Meeting Title: AI and Legal Research Discussion Date: 2026-03-02 Meeting participants: Robert Tseng, Chuck Arvin
WEBVTT
1 00:00:20.350 ⇒ 00:00:22.839 Chuck Arvin: Hello, can you hear me?
2 00:00:23.220 ⇒ 00:00:24.600 Robert Tseng: Yeah. Hey.
3 00:00:24.930 ⇒ 00:00:25.699 Chuck Arvin: Oh, yeah.
4 00:00:25.830 ⇒ 00:00:26.420 Chuck Arvin: I agree.
5 00:00:26.420 ⇒ 00:00:26.920 Robert Tseng: Good, how are you?
6 00:00:26.920 ⇒ 00:00:28.070 Chuck Arvin: Hey, Rob, how are you?
7 00:00:29.200 ⇒ 00:00:30.329 Chuck Arvin: Doing well. Good.
8 00:00:30.830 ⇒ 00:00:34.580 Robert Tseng: Yeah, you’re new… wait, you’re in DC area.
9 00:00:34.580 ⇒ 00:00:35.160 Chuck Arvin: Yep.
10 00:00:35.460 ⇒ 00:00:39.829 Robert Tseng: DC area, saw the badger at Amazon.
11 00:00:40.810 ⇒ 00:00:42.479 Chuck Arvin: You’re good, yep,
12 00:00:42.480 ⇒ 00:00:43.390 Robert Tseng: Yeah.
13 00:00:43.390 ⇒ 00:00:44.430 Chuck Arvin: Yeah.
14 00:00:44.610 ⇒ 00:00:51.759 Robert Tseng: And, yeah, I mean, thanks for taking the time. I guess, like, I know it’s been a while since I first reached out to you, so I guess…
15 00:00:51.760 ⇒ 00:00:52.650 Chuck Arvin: Of course!
16 00:00:52.650 ⇒ 00:00:58.679 Robert Tseng: How did I even find… I guess, so I’m a part-time evening law student, and,
17 00:00:58.680 ⇒ 00:00:59.170 Chuck Arvin: Oh, okay.
18 00:00:59.430 ⇒ 00:01:10.889 Robert Tseng: yeah, I go to Fordham, I live, like, just a couple blocks, I’m in New York City, and one of my professors is, like, an AI researcher. He, like, I don’t know, I guess he was doing some…
19 00:01:11.070 ⇒ 00:01:23.070 Robert Tseng: Spoke at some conference, and then I was poking around the conference, looking at interesting, kind of, like, white papers. Your name popped up a couple times of, like, hey, this guy publishes a lot of stuff. Seems like he’s a data guy, and also in law.
20 00:01:23.070 ⇒ 00:01:23.530 Chuck Arvin: Cool.
21 00:01:23.530 ⇒ 00:01:38.939 Robert Tseng: And I was like, I would love to talk to him and just kind of see what that’s like. So, that was really the whole, like, catalyst for why I reached out to you, read that specific paper, and yeah, I guess that’s kind of how I got in touch.
22 00:01:39.350 ⇒ 00:01:47.009 Chuck Arvin: Yeah, no, I appreciate it. There are not that many people interested in both of those topics, so I’m, I love to make the connection.
23 00:01:47.010 ⇒ 00:01:47.650 Robert Tseng: Yeah.
24 00:01:48.570 ⇒ 00:01:56.990 Robert Tseng: Well, I’d love to kind of first hear about, like, your experience. It seems like you did a law program, too. Like, kind of, what’s your interest in, like, the legal space? And… yeah.
25 00:01:57.880 ⇒ 00:02:13.419 Chuck Arvin: Yeah, sure, so, I mean, professionally, and I think that’s on the LinkedIn, you know, I work at Amazon, I’m a data scientist there. I’ve been in the data science industry for, probably a decade or so.
26 00:02:15.060 ⇒ 00:02:23.480 Chuck Arvin: going back before then, you know, I was actually interested in law and almost went to a JD program back in the day.
27 00:02:24.090 ⇒ 00:02:25.980 Chuck Arvin: And so…
28 00:02:26.770 ⇒ 00:02:39.300 Chuck Arvin: kind of recently have taken more of an interest in that again, now that it’s starting to become clear how the two topics kind of overlap again, in a way that, you know, they used to just be different things.
