Meeting Title: Uttam <> Ethan Date: 2024-12-14 Meeting participants: Uttam Kumaran, Ethanpetersen
WEBVTT
1 00:00:24.420 ⇒ 00:00:25.780 ethanpetersen: There you go.
2 00:00:26.120 ⇒ 00:00:29.410 ethanpetersen: One second here. Oh, good! Can you hear me? Okay.
3 00:00:34.300 ⇒ 00:00:38.793 ethanpetersen: yeah. I can’t hear you for some reason.
4 00:00:39.580 ⇒ 00:00:40.830 Uttam Kumaran: Better, maybe.
5 00:00:40.830 ⇒ 00:00:42.490 ethanpetersen: Yep, yep, now I can hear.
6 00:00:42.490 ⇒ 00:00:43.570 Uttam Kumaran: Okay, cool.
7 00:00:43.680 ⇒ 00:00:47.179 Uttam Kumaran: How’s it going? Thank you for taking the time. On Saturday morning.
8 00:00:48.740 ⇒ 00:00:52.790 ethanpetersen: It’s coming up pretty good kind of a weird.
9 00:00:53.940 ⇒ 00:00:58.560 ethanpetersen: My building’s alarm was going off this morning, and
10 00:00:58.890 ⇒ 00:01:03.650 ethanpetersen: I like I don’t know what the heck is going on. And then they finally came over
11 00:01:03.840 ⇒ 00:01:11.670 ethanpetersen: the intercom. And we’re just directing people. It’s like super high winds
12 00:01:11.780 ⇒ 00:01:17.560 ethanpetersen: right now, and and some people have their their windows open, which I don’t know how that
13 00:01:17.930 ⇒ 00:01:26.995 ethanpetersen: could be, because it’s like been pouring rain for the last like 1214 h, so.
14 00:01:27.620 ⇒ 00:01:28.290 Uttam Kumaran: Yeah.
15 00:01:28.290 ⇒ 00:01:28.585 ethanpetersen: Yeah.
16 00:01:29.360 ⇒ 00:01:30.750 Uttam Kumaran: Are you still in Vancouver?
17 00:01:32.088 ⇒ 00:01:33.679 ethanpetersen: Back in San Francisco.
18 00:01:34.340 ⇒ 00:01:34.800 Uttam Kumaran: Okay.
19 00:01:35.980 ⇒ 00:01:43.060 Uttam Kumaran: I mean, when when you guys are used to wind out, I was just there. Yeah, I mean, I was just there for Thanksgiving. And yeah, it’s always like
20 00:01:43.540 ⇒ 00:01:49.755 Uttam Kumaran: insane wind. But tornado, I mean, I don’t know. I’m I’m here in Texas now.
21 00:01:50.330 ⇒ 00:01:58.469 Uttam Kumaran: and like, this is where it’s like, okay, there’s like, could be tornado touchdown because we’re like close like Oklahoma and stuff. But
22 00:01:59.108 ⇒ 00:02:03.709 Uttam Kumaran: I don’t know if you like tornado force winds. Probably it could happen.
23 00:02:04.680 ⇒ 00:02:11.556 ethanpetersen: Yeah, I I had never gotten an alert like that. And I was like, well, I don’t know. I
24 00:02:12.480 ⇒ 00:02:19.240 ethanpetersen: I I guess Sf. Can have tornadoes. Actually. Oh, years ago I was
25 00:02:20.990 ⇒ 00:02:31.870 ethanpetersen: arguing with my co-founder like a a friendly argument that they’re about tornadoes in Chicago.
26 00:02:31.870 ⇒ 00:02:32.680 Uttam Kumaran: Hmm.
27 00:02:32.680 ⇒ 00:02:41.154 ethanpetersen: And tornadoes, like like actually in the in the city of Chicago. So we we both grew up in in Illinois.
28 00:02:41.540 ⇒ 00:02:41.980 Uttam Kumaran: Okay.
29 00:02:41.980 ⇒ 00:02:49.250 ethanpetersen: And I just I was like, there’s there’s no way that there’s a
30 00:02:49.822 ⇒ 00:03:02.957 ethanpetersen: the the tornadoes that I’m I’m used to like that’d be around my. My hometown are just like massive, completely destroy the town like there’s no like it’s
31 00:03:03.540 ⇒ 00:03:06.439 ethanpetersen: absolutely like catastrophic thing.
32 00:03:07.695 ⇒ 00:03:14.319 ethanpetersen: And and then, I know, like sometimes there’ll be, you know, by the definition of a tornado.
33 00:03:14.320 ⇒ 00:03:14.959 Uttam Kumaran: Yes, yes.
34 00:03:14.960 ⇒ 00:03:16.469 ethanpetersen: Something that that touches down.
35 00:03:16.470 ⇒ 00:03:20.604 Uttam Kumaran: What do you think that is? That’s just researchers being like there was one.
36 00:03:20.900 ⇒ 00:03:30.470 ethanpetersen: I I think so. I was just like there’s no way I I just don’t buy it. That Chicago could have like the the same type of tornado that
37 00:03:30.620 ⇒ 00:03:31.730 ethanpetersen: I’ve seen it grow.
38 00:03:32.071 ⇒ 00:03:41.978 Uttam Kumaran: Like Geo. Storm, and like stuff like that, because that’s where it’s like, Oh, my God! A tornadoes hitting like the biggest tornadoes ever hitting new York city today.
39 00:03:42.320 ⇒ 00:03:42.690 ethanpetersen: Yeah.
40 00:03:43.152 ⇒ 00:03:54.727 Uttam Kumaran: Otherwise. Yeah, there’s no, there’s no way. And those cities all have their own problems. So they don’t need like tornadoes. Chicago, already so cold.
41 00:03:56.334 ⇒ 00:04:00.289 ethanpetersen: Yeah, exactly. So. Where in Texas are you?
42 00:04:00.530 ⇒ 00:04:01.679 Uttam Kumaran: I’m in Austin.
43 00:04:02.240 ⇒ 00:04:06.190 ethanpetersen: Awesome. Yeah, I one of my friends, actually.
44 00:04:06.300 ⇒ 00:04:08.365 ethanpetersen: Guy, that introduced me to Craig.
45 00:04:08.710 ⇒ 00:04:09.140 Uttam Kumaran: Oh, okay.
46 00:04:09.140 ⇒ 00:04:13.819 ethanpetersen: Just moved to Austin. Maybe a few months ago.
47 00:04:13.820 ⇒ 00:04:29.319 Uttam Kumaran: That’s great. That’s a great place. I was in New York before this. And then I grew up in the Bay area. But yeah, I love it here. I think for me it was more like moving here, just for life, like it’s a lot more relaxing here, and I’m
48 00:04:29.510 ⇒ 00:04:46.390 Uttam Kumaran: I’m like a workaholic, anyway. So I don’t need like the external pressure. And I travel enough like I travel to the Bay Area or to New York, people and stuff like that. So that’s always like, that’s my reprieve is just through travel. So yeah.
49 00:04:47.190 ⇒ 00:04:50.759 ethanpetersen: Yeah, yeah, probably enough to get your fix of of.
50 00:04:50.760 ⇒ 00:04:59.130 Uttam Kumaran: Exactly here, and there’s like nobody talks about work. It’s like very relaxing. There’s like no traffic. It’s nice.
51 00:04:59.780 ⇒ 00:05:04.920 ethanpetersen: Yeah, how? What’s the does it skew? Pretty
52 00:05:05.100 ⇒ 00:05:08.269 ethanpetersen: like young professional, as far as the.
53 00:05:08.790 ⇒ 00:05:14.619 Uttam Kumaran: Yeah, it depends. So it depends like in so there’s Austin proper. But I mean.
54 00:05:14.760 ⇒ 00:05:29.859 Uttam Kumaran: like, Austin, like, Metro is like very large. So there’s like, basically like different cities, sort of on the outskirts. I would say, yeah, it’s more like 28 to like 35 professionals.
55 00:05:30.180 ⇒ 00:05:32.069 Uttam Kumaran: not really a lot of.
56 00:05:32.310 ⇒ 00:05:39.450 Uttam Kumaran: I mean, I think during Covid there’s a lot of very young people like 2223 but you don’t have a lot of like
57 00:05:39.940 ⇒ 00:06:07.870 Uttam Kumaran: the schools feeding into Austin. You have ut. But you know, I think, compared to New York or Chicago people that moved directly there after school, I think here it’s more like people get their footing, and then they come here. And then there’s this particular company or particular sector they work in. But here it’s a lot of remote, frankly, like Austin benefited the most from remote work. And I think also, just the permitting and building process in Texas is a lot different. So they were able to build
58 00:06:07.870 ⇒ 00:06:14.882 Uttam Kumaran: very fast to accommodate the demand, and actually, like prices are going down here.
59 00:06:15.350 ⇒ 00:06:15.920 ethanpetersen: Wow!
60 00:06:15.920 ⇒ 00:06:29.479 Uttam Kumaran: You know, it’s it’s 1 of the few places in the country where we’re back to basically trend and like the price for like a 1. Bed, you know, as a barometer is, is very, very competitive here. Which is great.
61 00:06:29.700 ⇒ 00:06:37.780 Uttam Kumaran: and then I think also for me, I think this is a place that probably similar to like Miami, like in terms of tech, like, I think Sf, is about
62 00:06:37.870 ⇒ 00:06:51.520 Uttam Kumaran: far and large, like Number one. I think New York is probably number 2, and then, I think, depending on what sector you’re in. You like biomeds in Boston. There’s probably stuff everywhere, but I think Austin will find its footing to have.
63 00:06:51.530 ⇒ 00:07:18.349 Uttam Kumaran: You know, certain things. There’s a there’s a lot of like startups in like Cpg Ecom stuff like that. And of course, here you have. Like the big you have, like the super large Enterprise Tech, you have Oracle Samsung Apple. Google, but a lot of the people here are like sales marketing things like that. So I think the community here in terms of like tech, like, I’m a data engineer. So for me, I’m like, are there meetups for data. And now I’m getting into AI stuff. But I think still, it skews more like
64 00:07:18.770 ⇒ 00:07:31.109 Uttam Kumaran: older and and more business. But I think that’ll change over time. Like I, I think it’s getting better. And I think the city is definitely like wants to support that. So yeah.
