WEBVTT 1 00:00:02.310 --> 00:00:03.400 David Bau: Allowed. 2 00:00:03.530 --> 00:00:05.020 David Bau: Okay, great. 3 00:00:08.490 --> 00:00:10.140 David Bau: And it's… 4 00:00:18.120 --> 00:00:19.190 David Bau: See? 5 00:00:19.450 --> 00:00:20.350 David Bau: Is that right? 6 00:00:22.190 --> 00:00:23.280 David Bau: I don't know why. 7 00:00:23.670 --> 00:00:27.030 David Bau: And all this random furniture here. Let's see here. 8 00:00:30.790 --> 00:00:32.150 David Bau: Silence. 9 00:00:34.570 --> 00:00:37.630 David Bau: It's gonna go. 10 00:00:40.100 --> 00:00:40.920 David Bau: Is that happening? 11 00:00:46.220 --> 00:00:47.729 David Bau: Yeah. H. 12 00:00:56.270 --> 00:00:57.020 David Bau: dominant. 13 00:01:02.550 --> 00:01:03.280 David Bau: Thank you. 14 00:01:05.450 --> 00:01:07.800 David Bau: Let's do on my business. 15 00:01:12.030 --> 00:01:12.900 David Bau: Excuse me. 16 00:01:37.110 --> 00:01:38.100 David Bau: God. 17 00:01:39.890 --> 00:01:40.680 David Bau: Bye. 18 00:01:42.680 --> 00:01:43.940 David Bau: That's not fair. 19 00:01:54.260 --> 00:01:55.380 David Bau: Okay, guys. 20 00:01:58.910 --> 00:01:59.950 David Bau: Here's ready. 21 00:02:01.190 --> 00:02:01.930 David Bau: Bill. 22 00:02:02.070 --> 00:02:13.719 David Bau: We're giving you some time to get your papers ready for the project of the class, to try to get them in good shape. Thank you, Dr. Thing. So I think that during this last phase of the class. 23 00:02:13.930 --> 00:02:17.210 David Bau: We'll try to schedule more exercises. 24 00:02:18.070 --> 00:02:20.930 David Bau: write parts of your paper. Of course, you're still 25 00:02:21.160 --> 00:02:30.980 David Bau: doing your experiments and trying to slot them into your paper, but I would like to do some exercises where we, you know, write up a draft of parts. 26 00:02:31.200 --> 00:02:40.649 David Bau: And then we exchange them with other students to get ideas and feedback. And so what I'm going to actually ask you to do for next 27 00:02:41.050 --> 00:02:42.850 David Bau: time, this Thursday. 28 00:02:43.140 --> 00:02:59.259 David Bau: is, to also… so, like, today, you guys already, as individuals, wrote a whole bunch of introductions that you submitted to the form. That's great. I think that it's, like, I've read them over, they're… they're in good shape for, you know, first drafts, it's great, you know, people are… are thinking. 29 00:02:59.510 --> 00:03:01.130 David Bau: What story they want to tell. 30 00:03:01.350 --> 00:03:04.730 David Bau: So what I'd like to ask you to do, just for Thursday. 31 00:03:04.970 --> 00:03:08.080 David Bau: And to do a similar thing, maybe as a team. 32 00:03:08.180 --> 00:03:25.200 David Bau: To, to think about what your, your first couple sections of your paper are. Your introduction, and maybe the methods section, where you introduce some of the technical things that you want to explain, whether it's your methods, your data set, your model, your… 33 00:03:25.320 --> 00:03:31.119 David Bau: You're setting, you know, what formulas you use, what symbols you use to talk about this and that, you know. 34 00:03:31.380 --> 00:03:40.900 David Bau: if you've got different users, or different models, what letters do you use, different formulas to talk about them, or whatever. Whatever technical thing you want to describe in your methods section. So right… 35 00:03:41.110 --> 00:03:46.550 David Bau: Write, you know, do another draft of your introduction, write your methods section, and then come and bring, like. 36 00:03:47.400 --> 00:03:50.380 David Bau: 5 physical paper printouts. We'll try 37 00:03:50.600 --> 00:03:55.659 David Bau: Physical paper. Next time, we'll see if it works. If it doesn't, we can go back to electronic. 