29 00:02:40.820 ⇒ 00:02:50.010 Chuck Arvin: And so, yeah, you saw a little bit of published work, kind of like a hobby on the side, and then I’m taking an,
30 00:02:50.590 ⇒ 00:02:57.319 Chuck Arvin: MSL degree, at USC, remotely, to kind of
31 00:02:57.480 ⇒ 00:03:02.240 Chuck Arvin: learn more about the law, because I really know very little of it at this point.
32 00:03:02.630 ⇒ 00:03:03.370 Robert Tseng: Sure.
33 00:03:05.610 ⇒ 00:03:27.680 Robert Tseng: Cool. Well, I mean, I guess, I actually didn’t… I mean, I graduated from USC undergrad. I didn’t know that they were… they offered a, like, a remote MSL program, but… and honestly, now I’m going through a JD program, I think an MSL might actually be a better fit for me, because I don’t think I really want to practice. I’m just more interested in, kind of, the… the subject matter.
34 00:03:27.680 ⇒ 00:03:28.250 Robert Tseng: Yeah.
35 00:03:28.250 ⇒ 00:03:36.469 Chuck Arvin: Fair enough. Yeah, it’s how high level do you want to get? Mine’s a little too high level, and the JD’s probably a little too detailed.
36 00:03:36.760 ⇒ 00:03:45.080 Robert Tseng: Yeah, like, yesterday, I was, like, kind of doing formatting for, like, this paper I submitted, and I was like, I don’t think I see myself doing paralegal work.
37 00:03:45.080 ⇒ 00:03:46.840 Chuck Arvin: Why did I do this?
38 00:03:47.060 ⇒ 00:03:47.740 Robert Tseng: Yeah, yeah.
39 00:03:47.740 ⇒ 00:03:48.350 Chuck Arvin: edging.
40 00:03:49.130 ⇒ 00:03:49.520 Robert Tseng: Yeah.
41 00:03:49.520 ⇒ 00:03:53.520 Chuck Arvin: Are you similarly, like, a data science background and looking to…
42 00:03:53.520 ⇒ 00:03:56.599 Robert Tseng: do more of that, or… You’re probably more together, Cole
43 00:03:56.950 ⇒ 00:04:11.690 Robert Tseng: than I am. I… I am… I think more of, like, a data engineer at this point. So, I mean, this just… I just use SQL. SQL Python is kind of my… is kind of where my technical skill set is. Right now, I run, like, a…
44 00:04:11.740 ⇒ 00:04:21.669 Robert Tseng: I run a data engineering consultancy that does, kind of, AI implementations as well. So, yeah, I mean, it’s been cool. Work a wide range of industries,
45 00:04:21.720 ⇒ 00:04:32.199 Robert Tseng: I started my career in logistics, so maybe similar kind of supply chain stuff to what maybe you see at Amazon, and then kind of went in-house at a consumer product
46 00:04:32.200 ⇒ 00:04:42.669 Robert Tseng: goods brand, and, you know, my current business we work with, it’s actually a mix of, mostly CPG and healthcare. So, yeah, it’s a wide range of disciplines.
47 00:04:43.880 ⇒ 00:05:00.500 Robert Tseng: But yeah, I am interested in building in the legal world. I guess, like, a few… I mean, if you’re a data person, I guess there’s a few, like, topics, you know, broad things that I’m interested in. I can go as deep as you’re interested, but,
48 00:05:00.690 ⇒ 00:05:04.240 Robert Tseng: Actually, like, are you familiar with dbt?
49 00:05:07.920 ⇒ 00:05:13.570 Robert Tseng: Okay, yeah. Yeah, it’s… it’s like a… it’s basically…
50 00:05:14.420 ⇒ 00:05:27.609 Robert Tseng: like, a DevOps workflow for data people. It’s called Data Build Tool. It just adds version control and, basically, like.
51 00:05:27.620 ⇒ 00:05:45.070 Robert Tseng: the type of lineage that you get in traditional software engineering and applies to data work. It kind of picked up some steam over the past 3 years, there was a big acquisition and everything. But yeah, I guess, like, I connected kind of the breakthroughs that it’s been making in the data world there to, like, in the legal world, where they’re way behind, and
52 00:05:45.070 ⇒ 00:05:52.810 Robert Tseng: In one of my classes right now that I’m taking, it’s called, like, Legislation Regulation,
53 00:05:52.990 ⇒ 00:05:54.120 Robert Tseng: they’re…
54 00:05:54.360 ⇒ 00:06:19.320 Robert Tseng: so many cases are kind of deliberated over, like, the meaning of terms, and, like, there’s, like, plain view meaning, or there’s plain meaning, and there’s, like, all these different things, and it’s really hard to, like, gauge the intent of, like, the author of the original statute versus, like, kind of the interpretation of, like, judges through different circuits. It’s, like, not very consistent. And, like, there isn’t just, like, a good repository to, like.
