,712 --> 00:14:06,392 [John] Right, right. 00:14:06,392 --> 00:14:11,192 [Eric] Um, and then on the flip side, I think the bad thinking for their competitors 00:14:12,252 --> 00:14:17,092 [Eric] was not questioning their assumptions, right? Like- 00:14:17,092 --> 00:14:22,072 [John] Yeah. Well, and then I assume they... And again, I assume they looked around to their perceived competitors- 00:14:22,072 --> 00:14:22,412 [Eric] To the... Yeah 00:14:22,412 --> 00:14:23,452 [John] ... like Kmart or- 00:14:23,452 --> 00:14:23,732 [Eric] Yeah 00:14:23,732 --> 00:14:27,192 [John] ... Kmart, Walmart or, or whoever, and they're like, "Eh, like, they're not really doing it either." 00:14:27,192 --> 00:14:27,262 [Eric] Right. 00:14:27,262 --> 00:14:27,672 [John] "We're fine." 00:14:27,672 --> 00:14:28,312 [Eric] Yeah, yeah. So 00:14:29,952 --> 00:14:32,712 [Eric] one, first principles of, like, understanding human nature- 00:14:32,712 --> 00:14:32,782 [John] Right 00:14:32,782 --> 00:14:33,702 [Eric] ... and translating that. 00:14:33,702 --> 00:14:33,712 [John] Right. 00:14:33,712 --> 00:14:39,972 [Eric] And then two, like, the failure to question your assumptions and take a lot of things for granted based on what you can see right in front of you, not- 00:14:39,972 --> 00:14:40,132 [John] Right 00:14:40,132 --> 00:14:44,471 [Eric] ... you know, necessarily looking to the future. So AI 00:14:45,512 --> 00:14:45,832 [Eric] is, 00:14:47,592 --> 00:14:49,292 [Eric] is problematic here, I think. 00:14:49,292 --> 00:14:49,772 [John] Right. 00:14:49,772 --> 00:14:52,092 [Eric] Um, not I think. I, I know that it can be- 00:14:52,092 --> 00:14:52,142 [John] Right 00:14:52,142 --> 00:14:54,052 [Eric] ... very problematic, because 00:14:55,592 --> 00:14:59,952 [Eric] if you bring bad thinking to the table, 00:15:01,172 --> 00:15:02,012 [Eric] right? So 00:15:03,272 --> 00:15:27,072 [Eric] let's look at both of those examples that we just talked about, right? So a bad assumption about human nature, right? People need to... And if we play this out, actually, it's kind of funny. I think a lot of our listeners can imagine what a conversation would be, right? So I open up GPT or Claude, and I start saying, you know, "I think people really need to, like... to really be convicted about a purchase decision." 00:15:27,072 --> 00:15:27,272 [John] Mm-hmm. 00:15:27,272 --> 00:15:30,852 [Eric] Like, it's really powerful for them to have a physical experience with the product, right? 00:15:30,852 --> 00:15:31,492 [John] Yeah. Yeah, exactly. 00:15:31,492 --> 00:15:37,852 [Eric] Is AI going to say, "Actually, you need to step back and think about-" 00:15:37,912 --> 00:15:37,922 [John] Right 00:15:37,922 --> 00:15:38,872 [Eric] "... you know, what is the threshold-" 00:15:38,872 --> 00:15:38,882 [John] Right 00:15:38,882 --> 00:15:40,032 [Eric] "... of validation that someone needs?" 00:15:40,032 --> 00:15:40,251 [John] Right. 