38 00:03:55.890 --> 00:03:57.080 David Bau: Bring a pen. 39 00:03:57.640 --> 00:04:05.970 David Bau: And we'll literally, you know, do some paper shuffling, where you spend some time reading somebody else's 40 00:04:06.060 --> 00:04:25.389 David Bau: a couple things, and, like, mark it up and say, you know, this was confusing to me, or whatever. And so I'm hoping that that would be useful. We'll do that on Thursday. After that, we'll probably go back on Tuesday to do project presentations, again, to brainstorm about ideas on the project, but I thought, let's do a little bit of writing this weekend. 41 00:04:25.930 --> 00:04:26.959 David Bau: That make sense? 42 00:04:27.220 --> 00:04:33.670 David Bau: Okay, so let's talk a little bit about writing. We'll do a little bit of this, exchanging… 43 00:04:33.820 --> 00:04:46.169 David Bau: a writing exercise today, and that's what the spreadsheet's for, but we're not going to do it quite yet. But there's a spreadsheet link on the project website that you guys might have seen that will, that we'll do. 44 00:04:46.370 --> 00:04:50.099 David Bau: And the idea from that spreadsheet is going to be that you, 45 00:04:50.360 --> 00:05:04.140 David Bau: You just randomly put your name next to one of the rows. I erased, like, who submitted each one, so it's all anonymous. It won't be anonymous for long, but, you know, don't put your name next to your own. That's not the point of it, you know? So put your name next to somebody else's. 46 00:05:04.250 --> 00:05:16.490 David Bau: Thing. And then later on, you'll… you'll make a minor edit to it, or some suggestions. And then after you've done that, then you'll go find who it is who suggested on yours. 47 00:05:16.660 --> 00:05:20.199 David Bau: And then you'll talk to them, and they'll explain what they're thinking. 48 00:05:20.420 --> 00:05:37.930 David Bau: That makes sense? Okay. So, alright. So, I have a lecture I sometimes do about how to do research. Actually, I do have the first couple parts of it near the beginning of the class, which is, like, how to pick a problem, how to conduct experiments, and so, so I'm gonna talk a little bit about 49 00:05:38.160 --> 00:05:42.390 David Bau: Thoughts on writing a paper. If I have time, I'll talk about thoughts on giving a talk. 50 00:05:42.680 --> 00:05:53.469 David Bau: If I don't have time, I can talk about that later, but that's… that's basically what you guys are going to be doing for the rest of the time here. We're trying to give you some time to write a paper and, like, a nice community to… 51 00:05:53.890 --> 00:05:55.109 David Bau: Talk about it with, and then… 52 00:05:55.320 --> 00:06:03.270 David Bau: And then every time you guys are presenting, I want you to get better at presenting until, at the end, you have a pretty nice talk. Make sense? 53 00:06:03.930 --> 00:06:04.720 David Bau: Okay. 54 00:06:05.160 --> 00:06:16.069 David Bau: So, so yeah, this is a link. It's easier to find it on the course website. But there's a spreadsheet that you can load into your data. So, how to write a technical paper. So, I'm… I'm stealing this. 55 00:06:16.330 --> 00:06:24.250 David Bau: material from… a talk that Bill Freeman likes to give near the end of his technical classes. 56 00:06:24.540 --> 00:06:28.369 David Bau: Sometimes before, before the students get into, into, 57 00:06:28.810 --> 00:06:32.270 David Bau: research. I really like his perspective, so I share the same… 58 00:06:32.500 --> 00:06:38.060 David Bau: ideas that he likes to… he likes to teach. So he's one of my… One of my advisors. 59 00:06:38.730 --> 00:06:41.790 David Bau: And, And so this is a… so… 60 00:06:42.470 --> 00:06:45.220 David Bau: So how important is it to write well? 61 00:06:45.760 --> 00:06:48.730 David Bau: How important is it to, like, write a good paper? 62 00:06:49.900 --> 00:06:55.419 David Bau: And you might think, Might think, oh, you know, 63 00:06:55.970 --> 00:07:03.600 David Bau: At one point, it is to write really brilliantly, but I should really avoid writing a stinker on a paper, because, gosh, it's embarrassing. 64 00:07:03.910 --> 00:07:05.520 David Bau: Right, right, right, a real bad one. 65 00:07:06.010 --> 00:07:09.959 David Bau: Actually, what… Phil has observed. 