55 00:06:19.320 ⇒ 00:06:28.820 Robert Tseng: be like, hey, this is what the term meant, and, like, follow the lineage across all the different… all the different cases. So, it’s more of an infrastructure kind of, like.
56 00:06:28.880 ⇒ 00:06:43.110 Robert Tseng: you know, topic, but I was like, why do they not have this in law? And I was interested in, like, trying to pick one domain specifically where this is, like, you know, happening atrociously, and I would… I wanted to just, like, use that as a sample
57 00:06:43.110 ⇒ 00:07:02.770 Robert Tseng: kind of, like, set of, like, what would happen if we applied DBT practices to… to law in this specific way. So that was, like, one, like, research thing that I have on my mind. But anyway, I have a few other ones, I just… I don’t… but anyway, that’s, like, one example of the kind of stuff I think about while I’m in… while I’m in, school.
58 00:07:03.190 ⇒ 00:07:08.019 Chuck Arvin: And if I, if I can’t, so I don’t know the dbt, it sounds kind of like Git for software.
59 00:07:08.020 ⇒ 00:07:08.339 Robert Tseng: It’s like.
60 00:07:08.340 ⇒ 00:07:08.980 Chuck Arvin: Familiar with that.
61 00:07:08.980 ⇒ 00:07:09.420 Robert Tseng: Yeah.
62 00:07:09.420 ⇒ 00:07:17.029 Chuck Arvin: And so, so what this is, like, I don’t know, Git for… for the diction… almost basically for the dictionary.
63 00:07:17.030 ⇒ 00:07:19.880 Robert Tseng: Yeah, for the dictionary, sure. The legal, the legal dictionary.
64 00:07:19.880 ⇒ 00:07:27.900 Chuck Arvin: the definition of this term, you know, at some point in the 80s, and so, like, that becomes a thing, and it’s actually a structured…
65 00:07:27.900 ⇒ 00:07:28.949 Robert Tseng: Yeah.
66 00:07:29.150 ⇒ 00:07:31.219 Chuck Arvin: Way to map that out.
67 00:07:31.610 ⇒ 00:07:32.210 Robert Tseng: Yeah.
68 00:07:32.610 ⇒ 00:07:44.760 Robert Tseng: Because I actually think, like, from an AI perspective, kind of further down the line, like, it’s so hard to, like, plug these, you know, LLM tools on top of it, because, you know, it’s… yeah, it’s, like, taking…
69 00:07:45.060 ⇒ 00:07:46.260 Robert Tseng: the…
70 00:07:46.380 ⇒ 00:07:57.179 Robert Tseng: the average person’s usage of… or, like, you know, it’s a very specific… the way it defines things is not super nuanced, right? It’s just taking kind of, like, an average of, like, the…
71 00:07:57.190 ⇒ 00:08:08.789 Robert Tseng: the kind of… of the data set that it has. Whereas, like, legal interpretation requires something that’s really precise, and it only… not everybody’s opinion matters, only a few, like, special people matter. Like, the author
72 00:08:08.790 ⇒ 00:08:26.569 Robert Tseng: the judge, usually not the attorney. The attorney stuff is not part of the record, so it’s really whatever the court publishes. And so, I just think that it kind of helps distill some signal from all the noise that’s in there, and, like, if you could kind of dial this in a bit more, just building, like, a better data set of, like.
73 00:08:26.570 ⇒ 00:08:37.079 Robert Tseng: terms, like, then… then maybe, a lot of these AI kind of, like, tools would be… would be able to work better with, like, legal, data, I guess.
74 00:08:38.950 ⇒ 00:08:53.679 Chuck Arvin: Yeah, I mean, you could… you can look up the dictionary definition or whatever, but obviously that’s missing the interpretation and the context, and hey, people are starting to cast doubt on it, maybe they… so, yeah, I could see that. And I know LexisNexis has, like.
75 00:08:54.000 ⇒ 00:09:01.009 Chuck Arvin: Here’s how, you know, there’s relevant case law to a particular thing, but it doesn’t have, maybe, the temporal
76 00:09:01.410 ⇒ 00:09:07.139 Chuck Arvin: structure that you’re talking about. Exactly. Go back and replay it at a particular point.