00:15:40,252 --> 00:15:43,312 [Eric] No. It's going to say, "Absolutely," like- 00:15:43,312 --> 00:15:43,602 [John] Right 00:15:43,602 --> 00:15:52,572 [Eric] ... and maybe even pull a statistic of, like, if someone can try something on, they're 10 times more likely to purchase it, you know, whatever. Which is like, okay, there may be situations where that's true- 00:15:52,572 --> 00:15:52,682 [John] Right 00:15:52,682 --> 00:15:58,592 [Eric] ... but in aggregate across, like, Amazon scale e-commerce, you know, it's not gonna ask you... It's not gonna question that. 00:15:58,592 --> 00:16:14,912 [John] Right. Well, and same thing with, um, because we just talked about competitors with, you know, Walmart, Kmart. And Walmart's like, "All right, fine, we'll do a s- survey. We'll survey all of our customers." And, and, and then, you know, they have these questions in there like, "Would you rather shop online or in person?" Like, you know, things like that. 00:16:14,912 --> 00:16:15,332 [Eric] Totally. 00:16:15,332 --> 00:16:18,232 [John] That from a survey construction, like, it's a bad survey. 00:16:18,232 --> 00:16:18,392 [Eric] Yep. 00:16:18,392 --> 00:16:21,892 [John] But for them, it's like, "Look, all of our customers said they'd rather shop in person." 00:16:21,892 --> 00:16:36,212 [Eric] Yep. Yeah, totally. It's... And so this, uh, I'm, I'll be very interested to see how this plays out, uh, because I think it will probably be more problematic than we think it is. 00:16:36,212 --> 00:16:36,952 [John] Right. 00:16:36,952 --> 00:16:37,272 [Eric] Um, 00:16:39,152 --> 00:16:43,772 [Eric] because it's a lot of really hard work, and I think there are opposing forces here. 00:16:43,772 --> 00:16:43,992 [John] Right. 00:16:43,992 --> 00:16:48,632 [Eric] It's a lot of hard work to get the thinking correct upfront. 00:16:48,632 --> 00:16:49,732 [John] Right. 00:16:49,732 --> 00:16:59,902 [Eric] And because AI is so high velocity and, like, low barrier to, like, produce something or, like, f- 00:16:59,902 --> 00:16:59,902 [John] Right 00:16:59,902 --> 00:17:10,922 [Eric] ... or, or feel like you're making progress, it doesn't reward... In the actual experience of using AI, it does not reward critical thinking, 00:17:11,972 --> 00:17:16,092 [Eric] right? The, you get the same path whether or not you bring that to the table, right? 00:17:16,092 --> 00:17:16,552 [John] Correct. Right. 00:17:16,552 --> 00:17:16,642 [Eric] Um, 00:17:17,872 --> 00:17:19,572 [Eric] and I'll just give one example. 00:17:20,892 --> 00:17:33,272 [Eric] Uh, I'll just give one personal example. Uh, you know, we have a content agent that we've built. It's really cool and super helpful and does all these awesome things, and using it almost every day. 00:17:34,292 --> 00:17:41,192 [Eric] And the thing that I've realized is it's most powerful when I come with a very well-thought-out thesis. 00:17:41,192 --> 00:17:41,572 [John] Sure. 00:17:41,572 --> 00:17:49,032 [Eric] Right? And things move much more quickly, and in fact, like, the cost of how many tokens I burn for, like, a project- 00:17:49,032 --> 00:17:49,392 [John] Mm-hmm 00:17:49,392 --> 00:17:52,282 [Eric] ... goes down dramatically. Because, 00:17:53,292 --> 00:17:56,472 [Eric] because of the sycophantic nature of AI, 00:17:57,552 --> 00:18:02,532 [Eric] using it to get to the thesis is actually time-consuming and cumbersome. 00:18:02,532 --> 00:18:03,092 [John] Right. 