65 00:07:32.140 ⇒ 00:07:42.770 ethanpetersen: Yeah, yeah, it’s also i 1. So my friend, that moved there, I I was surprised to hear from him. One of the 1st things he talked about is how much more he
66 00:07:43.648 ⇒ 00:07:46.560 ethanpetersen: preferred Austin over Denver.
67 00:07:46.860 ⇒ 00:07:47.499 ethanpetersen: Oh, got it!
68 00:07:47.500 ⇒ 00:07:54.020 ethanpetersen: I know that he he really liked Denver when he, when he moved there. So
69 00:07:54.770 ⇒ 00:07:57.990 ethanpetersen: yeah, I I really need to visit, and at least like.
70 00:07:57.990 ⇒ 00:08:25.849 Uttam Kumaran: What is it I mean? I’ve been to Denver. I think this is. I think this is very similar in that. I think most people are interested in spending time outside, or, you know, not 100 occupied with work like when I was in New York. I don’t even go home like it’s all work. And you after work you get food. And it was very like much for someone like me. I think it was a lot of great like pressure. But then I was always like, Okay, I’m gonna I’m gonna
71 00:08:25.900 ⇒ 00:08:35.039 Uttam Kumaran: do what I can there, and then leave and go somewhere where it’s like. I don’t need the external pressure. I like to just be able to go to a coffee shop and work and
72 00:08:35.169 ⇒ 00:08:45.690 Uttam Kumaran: kind of relax. And then, you know I work hard at at Brain Forge every day, which is a grind. So it’s like nice that the rest of I could like drive my car to the grocery store drive back
73 00:08:45.910 ⇒ 00:09:04.883 Uttam Kumaran: like it’s more resembles like how I grew up in like the suburbs. And then also, Austin downtown is 5 6 min, you know I’m a little bit like maybe 5, 6 min outside the city. But Austin downtown is very close, so it’s kind of weird, so you can be right downtown in like 10 min from, like most places in the city.
74 00:09:05.720 ⇒ 00:09:12.739 Uttam Kumaran: which I feel like for both cities, is not the case. You may even just be a few miles away, but it still will take 2030 min.
75 00:09:12.970 ⇒ 00:09:21.959 Uttam Kumaran: The other tough part here is. No, there’s no public transport, though there’s like buses and no rail, no like. So no like Bart or or like community system.
76 00:09:22.341 ⇒ 00:09:27.570 Uttam Kumaran: so that I think they’re gonna have to invest a lot in. But also Texas is like a big car.
77 00:09:27.910 ⇒ 00:09:29.819 Uttam Kumaran: Everybody drives everywhere.
78 00:09:29.970 ⇒ 00:09:42.229 Uttam Kumaran: So maybe, like, the only thing I’m hoping for is like, if Waymo can really make a dent here. They can start deleting a lot of the parking garages and you know, make open space and stuff like that. So.
79 00:09:43.986 ⇒ 00:09:44.513 ethanpetersen: Yeah.
80 00:09:45.330 ⇒ 00:09:54.349 ethanpetersen: yeah, it’s definitely I I sold my car before I moved to to San Francisco. And I
81 00:09:54.460 ⇒ 00:09:57.459 ethanpetersen: I’ve just been like, if I have to do it
82 00:09:57.640 ⇒ 00:10:09.230 ethanpetersen: trip, or whatever I rent something, and it’s it was just insane to me when I was moving. How much apartment spot cost.
83 00:10:09.730 ⇒ 00:10:13.899 ethanpetersen: and I’m having to deal with just like
84 00:10:14.390 ⇒ 00:10:27.870 ethanpetersen: I mean having a car. And that stuff would be kind of a nightmare and then you got a budget, for you know all the the expected occurrences of broken windshields and everything like that. So
85 00:10:29.440 ⇒ 00:10:33.420 ethanpetersen: that’s a it’s it’s a lot. But.
86 00:10:35.732 ⇒ 00:10:38.400 Uttam Kumaran: How did you? How did you get put in touch with Greg.
87 00:10:40.266 ⇒ 00:10:42.453 ethanpetersen: So that was through
88 00:10:43.550 ⇒ 00:10:48.150 ethanpetersen: so that that friend of mine Mike, he was
89 00:10:49.730 ⇒ 00:11:02.030 ethanpetersen: actually yeah, it’s a very random story Mike was at I think, like a Rockies game, and met Craig there.
90 00:11:02.620 ⇒ 00:11:09.810 ethanpetersen: and knew Craig was in in AI. And then I so my my background was
91 00:11:10.050 ⇒ 00:11:14.729 ethanpetersen: prior to Crusoe. I ran an AI startup for about 5 years.
92 00:11:15.410 ⇒ 00:11:29.729 ethanpetersen: We specialized in facial expression recognition models and integrated those into like conversational analysis products. And we primarily had traction within defense contracting
93 00:11:30.372 ⇒ 00:11:36.650 ethanpetersen: so did that for the latter, like 3 years of the company
94 00:11:37.452 ⇒ 00:11:40.529 ethanpetersen: we contracted out for bunch of
95 00:11:40.750 ⇒ 00:11:47.799 ethanpetersen: awesome groups in the Air Force and the army. But long story short
96 00:11:48.345 ⇒ 00:11:55.070 ethanpetersen: the the last year of it. It was just like one thing after the other our
97 00:11:55.627 ⇒ 00:12:12.659 ethanpetersen: so I I was bringing on close friend of mine that was gonna be coo it was a pitch that took over the course of like 6 months, where he was contracting like consulting for us, and then try to get him to join full time, and then he finally
98 00:12:14.120 ⇒ 00:12:17.000 ethanpetersen: finally approved. So I had
99 00:12:19.460 ⇒ 00:12:24.509 ethanpetersen: it it just it took a lot to finally get it on board, and I had a
100 00:12:24.740 ⇒ 00:12:27.570 ethanpetersen: co-founder that was kind of starting to like disengage.
101 00:12:27.570 ⇒ 00:12:28.300 Uttam Kumaran: Oh, okay.
102 00:12:28.300 ⇒ 00:12:37.610 ethanpetersen: From the business. So he was gonna be brought on to kind of take over a lot of her responsibilities that, you know. Still need to like.
103 00:12:37.610 ⇒ 00:12:38.050 Uttam Kumaran: Yeah.
104 00:12:38.270 ⇒ 00:12:41.100 ethanpetersen: Need to happen. And then
105 00:12:42.054 ⇒ 00:12:58.540 ethanpetersen: he that there was an incident at at an investor’s wife’s show, and I had to fire my friend that I was bringing on board to be coo that was one domino that just kind of
106 00:12:58.690 ⇒ 00:13:06.230 ethanpetersen: like fell. And then there was so a lot of our defense relationships had already been
107 00:13:06.690 ⇒ 00:13:08.309 ethanpetersen: kind of hand it over to him.
108 00:13:08.470 ⇒ 00:13:15.820 ethanpetersen: And then that happened. And defense is very. You know, it’s it’s a everything is built on trust and and relationships. And
109 00:13:16.722 ⇒ 00:13:20.869 ethanpetersen: it’s really hard to try to recover.
110 00:13:21.690 ⇒ 00:13:29.129 ethanpetersen: And then we recovered. But our source of funding for the next contract.
111 00:13:29.774 ⇒ 00:13:36.650 ethanpetersen: Fell through. So then it went from we like we were expecting a a 50 mill contract
112 00:13:37.310 ⇒ 00:13:40.720 ethanpetersen: that got bumped down to like a final contract.
113 00:13:41.110 ⇒ 00:13:46.519 ethanpetersen: and then it just get progressively bumped down where they finally were like, well, we can.
114 00:13:46.920 ⇒ 00:13:49.289 ethanpetersen: We can do something, for like 200 k.
115 00:13:49.480 ⇒ 00:13:53.340 ethanpetersen: We think that eventually we can get to oh.
116 00:13:53.610 ⇒ 00:13:59.249 ethanpetersen: funding the full contract. But we can do 200 K. For for now.
117 00:14:00.010 ⇒ 00:14:02.270 ethanpetersen: But then the problem was like, there’s no
118 00:14:03.120 ⇒ 00:14:07.489 ethanpetersen: oh, we’re we’re putting tons of trust in in them for.
119 00:14:08.086 ⇒ 00:14:13.500 ethanpetersen: okay, what? When are we gonna move beyond this like life support amount
120 00:14:14.196 ⇒ 00:14:18.479 ethanpetersen: and at the same time, like the actual work of the contract
121 00:14:18.600 ⇒ 00:14:24.149 ethanpetersen: wasn’t really changing substantially. So it’s like you’re still gonna be demanding
122 00:14:24.250 ⇒ 00:14:31.820 ethanpetersen: crazy amount from us, but paying, like, you know, less than a 10th Adam.
123 00:14:31.950 ⇒ 00:14:36.930 ethanpetersen: And and so that kind of kicked off just this
124 00:14:37.620 ⇒ 00:14:40.460 ethanpetersen: dialogue with my my co-founder of.
125 00:14:40.920 ⇒ 00:14:44.789 ethanpetersen: Oh, we’ve been doing this for 5 years, you know. Do we?
126 00:14:46.510 ⇒ 00:14:49.079 ethanpetersen: Try to just kind of survive
127 00:14:49.220 ⇒ 00:14:53.789 ethanpetersen: with this? Or or is it? Is it time to kind of shut things down? Move on to the next thing.
128 00:14:53.910 ⇒ 00:14:56.740 ethanpetersen: and and earlier so last year
129 00:14:57.660 ⇒ 00:15:06.559 ethanpetersen: I also had shingles due to just like a bunch of things going on, and that really
130 00:15:07.990 ⇒ 00:15:13.789 ethanpetersen: pretty much took took me out of the game. I had shingles, and then, later in the year I I had Covid, and and still.
131 00:15:13.920 ⇒ 00:15:17.709 ethanpetersen: when I had Covid, I was dealing with the all of them
132 00:15:18.010 ⇒ 00:15:22.450 ethanpetersen: kind of ripple effects of of shingles. So Covid hit me a lot harder than
133 00:15:22.740 ⇒ 00:15:24.930 ethanpetersen: that I probably should have
134 00:15:25.352 ⇒ 00:15:32.219 ethanpetersen: and so I I kind of figured like I’m I’m at the end of my rope with it just physically.
135 00:15:32.220 ⇒ 00:15:32.620 Uttam Kumaran: Totally.
136 00:15:33.285 ⇒ 00:15:36.689 ethanpetersen: And I took like a 4 month break
137 00:15:37.243 ⇒ 00:15:50.609 ethanpetersen: just went back home and played a a crazy amount of valorants, and then hung out with friends and family and just try to take a lot of time to.