77 00:09:07.660 ⇒ 00:09:14.470 Robert Tseng: Yeah, I don’t know if you’ve played around Alexis or Westlaw, but I think their semantic search is, like, terrible. So, like.
78 00:09:14.470 ⇒ 00:09:18.139 Chuck Arvin: You need to know what you’re doing, for sure. Yeah.
79 00:09:18.140 ⇒ 00:09:26.549 Robert Tseng: Like, I’ve had to take a whole class on just legal research, and it, like, makes absolutely no sense. And they’re really smart people that are building better, kind of.
80 00:09:26.550 ⇒ 00:09:41.340 Robert Tseng: things to, like, slap on top of Lexus and Westlaw, I guess, but… so I think that problem will get better. There will be a better solution for that. But yeah, you’re right, I think it just, you know, it just… it… however they’re linking, kind of,
81 00:09:41.380 ⇒ 00:09:47.420 Robert Tseng: however they’re doing their makes this… it’s just not really… not really good. Yeah, so…
82 00:09:47.580 ⇒ 00:09:56.389 Robert Tseng: I guess I’m… I’m curious, you know, for you, kind of, anything that… it seems like you’re working on a few things as well,
83 00:09:56.490 ⇒ 00:09:59.460 Robert Tseng: Yeah, like, kind of what… is there a particular, like.
84 00:09:59.590 ⇒ 00:10:05.229 Robert Tseng: specialty, or… I don’t know how to talk about it. Like, is there something that you’re interested in kind of diving deeper into?
85 00:10:05.890 ⇒ 00:10:07.140 Chuck Arvin: Yeah,
86 00:10:08.380 ⇒ 00:10:18.499 Chuck Arvin: Honestly, work has been keeping me busy enough that I haven’t done a whole lot of this, recently. I think one of the things that…
87 00:10:18.650 ⇒ 00:10:22.719 Chuck Arvin: I was pretty interested in. I don’t know why the thread ended up.
88 00:10:22.840 ⇒ 00:10:29.420 Chuck Arvin: But, A pain point for, like, a developer, including myself.
89 00:10:29.570 ⇒ 00:10:30.750 Chuck Arvin: is, like…
90 00:10:31.370 ⇒ 00:10:48.279 Chuck Arvin: you get some vague statute, hey, you know, you need to adopt reasonable data security protections against blah blah blah. Yeah. Like, I have no idea what that means. I need to tell me what… and in reality, like, in practice, what happens is, like, you sit down with a bunch of attorneys, you say, here’s what I’m doing, is that good enough? No, okay, okay.
91 00:10:49.760 ⇒ 00:10:52.769 Chuck Arvin: It’s a very inefficient workflow, and .
92 00:10:52.770 ⇒ 00:10:53.350 Robert Tseng: Hmm.
93 00:10:53.350 ⇒ 00:10:54.030 Chuck Arvin: doesn’t…
94 00:10:54.530 ⇒ 00:10:59.779 Chuck Arvin: Yeah, and you obviously don’t end up in the right place anyway, because you have all these legal issues.
95 00:10:59.940 ⇒ 00:11:01.020 Chuck Arvin: And so…
96 00:11:01.840 ⇒ 00:11:19.349 Chuck Arvin: you know, one… one thing that is kind of solved again in the software world, but obviously not in the legal world, is like, how would I know that a particular stack is compliant with a law? Well, you need to write tests, right? Like, here’s… if this kind of workflow happens, like, I need to be able to prove that
97 00:11:21.170 ⇒ 00:11:30.920 Chuck Arvin: the consent that you offered will be respected, or something like that. And those are, again, software things. You know how to write a test, you know how to check all the code paths, or whatever.
98 00:11:31.170 ⇒ 00:11:36.599 Chuck Arvin: But the legal profession hasn’t written down, like, here’s the requirements for what
99 00:11:36.760 ⇒ 00:11:38.520 Chuck Arvin: this thing needs to do, it’s all very.
100 00:11:38.950 ⇒ 00:11:46.469 Chuck Arvin: English. Yeah. Interpreted through case law and whoever your general counsel is, which might be a different interpretation, too.
101 00:11:46.530 ⇒ 00:11:49.470 Robert Tseng: Yeah. So that one…
102 00:11:49.470 ⇒ 00:11:54.430 Chuck Arvin: At least as a very high-level idea. Seems kind of interesting.
