00:46:24 · JC Lozano, Ryan Deng
[00:46:24] S1: man. [00:46:24] S0: I gotta I gotta level my own startup. I mean, we're still, like, at the pre seed level. I mean, you gotta get some coach on my way. But, like, you know, my question is, do you think that skills... I mean, I obviously said that, like, you know, new value after the growth skills, but if... I'm a generalist. Right? I can throw [00:46:42] S1: on a generalist. [00:46:42] S0: If I wanted to join your company Sure. And I didn't have the technical acumen Sure. To perform well, but I was really good at Cloud Co. Would you give would you would you give me a shot? Would you give me a shot, hon? Like, I'm really good with agents, you know? I'm fanning out these sub agents, you know? That's a good question. [00:47:02] S1: I mean, depends. Okay. There's all the other bars. But... So we... It's a good question. Uh-huh. You'd be considered. Really? [00:47:14] S0: For sure. [00:47:15] S1: Wow. Yeah. [00:47:17] S0: So hop on the cloud code everybody. Yeah.
00:47:17 · Ryan Deng, JC Lozano
[00:47:17] S0: So hop on the cloud code everybody. Yeah. [00:47:25] S1: You would be considered. Look, like, the the cool thing about, like, all these large language models, right, is like, in some sense, it is, like, commoditizing some form of intelligence. Right? There's, like, all these debates about, like, what that actually means and all that good stuff. But I think everybody can agree. For most people, sensible people, can agree that it's, this is pretty crazy. Mhmm. And so that does shift the bar, I think, to like what else are you bringing to the table and that's maybe like you're very hardworking, you're very genuine, you're like very creative, like all these things matter a ton. That's not to say that technical skills don't matter, I actually think they matter even more. [00:48:01] S0: Explain to me why because you know... [00:48:03] S1: Because you gotta tell if the thing's lying or hallucinating. Mhmm. [00:48:06] S0: Okay. Like... What if the models get so good that it's just strictly based on like human nuance of how you interpret the answer? No longer is it like incorrect. Don't think that's... I think the models will get to a point where, hey, like, they're they're so good that like, this not a faction incorrect answer, it's just how you're interpreting the answer. Right? [00:48:26] S1: At some point, I'm sure they will. I don't think they're there yet. [00:48:29] S0: Oh, yeah.
00:48:29 · Ryan Deng, JC Lozano
[00:48:29] S0: Oh, yeah. [00:48:29] S1: That makes sense. I think, like, at some point, they could be. Mhmm. But I think right now, like, problem that you have is just, like, the reason you still need skills and expertise and all this good stuff Right. Is because they sound exceptionally convincing and you still have to like understand like Mhmm. So at the end of the day, right, like you're gonna put up a building. There's like a... There's like a bunch of laws of engineering and... Sorry. There's a bunch of engineering rules and a bunch of laws of physics that are gonna tell you whether or not that thing's gonna fall or not. [00:48:56] S0: Right. [00:48:57] S1: Right? And so like They [00:48:58] S0: don't wanna be body coding any any buildings any time soon. [00:49:01] S1: So like, you... I really do think we all would feel better if like the people who are doing that, like, you know, they can use Cloud Code, they can use [00:49:07] S0: all these tools, [00:49:08] S1: but they should definitely know the underlying like principles Uh-huh. That they [00:49:12] S0: can do We we [00:49:13] S1: a check for them. [00:49:14] S0: We we call it slop cannon. [00:49:16] S1: Okay. Yeah. Slop cannon. [00:49:19] S0: Slop cannon. Yeah. It's when somebody uses an LLM and just takes it for face value all the way through. [00:49:24] S1: Got it. And then [00:49:25] S0: they start, engaging in psychofantasy with the agent and they start convincing it that it's correct. Like, no, no, you're wrong. This is the right way. And it's like, oh, actually, you're right.
00:49:25 · Ryan Deng, JC Lozano
[00:49:25] S0: they start, engaging in psychofantasy with the agent and they start convincing it that it's correct. Like, no, no, you're wrong. This is the right way. And it's like, oh, actually, you're right. [00:49:33] S1: Yeah. Understood. Yeah. Okay. [00:49:35] S0: So I've seen that too with some engineers, to be honest. [00:49:38] S1: Yeah. No. So it's like, it's... I think we we would all feel better and we all, you know, we need to take, like, go understand what its, like, basis of assumptions are. Is it running the right equations? Like, all that kind [00:49:49] S0: of good stuff.
00:49:49 · Ryan Deng, JC Lozano
[00:49:49] S0: of good stuff. [00:49:49] S1: Right. And I think, like, the interesting part about, like, engineering in general, and it's the same with computer science, by the way, and, like, coding. Right? Like, it's actually about, like, your choice of problems, your choice of assumptions, and, like, the taste that you put into that. Like, that is, like, ultimately, like, an LN that can generate a million lines of code. If it's not solving the right problem, if it's not abstracted the right way, the problem isn't decomposed the right way, like, it's still not gonna work. Mhmm. And so, like, you shouldn't, like, confuse, like, output with outcome. Right? And I think it's the same with everything else. Right? Like, you can generate a lot of output, but, like, if the outcome isn't, like, a good outcome for what you're shooting for, like, it doesn't actually matter. So, yeah, I think, like, you know, the agents will keep getting smarter, which is awesome. I'm super excited by it, by the way. Like, it's amazing what you can do these days. It's amazing the leverage you can get on intelligence, the leverage you can get on... It's awesome. I'm super excited about it. But personally, I don't think they're there yet. [00:50:47] S0: Mhmm. When do you think they'll be good at? When do I think they'll be When do you think [00:50:50] S1: They're good right now, They're to be awesome. They got me, [00:50:54] S0: you know, building out some system architecture. So I think the models are pretty good.
00:50:54 · Ryan Deng, JC Lozano
[00:50:54] S0: you know, building out some system architecture. So I think the models are pretty good. [00:51:00] S1: They're awesome. No. No. No. They're super good. That's not that's not what I'm saying. I think there's still a lot of, like, human involvement needed. Oh, I agree. I agree. That's what [00:51:09] S0: I'm Computer use has to get better. [00:51:12] S1: Well, I feel like the new ones are are pretty... Quite good computer use. Yeah. [00:51:15] S0: It's like this, like, on my screen. It's, like, two seconds. Two seconds. [00:51:19] S1: What's wrong with it moving slow? [00:51:21] S0: Gotta move faster. That that inference we talked about earlier. You know? [00:51:24] S1: That's what's Faster [00:51:25] S0: inference, Anthropically sponsored. It's probably moving slow so it [00:51:29] S1: can confuse those, like, captcha detectors. [00:51:31] S0: Yeah. Probably. [00:51:31] S1: Right? Because that's how they detect if it's a robot or not. It's like, quickly do you find the button? Well, Juan, [00:51:36] S0: thank you so much for coming on. Where we're [00:51:37] S1: gonna end on. I love it. [00:51:38] S0: Appreciate it so much. [00:51:39] S1: Yeah. Absolutely. This is a [00:51:40] S0: lot of fun. Hopefully, we're here for a series c. Hop on the show again. Sure. That sounds awesome. [00:51:44] S1: Let's see. Hopefully, we get another great dinner. This place is great. [00:51:47] S0: What is [00:51:48] S1: the name of this place, by the way?