00:18:03,092 --> 00:18:11,708 [Eric] I think you can actually do that well. I think it's possible to do that well, but it's a very specific skill set. 00:18:11,708 --> 00:18:15,478 [Eric] ... to leverage AI to, like, push back on you and- 00:18:15,478 --> 00:18:15,478 [John] Yeah 00:18:15,478 --> 00:18:17,308 [Eric] ... go back and forth on sort of- 00:18:17,308 --> 00:18:17,318 [John] Right 00:18:17,318 --> 00:18:26,418 [Eric] ... getting to a thesis. Not exploring an idea or even brainstorming, right? I'm talking about sort of like building a concrete argument. Again, you can get it to do that, but that's not the natural path. You have to- 00:18:26,418 --> 00:18:26,418 [John] Right 00:18:26,418 --> 00:18:32,908 [Eric] ... proactively sort of steer it in that direction. You had an experience just this week, 00:18:34,008 --> 00:18:35,788 [Eric] uh, with someone on your team- 00:18:35,788 --> 00:18:35,858 [John] Right 00:18:35,858 --> 00:18:41,187 [Eric] ... that's sort of a prime example of this thinking and then agreeable AI- 00:18:41,188 --> 00:18:41,308 [John] Yeah 00:18:41,308 --> 00:18:42,728 [Eric] ... leading you down the path. 00:18:42,728 --> 00:18:46,948 [John] Right. So actually, I actually had two. I haven't told you about one of them. But I'll start- 00:18:46,948 --> 00:18:46,998 [Eric] Oh, great 00:18:46,998 --> 00:18:47,828 [John] ... with the one we talked about. 00:18:47,828 --> 00:18:48,148 [Eric] Okay. 00:18:48,148 --> 00:18:48,328 [John] Um, 00:18:49,388 --> 00:18:56,948 [John] and th- this was a funny one. Um, so one of my team members was working. Somebody reported a, a, a bug or an issue of like, "Hey, like these numbers look wrong." 00:18:56,948 --> 00:18:57,788 [Eric] Mm-hmm. 00:18:57,788 --> 00:19:08,488 [John] Um, stated with like, "And I think it's this." Like, "I think this is the problem." Um, some kind of like a matching algorithm. It's like a, "These numbers look wrong. I think our matching algorithm is off." 00:19:08,488 --> 00:19:09,048 [Eric] Yep. 00:19:09,048 --> 00:19:09,808 [John] Um- 00:19:09,808 --> 00:19:12,008 [Eric] Norm- just a normal like, "Okay, yeah, this looks off." 00:19:12,008 --> 00:19:13,888 [John] Yeah. And it was a reasonable assumption. It- 00:19:13,888 --> 00:19:14,248 [Eric] Mm-hmm 00:19:14,248 --> 00:19:17,208 [John] ... like, like I, I looked at it personally, yeah, probably so. 00:19:17,208 --> 00:19:17,687 [Eric] Mm-hmm. 00:19:17,688 --> 00:19:26,148 [John] Um, so team member like grabs all that context from the reported issue, puts that in AI, pulls up the code base, starts working, blah, blah, blah. 00:19:26,148 --> 00:19:26,338 [Eric] Mm-hmm. 00:19:26,338 --> 00:19:29,248 [John] Gets through to like adjusting that algorithm- 00:19:29,248 --> 00:19:29,688 [Eric] Mm-hmm 00:19:29,688 --> 00:19:30,858 [John] ... blah, blah, blah. And, you know, turns out 00:19:32,008 --> 00:19:34,668 [John] that didn't help, like didn't, didn't, didn't solve the problem, didn't help. 00:19:34,668 --> 00:19:35,648 [Eric] Hmm. 00:19:35,648 --> 00:19:42,908 [John] Um, and then I jumped in with him, was like, "All right. Let's look at this." I did, and I kinda did the same thing. "Yeah, we probably just missed something," like worked on the same thing. 00:19:42,908 --> 00:19:43,008 [Eric] Yep. 00:19:43,008 --> 00:19:45,438 [John] Like, "Let's look at the algorithm again," blah, blah, blah. 00:19:45,438 --> 00:19:45,488 [Eric] Mm-hmm. 00:19:45,488 --> 00:20:02,248 [John] Like still, like didn't get to the right answer. Then I went back, and I felt dumb. Like here's, like we did something wrong here. Like there's a problem stated with a proposed solution. We didn't investigate the problem outside of the proposed solution. 