103 00:11:54.530 ⇒ 00:12:00.530 Chuck Arvin: The reason that I mentioned it, I know there was a paper at,
104 00:12:01.360 ⇒ 00:12:05.699 Chuck Arvin: I don’t know which conference you were talking about, but the one in Chicago last year, where they were still.
105 00:12:05.700 ⇒ 00:12:06.490 Robert Tseng: I think that’s the one.
106 00:12:06.490 ⇒ 00:12:09.729 Chuck Arvin: Something like that? Okay. Yeah.
107 00:12:10.530 ⇒ 00:12:17.570 Chuck Arvin: Where they were starting to do something like that, and I sent them a message. I don’t think… I don’t think they ever got back to me, which is okay, but…
108 00:12:18.100 ⇒ 00:12:34.450 Chuck Arvin: I think they were similarly interested in that problem of, like, how do we actually write down the logic that needs to be tested to verify compliance with some law? And then, once you do that, the code, it kind of follows, you know, right out of the box.
109 00:12:34.690 ⇒ 00:12:36.369 Chuck Arvin: Which would be pretty cool.
110 00:12:37.390 ⇒ 00:12:51.810 Robert Tseng: Yeah. I mean, I guess, like, I mean, I don’t know how related this is, but you had created a… in your… in the study that I had read from you, you had built a legal benchmark data set, for case holding specifically, so a.
111 00:12:51.810 ⇒ 00:12:54.550 Chuck Arvin: I didn’t build that one. But I did use it, yes.
112 00:12:54.550 ⇒ 00:13:09.910 Robert Tseng: Oh, you used it. Oh, right. Okay, yeah, you, you, like, did some research on it, just to kind of measure the performance of it. Yeah. But yeah, you’re… I mean, it seems like you’re interested in this topic of, like, kind of taking, you know, complicated legal jargon and trying to, like, you know, normalize it for some… for, like, a
113 00:13:10.070 ⇒ 00:13:21.710 Robert Tseng: For a wider audience in some way, and then, kind of, you’re trying to assess, like, whether the, you know, you’re able to distill enough, kind of, of the…
114 00:13:21.940 ⇒ 00:13:31.259 Robert Tseng: I guess the meaning out of it, so that LLMs can use it for, like, automate… automation tasks. Is that kind of a good summary of, like, kind of what you did there?
115 00:13:32.140 ⇒ 00:13:34.599 Chuck Arvin: Yeah, I, I think…
116 00:13:35.060 ⇒ 00:13:42.860 Chuck Arvin: And again, that one was very much a hobby, like, so, don’t read too much into it. But yeah, I think that’s right, it’s like…
117 00:13:43.350 ⇒ 00:13:45.820 Chuck Arvin: We, we obviously see that
118 00:13:45.950 ⇒ 00:13:52.359 Chuck Arvin: language models can do so much. You know, industry, I have other papers that are a little more rigorous.
119 00:13:52.470 ⇒ 00:13:59.770 Chuck Arvin: For my actual day job. Like, they can, they can obviously do so much, can…
120 00:14:00.170 ⇒ 00:14:05.320 Chuck Arvin: Where are the areas where they can actually make the legal practice better, more efficient, and stuff like that?
121 00:14:05.480 ⇒ 00:14:09.220 Chuck Arvin: Yeah. And then the flip side of that is, like.
122 00:14:10.840 ⇒ 00:14:19.839 Chuck Arvin: how would you measure that, right? If it gets all these questions right in the benchmark, like, is that actually evidence that it understands the concepts, or is it, like…
123 00:14:20.980 ⇒ 00:14:26.529 Chuck Arvin: already read the test, or something like that. Sure. It’s a more robust, you know, kind of measurement.
124 00:14:26.690 ⇒ 00:14:30.289 Chuck Arvin: To convince yourself that there’s something actually there.
125 00:14:30.670 ⇒ 00:14:31.280 Robert Tseng: Yeah.
126 00:14:31.380 ⇒ 00:14:35.910 Robert Tseng: you mentioned that, I mean, just a hobby, publishing research, just a hobby, I mean, that’s…
127 00:14:36.010 ⇒ 00:14:39.870 Robert Tseng: It’s a pretty, pretty, impressive hobby, if you ask me, but curious, like…
128 00:14:39.870 ⇒ 00:14:40.360 Chuck Arvin: Thank you.