66 00:07:10.480 --> 00:07:13.630 David Bau: In his many decades of advising, longer than me. 67 00:07:14.020 --> 00:07:17.000 David Bau: Is that this is… this is, like, the last function. 68 00:07:17.330 --> 00:07:20.400 David Bau: Okay, this is… this is the impact of… 69 00:07:20.660 --> 00:07:24.750 David Bau: How much a paper impacts your career? 70 00:07:25.350 --> 00:07:29.009 David Bau: As determined by how good the paper is. 71 00:07:29.350 --> 00:07:36.619 David Bau: Right? And so if you write, like, a really stinker of a paper, it's over here, and if you write a really brilliant paper. 72 00:07:36.860 --> 00:07:41.259 David Bau: it's over here, right? And Bill's point is sort of funny. 73 00:07:41.450 --> 00:07:43.470 David Bau: He's like, Actually… 74 00:07:44.090 --> 00:07:49.490 David Bau: Actually, even, like, if you write… if you write an okay paper, of course, you know, it doesn't help, doesn't hurt. 75 00:07:49.680 --> 00:07:53.399 David Bau: Right? That maybe helps a little bit, that you wrote a paper. Better than nothing. 76 00:07:53.500 --> 00:07:54.220 David Bau: Right? 77 00:07:54.520 --> 00:08:04.019 David Bau: If you write a pretty good paper, pretty good, like, solid, great paper, gets through peer review, you know, they invite you to give a poster, it's nice, right? 78 00:08:04.330 --> 00:08:07.670 David Bau: That's a… that's a big accomplishment as a student. 79 00:08:07.820 --> 00:08:10.630 David Bau: But his observation is, 80 00:08:11.520 --> 00:08:17.220 David Bau: It's not… even if it's much better than an OK paper, it doesn't… also doesn't move the needle much. 81 00:08:17.470 --> 00:08:23.289 David Bau: He says, he says, the thing that moves the needle is, like, really good paper. 82 00:08:23.500 --> 00:08:24.260 David Bau: Right. 83 00:08:24.420 --> 00:08:41.729 David Bau: And so… so that's… so that's his advice, is that, you know, whenever you're writing a paper, you're going for that. You don't know, it's hard to hit this. It's a little bit like the lottery. You have to… a lot of things have to align, right? You have to be lucky a bit, and probably not a good idea. You have to… you have to… 84 00:08:41.850 --> 00:08:44.190 David Bau: you know, Be doing well. 85 00:08:44.310 --> 00:08:58.500 David Bau: In terms of your writing, you have to execute your experiments well, it has to be at the right moment, where the community is receptive to the idea. You know, a bunch of things sort of have to… have to align, and then this could happen. But his point is. 86 00:08:58.630 --> 00:09:04.170 David Bau: That it, That this is where all the value is. 87 00:09:04.320 --> 00:09:08.190 David Bau: And… and you're… and really, you can think of what you're trying to do as, like. 88 00:09:08.490 --> 00:09:11.629 David Bau: Trying to increase your probability of writing a really good one. 89 00:09:12.980 --> 00:09:30.409 David Bau: And that's what we're all going for, every time we write a paper. Like, how bad is it if you write a stinker of a paper? Well, it's not good. It might be negative. It might, like, somebody reads it, they might have a negative opinion of you, or something like that. But not too bad, also. 90 00:09:30.560 --> 00:09:32.830 David Bau: Like, more likely, people just won't read it. 91 00:09:32.960 --> 00:09:33.970 David Bau: Does that make sense? 92 00:09:34.580 --> 00:09:41.610 David Bau: And so, so, so yeah, so, so I don't know why, you know, I think that… 93 00:09:42.420 --> 00:09:47.540 David Bau: Well, he starts off with this graph, Bill, but I think that maybe most students are too risk-averse. 94 00:09:47.790 --> 00:10:06.350 David Bau: in writing papers that are, like, you know, worried a lot about the mechanics and trying to avoid a bad one, when really they should be thinking about, like, oh, how do we make it brilliant? And so maybe it goes into the writing, too. So if your writing is too risk-averse, and it's just very wordy, and, you know, you're covering your ass a lot, you know, and you're trying to say too much. 95 00:10:06.380 --> 00:10:14.709 David Bau: But you're not letting the brilliance come through, then you might actually be reducing the probability that you have brilliant paper, just because you're afraid of not writing a bad one. 96 00:10:14.810 --> 00:10:15.770 David Bau: Any sense? 