00:20:02,248 --> 00:20:02,648 [Eric] Hmm. 00:20:02,648 --> 00:20:07,368 [John] And AI further encouraged us to go down the, the, the matching algorithm- 00:20:07,368 --> 00:20:07,718 [Eric] Yep 00:20:07,718 --> 00:20:08,308 [John] ... route. 00:20:08,308 --> 00:20:08,448 [Eric] Yep. 00:20:08,448 --> 00:20:16,468 [John] But that wasn't the only route. The... And, and this is true before AI. Like when a problem's stated, you should always try to reproduce it. Always, every time. 00:20:16,468 --> 00:20:16,908 [Eric] Yes. 00:20:17,608 --> 00:20:17,808 [John] You know? 00:20:18,868 --> 00:20:21,648 [Eric] Especially if you're thinking technical, anything technical- 00:20:21,648 --> 00:20:21,687 [John] Yeah, yeah, yeah 00:20:21,687 --> 00:20:23,178 [Eric] ... right? Like I can't reproduce the issue. 00:20:23,178 --> 00:20:29,748 [John] And, and it was a data problem that you could visually see, so I didn't need to reproduce it because I didn't believe it was an issue. 00:20:29,748 --> 00:20:29,868 [Eric] Mm-hmm. 00:20:29,868 --> 00:20:33,708 [John] I believed it was an issue, so that's one thing. There's some bug, and like, eh, doesn't... Like, is it- 00:20:33,708 --> 00:20:33,768 [Eric] Mm 00:20:33,768 --> 00:20:34,378 [John] ... really a bug? 00:20:34,378 --> 00:20:34,408 [Eric] Mm. 00:20:34,408 --> 00:20:46,468 [John] Didn't need to validate that. Knew, knew there was something wrong. But there's still power in, in reproducing it, not because you wanna know if it's an issue, 'cause you know it is, but because of the steps that happened prior to that- 00:20:46,468 --> 00:20:47,268 [Eric] Yep 00:20:47,268 --> 00:20:52,928 [John] ... that you're building this context of like when we investigate this, we have all five of the things in context it takes- 00:20:52,928 --> 00:20:52,998 [Eric] Mm-hmm 00:20:52,998 --> 00:20:53,917 [John] ... to calculate this number- 00:20:53,917 --> 00:20:53,938 [Eric] Mm-hmm 00:20:53,938 --> 00:20:54,768 [John] ... for example. 00:20:54,768 --> 00:20:55,228 [Eric] Yep. 00:20:55,228 --> 00:20:59,048 [John] Um, so I thought that was really interesting, and I, I think it's gonna be a problem for everybody. 00:20:59,048 --> 00:20:59,218 [Eric] Yep. 00:20:59,218 --> 00:21:11,668 [John] 'Cause if, if the instruction is to the human or to the AI loop or what- however you do troubleshooting nowadays, um, that- that's gonna lead humans and AIs down the wrong rabbit hole every time- 00:21:11,668 --> 00:21:12,168 [Eric] Yep 00:21:12,168 --> 00:21:15,868 [John] ... if, if you are s- steering or stating the context wrong. 00:21:15,868 --> 00:21:16,328 [Eric] Yep. 00:21:16,328 --> 00:21:39,948 [John] Um, all right. So in the positive example, had this project which I still don't know what to think about this. Um, the... Okay, so if you're building something that new, pretty, pretty much in any technology, um, I'll speak about data, is you always want to leverage as much of existing available work as possible and then add your two cents. 00:21:39,948 --> 00:21:39,988 [Eric] Hmm. 00:21:39,988 --> 00:21:41,208 [John] That's always the best way to work. 00:21:41,208 --> 00:21:41,908 [Eric] Yep. 00:21:41,908 --> 00:21:50,308 [John] Um, so if you're in open source land, it's like, all right, let's do the research. Um, we wanna leverage like this library or this bit of code or whatever. 