129 00:14:40.360 ⇒ 00:14:59.199 Robert Tseng: what is a more robust research… like, what’s the… what’s, like, something that you’re… I don’t know if you’re allowed to speak, that… maybe you could point me to something that you feel like is more representative of, like, you know, you put your full kind of effort into that, and I’d be curious to read into it, and… I mean, I don’t… I’ve never published research in an academic setting, I just, like…
130 00:14:59.830 ⇒ 00:15:16.139 Robert Tseng: in the private sector, you just turn out white papers, and, like, people are okay with that, so I do that with my team, to… for sales purposes, but I’d be curious on, like, how do I really participate in research more? And if you have any guidance as well for me on, like, hey, like.
131 00:15:16.770 ⇒ 00:15:26.369 Robert Tseng: what you learned about, you know, publishing a short paper at a legal conference versus, like, some of the other things that you’ve done. I know that was, like, there’s multiple questions there. One is really just.
132 00:15:26.370 ⇒ 00:15:26.980 Chuck Arvin: Yeah.
133 00:15:27.200 ⇒ 00:15:40.259 Robert Tseng: your larger research kind of initiatives, things that I… something I should look at, and then just kind of advice on, like, hey, is it even worth publishing? And, like, what’s the level of effort to, like, really publish stuff, especially in the legal… in the legal circle?
134 00:15:41.450 ⇒ 00:15:45.950 Chuck Arvin: Yeah, so, one, are you familiar with Google Scholar?
135 00:15:46.340 ⇒ 00:15:46.880 Robert Tseng: Yeah.
136 00:15:46.880 ⇒ 00:15:55.699 Chuck Arvin: Yeah. So, you should be able to find, I think there’s, there’s 4 papers that I have, externally. You can, you can read all of those.
137 00:15:55.860 ⇒ 00:15:56.660 Chuck Arvin: Okay.
138 00:15:56.820 ⇒ 00:16:01.469 Chuck Arvin: all the Amazon stuff, you unfortunately can’t, but,
139 00:16:01.800 ⇒ 00:16:13.160 Chuck Arvin: So, like, some of the more detailed stuff is like, hey, we live in the supply chain, we want to make better predictions, and we want to, you know, do stuff in a supply chain, and we’re using language models to help make that happen.
140 00:16:14.140 ⇒ 00:16:18.280 Chuck Arvin: The legal stuff is more of a, like.
141 00:16:18.910 ⇒ 00:16:22.500 Chuck Arvin: Personal interest and sort of, thing there.
142 00:16:22.500 ⇒ 00:16:23.290 Robert Tseng: I see.
143 00:16:23.290 ⇒ 00:16:27.689 Chuck Arvin: So, you’ll probably see a quality difference between the two, I guess, the warning.
144 00:16:27.690 ⇒ 00:16:29.440 Robert Tseng: But that’s okay.
145 00:16:29.780 ⇒ 00:16:32.480 Chuck Arvin: Yeah, I mean…
146 00:16:34.490 ⇒ 00:16:41.740 Chuck Arvin: To your second question, which is, you know, like, what’s the level of effort to get started?
147 00:16:41.990 ⇒ 00:16:43.290 Chuck Arvin: I mean, I, I…
148 00:16:44.180 ⇒ 00:16:49.919 Chuck Arvin: Certainly, if you’re more technically minded, which I think you are, you know, we both are, like.
149 00:16:50.390 ⇒ 00:16:55.739 Chuck Arvin: It’s a fun and interesting exercise to kind of push these things and see what happens, and .
150 00:16:55.740 ⇒ 00:16:56.280 Robert Tseng: Hmm.
151 00:16:56.790 ⇒ 00:16:59.970 Chuck Arvin: How far the technology can get you or not.
152 00:17:00.080 ⇒ 00:17:01.340 Robert Tseng: And…
153 00:17:02.100 ⇒ 00:17:07.620 Chuck Arvin: like, the paper that you read literally started as that. It was like.
154 00:17:07.780 ⇒ 00:17:19.709 Chuck Arvin: I am genuinely just curious whether or not this could work, and then started working, found these benchmarks, and said, okay, I’m gonna use that, and like, okay, it seems to be working, let me kind of package it up a little better, and…
155 00:17:19.829 ⇒ 00:17:26.630 Chuck Arvin: send it off. I didn’t, frankly, expect for it to get accepted, so I was pretty surprised by that.
156 00:17:27.040 ⇒ 00:17:27.839 Chuck Arvin: But…
157 00:17:28.220 ⇒ 00:17:38.539 Chuck Arvin: even if that whole process kind of doesn’t work in the end, like, you still learned a lot and have some interesting, knowledge for next time out of it. So I think that’s…
158 00:17:38.650 ⇒ 00:17:43.789 Chuck Arvin: A really beneficial thing to do, and I would encourage it either way.