97 00:10:16.090 --> 00:10:26.620 David Bau: Okay, so… okay, so the other thing that he introduces with is that she says, especially for, like, first-year students, right, you come in and you think. 98 00:10:26.950 --> 00:10:28.180 David Bau: Oh, I'm an academic. 99 00:10:29.080 --> 00:10:35.439 David Bau: And this process that I'm going to participate in is, like, joining… 100 00:10:35.770 --> 00:10:40.749 David Bau: Joining the, joining the community of scholars, right? 101 00:10:40.840 --> 00:10:49.029 David Bau: Where, for thousands of years, scholars have poured over their manuscripts, and, you know, you mentioned the monks in the… 102 00:10:49.080 --> 00:11:03.200 David Bau: you know, in the Abbey, you know, taking years and years to do the calligraphy and transcribe things and looking at every word and scrutinizing everything. And you think, okay, now I'm joining this long tradition of scholarly studies. 103 00:11:03.310 --> 00:11:09.499 David Bau: And when I write my paper, you kind of imagine all the monks, like, crowding around the paper and saying. 104 00:11:09.860 --> 00:11:16.689 David Bau: This is… so this is, like, what, you know, sort of what you might imagine. 105 00:11:16.890 --> 00:11:18.139 David Bau: As an outsider? 106 00:11:18.240 --> 00:11:23.599 David Bau: That, you know, the paper process… But who says no? 107 00:11:24.030 --> 00:11:27.739 David Bau: That's… that's not… at least not in the 21st century. 108 00:11:27.990 --> 00:11:29.939 David Bau: This is not how it is at all. 109 00:11:30.200 --> 00:11:34.209 David Bau: You have this massive surplus of information out there. 110 00:11:34.420 --> 00:11:38.659 David Bau: And so the reality is that when you create a paper and you put it out in the world. 111 00:11:38.900 --> 00:11:50.200 David Bau: it is entering this very busy, crowded marketplace of ideas. It's more like this kind of picture. It's like a bazaar. And you're here honking your wares. 112 00:11:50.500 --> 00:11:55.569 David Bau: hey, check out my 5-cent paper, right? And then people are streaming by and, you know… 113 00:11:55.740 --> 00:12:03.079 David Bau: you know, most people will ignore you, and if you're lucky, somebody will say, well, just 5 cents, maybe I'll pay for not this, right? 114 00:12:03.480 --> 00:12:07.090 David Bau: And that's, that's the, you know, that's the picture. 115 00:12:07.480 --> 00:12:08.440 David Bau: Okay. 116 00:12:08.770 --> 00:12:10.210 David Bau: And so, okay. 117 00:12:10.440 --> 00:12:19.779 David Bau: So, let's, let's talk about what makes a good paper. Oh, you know, Bill says, okay, Bill said, my advisor was Ted Ailsman, and this is what Ted Ailsman said. 118 00:12:19.900 --> 00:12:28.930 David Bau: Okay, well, I don't know. What is the sense? So basically, he says, you know, there's… there's… there's a… he says, it's easy to write a good paper, said Ted. 119 00:12:29.410 --> 00:12:31.370 David Bau: Just do 4 things. 120 00:12:32.040 --> 00:12:40.839 David Bau: And if you do these four things, then you will get your paper in and through peer review. He says, since I've been following this formula. 121 00:12:41.650 --> 00:12:45.029 David Bau: Every paper I've written has been accepted since the 1960s. 122 00:12:45.540 --> 00:12:51.470 David Bau: So, I don't know if this, like, if this formula still works, but I'll share it with you. 123 00:12:51.720 --> 00:12:53.129 David Bau: He said it's that. 124 00:12:53.260 --> 00:13:00.500 David Bau: He says the key thing is to do 3 key things, right? He has to be really clear 125 00:13:00.620 --> 00:13:05.800 David Bau: About what the problem is, and why the audience Cares about it. 126 00:13:06.160 --> 00:13:08.269 David Bau: Right, so that, that'll kill a paper if… 127 00:13:08.470 --> 00:13:11.040 David Bau: Somebody can't tell what it is that you're trying to solve. 