138 00:15:50.610 ⇒ 00:15:51.100 Uttam Kumaran: Oh!
139 00:15:52.070 ⇒ 00:15:56.630 ethanpetersen: Do nothing. And then I I was starting to interview
140 00:15:57.369 ⇒ 00:16:03.359 ethanpetersen: for places again, and I chatting with that that friend Mike.
141 00:16:03.470 ⇒ 00:16:05.720 ethanpetersen: and he introduced me to some
142 00:16:06.240 ⇒ 00:16:08.830 ethanpetersen: customers of Crusoe where I work now.
143 00:16:09.010 ⇒ 00:16:29.889 ethanpetersen: and so I interviewed with them, and then one of them asked me, Is why not work with Mike at Crusoe, and said, Well, that’s a good question I would love to. I hadn’t really thought about it. So I ended up interviewing. And and Crusoe is so we’re in AI cloud.
144 00:16:31.020 ⇒ 00:16:38.559 ethanpetersen: But so AI clown is is effectively a meaningless term. So what it actually is is
145 00:16:38.990 ⇒ 00:16:52.159 ethanpetersen: where we build and manage data centers. And then we have the software platform on top for managing like virtual machines, and all virtual machines have have Gpus and and everything set up for AI.
146 00:16:52.160 ⇒ 00:16:55.038 Uttam Kumaran: It’s kind of like modal labs, almost.
147 00:16:55.780 ⇒ 00:17:02.680 ethanpetersen: Yeah, it’s similar. Yeah, very similar. The only thing is we, we aren’t serverless. So you have to
148 00:17:03.350 ⇒ 00:17:25.530 ethanpetersen: more management. But same idea where it’s, it’s it’s a focus product. And it’s not like like Aws or Tcp, you know, you’re like people use it to train AI models. But you’re getting charged for potentially a whole bunch of other services.
149 00:17:25.589 ⇒ 00:17:36.840 ethanpetersen: And then the infrastructure is like, if it’s a general purpose cloud. There they have to. Make a lot of trade offs in order to serve all of those use cases right? So
150 00:17:37.030 ⇒ 00:17:38.410 ethanpetersen: for us.
151 00:17:38.960 ⇒ 00:17:56.279 ethanpetersen: we make the trade offs that, you know, only benefit AI training and inference. And we, we don’t care about any other work workloads so you can’t you? You could host a web app on our cloud. It’s probably gonna be a terrible experience.
152 00:17:56.280 ⇒ 00:17:56.620 Uttam Kumaran: Yeah.
153 00:17:56.989 ⇒ 00:18:06.970 ethanpetersen: But those are the things that we, you know we don’t optimize for and and then certainly, when it when it comes to like product. Roadmap.
154 00:18:07.610 ⇒ 00:18:12.629 ethanpetersen: AI 1st is the like lens that we look at everything through.
155 00:18:13.262 ⇒ 00:18:34.517 ethanpetersen: But the the issue is that Crusa doesn’t have really any AI people. So that’s why I was brought in somebody with a lot of recent recent experience. And I like on on paper. I lead developer relations and it’s a
156 00:18:36.400 ⇒ 00:18:40.040 ethanpetersen: it’s that’ll that’ll probably change pretty soon.
157 00:18:40.040 ⇒ 00:18:41.006 Uttam Kumaran: Okay, actually.
158 00:18:41.490 ⇒ 00:18:49.545 ethanpetersen: I I spoke with our as CEO yesterday. We had our our holiday party, and he
159 00:18:51.430 ⇒ 00:19:08.278 ethanpetersen: yeah, I I was brought in by Mike into this like devrel role, because that was what they had headcount for at the time. Now, in like, in reality, what I I’ve been doing is is partnerships, product and
160 00:19:08.980 ⇒ 00:19:13.869 ethanpetersen: and then some some marketing, but very, very focused
161 00:19:14.460 ⇒ 00:19:21.991 ethanpetersen: marketing that effectively. The problem is that nobody knows about Crusoe in in the space. And so I
162 00:19:22.770 ⇒ 00:19:34.900 ethanpetersen: I’ve been working to working to fix that and and and then building out partnerships with specific.
163 00:19:35.160 ⇒ 00:19:44.218 ethanpetersen: you know, teams and and researchers in in the space as we start to move from just managing infrastructure to actually launching
164 00:19:44.850 ⇒ 00:19:50.919 ethanpetersen: more like true AI products. And then I said.
165 00:19:51.060 ⇒ 00:19:54.079 ethanpetersen: that’s been kind of what what I’ve been doing, and
166 00:19:54.220 ⇒ 00:20:02.490 ethanpetersen: and Mike Matt Craig and and intro me to Craig just bye
167 00:20:03.480 ⇒ 00:20:11.800 ethanpetersen: both being in in AI. And then, Craig, you know, it’s always looking for for new perspectives on on things. And
168 00:20:11.970 ⇒ 00:20:20.120 ethanpetersen: and then I also I very interested in video and have been right
169 00:20:20.120 ⇒ 00:20:22.130 ethanpetersen: and a lot over the last
170 00:20:23.790 ⇒ 00:20:28.039 ethanpetersen: like about about 6 months on on video models and
171 00:20:29.344 ⇒ 00:20:32.656 ethanpetersen: partnerships for for video and and content creation
172 00:20:33.470 ⇒ 00:20:37.460 ethanpetersen: like, what? What sort of video like, what give me? Come a couple of examples.
173 00:20:38.280 ⇒ 00:20:53.269 ethanpetersen: Yeah. So one of the things that we’re trying to bring to market is a video to video type service where you could do you could do 2 things. So one is, you know, upload a video
174 00:20:53.380 ⇒ 00:20:58.560 ethanpetersen: and then prompted and say, You know, replace
175 00:21:00.010 ⇒ 00:21:21.880 ethanpetersen: this this car with the truck, or or restyle the video and then another would be a streaming product where you could do that effectively in in real time. Where then, I think that there’s opportunity in the advertising space for like product placement. And
176 00:21:22.100 ⇒ 00:21:26.560 ethanpetersen: and then let’s really interesting to me.
177 00:21:26.710 ⇒ 00:21:29.590 ethanpetersen: Talking to Craig about this is like.
178 00:21:29.770 ⇒ 00:21:36.090 ethanpetersen: you know, in the in the creative like ideation process, you know if if you could do real time video
179 00:21:36.624 ⇒ 00:21:45.079 ethanpetersen: and then you can just kind of start branching path of the content like, live as as you’re interacting with the person.
180 00:21:45.280 ⇒ 00:21:47.439 ethanpetersen: I think that it could cut down that
181 00:21:47.730 ⇒ 00:21:53.960 ethanpetersen: feedback loop to, you know, ending on on really effective, you know, ads and copy
182 00:21:54.100 ⇒ 00:21:58.550 ethanpetersen: pretty significantly. But I’ve been working on just the like.
183 00:22:00.530 ⇒ 00:22:03.882 ethanpetersen: Kind of foundational research side of like, what do? What do we need?
184 00:22:04.370 ⇒ 00:22:10.049 ethanpetersen: What do we need to figure out in order to make that work, and then and then
185 00:22:11.890 ⇒ 00:22:20.240 ethanpetersen: coming in with the compute to really scale it up and and then on the partnership side, like, Okay, how do we actually
186 00:22:21.060 ⇒ 00:22:23.590 ethanpetersen: bring a product like that to market?
187 00:22:23.590 ⇒ 00:22:23.980 Uttam Kumaran: Yeah.
188 00:22:24.975 ⇒ 00:22:29.289 ethanpetersen: And yeah, so it’s kind of a lot of.
189 00:22:29.290 ⇒ 00:22:31.520 Uttam Kumaran: A lot of work. Big pieces. Yeah.
190 00:22:32.240 ⇒ 00:22:42.899 Uttam Kumaran: yeah. I mean for me, the extent of video that I’ve been following is like all this stuff on Sora. You know, I I followed all the audio generation
191 00:22:43.030 ⇒ 00:22:56.309 Uttam Kumaran: sort of big companies, of course, all the image. But the video stuff has been really cool. I think. You know, I think for me in the past 6 months it’s gone way better, but I’m still only seeing scenes that are like
192 00:22:56.900 ⇒ 00:22:58.350 Uttam Kumaran: seconds long.
193 00:22:58.990 ⇒ 00:23:06.640 Uttam Kumaran: And I have watched a couple of like short AI films. I do think that they’re heavily
194 00:23:06.810 ⇒ 00:23:24.259 Uttam Kumaran: like there’s heavily. I don’t want to know what percentage is like generated by, I think there’s a lot of like heavy after the fact. But I mean, I think there’s a whole. I mean, this is where I tell a lot of people like I have a friend that works he produces, for, like Netflix, and I talk to him a lot
195 00:23:24.330 ⇒ 00:23:45.169 Uttam Kumaran: because he wants to start to do something in in video AI and I, kinda we chat, we chat every few months to kind of give him what the latest is. And basically, I said, is like, Look, people in AI have been pitching like it for it to take over like entire workflows. And that’s not how anything works. And for us we’ve seen success
196 00:23:45.230 ⇒ 00:24:08.899 Uttam Kumaran: internally, like automating our own business, and then, as well as helping clients when we have a clearly defined workflow, and we augment a part of it. Part of it. That’s the most cumbersome. That’s probably the most boring, or that’s the most like ripe for automation. I think that will get wider. But it’s not like a point and shoot like. That’s why even prompting as a concept.
197 00:24:08.920 ⇒ 00:24:16.259 Uttam Kumaran: it’s tough, because people are like. I asked her to do something and do it. I’m like you have 5 words like you didn’t give any requirements.
198 00:24:16.623 ⇒ 00:24:17.349 Uttam Kumaran: So really.
199 00:24:17.350 ⇒ 00:24:31.859 Uttam Kumaran: the game is like requirements building and then the second thing is like, it has to be like a human in the loop, or like a co-pilot system. Right now. I I will say, maybe in the future it becomes more of a like.
200 00:24:32.120 ⇒ 00:24:48.800 Uttam Kumaran: get very, very structured requirements, like very detailed and then kind of like, let it iterate almost kind of like. How I don’t know if you’ve played around with bolt like something that we’re kind of builds on top like, and it kind of figures. It’s it like washes itself, like.