159 00:17:44.050 ⇒ 00:17:50.420 Chuck Arvin: The other thing that I would say is, like, there are a couple of…
160 00:17:51.890 ⇒ 00:17:58.589 Chuck Arvin: venues, I mean, you… sounds like you saw that conference, I know there’s one or two other things that… like, some web…
161 00:17:58.850 ⇒ 00:18:07.950 Chuck Arvin: seminar where people give a talk every month, that I sometimes dial into. I’m always happy, like, if you have an idea for an abstract or whatever.
162 00:18:08.150 ⇒ 00:18:14.570 Chuck Arvin: Love to read it, even if I can’t work on it. Sure. Interesting. But…
163 00:18:14.750 ⇒ 00:18:17.249 Chuck Arvin: I’ll stop there. Does that… does that help answer the question? Are there.
164 00:18:17.250 ⇒ 00:18:29.810 Robert Tseng: Yeah, that helps. I’d be curious in that webinar, you know, and just kind of… yeah, I’m not really around that many academics, I think I… but I do feel like I want to be closer to, like, what people are publishing, because, you know, I think,
165 00:18:30.700 ⇒ 00:18:46.609 Robert Tseng: I’m interested in, kind of, all the stuff that’s not necessarily, like, all commercialized yet, and, like, what… what are… what are the true capabilities that people are assessing? Like, I’ll definitely take a look at, I think your… your forecast… your forecasting research, thanks for pointing that out.
166 00:18:46.610 ⇒ 00:19:00.799 Robert Tseng: And then as far as, like… yeah, I mean, it seems like you have a really structured way of, like, when you have a question, like, how you explore your curiosity, and it, you know, you’re finding benchmarks, you’re able to basically structure it into something that’s, like, submittable as research, which is…
167 00:19:00.800 ⇒ 00:19:02.690 Chuck Arvin: I mean, with Livermore, but yeah.
168 00:19:02.690 ⇒ 00:19:04.580 Robert Tseng: skill, great skill to have, I guess.
169 00:19:04.580 ⇒ 00:19:06.020 Chuck Arvin: Sure, sure.
170 00:19:06.020 ⇒ 00:19:15.060 Robert Tseng: Yeah. So, yeah, I mean, I guess I… I just have to… have to try, and, we’ll see, see where that goes, but.
171 00:19:15.060 ⇒ 00:19:19.300 Chuck Arvin: The one thing that I will say, because I can tell you we’re technical, is like,
172 00:19:19.990 ⇒ 00:19:28.630 Chuck Arvin: AI is really powerful for this, and I think you’ll see acknowledgements in there of, like, use AI to help write all this. Like…
173 00:19:29.510 ⇒ 00:19:41.789 Chuck Arvin: As a thinking partner to, like, hey, let’s start fleshing out the, hypothesis that we want to test, and what kind of analyses will we need to run, and okay, let’s start writing some of that code, and actually analyze it, and…
174 00:19:41.970 ⇒ 00:19:46.780 Chuck Arvin: like, I basically and quite literally had a big repo where
175 00:19:46.880 ⇒ 00:19:56.820 Chuck Arvin: here’s the code, here’s the markdown files that I’m writing along the way, and they’re all sort of talking to each other, which…
176 00:19:57.510 ⇒ 00:20:04.249 Chuck Arvin: helps, I think, accelerate that. Like, I’m curious, too. I have a nice White paper of the thing?
177 00:20:04.250 ⇒ 00:20:04.950 Robert Tseng: Yeah.
178 00:20:04.950 ⇒ 00:20:09.929 Chuck Arvin: At the end of it. And that’s only really been possible in, like, the last year, so…
179 00:20:10.120 ⇒ 00:20:14.819 Chuck Arvin: Yeah. It’s like a superpower if you, can nail that workflow.
180 00:20:15.680 ⇒ 00:20:21.699 Robert Tseng: Yeah, no, that’s… I mean, we do that internally, so I feel like I get the workflow that you’re describing.
181 00:20:21.890 ⇒ 00:20:25.289 Robert Tseng: I’m… I’m… I’m curious, like, how… how… I mean…
182 00:20:26.030 ⇒ 00:20:37.710 Robert Tseng: maybe before it was… before AI, it would have taken you something like that, X number of hours, and then with AI, it’s taking you, like, Y number of hours, like, do you have a sense of, like, how fast this accelerated that for you?