128 00:13:11.350 --> 00:13:14.099 David Bau: In the paper, so that's absolutely the first thing. 129 00:13:14.770 --> 00:13:17.240 David Bau: But then, to be a scholar. 130 00:13:17.430 --> 00:13:19.369 David Bau: You have to give a little context. 131 00:13:21.020 --> 00:13:25.840 David Bau: you can't just say, hey, I'm gonna solve this problem. Like, that would be more of a marketing pitch. 132 00:13:26.280 --> 00:13:39.969 David Bau: like, you've never seen this problem before, amazing, right? You know, we're gonna solve this. No, a scholar would give you a little context of, like, oh, there are other solutions here, and here's where we fit in the context. 133 00:13:40.080 --> 00:13:45.979 David Bau: And then here's… here's what our solution is. And then, so this is the meat of the paper when you explain what you've done. 134 00:13:46.200 --> 00:13:54.080 David Bau: Right? And, and I don't know why it's more interesting, or why it's worth… reading about. 135 00:13:54.510 --> 00:13:55.410 David Bau: Right? 136 00:13:55.520 --> 00:14:00.799 David Bau: And then… and then at the end, you… you say, you know, you might have been hoping. 137 00:14:01.330 --> 00:14:03.370 David Bau: To hear about something else. 138 00:14:03.950 --> 00:14:06.120 David Bau: When you write your paper, 139 00:14:06.430 --> 00:14:24.820 David Bau: And sorry, that's not what our paper is about. We'll talk about, you know, a little bit of related work and other things and say there are other interesting things out there as well. They're all interesting, but they're different from our thing. Our thing is worthwhile anyway, right? So… so there's… so… so there's sort of four things you gotta do. 140 00:14:25.330 --> 00:14:30.929 David Bau: That kind of protect your paper against misunderstanding, and make sure that it's clear. 141 00:14:31.060 --> 00:14:35.140 David Bau: And if you do that, says Ted Adelson, Wow. 142 00:14:35.710 --> 00:14:42.129 David Bau: I'm sure he's also a brilliant professor, right? But he, you know, these are the key things from his point of view. 143 00:14:42.390 --> 00:14:47.900 David Bau: So what is built? Okay, so… okay, so for… for a machine… so if we're… 144 00:14:48.230 --> 00:14:51.039 David Bau: If you're gonna put your paper into NURPS or something. 145 00:14:51.370 --> 00:14:55.430 David Bau: I'm just gonna share. So, every field has different conventions. 146 00:14:55.640 --> 00:14:58.129 David Bau: For what goes into a paper. 147 00:14:58.310 --> 00:15:05.040 David Bau: and… Like, if you're… if you're… if you publish in, Nature… 148 00:15:05.690 --> 00:15:22.259 David Bau: And the sections, and the ordering of the sections, and the length of the sections, and the language that they use, and the style of the figures, and the style of the presentation of the data, and everything is different from if you, like, submit to ICML or NURIPS, and that's different if you submit to, like, a philosophy journal or something. 149 00:15:22.450 --> 00:15:26.860 David Bau: And so, you know, you should… before you submit, you should take a look. 150 00:15:27.030 --> 00:15:37.490 David Bau: at the culture of the community that you're writing into. What I'm gonna do here is I'm gonna give you a little overview for what, like, a NURPS-style paper looks like. So there's this machine learning 151 00:15:37.600 --> 00:15:46.490 David Bau: community that now is so huge, it goes over ICML, and ICLR, and nerves, and, you know, ACL, a whole bunch of 152 00:15:46.500 --> 00:15:58.349 David Bau: you know, natural language conferences, a whole bunch of computer vision conferences, CGPR, robotics, you know, and there's this massive unification of these kind of communities where they're all doing very similar work, and a lot of the papers look similar. 153 00:15:58.510 --> 00:16:03.030 David Bau: in this machine learning community. And so… so this is… so this is, like. 154 00:16:03.240 --> 00:16:06.819 David Bau: This is, like, the outline of what all the papers look like. 155 00:16:07.110 --> 00:16:08.500 David Bau: You know, introduction? 