201 00:24:48.880 ⇒ 00:25:02.529 Uttam Kumaran: I could see something like that for video. But I do think that probably what will be happening is, it’s like you can. You can affect clips or short frames. And then that takes on that takes off like 80% of like the work
202 00:25:02.650 ⇒ 00:25:13.604 Uttam Kumaran: for the editor, and instead they focus on getting the timings right and getting the flow right. And that’s what I’ve been telling him is like, I think
203 00:25:14.310 ⇒ 00:25:17.500 Uttam Kumaran: stuff like runway and things like that. I think it’s probably more important
204 00:25:17.610 ⇒ 00:25:20.569 Uttam Kumaran: to just like edit the clips, or like
205 00:25:20.850 ⇒ 00:25:37.519 Uttam Kumaran: Denoise stuff, or like, do the things that the editors probably spend a lot of redundant time doing, and that’s the real effect. And then, in terms of like for him, you know where I was telling it was like dude you, I mean, you just hire less people, or you do more with less. You know whether it’s like a budget effect.
206 00:25:37.992 ⇒ 00:25:59.240 Uttam Kumaran: Or it’s like a hey, we can actually make way more profit because we have produced this. We we brought in someone. We threw every AI tool at them. And then we said, like, you’re the only person on the team. And that’s basically what I said is like, probably gonna happen is, individuals will just have way more leverage. You know.
207 00:26:00.250 ⇒ 00:26:03.440 ethanpetersen: Oh, yeah, yeah, I I think
208 00:26:03.920 ⇒ 00:26:10.624 ethanpetersen: I pretty much see all. I’m not super bullish yet.
209 00:26:11.979 ⇒ 00:26:20.220 ethanpetersen: but probably soon on on some of the agent stuff, but until
210 00:26:21.920 ⇒ 00:26:25.090 ethanpetersen: until a couple of techniques are are scaled up.
211 00:26:25.380 ⇒ 00:26:27.745 ethanpetersen: I I think that it’s all
212 00:26:28.400 ⇒ 00:26:35.800 ethanpetersen: you know, effectively, just 10 x augment augmentation. I do think I mean, the big
213 00:26:36.090 ⇒ 00:26:39.633 ethanpetersen: big problem with video right now is,
214 00:26:41.130 ⇒ 00:26:45.589 ethanpetersen: prompt adherence is just super low compared to other
215 00:26:46.083 ⇒ 00:27:10.810 ethanpetersen: like images, for example, like images. Now you know you you have. That they just they follow the prompt much better. If you want something in a particular region. You can. You can probably get that, or there are techniques to be able to in paint and then have something generated like, you have a lot of control there. And video doesn’t have that control yet. But there, there are.
216 00:27:11.030 ⇒ 00:27:13.069 ethanpetersen: There are a couple of approaches. I’m
217 00:27:13.530 ⇒ 00:27:16.979 ethanpetersen: I’m investing in that. I think I think that’ll get fixed
218 00:27:17.550 ⇒ 00:27:19.941 ethanpetersen: certainly within the next 6 months.
219 00:27:20.500 ⇒ 00:27:24.559 ethanpetersen: I ideally within the next 3 cause.
220 00:27:26.620 ⇒ 00:27:34.840 ethanpetersen: Yeah, I’m gonna try to be scaling up some some approaches that I think will enable like
221 00:27:36.630 ⇒ 00:27:41.200 ethanpetersen: you you to ground events in video like right right now, the
222 00:27:43.070 ⇒ 00:27:53.539 ethanpetersen: I I think one of the weaknesses is that they can’t. It’s very, very hard to tell story with generated video where things flow from action to action, to action.
223 00:27:53.740 ⇒ 00:27:58.480 ethanpetersen: And it’s there are.
224 00:27:59.020 ⇒ 00:28:05.040 ethanpetersen: There are 2 approaches. I’ve seen that I that I like that, I think, can fix that
225 00:28:06.630 ⇒ 00:28:15.706 ethanpetersen: But it. It’s very much, in my opinion. It’s where like Gpt. 3 was that before it?
226 00:28:16.580 ⇒ 00:28:21.699 ethanpetersen: it got much better, and made the jump to 3.5 and chat qpt. And
227 00:28:21.810 ⇒ 00:28:25.280 ethanpetersen: and as soon as it makes that jump, then it’s gonna be insane.
228 00:28:25.440 ⇒ 00:28:25.870 Uttam Kumaran: Yeah.
229 00:28:26.331 ⇒ 00:28:32.790 ethanpetersen: It’s still not gonna be like replacing anything but but definitely like super ubiquitous.
230 00:28:32.790 ⇒ 00:28:33.749 Uttam Kumaran: Those are available.
231 00:28:34.870 ⇒ 00:28:35.375 ethanpetersen: Yeah.
232 00:28:36.170 ⇒ 00:28:47.271 ethanpetersen: have have you seen? What? What’s the adoption been been like for video amongst folks in in your world? Are are people just using it in like a
233 00:28:48.258 ⇒ 00:28:52.830 ethanpetersen: ideation and and expiration mode? Or have you seen
234 00:28:53.437 ⇒ 00:28:59.780 ethanpetersen: anyone actually ship like, you know, advertisements or or even just like.
235 00:29:00.080 ⇒ 00:29:01.119 Uttam Kumaran: Yeah, so in.
236 00:29:01.120 ⇒ 00:29:01.880 ethanpetersen: Soccer.
237 00:29:01.880 ⇒ 00:29:13.420 Uttam Kumaran: Yeah, in my world. Where. So we work with a lot of like, e-commerce. And so I follow a lot of like Ecom advertising, or like Cpg advertising people, I think the biggest thing is like.
238 00:29:14.211 ⇒ 00:29:23.859 Uttam Kumaran: image is still like very much where there’s a lot of action. I think. Now, you basically have the ability to input in your product and
239 00:29:23.880 ⇒ 00:29:45.819 Uttam Kumaran: get out stock images of it placed like you don’t need to do shoots like that and I’m happy to. I’ll I’ll send you a couple of things after this where basically they have like you could you could have like, for example, you’re selling like something like this. You could have it stay with like whatever you need, and you can run that entire ideation process without ever needing
240 00:29:46.173 ⇒ 00:30:11.269 Uttam Kumaran: to hire like a designer, basically, or what you do is you get a 90% there and then you have the designer fix it in in Photoshop. So that’s where I’m like, you basically don’t need I think if you’re if you’re a founder or you’re like ahead of marketing, and you have the tools. I think you can start to design that. The problem is nobody on the business side can code. So anything that’s like.
241 00:30:11.270 ⇒ 00:30:11.760 ethanpetersen: Me!
242 00:30:11.760 ⇒ 00:30:36.040 Uttam Kumaran: You have to write. A lick of code is like out the window, and there’s no marketing engineers. So there’s not many people like you, or maybe even like me, who are like, I’m down to write code to generate images like that’s not a role in the enterprise, really, or in the marketing department. Like designers don’t know to code frequently, if they do, they probably switch jobs by now, you know. So.
243 00:30:36.040 ⇒ 00:30:36.430 ethanpetersen: Hmm.
244 00:30:36.430 ⇒ 00:30:50.450 Uttam Kumaran: I think, like one is like these. All need to be no code, and then maybe low code. After a while, and then I think one. But like stock images and stuff, I’m seeing it a lot. I think an advert on video. It’s still short, like it’s like
245 00:30:51.170 ⇒ 00:31:01.010 Uttam Kumaran: stock people, things like that. But I still think we haven’t gotten over the maybe it’s because I just watch this stuff all day. It’s like I can kind of tell. But I do think that for
246 00:31:01.060 ⇒ 00:31:23.720 Uttam Kumaran: for, like nature scapes and like large things, it’s definitely can do it. The only thing is like, I still think none of this is distributed like nobody knows about all this stuff, and until it like makes its way off of twitter I guess, or like outside of like people like us talking. It doesn’t get any adoption. So that’s why for us, you know.
247 00:31:23.780 ⇒ 00:31:45.216 Uttam Kumaran: I pray for for me. The thing is like we try to go at the executive level. And like, I just just try to show people what the sort of technology works. But a lot of the work we do is actually like education. You know, a lot of the time like. So so my background is in data engineering. We started off about a year and a half ago, primarily doing
248 00:31:45.710 ⇒ 00:31:49.869 Uttam Kumaran: data engineering data modeling bi work, mostly
249 00:31:50.040 ⇒ 00:32:10.530 Uttam Kumaran: data engineering data modeling on snowflake and then I started using AI just to automate my own business. So, like anytime, I had to do copy. I wanted to build like more sales functions with with AI. We started building some internal agents to handle like lead research. And you know, writing blogs and stuff like that. And then I was like, this stuff is like
250 00:32:10.670 ⇒ 00:32:16.326 Uttam Kumaran: it’s it’s very flashy on that surface. But then there’s not like a well defined stack.
251 00:32:17.020 ⇒ 00:32:23.990 Uttam Kumaran: there’s definitely for production use cases. There’s only a couple of big players for evaluation and testing that work.
252 00:32:24.434 ⇒ 00:32:29.020 Uttam Kumaran: And then I’m like, if I’m having, I’m fairly technical. If I have to figure this out, then
253 00:32:29.240 ⇒ 00:32:47.080 Uttam Kumaran: people who actually want to adopt this at scale, which are the companies that may not have staffing towards this, they’re gonna need a help. So we started doing more AI work in the last 3 months. A lot of what we’ve been doing is more on like knowledge based rag like more like web research.
254 00:32:47.389 ⇒ 00:33:10.879 Uttam Kumaran: Sort of like on the sales and marketing side. But for us. And the reason why I I had of Craig is, I was like, we’re gonna continue to niche down in our expertise like we’re gonna get logos and successes from just basically automating like stuff you probably could have done in Zapier, like maybe like zapier plus, plus, you know, like really hooking things together and like using lms to translate and summarize and
255 00:33:10.880 ⇒ 00:33:30.530 Uttam Kumaran: and rich. But the stuff we want to go towards is things in the department of like fine tuning. I want to get into content image and video, really enabling, like helping the non technical person. Adopt some of these tools right? And then, you know, kind of almost get to the point where we’re able to
256 00:33:30.835 ⇒ 00:33:44.869 Uttam Kumaran: host our own open source. Llms train our own Llms and kind of just get more technical like, I wanna kind of leave this sort of surface level agent building because there’s a lot of people doing that. And it’s very easy to grasp like
257 00:33:45.195 ⇒ 00:33:52.520 Uttam Kumaran: it’s just chaining functions together. It’s like a very easy thing. So for me. It’s like, I think about, where are the areas where
258 00:33:52.600 ⇒ 00:34:20.489 Uttam Kumaran: a company like ours, who brings that technology and expertise to clients? We basically we don’t build the shovels, we just dig holes. And so for me, the the thing is like, can I? I go to clients? I’m like, I’m in the market, find the best shovels for you. I don’t get paid by any shovel company. I have my favorite that make my job easier. You may have bring yours. But for us it’s like, How do we dig holes fast?