183 00:20:45.250 ⇒ 00:20:57.999 Chuck Arvin: It’s hard to say, because, I mean, I’ll be honest, like, I just have never even bothered to try, and in part because it’s like, that feels very bitively difficult, even just doing a literature review for some of these things.
184 00:20:58.000 ⇒ 00:20:58.500 Robert Tseng: safe.
185 00:20:59.130 ⇒ 00:21:12.430 Chuck Arvin: huge amount of time, and, like, I don’t have that time, so… Yeah. I don’t bother. In a world where, like, okay, now I can do a literature review and get a couple of interesting papers in
186 00:21:12.560 ⇒ 00:21:15.409 Chuck Arvin: 20 minutes on Google,
187 00:21:16.080 ⇒ 00:21:19.950 Chuck Arvin: it becomes feasible. It’s more of a zero-to-one thing, I guess, than it is, like.
188 00:21:19.950 ⇒ 00:21:20.440 Robert Tseng: I see.
189 00:21:20.440 ⇒ 00:21:23.739 Chuck Arvin: Number of hours reduction, and it was just…
190 00:21:23.910 ⇒ 00:21:26.980 Chuck Arvin: infeasible prior to that, I think. At least for me.
191 00:21:28.110 ⇒ 00:21:44.239 Robert Tseng: Yeah, okay, that’s… that’s… that’s a helpful direction. I’m just curious, like, you know, I mean, you’re somebody who straddles both, like, you know, commercial work and Amazon, and also in academia. I’m just curious, like, how… how you… how does your AI use in your… in your, like.
192 00:21:44.390 ⇒ 00:21:50.850 Robert Tseng: I guess your private work, also, like, kind of inform, like, your public research work as well, and, it seems like you’re…
193 00:21:51.340 ⇒ 00:21:54.970 Robert Tseng: Yeah, a lot of trans… transferable, transferable stuff, and yeah.
194 00:21:55.130 ⇒ 00:22:07.039 Chuck Arvin: Yeah, that’s been kind of the bet, is I’m like, at work, I’m going to focus on AI applications, and at home, I’m going to try to do the same thing. So, a lot of overlap there, which is… which is helpful for getting it done, yes.
195 00:22:07.650 ⇒ 00:22:21.809 Robert Tseng: Nice. Well, do you ever come to New York? I’d love to try to meet some… I mean, I… if, you know, I’m sure you’re busy, but, you know, I’d love to grab a coffee with you sometime if you ever come in, and if not, like, I’m in DC every now and then, so I’ll definitely keep you in mind.
196 00:22:21.810 ⇒ 00:22:32.360 Chuck Arvin: That sounds fun. Yeah, let me know when you’re done here. I think I will probably be up there in a few weeks, to visit the team up there, so I can let you know, for sure.
197 00:22:32.360 ⇒ 00:22:32.730 Robert Tseng: Cool.
198 00:22:32.730 ⇒ 00:22:34.130 Chuck Arvin: when I’m up there as well.
199 00:22:34.520 ⇒ 00:22:44.989 Robert Tseng: Yeah, I’d love to kind of have a… yeah, chat with you more, kind of go deeper into some of these things, but really appreciate you taking the time, especially in the middle of a workday, so,
200 00:22:44.990 ⇒ 00:22:45.880 Chuck Arvin: That’s all good, listen.
201 00:22:45.880 ⇒ 00:22:46.830 Robert Tseng: entertaining.
202 00:22:46.830 ⇒ 00:22:48.389 Chuck Arvin: More fun than the other stuff.
203 00:22:48.390 ⇒ 00:23:05.940 Robert Tseng: Yeah, I’m curious, like, kind of how you, I mean, obviously being a very technical person at a giant company like there, I would love to learn more about, kind of the AI developments that you’re seeing, and you say you build AI applications, too, so I’d love to kind of pick your brain on that next time.
204 00:23:06.440 ⇒ 00:23:07.770 Chuck Arvin: Yeah, that sounds fun.
205 00:23:08.250 ⇒ 00:23:12.240 Chuck Arvin: Great. Alright, thank you so much for your time, Chuck, and great to see you soon.
206 00:23:12.240 ⇒ 00:23:12.770 Robert Tseng: Yeah.
207 00:23:12.770 ⇒ 00:23:13.259 Chuck Arvin: Yeah, bye.