00:21:50,308 --> 00:21:50,448 [Eric] Yep. 00:21:50,448 --> 00:21:54,708 [John] And obviously you wanna use good libraries and people that have done things, solve things the right way. 00:21:54,708 --> 00:21:55,068 [Eric] Yep. 00:21:55,068 --> 00:21:59,348 [John] But, but especially in tech communities, people kinda know. Like- 00:21:59,348 --> 00:21:59,388 [Eric] Yeah 00:21:59,388 --> 00:22:02,108 [John] ... you can look at how many stars something has to get a sense for it. 00:22:02,108 --> 00:22:02,648 [Eric] Sure. 00:22:02,648 --> 00:22:06,287 [John] And then obviously inspect the code. But the, but that's, that's the right way to work- 00:22:06,288 --> 00:22:06,808 [Eric] Mm-hmm 00:22:06,808 --> 00:22:07,608 [John] ... and, and pretty established. 00:22:08,868 --> 00:22:18,248 [John] Interesting, I... So working on a project with a client trying... And that's what we were doing, like, all right, we wanna use this library or this component for like this piece of the tech, and use this- 00:22:18,248 --> 00:22:18,268 [Eric] Mm-hmm 00:22:18,268 --> 00:22:20,988 [John] ... and this, and we're gonna like stitch together like we normally would. 00:22:20,988 --> 00:22:21,368 [Eric] Mm-hmm. 00:22:21,368 --> 00:22:25,088 [John] Here's the edges that we know that we need to do ourselves 'cause it's a unique situation, et cetera. 00:22:25,088 --> 00:22:25,728 [Eric] Yep, yep. 00:22:25,728 --> 00:22:50,148 [John] So went down that road using AI around it. Like, uh, results were like, eh, like they're fine, but we're not very happy with them. We feel like there's like some extra code, and it's not quite behaving right. So one of the things we did, um, this week is w- A, we drastically strip- um, stripped down the use case, stripped out a bunch of like features and things we wanted to do, made it very specific, like we wanna move data from A to B. 00:22:50,148 --> 00:22:50,808 [Eric] Mm-hmm. 00:22:50,808 --> 00:22:55,208 [John] Um, and, um, stripped back, like, well, it doesn't have to do this, doesn't have to do this. 00:22:55,208 --> 00:22:55,768 [Eric] Yep. 00:22:55,768 --> 00:22:55,928 [John] Um, 00:22:57,068 --> 00:23:09,108 [John] and then stated all that very clearly and had it just work from moving data to A, from A to B and gave it some like, like use Python, like pretty basic scaffolding. 00:23:09,108 --> 00:23:10,008 [Eric] Yep. 00:23:10,008 --> 00:23:12,188 [John] Um, and it was awesome. 00:23:12,188 --> 00:23:12,688 [Eric] Hmm. 00:23:12,688 --> 00:23:18,288 [John] Like, it, it didn't lever... Like the right thing to do is to leverage, like, oh, well, this library is really good at this and this. Like- 00:23:18,288 --> 00:23:18,988 [Eric] Yep 00:23:18,988 --> 00:23:22,008 [John] ... but at least in this u- instance, it was awesome. 00:23:22,008 --> 00:23:22,268 [Eric] Yeah. 00:23:22,268 --> 00:23:23,448 [John] Like- 00:23:23,448 --> 00:23:23,708 [Eric] Yeah, it's- 00:23:23,708 --> 00:23:25,568 [John] And it's a different way to think. 00:23:25,568 --> 00:23:28,448 [Eric] It, it's a different way to... Well, I think 00:23:29,508 --> 00:23:31,848 [Eric] my big takeaway from the second story 00:23:33,028 --> 00:23:38,288 [Eric] is that... Well, I have a takeaway from the first story, but takeaway from the second story is that the 00:23:40,608 --> 00:23:47,828 [Eric] mileage varies a lot when there are areas of ambiguity in a plan and AI fills in the gaps. 00:23:47,828 --> 00:23:48,548 [John] Yep. 00:23:48,548 --> 00:23:49,048 [Eric] Right? 