259 00:34:21.006 ⇒ 00:34:22.670 Uttam Kumaran: For you? And so
260 00:34:22.780 ⇒ 00:34:37.600 Uttam Kumaran: content is something that we’ve been trying to do internally, like we’ve almost automated a lot of our text based content, pipeline with. But I’m still involved to review and do and ideate like kind of on the on the ends, but the middle part is mostly taken care of.
261 00:34:37.630 ⇒ 00:34:59.180 Uttam Kumaran: We also we have a full time designer, and she uses AI tools for all of our illustrations which which have come out really, really great. And I also have our, the one of our guys who’s doing content. He also uses a lot of AI tools. So I learned how to train some of the folks to do that who are non technical themselves. The next thing we want to get into is audio.
262 00:34:59.440 ⇒ 00:35:03.719 Uttam Kumaran: So we’re gonna try to do some of the 11 labs podcasts, like, turn.
263 00:35:03.720 ⇒ 00:35:04.050 ethanpetersen: To hear.
264 00:35:04.050 ⇒ 00:35:29.500 Uttam Kumaran: Blog, content to audio content. I definitely wanna try video. I don’t know like we don’t run any ads. So I don’t know exactly like where it’s gonna fit into our strategy. But I want to try to run video, but it’s work. It’ll work for us, you know. It’s all the tools are there. I just think it’s not easily distributed, and they’re all a little bit tough to use. So that’s kind of like what what we’re what we’re seeing.
265 00:35:30.720 ⇒ 00:35:33.826 ethanpetersen: Yeah, what? What’s the tooling look like?
266 00:35:34.450 ⇒ 00:35:39.569 ethanpetersen: they use currently for? Let’s, let’s say, like images. For for example.
267 00:35:40.040 ⇒ 00:35:54.479 Uttam Kumaran: Yeah, so right now we’re using I mean, mainly everything we have stored in figma. And then so we use let me get you the tool that we use for actually like building these like illustrations.
268 00:35:56.820 ⇒ 00:36:03.689 Uttam Kumaran: it’s called yeah, I mean, we’re using. I think our team is using free pick
269 00:36:04.487 ⇒ 00:36:14.120 Uttam Kumaran: which is basically like kind of like a quick design tool for generating stock images and and AI sort of images.
270 00:36:14.940 ⇒ 00:36:17.780 Uttam Kumaran: I don’t know if you’ve checked them out before, but
271 00:36:18.330 ⇒ 00:36:19.970 Uttam Kumaran: I think they’re kind of like
272 00:36:20.260 ⇒ 00:36:24.082 Uttam Kumaran: sort of growing among the the designer community.
273 00:36:24.890 ⇒ 00:36:38.319 Uttam Kumaran: but again, you’re gonna go to that. And you’re gonna see, like a lot of it is just like making it accessible for designers. And then also making the integrations nice, like you can export this as Psd files like.
274 00:36:38.736 ⇒ 00:37:05.310 Uttam Kumaran: I don’t know. It’s like a tool like this that’s gonna come up and help people and then also, because again, not they’re not go. Real designers are not like going in here and being like cool. I got it. I’m done. They’re gonna they’re gonna get 10. And then they’re gonna go to figma. And they’re gonna work, or they’re gonna go into Photoshop and work. And so I think a tool like free pick where it has basically like, you start to ideate you input your assets, and then you can get out.
275 00:37:05.702 ⇒ 00:37:10.387 Uttam Kumaran: You know the the right file formats. You need that this has been working for us.
276 00:37:10.830 ⇒ 00:37:22.880 Uttam Kumaran: we haven’t rolled anything else like really custom on our side. But what what we do have is we have a very specific design, like A, basically, we have, like a
277 00:37:24.280 ⇒ 00:37:33.690 Uttam Kumaran: like a design file with all of our fonts, all of our colors, everything. So like a very specific. So that that becomes a requirements file for like a lot of this. So the AI doesn’t
278 00:37:34.020 ⇒ 00:37:49.330 Uttam Kumaran: knows like. And we also have, like a brand book which is like, how do we speak the type of content? So all of that is like the foundation building that helps like any AI content generation get way better. So yeah, I think our team is using Pre pick for stuff mostly.
279 00:37:51.260 ⇒ 00:37:53.950 ethanpetersen: Gotcha. Yeah, I’m curious. What?
280 00:37:54.140 ⇒ 00:38:01.100 ethanpetersen: So are, are you guys contracting for for crocs? Is that how you know Craig or.
281 00:38:01.100 ⇒ 00:38:28.500 Uttam Kumaran: Yeah. So I got put in touch with Craig through a friend of mine in New York. He runs a company called Wild. They are a like a data clean room solution. And no, I I he just was like yo call it’s wilde.ai it’s run by my friend Clint Dunn. And yeah, no, he just put me in touch with Craig, and we just chatted. Originally, I was like, Okay, maybe they’re interested. And I met Craig, and I could tell. Craig is just like a
282 00:38:28.600 ⇒ 00:38:33.520 Uttam Kumaran: you know. He’s kind of like a super connector, and also very interested in just like how to use AI.
283 00:38:33.570 ⇒ 00:39:00.340 Uttam Kumaran: And you know, I I think initially, I was like, Okay, cool, like, maybe there’s some good to go do for crocs. But honestly, it’s just been helpful to meet people through him and sort of like, help him get a sense of like, what’s going on. So yeah, we’re just friends like I. We’re just kind of like, I kind of explain, sort of like what we’re seeing. And you know, he runs a couple of different, like networking sort of groups and things like that. So he’s also trying to figure out like what he wants to do next.
284 00:39:00.420 ⇒ 00:39:05.809 Uttam Kumaran: He kind of got into clock and was like crocs, and he’s like they tasked me with doing. AI. I’m trying to figure out
285 00:39:06.230 ⇒ 00:39:14.920 Uttam Kumaran: so he’s like now that I have crocs behind me. I can get any meeting I want, and I think he’s taking advantage of that, and it’s helped him accelerate. And
286 00:39:15.180 ⇒ 00:39:20.040 Uttam Kumaran: so but for me, it’s like, I want to talk to the people who are at his level, who are like.
287 00:39:20.250 ⇒ 00:39:24.280 Uttam Kumaran: I need to allocate budget and actually make impact.
288 00:39:24.430 ⇒ 00:39:42.779 Uttam Kumaran: But there’s nobody in my, not only is there nobody in my org? You can’t like Google these problems like, you know. And and even as you mentioned that there’s not a stack for anything like even for us, on where I’m trying to get the team to run anytime. We build an agent or something. I’m like we need to run evals. We need to have testing trace. There’s like
289 00:39:42.890 ⇒ 00:40:00.150 Uttam Kumaran: there’s still not like a really like defined winner in that space, like, who? Where’s the data dog for trace. Like, you know, there’s like, there’s a couple of people that are winning. But then I’m like, are we gonna invest in some in like the newest startup that’s like.
290 00:40:00.200 ⇒ 00:40:20.090 Uttam Kumaran: you know. So I’m like, No, so is there anybody that’s been doing this for like at least a year. And then I could consider because we’re making decisions on behalf of clients. So that’s the only thing I tell clients like we’re not gonna go for just the flashy stuff, because all a lot of these startups are not gonna exist next year for one reason or another. And so for us, it’s I’m trying to get a sense
291 00:40:20.990 ⇒ 00:40:37.550 Uttam Kumaran: for what parts of the AI industry are like reaching any level of maturity. Like, I think, for text generation, we’re in a really solid place. Text generation summarization, writing, content at scale. All that stuff, I think, also for research, like being able to
292 00:40:37.840 ⇒ 00:40:54.153 Uttam Kumaran: hook into a bunch of Apis to get information to summarize that which is a huge problem for folks. I think voice is getting figured out so you can think like call centers, or just building, like inbound people to take calls. That’s, I think, pretty really, really solid, too.
293 00:40:54.760 ⇒ 00:40:57.130 Uttam Kumaran: I think. Still, to be figured out is like.
294 00:40:57.690 ⇒ 00:41:03.159 Uttam Kumaran: I think the really, the stars with a lot of budget probably build their own testing framework and eval framework.
295 00:41:03.350 ⇒ 00:41:07.100 Uttam Kumaran: I can’t like afford to do that. So we have to get something off the shelf.
296 00:41:07.230 ⇒ 00:41:11.629 Uttam Kumaran: And so that’s like, kind of more about what I’m thinking about is, how do we actually deploy some of this stuff
297 00:41:11.740 ⇒ 00:41:14.529 Uttam Kumaran: to production. You know that that works
298 00:41:16.900 ⇒ 00:41:21.939 ethanpetersen: Yeah, do you use brain trust for emails or.
299 00:41:21.940 ⇒ 00:41:25.129 Uttam Kumaran: No, we were thinking of using arise
300 00:41:27.260 ⇒ 00:41:29.910 Uttam Kumaran: arise a RIZE
301 00:41:32.096 ⇒ 00:41:44.129 Uttam Kumaran: but I would. I wanted to ask you like what you think is like a stack for Eval and testing cause. I I we need to make a vendor decision soon. We were also talking to trace loop
302 00:41:47.340 ⇒ 00:41:52.349 Uttam Kumaran: You may have heard of them, and then arise.
303 00:41:52.880 ⇒ 00:41:59.920 Uttam Kumaran: I arise. A friend of mine mentioned that he was. It was really really solid for them, and they have an open source. Eval framework called Phoenix.
304 00:42:01.529 ⇒ 00:42:02.569 Uttam Kumaran: Arise.
305 00:42:02.570 ⇒ 00:42:02.900 ethanpetersen: Hmm.
306 00:42:02.900 ⇒ 00:42:07.990 Uttam Kumaran: Just based on their website. I trust them a lot more seem a lot more boring and a lot more enterprise. So I like.