156 00:16:10.260 --> 00:16:15.920 David Bau: And then usually people have a special section for literature review, you know, maybe called Related Work. 157 00:16:16.360 --> 00:16:19.700 David Bau: Sometimes it'll be right at the beginning of the paper. 158 00:16:20.180 --> 00:16:21.759 David Bau: And sometimes it'll be at the end. 159 00:16:21.990 --> 00:16:22.780 David Bau: Right? 160 00:16:22.950 --> 00:16:28.489 David Bau: And then… and then because the field is getting mature now, there's usually some background to talk about. 161 00:16:28.660 --> 00:16:39.330 David Bau: It's an opportunity to introduce a little bit of notation, to give a little context for what… how you're thinking about the problems in relation to everybody else. And then… and then there's the technical, 162 00:16:39.540 --> 00:16:43.090 David Bau: explanation of what's going on. 163 00:16:43.200 --> 00:16:53.569 David Bau: Like, what is… what is the method? And then… and then usually have a, you know, an experiment section where you… you… you convey all your empirical evidence that what you've got is interesting. 164 00:16:53.940 --> 00:17:05.179 David Bau: And, you know, worth learning about. And then you write a very, very brief conclusion section to wrap it up, that usually you don't say too much in conclusions in, in late hours. 165 00:17:05.380 --> 00:17:10.040 David Bau: And so… so this is, like, a style of a paper that's very common, because… 166 00:17:10.170 --> 00:17:20.669 David Bau: Because really what the thing that it puts at the center of the paper is the meth… is, like, the method, like, this is a methods-oriented paperwork. We're like, we're gonna teach you how to do a thing. 167 00:17:20.940 --> 00:17:26.539 David Bau: So, so we're gonna, you know, we're gonna introduce the thing, and we're gonna spend some time talking about it and support it. 168 00:17:27.050 --> 00:17:27.900 David Bau: Sense? 169 00:17:28.150 --> 00:17:41.519 David Bau: And so, so I just… you don't have to obviously read this paper, but this is an example paper. This one's from one of my students, Roge Kandakota, erasing concepts from diffusion models so you can just kind of get a sense. 170 00:17:41.660 --> 00:17:50.530 David Bau: for what filled out in the paper is in Europe's paper, is, what's the current patient amount in NRIPS paper? Is it 9 pages now? 8 pages? 9 pages? They've changed it recently. 171 00:17:50.560 --> 00:18:06.240 David Bau: But it's, like, it's an 8-page format. Oh, this is not yours, this is, this might be CBPR. CDPR is a double common format, so they all… they all have slightly different formats that they use. But you can see here, there's an abstract, introduction, he goes into related work. 172 00:18:06.430 --> 00:18:16.689 David Bau: And after about, you know, a couple pages of introduction-related work, now he's into the meat of the paper. He's gonna introduce a bunch of their patient, background. 173 00:18:16.800 --> 00:18:22.139 David Bau: And then… and then he gets into his method, and talks about… he'll teach you, like, what it is that he's doing. 174 00:18:22.270 --> 00:18:39.430 David Bau: There, got some figures to describe what's going on. And then after he's described what he's doing, then there's all these experiments where he's like, okay, so how well does it actually work in practice? And he's sort of, you know, measuring everything all the way till he runs out of pages. 175 00:18:39.620 --> 00:18:48.609 David Bau: And then… and then he has this conclusion. See this, like, really short conclusion at the end, which basically says, yeah, we did what we did, and Zend will reset, right? 