00:23:49,048 --> 00:23:49,508 [John] Right. 00:23:49,508 --> 00:23:54,578 [Eric] Be- and especially when you're talking about a complex system that's gonna use components from different open source projects- 00:23:54,578 --> 00:23:54,638 [John] Yeah, yeah 00:23:54,638 --> 00:23:59,708 [Eric] ... and other things like that. And so it's like you get to the end, and it doesn't feel quite right. It's like, okay, well, 00:24:01,088 --> 00:24:07,008 [Eric] in a complex system, generally they work really well if you're very, very opinionated about it. 00:24:07,008 --> 00:24:07,828 [John] Right. 00:24:07,828 --> 00:24:08,328 [Eric] And, 00:24:09,348 --> 00:24:13,514 [Eric] you know, that is not the natural disposition of AI. Right? 00:24:13,514 --> 00:24:13,544 [John] Right. 00:24:13,544 --> 00:24:15,664 [Eric] Is to handle ambiguity that way. It's gonna tend- 00:24:15,664 --> 00:24:15,674 [John] Right 00:24:15,674 --> 00:24:17,944 [Eric] ... towards sort of an amalgamation of, like- 00:24:17,944 --> 00:24:17,954 [John] Right 00:24:17,954 --> 00:24:19,734 [Eric] ... what it believes your intent to be- 00:24:19,734 --> 00:24:19,744 [John] Right 00:24:19,744 --> 00:24:21,264 [Eric] ... and then follow that path, right? 00:24:21,264 --> 00:24:28,484 [John] Right. W- which, to be fair, I did change two variables. One is I reduced and sharpened the goal. 00:24:28,484 --> 00:24:28,724 [Eric] Yep. 00:24:28,724 --> 00:24:29,754 [John] Which is super important. 00:24:29,754 --> 00:24:31,524 [Eric] Which is, which is called good thinking. 00:24:31,564 --> 00:24:37,384 [John] Yeah, yeah. Reduced and sharpened the goal and, and, and didn't truly do a, an AB test of like- 00:24:37,384 --> 00:24:37,394 [Eric] Mm-hmm 00:24:37,394 --> 00:24:38,944 [John] ... well, what if we went down the same path- 00:24:38,944 --> 00:24:39,114 [Eric] Mm-hmm 00:24:39,114 --> 00:24:39,914 [John] ... with the tech before it- 00:24:39,914 --> 00:24:39,914 [Eric] Mm-hmm 00:24:39,914 --> 00:24:41,384 [John] ... with a reduced, sharpened goal. Sure. 00:24:41,384 --> 00:24:41,404 [Eric] Yeah. 00:24:41,404 --> 00:24:54,044 [John] I'm sure I would've gotten better results. But, um, I mean, one of the, one of the metrics that I use in, like, good quality software is, like, how do we accomplish the, the, a thing we want to do in the least number of lines of code? 00:24:54,044 --> 00:24:54,404 [Eric] Yep. 00:24:54,404 --> 00:24:56,914 [John] Right? Um, and 00:24:57,984 --> 00:25:02,764 [John] based on what we were able to do, I can't imagine it being net, like, less lines of code. 00:25:02,764 --> 00:25:03,044 [Eric] Mm. 00:25:03,044 --> 00:25:04,424 [John] Even with the abstractions. 00:25:04,424 --> 00:25:04,943 [Eric] Mm-hmm. Mm-hmm. 00:25:04,944 --> 00:25:14,144 [John] Um, but, and, and, and then looking back on it, we might not have actually had quite the right toolkit as far as, like, these weren't probably the ideal, like, libraries and stuff to use. 00:25:14,144 --> 00:25:14,744 [Eric] Yep. 00:25:14,744 --> 00:25:16,384 [John] Um, but it was such an interesting 00:25:17,444 --> 00:25:31,124 [John] exercise because you think, like, you think you know how hard something is or you think you know how to do something, and, um, like, fundamentally, like, would've always worked that way- 00:25:31,124 --> 00:25:31,284 [Eric] Mm-hmm 00:25:31,284 --> 00:25:33,064 [John] ... like, for the last 15 years. 