307 00:42:09.649 ⇒ 00:42:17.549 Uttam Kumaran: I’m like, the less the less flashy the website is. I’m actually way more interested because
308 00:42:18.130 ⇒ 00:42:25.549 Uttam Kumaran: I don’t know. I work. You know, we have design, too. You can make any sort of landing page look really good, and the product sucks so.
309 00:42:27.490 ⇒ 00:42:31.839 ethanpetersen: Yeah, I I think for for emails.
310 00:42:32.380 ⇒ 00:42:43.530 ethanpetersen: Excuse me, the folks that I oh, I typically, I write.
311 00:42:46.280 ⇒ 00:42:59.460 ethanpetersen: we could evaluary testing framework but if I if I were running a company and I needed to actually ship something and to use it I would.
312 00:42:59.940 ⇒ 00:43:03.619 ethanpetersen: Probably I know these guys?
313 00:43:05.192 ⇒ 00:43:07.450 ethanpetersen: And I know.
314 00:43:07.640 ⇒ 00:43:16.110 ethanpetersen: like in in my circles, I’ve heard great things about their developer experience. And then they’re done.
315 00:43:16.270 ⇒ 00:43:20.779 ethanpetersen: Question really would be, well, what’s what’s their longevity look like?
316 00:43:21.130 ⇒ 00:43:30.500 ethanpetersen: And they I think they just recently raised their last round. But the the main signal is that
317 00:43:31.110 ⇒ 00:43:34.996 ethanpetersen: I believe that open AI is a big user.
318 00:43:35.430 ⇒ 00:43:36.540 Uttam Kumaran: Oh, okay.
319 00:43:36.540 ⇒ 00:43:42.907 ethanpetersen: So as long as like, you know, open AI uses it. Then, you know, they’re gonna be.
320 00:43:44.260 ⇒ 00:43:46.469 ethanpetersen: They’re gonna be fine.
321 00:43:46.720 ⇒ 00:43:55.669 ethanpetersen: They, I mean, they’re very responsive to people as far as like
322 00:43:59.817 ⇒ 00:44:04.119 ethanpetersen: if if anyone is having issues on Twitter.
323 00:44:04.120 ⇒ 00:44:23.979 Uttam Kumaran: Okay. I mean, that’s a good sign like, that’s the stuff I’m looking for is like. So in the data world, I’m like, incredibly plugged in that I know what tools are like, Fugazi, and like what tools are real, I know, like a lot of the people in the AI world for me. All my knowledge is from Twitter. So I’ve learned everything from Twitter
324 00:44:24.342 ⇒ 00:44:34.890 Uttam Kumaran: but still, like I feel like the stack isn’t like super defined. I mean. The people that really want on Twitter are like laying chain and stuff like that. But it’s like a big hobbyist group.
325 00:44:35.050 ⇒ 00:44:40.539 Uttam Kumaran: And I’m like this is, we’re not. I’m not building stuff that can break every 5.
326 00:44:41.520 ⇒ 00:44:49.839 Uttam Kumaran: So and that’s also, you know, that’s the trouble is like, even in hiring an AI. The people have on my team. They’re used to building stuff, but
327 00:44:50.000 ⇒ 00:45:03.589 Uttam Kumaran: they’ve never worked on like a strict engineering team because a lot of backgrounds in AI aren’t typically software engineers. They’re like ex marketers or ex zapier people who kind of like figure it out stuff.
328 00:45:03.790 ⇒ 00:45:20.049 Uttam Kumaran: So that’s that is the fun. And the challenge is that I have, like a few people internally that are really really good, with building agents like understanding, like everything about building AI, but ha! Have never worked in like a strict engineering team with like
329 00:45:20.310 ⇒ 00:45:26.699 Uttam Kumaran: testing and like Pr reviews. And a lot of them don’t know how to code, really. Well, you know, they’re more like.
330 00:45:26.700 ⇒ 00:45:27.050 ethanpetersen: Whoa!
331 00:45:27.050 ⇒ 00:45:29.999 Uttam Kumaran: People. And so for me, I’m like
332 00:45:30.340 ⇒ 00:45:50.462 Uttam Kumaran: this is, gonna be a nightmare if we try to ship something in into production. So I need to. I’m trying to teach everybody like, because on the data side this week we do this all the time. But it’s like we’re trying to. And I wanted to have a stack that works where everything we ship goes through this process of testing and Eval. And then we have observability.
333 00:45:50.890 ⇒ 00:46:07.630 Uttam Kumaran: and then, as we start to get more technical and do things like image voice like, how do we think about evaluation and testing for those products? And it’s like, it’s not good enough for us to just say that the stuff works. It’s like, I want data to back
334 00:46:08.160 ⇒ 00:46:13.780 Uttam Kumaran: accuracy, and and then also that forces our requirements gathering to get way better, you know, like
335 00:46:15.530 ⇒ 00:46:29.169 Uttam Kumaran: So there’s a couple of things that are like on my mind. I mean, yeah, I guess, like, you know, to kind of tell you like where we’re at like where one like, I love having conversations like this so would love to keep in touch and kind of even, just like share
336 00:46:29.530 ⇒ 00:46:38.733 Uttam Kumaran: the types of clients. We’re working on things we’re doing. But I also know that, like you started a company before. You know, like how challenging it is, and
337 00:46:39.615 ⇒ 00:46:40.189 Uttam Kumaran: right?
338 00:46:40.190 ⇒ 00:46:54.520 Uttam Kumaran: It’s a it’s like insane. I’m thankful that, like I work in it like I’m an engineer and I work mostly with engineers. I unfortunately have to become more of like a business person recently, but I would love for us
339 00:46:54.610 ⇒ 00:47:20.299 Uttam Kumaran: to get from, you know, working on, you know, in on the, in, the, on the data side, we’re working on more technical stuff on AI side, I want to start working on stuff that’s actually like really challenging and take on problems for people that are not things that you can find a little twitter template for and get your way through, and then also be able to recruit the best people right? And so like making sure we have a stack and a framework that works is super important. But
340 00:47:20.380 ⇒ 00:47:33.629 Uttam Kumaran: that’s kind of like where we’re at. So you know, one thing is I wanna I think, hearing from you that like content and video. And even this discussion makes me feel like there’s a lot of meat in like content and video. Still to do.
341 00:47:33.750 ⇒ 00:47:51.260 Uttam Kumaran: But yeah, like curious like, are you? Are you like in the market? Are you still focused on like Rousseau stuff like, are you open to like part time stuff like if I I don’t think we have immediate stuff. But I think I just asked. I’ll just ask the question out loud. So yeah.
342 00:47:51.647 ⇒ 00:48:00.169 ethanpetersen: Yeah, no, I I appreciate it. I to be honest, I so I’m trying to get a couple of
343 00:48:00.350 ⇒ 00:48:12.720 ethanpetersen: projects over over the board or over the finish line. And then I my, my one year mark is is March and coming up to March, I’m
344 00:48:12.880 ⇒ 00:48:23.740 ethanpetersen: I’m gonna be exploring a bit. I I like what I currently do at Crusoe. But I I am starting to get a little
345 00:48:26.240 ⇒ 00:48:27.240 ethanpetersen: no, the
346 00:48:28.760 ⇒ 00:48:37.019 ethanpetersen: I I’m in an interesting spot where I somehow I I’ve become. I’m teamless within the org.
347 00:48:37.180 ⇒ 00:48:46.269 ethanpetersen: and I do a lot of different things. And I’m working with like our our senior team.
348 00:48:46.674 ⇒ 00:48:50.475 ethanpetersen: But nobody knows what to do with me. And then, well.
349 00:48:51.254 ⇒ 00:48:57.079 ethanpetersen: guys, if if I want to have this level of autonomy and things. I would do it. I would do a startup.
350 00:48:57.410 ⇒ 00:49:02.210 ethanpetersen: And then, if I’m if I’m taking on this amount of stress and responsibility
351 00:49:02.880 ⇒ 00:49:05.820 ethanpetersen: like I, I would rather participate in
352 00:49:06.180 ⇒ 00:49:12.590 ethanpetersen: the, you know, commensurate amount of upside again, and.
353 00:49:12.590 ⇒ 00:49:21.319 Uttam Kumaran: 100 dude. Yeah, I mean, but you know also, like I don’t know. I I feel like I’m sure you have the itch again to kind of do stuff, especially
354 00:49:21.550 ⇒ 00:49:35.930 Uttam Kumaran: for me. Like it. In some ways. I would love to go now that I’ve spent a lot of time in AI over the past year and a half. I’m like shit. It’d be nice to go work at this one of these companies for like 6 months, just to get a sense of like what’s really working, and then come back.
355 00:49:36.797 ⇒ 00:49:41.349 Uttam Kumaran: But for me, I get that. They’re like just chatting with people and sort of being like
356 00:49:41.660 ⇒ 00:49:58.889 Uttam Kumaran: any interest in coming, working for us and or like, and but the nice thing about you know I I never wanted to be a consultant. I actually, I’ve hired a bunch of, and I always had a bad, you know. I’ve always like never liked consulting so it’s very funny that I’m in like services business.
357 00:49:59.591 ⇒ 00:50:10.999 Uttam Kumaran: But I don’t know. I I feel like we’re an engineering company. I think our goal is to be the best communicators and the best engineers that any of our clients have on their team.
358 00:50:11.300 ⇒ 00:50:29.300 Uttam Kumaran: And so for me, I think a lot about like the quality of our work. I think also, like we can be on the cutting edge, and especially when there’s a big wave like AI where there’s so much going on, it pays dividends to be on the edge. It takes a lot of work, and you have to read super widely. Have to be very plugged in, but
359 00:50:29.480 ⇒ 00:50:53.340 Uttam Kumaran: it pays to kind of be on the edge. And then there’s certainly people that need this sort of stuff that nowhere to go. The vendors are gonna sell you that they can do everything right. And then the only other people. And we wrote a blog post about this is like Mckinsey Bain. They’re gonna and they’re gonna sell you. And Craig even told me like they they hired one of those guys and they come and they build a bunch of strategy decks, and they don’t do any implementation
360 00:50:53.710 ⇒ 00:50:57.909 Uttam Kumaran: and and then the other last thing is like all the talent is locked up in
361 00:50:58.170 ⇒ 00:50:59.919 Uttam Kumaran: the startup sort of thing.
362 00:51:00.210 ⇒ 00:51:25.410 Uttam Kumaran: right? And still, none of those guys are building like they’re not implementing. There’s nobody like implementing this stuff or like working with clients to like, do that. None of these companies that professional services arms you call them. They’ll help you implement it. But it’s like a software engineer on their team who’s like taking time out of this day to like help you implement. So that’s how I kind of think about. You know this stuff and and our company. But sorry I interrupted a bit.