00:25:33,124 --> 00:25:33,844 [Eric] Mm-hmm. 00:25:33,844 --> 00:25:36,904 [John] And, um, then AI changed that, like- 00:25:36,904 --> 00:25:36,954 [Eric] Yeah 00:25:36,954 --> 00:25:38,064 [John] ... overnight, you know? 00:25:38,064 --> 00:25:43,384 [Eric] Right. Right. Yeah. I think, so let's, let's just talk through some lessons here. The, 00:25:44,404 --> 00:25:48,864 [Eric] so one is don't outsource your core thinking to AI. 00:25:48,864 --> 00:25:49,484 [John] For sure. 00:25:49,484 --> 00:26:03,164 [Eric] You know? I mean, a- and even if you use AI, because I... You know, it's interesting, I see, I've seen multiple posts, like, read these really interesting long posts from people, you know, who write on whatever topic- 00:26:03,164 --> 00:26:03,264 [John] Right 00:26:03,264 --> 00:26:04,144 [Eric] ... a little bit about AI. 00:26:04,144 --> 00:26:04,304 [John] Right. 00:26:04,304 --> 00:26:05,204 [Eric] But they'll say, 00:26:06,244 --> 00:26:12,124 [Eric] uh, "You know, I used AI to help me, like, think through this post"- 00:26:12,124 --> 00:26:12,134 [John] Mm-hmm 00:26:12,134 --> 00:26:12,624 [Eric] ... right? 00:26:12,624 --> 00:26:13,224 [John] Mm-hmm. 00:26:13,224 --> 00:26:16,784 [Eric] It's really fascinating to me, and I've, I actually reached out to one of them- 00:26:16,784 --> 00:26:17,034 [John] Oh, cool 00:26:17,034 --> 00:26:17,894 [Eric] ... and asked them how they did it. 00:26:17,894 --> 00:26:17,904 [John] Yeah. 00:26:17,904 --> 00:26:18,984 [Eric] They didn't respond to me, but- 00:26:18,984 --> 00:26:19,084 [John] All right 00:26:19,084 --> 00:26:20,594 [Eric] ... um, they're very famous. 00:26:20,594 --> 00:26:21,124 [John] Maybe they will now. 00:26:21,124 --> 00:26:24,384 [Eric] Yeah, maybe they will. Yeah, exactly. Have them on the show. 00:26:24,384 --> 00:26:25,664 [John] Exactly. 00:26:25,664 --> 00:26:28,754 [Eric] Uh, but that's really interesting to me because 00:26:29,804 --> 00:26:38,944 [Eric] my guess would be that if they're a really good writer, they're actually really disciplined in... L- so actually, let me... I'll, I'll put a sharp point on this. 00:26:38,944 --> 00:26:39,904 [John] Mm-hmm. 00:26:39,904 --> 00:26:41,264 [Eric] The people who I know 00:26:42,844 --> 00:26:46,264 [Eric] who are really good writers who use AI heavily 00:26:47,444 --> 00:26:54,704 [Eric] tell AI it's wrong more often than they agree with it. 00:26:54,704 --> 00:26:55,744 [John] Okay. 00:26:55,744 --> 00:26:55,964 [Eric] Right? 00:26:55,964 --> 00:26:55,984 [John] Yeah. 00:26:55,984 --> 00:26:59,744 [Eric] Just constantly, "That is not right," or they're questioning assumptions, right? 00:26:59,744 --> 00:26:59,814 [John] Right. 00:26:59,814 --> 00:27:03,963 [Eric] They're, they're... Sorry, they're questioning output that the AI, you know, brings back. 00:27:03,964 --> 00:27:04,624 [John] Right. 