363 00:51:26.430 ⇒ 00:51:51.489 ethanpetersen: Oh, no, you’re you’re good. Yeah. Yeah. So might might be open definitely, and would love to chat on on that part, probably in in like Jan, Jan, Feb. And then on the on the content creation. I I got maybe some some additional insight that might be helpful, because everyone like the the big question is, is anyone making money in the
364 00:51:53.245 ⇒ 00:52:03.830 ethanpetersen: so runway is probably, I think, from what I’ve heard they’re the
365 00:52:04.680 ⇒ 00:52:16.189 ethanpetersen: the leader in revenue right now, and and they jumped from under 10 million arr to about 80 to 90, with the launch of the Gen. 3 model
366 00:52:17.043 ⇒ 00:52:20.260 ethanpetersen: and then everyone else’s.
367 00:52:21.810 ⇒ 00:52:28.559 ethanpetersen: you know, a bit a bit below that luma they
368 00:52:30.030 ⇒ 00:52:38.939 ethanpetersen: they got crushed by runningways. Gem 3 but it’s possible that they so so Luma just
369 00:52:39.200 ⇒ 00:52:48.040 ethanpetersen: announced at aws reinvent a deal with Amazon where it’s effectively like access to
370 00:52:48.160 ⇒ 00:52:55.130 ethanpetersen: Loomis models in exchange for for compute, and then everyone is speculating.
371 00:52:56.840 ⇒ 00:53:00.760 ethanpetersen: You know, and spotlight the.
372 00:53:00.900 ⇒ 00:53:07.139 ethanpetersen: It was probably because Luma struggled fundraising and needed.
373 00:53:08.850 ⇒ 00:53:18.789 ethanpetersen: Amazon probably took advantage of the the situation there. But but runway I know I have that that data point, that they’re round
374 00:53:19.730 ⇒ 00:53:25.630 ethanpetersen: somewhere in that 70 to 90 millionaire harp with a with a pretty high growth rate.
375 00:53:25.630 ⇒ 00:53:26.010 Uttam Kumaran: Okay.
376 00:53:26.623 ⇒ 00:53:30.570 ethanpetersen: And then the Oh.
377 00:53:31.050 ⇒ 00:53:35.720 ethanpetersen: you know I I think there’s a lot of expectations, especially now that Sora is finally released.
378 00:53:37.423 ⇒ 00:53:42.136 ethanpetersen: That next year with with video, it should be
379 00:53:43.350 ⇒ 00:53:54.929 ethanpetersen: yeah, it. It should be pretty massive. And things are finally starting to get to a quality that can be useful and and and drive value.
380 00:53:55.110 ⇒ 00:53:55.730 ethanpetersen: but it’s.
381 00:53:55.730 ⇒ 00:54:05.929 Uttam Kumaran: Yeah, for Sora. It’s more like it, for the developers having access to that is great, because open AI is not going to come up with the consumer facing
382 00:54:06.160 ⇒ 00:54:15.569 Uttam Kumaran: video editor tool. But they will release the best and most and cheapest Api available. Right? And so for that, it’s like.
383 00:54:16.030 ⇒ 00:54:21.309 Uttam Kumaran: I think it’ll be great. I totally agree. I mean, it’s I’m surprised that this delayed
384 00:54:22.630 ⇒ 00:54:24.889 Uttam Kumaran: but I mean it is hard.
385 00:54:25.090 ⇒ 00:54:36.340 ethanpetersen: I actually, I got some insight into that. So I met last week at the conference. tpm, at Openai. That does their red teaming.
386 00:54:36.460 ⇒ 00:54:41.129 ethanpetersen: And he said, Yeah, he’s the one that delayed Sora.
387 00:54:42.170 ⇒ 00:54:48.409 ethanpetersen: but it it was effectively, for I mean the election was part of it.
388 00:54:48.800 ⇒ 00:54:49.470 Uttam Kumaran: Yeah.
389 00:54:49.780 ⇒ 00:54:59.509 ethanpetersen: And then the some some broader, like geopolitical events.
390 00:55:00.147 ⇒ 00:55:02.859 ethanpetersen: So so potentially something like, you know.
391 00:55:03.200 ⇒ 00:55:05.940 ethanpetersen: when you see the models and different things. That’s
392 00:55:06.403 ⇒ 00:55:12.800 ethanpetersen: at least, that’s my understanding right now is like, that’s a big factor for.
393 00:55:12.980 ⇒ 00:55:14.580 Uttam Kumaran: Because they don’t want to get playing release.
394 00:55:14.580 ⇒ 00:55:18.880 Uttam Kumaran: We don’t wanna get yeah, I guess for them, they’re they’re in like playing politics. Basically. Now.
395 00:55:19.240 ⇒ 00:55:23.860 Uttam Kumaran: you know, they’re they’re they’re effectively like the biggest fish to fry
396 00:55:25.120 ⇒ 00:55:27.520 Uttam Kumaran: outside of like saying right,
397 00:55:28.000 ⇒ 00:55:34.079 ethanpetersen: And global. So like implications within like, like the Ukraine, Russia war.
398 00:55:34.080 ⇒ 00:55:34.790 Uttam Kumaran: Oh, yeah.
399 00:55:35.359 ⇒ 00:55:38.040 ethanpetersen: So that was autumn.
400 00:55:38.540 ⇒ 00:55:41.940 ethanpetersen: That was another element of it. From from what I understand.
401 00:55:41.940 ⇒ 00:55:42.880 Uttam Kumaran: Interesting.
402 00:55:43.580 ⇒ 00:55:48.319 ethanpetersen: And then unrelated to that. But I wanted to share another framework that might be
403 00:55:48.700 ⇒ 00:55:54.769 ethanpetersen: for you. So this is it’s it’s a bit earlier in in its life cycle.
404 00:55:55.525 ⇒ 00:56:01.200 ethanpetersen: But it’s the the guy that wrote it maintains it.
405 00:56:01.640 ⇒ 00:56:03.779 ethanpetersen: He hates Lane chain.
406 00:56:04.010 ⇒ 00:56:06.560 Uttam Kumaran: Yeah, I didn’t see and preach.
407 00:56:06.890 ⇒ 00:56:13.819 ethanpetersen: Yeah, he! He! His entire mission with that package is to kill Lane Ching.
408 00:56:15.228 ⇒ 00:56:18.811 ethanpetersen: And and he’s great. I I know him.
409 00:56:19.560 ⇒ 00:56:24.392 ethanpetersen: Actually, this is a complete tangent. But
410 00:56:25.530 ⇒ 00:56:29.310 ethanpetersen: I know him because an engineer that works for me
411 00:56:30.128 ⇒ 00:56:37.120 ethanpetersen: interviewed with him. And then and then he called me asking for a reference about the guy.
412 00:56:37.320 ⇒ 00:56:38.840 ethanpetersen: and I was like dude.
413 00:56:39.510 ⇒ 00:56:45.779 ethanpetersen: I can’t say anything and and that’s all I can say about it.
414 00:56:46.510 ⇒ 00:56:47.235 Uttam Kumaran: Good.
415 00:56:48.430 ⇒ 00:56:53.800 ethanpetersen: That engineer like that was another factor of imploding my company like the guy.
416 00:56:53.800 ⇒ 00:56:54.380 Uttam Kumaran: Checked.
417 00:56:54.760 ⇒ 00:56:56.200 ethanpetersen: It it was
418 00:56:56.680 ⇒ 00:57:01.008 Uttam Kumaran: You’re very kind to not saying anything like
419 00:57:03.060 ⇒ 00:57:04.670 Uttam Kumaran: you set it up. You set it up.
420 00:57:05.400 ⇒ 00:57:15.722 ethanpetersen: Yep, and and then ever since then, you know, we we’ve been friends, but he’s so. He was an ex research scientist at open AI
421 00:57:16.330 ⇒ 00:57:19.080 ethanpetersen: super super break guy
422 00:57:19.646 ⇒ 00:57:29.649 ethanpetersen: but his mission has been like I killed lang chain lang chain needs to die because it’s this over engineered piece of shit and
423 00:57:30.120 ⇒ 00:57:37.969 ethanpetersen: is just I mean it won because it was 1st right, like everyone that wanted to build kind of some sort of a
424 00:57:38.180 ⇒ 00:57:46.620 ethanpetersen: framework where you can codify your prompts, and and the flow of everything like link chain was just. It was the 1st
425 00:57:47.226 ⇒ 00:57:52.379 ethanpetersen: and then because other people start using and then writing documentation around it. It just got that like
426 00:57:52.550 ⇒ 00:57:53.400 ethanpetersen: critical. Now.
427 00:57:53.400 ⇒ 00:58:01.819 Uttam Kumaran: But it’s like not written. But no like no, like real engineers are like. It’s always like other people. I don’t even know who developed it like
428 00:58:02.060 ⇒ 00:58:03.899 Uttam Kumaran: it didn’t stop.
429 00:58:05.080 ⇒ 00:58:06.480 ethanpetersen: There was no.
430 00:58:06.880 ⇒ 00:58:30.200 ethanpetersen: there was no intentionality behind it, and you can feel that when you’re trying to write code with it. And so L is a, it’s, it’s very early, but it’s the the advantage of that is, it’s it’s really simple, really simple code base. You know, if if you guys wanted to adopt it for yourself and start really taking run.
431 00:58:30.320 ⇒ 00:58:39.139 ethanpetersen: I think it’s a it’s a good framework that’s that’s set up for that and then will the Maintainer super
432 00:58:39.360 ⇒ 00:58:50.099 ethanpetersen: friendly Guy like, if if you were using it and ran into any any issues you could reach out to him, and then as as well, he’s he’s working on flushing it out more right?
433 00:58:50.360 ⇒ 00:58:59.510 ethanpetersen: I I bet that at some point he’s probably gonna raise and then turn it into a company, but he’s working on everything, from evals to tracing.
434 00:58:59.860 ⇒ 00:59:01.630 Uttam Kumaran: Yeah, no. I see that in here. Yeah.
435 00:59:03.173 ⇒ 00:59:19.639 Uttam Kumaran: And this is the problem is like, we’ve we’ve gone from. We use like some of these agent building tools to like. We’re getting to more of like our own self hosted sort of like workflow building like there’s like flow wise. And any. Then the the only problem is like, and some of my guys, I’m like
436 00:59:20.220 ⇒ 00:59:38.210 Uttam Kumaran: pushing them towards like this model of like for us, internally like like, if it works, it works. It’s not like make or break. But I wanna have some level of testing Eval that we can review as a team and be like, how many? What was successful when we make changes? Can we test those changes?
437 00:59:38.250 ⇒ 00:59:53.650 Uttam Kumaran: And then, of course, like we’re gonna use something I want. So I want a set of tools to be available, whether people are doing local testing or we want to test in the cloud or want to test for clients. So this guy, I feel like I’ve come across this or something like this.
438 00:59:53.930 ⇒ 00:59:56.359 Uttam Kumaran: or maybe it was. Maybe honestly, it was this
439 00:59:58.150 ⇒ 01:00:02.330 ethanpetersen: It went like Mega viral on, on Twitter, for like a week.
440 01:00:02.330 ⇒ 01:00:03.110 Uttam Kumaran: Oh, okay.
441 01:00:03.110 ⇒ 01:00:07.409 ethanpetersen: So that’s when it got most of its github stars. So
442 01:00:08.500 ⇒ 01:00:19.179 ethanpetersen: odds are definitely high I got exposed to it. But literally, if you, if you DM. Will and say, Hey, I’m trying to move off of Lane chain.
443 01:00:19.801 ⇒ 01:00:22.609 ethanpetersen: and I came across this. You probably.
444 01:00:22.610 ⇒ 01:00:23.883 Uttam Kumaran: Call immediately.
445 01:00:24.520 ⇒ 01:00:31.819 ethanpetersen: Oh, he he would be! And then he would take a personal victory, and like yes, one more team. That’s not using Lane chain, even.
446 01:00:31.820 ⇒ 01:00:33.599 Uttam Kumaran: That’s the case. Super solid.
447 01:00:35.740 ⇒ 01:00:37.880 Uttam Kumaran: Yeah, it’s sharing. Yeah.
448 01:00:37.880 ⇒ 01:00:40.070 ethanpetersen: It’s like, actually pythonic.
449 01:00:40.070 ⇒ 01:00:40.600 Uttam Kumaran: Yeah.
450 01:00:42.120 ⇒ 01:00:49.929 ethanpetersen: I mean, you can. You can see like the ergonomics around it, and it’s great cause like like Will Will gets it.
451 01:00:50.220 ⇒ 01:01:02.770 ethanpetersen: You know he was a research scientist. He understands, and he’s actually opinionated about how a framework like this should be developed, which is, you know, it’s exactly the problem with with link chain is it’s not
452 01:01:04.150 ⇒ 01:01:09.939 ethanpetersen: it’s not opinionated. So it turns into this just like contradictory spaghetti of a.
453 01:01:09.940 ⇒ 01:01:14.749 Uttam Kumaran: No. And so this is a problem we have even for, like for us, like we have.
454 01:01:15.330 ⇒ 01:01:23.650 Uttam Kumaran: we have prompts floating around for everything. And I. I have a big believer in like documentation, and we have a big notion with a shitload of documentation in it. But
455 01:01:23.850 ⇒ 01:01:27.630 Uttam Kumaran: a lot of our processes now are like AI driven. And it’s like.
456 01:01:27.970 ⇒ 01:01:38.099 Uttam Kumaran: I’m trying to tell people that like you can’t just I mean, I don’t know. I’m like, Oh, we just remove something, and it works from the prompt I’m like that’s not like it doesn’t like work with doesn’t sit well with me
457 01:01:38.220 ⇒ 01:01:53.730 Uttam Kumaran: like we can’t do that. So I’m like, okay, we need a prompt repo. So that I’m like, Okay, what are we gonna do? Well, either we either. I’m gonna create a prompt repo in Github. And then maybe every time we change it, you gotta commit or like.
458 01:01:53.900 ⇒ 01:02:05.829 Uttam Kumaran: we’re gonna create a prompt basically like, set a database in notion which basically has all that. And then, at least, there’s some versioning history. But you’re totally right, is like
459 01:02:06.030 ⇒ 01:02:20.799 Uttam Kumaran: we’re gonna move everything from like these sorts of zapier sort of world into like actual code, because we’re gonna start running this even bigger. And I think something like this will work really, really well for us, whereas you’re totally right, like an instance of a string.
460 01:02:21.000 ⇒ 01:02:22.979 Uttam Kumaran: That’s the problem is, it’s like
461 01:02:23.010 ⇒ 01:02:51.700 Uttam Kumaran: that holds all the information in it, and every tool is built on like basically templating that. So it’s a good string formatting that with like, input variables. And then it’s like, yeah. Okay, then you have to. You have to put in like a specific thing in the front, and everything on Twitter is like it gets to that level where it’s like this is a perfect prompt. But nobody goes and like AV tests or like test anything out. And that’s where I’m like, this is not gonna work. I’m doing it now, because we need to move, but
462 01:02:51.730 ⇒ 01:02:53.960 Uttam Kumaran: not gonna work for us long term.
463 01:02:55.440 ⇒ 01:02:58.835 ethanpetersen: Oh, yeah, yeah, I’ll see, too. Like.
464 01:02:59.520 ⇒ 01:03:04.609 ethanpetersen: if there are any other especially around image and video.
465 01:03:04.610 ⇒ 01:03:21.720 Uttam Kumaran: Yeah, if there’s anything around images. Video, that’s interesting. I mean, we’ve been doing a lot on on Ocr, like, I’ve read a lot about Ocr, and basically, we do a lot of stuff that’s around knowledge like knowledge base for Pdfs. But now that the.
466 01:03:21.720 ⇒ 01:03:22.120 ethanpetersen: -
467 01:03:22.525 ⇒ 01:03:26.580 Uttam Kumaran: Now that the image Apis are like, really, really good.
468 01:03:26.780 ⇒ 01:03:32.549 Uttam Kumaran: we’re almost like thinking about not even scraping anymore and going straight to just like screenshots.
469 01:03:33.153 ⇒ 01:03:37.289 Uttam Kumaran: or Ocrs, whether we’re scraping a website or otherwise.
470 01:03:37.698 ⇒ 01:03:44.559 Uttam Kumaran: And so there we, I’m following a lot of different libraries that are helping to say, like you take a Pdf of like
471 01:03:44.730 ⇒ 01:03:49.590 Uttam Kumaran: random diagrams in it, and it can move all that to Markdown
472 01:03:49.750 ⇒ 01:04:01.319 Uttam Kumaran: really effectively stuff like that, but also totally anything that’s compelling on the image editing side or the video editing side. Not so much about like, hey, we made
473 01:04:01.580 ⇒ 01:04:06.760 Uttam Kumaran: this entire movie. But it’s like specific tools.
474 01:04:07.270 ⇒ 01:04:18.009 Uttam Kumaran: I’m super super curious about that’s where I’m like very much, not super plugged in for me. And stuff, I see is like the sexy stuff, you know, like in in in an image, or whatever
475 01:04:18.923 ⇒ 01:04:25.666 ethanpetersen: Yeah, yeah, definitely. happy to pass it along, especially
476 01:04:26.430 ⇒ 01:04:34.770 ethanpetersen: yeah. I just from from what I know well of the things I’m involved in. I know there should be
477 01:04:35.110 ⇒ 01:04:41.690 ethanpetersen: lot of cool things coming out. And then I I I tend to.
478 01:04:42.570 ⇒ 01:04:43.470 ethanpetersen: We get
479 01:04:43.620 ⇒ 01:04:51.649 ethanpetersen: get some early previews of of things. So hopefully, I just think that next year for video is going to be.
480 01:04:51.800 ⇒ 01:04:52.370 Uttam Kumaran: Okay.
481 01:04:52.370 ⇒ 01:04:53.530 ethanpetersen: So much fun.
482 01:04:54.865 ⇒ 01:04:58.909 ethanpetersen: I’m I’m excited for it. Yeah.
483 01:04:59.550 ⇒ 01:05:00.400 Uttam Kumaran: Also.
484 01:05:00.780 ⇒ 01:05:07.099 Uttam Kumaran: Alright. Well, I really, this is an amazing Saturday morning conversation. I’m jazzed. I like I have a lot.
485 01:05:07.100 ⇒ 01:05:08.300 ethanpetersen: Yeah, likewise.
486 01:05:08.610 ⇒ 01:05:23.332 Uttam Kumaran: I really appreciate the time. And yeah, let’s keep in touch. I mean, I’m gonna I’m gonna start looking at more video and content stuff, and, you know, would love like if if we start doing some research internally, or even my friend and Netflix, I’m gonna give him a call, probably right after this.
487 01:05:24.160 ⇒ 01:05:40.660 Uttam Kumaran: I would love to see. Maybe we can all have a conversation, or just do another one of these chats like talk about like what’s going on. And then, yeah, we’ll see where we’re at. And you know, Feb, March time. And like, maybe there’s an opportunity for us to work together. That’d be awesome. But regardless this was, this is really really great. So thank you.
488 01:05:41.721 ⇒ 01:05:52.629 ethanpetersen: Yeah. Likewise, I I also, I love conversations like this. And I’m just generally stoked about the space. And and in particular people that put it into actual usage.
489 01:05:53.150 ⇒ 01:05:54.190 Uttam Kumaran: Trying to.
490 01:05:55.280 ⇒ 01:06:00.410 ethanpetersen: Yeah, yeah, awesome. Yeah, definitely. Let me know.
491 01:06:01.431 ⇒ 01:06:03.870 ethanpetersen: if you ever wanna chat about video and and that stuff.
492 01:06:03.870 ⇒ 01:06:04.500 Uttam Kumaran: Cool.
493 01:06:04.500 ⇒ 01:06:05.714 ethanpetersen: All always down.
494 01:06:06.490 ⇒ 01:06:06.840 Uttam Kumaran: Cool.
495 01:06:07.236 ⇒ 01:06:07.603 ethanpetersen: Yeah.
496 01:06:08.170 ⇒ 01:06:12.369 ethanpetersen: Well, great to meet you. I appreciate your patience again trying to get this scheduled.
497 01:06:12.370 ⇒ 01:06:14.950 Uttam Kumaran: No, no, no problem. Thank you. Appreciate it.
498 01:06:15.210 ⇒ 01:06:16.529 Uttam Kumaran: Yeah, cool.
499 01:06:16.710 ⇒ 01:06:17.550 ethanpetersen: Yep.