00:27:04,624 --> 00:27:13,904 [Eric] Which I think is actually correct. Like, that's the, that's the right way if you're working through something, is for the conversation to be more... Adversarial is probably too strong of a word- 00:27:13,904 --> 00:27:14,064 [John] Right 00:27:14,064 --> 00:27:26,064 [Eric] ... um, but sort of not going down this path, right? But it's because you are being a very critical thinker and, and critical thinking pushes back against- 00:27:26,064 --> 00:27:27,304 [John] Right 00:27:27,304 --> 00:27:27,414 [Eric] ... sycophancy. 00:27:27,414 --> 00:27:27,414 [John] Right. 00:27:27,414 --> 00:27:27,784 [Eric] Right? 00:27:27,784 --> 00:27:37,744 [John] Which, which is interesting, and, uh, this is new for me. Totally agree with that. But I find if I'm doing that too much, then I probably frame something wrong, and I don't have the right goal- 00:27:37,744 --> 00:27:37,754 [Eric] Yes 00:27:37,754 --> 00:27:39,864 [John] ... and I don't have the right goal set for what I wanted to do. 00:27:39,864 --> 00:27:40,463 [Eric] Agreed. 00:27:40,464 --> 00:27:42,444 [John] And often it makes sense to start over- 00:27:42,444 --> 00:27:42,964 [Eric] Yes 00:27:42,964 --> 00:27:46,713 [John] ... reframe, and then you don't have to be constantly, like, steering. 00:27:46,713 --> 00:27:46,724 [Eric] Yep. 00:27:46,724 --> 00:27:57,764 [John] 'Cause I, 'cause I get caught in that, and I think people get caught in the, like, "I'm trying to be... I'm, I'm thinking hard. I'm interacting. I'm telling it it's wrong." Um, that's good, but if you're doing it too much, then, like, start over. Like, you're- 00:27:57,764 --> 00:27:57,844 [Eric] Yes 00:27:57,844 --> 00:27:58,964 [John] ... you're just... Y- 00:27:59,604 --> 00:27:59,784 [Eric] Totally agree. 00:27:59,784 --> 00:28:00,403 [John] Start over. 00:28:00,404 --> 00:28:00,884 [Eric] Totally agree. 00:28:00,884 --> 00:28:01,364 [John] Yeah. 00:28:01,364 --> 00:28:05,384 [Eric] I think the other thing, and this, this is... It sounds obvious, but I don't think it is. 00:28:07,094 --> 00:28:12,944 [Eric] Someone once said this to me, which was, uh, it's really stuck with me. They said, "The problem is the way we see the problem." 00:28:14,024 --> 00:28:17,183 [Eric] And that goes back to your first example, which was one of my- 00:28:17,184 --> 00:28:17,254 [John] Right 00:28:17,254 --> 00:28:19,324 [Eric] ... takeaways from the first example you gave, right? 00:28:19,324 --> 00:28:19,484 [John] Right. 00:28:19,484 --> 00:28:23,104 [Eric] Where you didn't separate the problem and the proposed solution. 00:28:23,104 --> 00:28:23,764 [John] Yep. 00:28:23,764 --> 00:28:30,324 [Eric] AI is actually a great tool if you're very explicit to say, "What are different ways to think about this problem?" 00:28:30,324 --> 00:28:30,763 [John] Right. Yeah. Right. 00:28:30,764 --> 00:28:33,073 [Eric] Completely devoid of a proposed solution. 00:28:33,073 --> 00:28:33,104 [John] Devoid of the... Yeah. 00:28:33,104 --> 00:28:33,284 [Eric] Right? 00:28:33,284 --> 00:28:34,123 [John] Exactly. 00:28:34,123 --> 00:28:35,573 [Eric] Uh, and it can be very helpful for that. 00:28:35,573 --> 00:28:35,583 [John] Yeah. 00:28:35,584 --> 00:28:43,004 [Eric] But again, that's going back to being very deliberate about steering the AI, um, you know, not just following the path of least resistance. 00:28:43,004 --> 00:28:43,104 [John] Yep. 00:28:44,144 --> 00:28:48,344 [Eric] All right. Don't outsource your thinking, and we'll catch you on the next show. 00:28:48,344 --> 00:28:49,284 [John] Yep.