176 00:18:48.800 --> 00:18:52.170 David Bau: And, and then, 177 00:18:52.360 --> 00:18:58.500 David Bau: And, you know, in typical PhD student fashion, And he ends… 178 00:18:58.740 --> 00:19:01.249 David Bau: At the last millimeter of the last page of this. 179 00:19:01.360 --> 00:19:03.500 David Bau: Right? Dumb. Which is… 180 00:19:03.860 --> 00:19:13.499 David Bau: if you've ever done peer review, you know, you'll notice if somebody rushed a paper and they're, like, halfway down the last page or something, like, whoa, you had an extra half page, why didn't you use it? 181 00:19:13.650 --> 00:19:19.199 David Bau: So it's just, it's just one of these things. Like, you know, a polished paper usually uses the last inch. 182 00:19:19.760 --> 00:19:23.380 David Bau: Okay, and then, and then after that, There's all this… 183 00:19:23.470 --> 00:19:31.099 David Bau: bibliography information, which usually doesn't count against the page limit, so it goes over. So depending on the conference, there's some things that 184 00:19:31.130 --> 00:19:43.649 David Bau: kind of go over the last page, they don't count. And so, for this conference, it was, like, some disclaimers, some acknowledgements, stuff like this. Usually in peer review, you wouldn't have an acknowledgement section, you've got to keep it anonymous. 185 00:19:44.050 --> 00:19:44.720 David Bau: Okay? 186 00:19:45.010 --> 00:19:45.970 David Bau: Make sense? 187 00:19:46.210 --> 00:19:49.620 David Bau: And so… So now that's one style paper. 188 00:19:49.770 --> 00:19:53.390 David Bau: Here's another style of paper, so… 189 00:19:54.240 --> 00:20:02.440 David Bau: The other side of paper here, this is a paper that's organized around analysis, instead of organized around introducing a method. 190 00:20:02.690 --> 00:20:11.839 David Bau: And so, the paper has a little bit of a different flavor, so I've noticed that my students' papers sort of come in these two categories. So some of the papers are like, I'm going to teach you how to do a thing. 191 00:20:12.230 --> 00:20:16.000 David Bau: And then some of the other papers are like, well, I'm gonna study a thing. 192 00:20:16.270 --> 00:20:24.839 David Bau: And, and we'll see what there is to find out about it, or I've discovered some phenomenon, I'm going to talk about this phenomenon. 193 00:20:25.030 --> 00:20:28.979 David Bau: And it's not… it's not like a method. And so, 194 00:20:29.170 --> 00:20:35.319 David Bau: So even though it's about this, you can see that the section… the sections are very similar. 195 00:20:35.340 --> 00:20:55.009 David Bau: to the other paper. So, because it's, like, a new thing that he's looking into, he doesn't have a special background method. He's got, like, other methods to do this, or like this. There's some new thing. And so he has a method section, but the method section is a little bit of an introduction of his terms, of, like, how to even talk about the thing that he's going to observe. 196 00:20:55.240 --> 00:21:07.240 David Bau: Right? And so he, he, it's like, he really goes in and teaches, you know, what, what leads you to think about this thing, you know, how to formulate it, what you do to kind of measure it. 197 00:21:07.300 --> 00:21:15.560 David Bau: And then his experiment section, goes and looks at this phenomenon, in, like, 4 or 5 different ways. 198 00:21:15.820 --> 00:21:21.530 David Bau: Or I had 3 bullets here, so he looks at it in 3 major different ways. Through some extra things, and then… 199 00:21:21.710 --> 00:21:28.559 David Bau: And then at the end, he says, okay, now that I brought you on this journey, here's other stuff you should know related to work and discussion. 200 00:21:28.790 --> 00:21:33.710 David Bau: Right, and so this is… oh, I think this is an ICLR, so this is a single column format. 201 00:21:33.900 --> 00:21:41.689 David Bau: But you can kind of see what this looks like. So, short introduction, no really work, just into the motivating observation in the method. 202 00:21:42.170 --> 00:21:47.869 David Bau: And and then after he describes what it is he's going to be looking at with you. 203 00:21:48.240 --> 00:21:52.170 David Bau: He goes and he tries to look at it in several different ways. 204 00:21:52.480 --> 00:21:54.789 David Bau: And, and then describes what he's seeing. 205 00:21:56.080 --> 00:22:02.779 David Bau: And then… Yeah, and then along the way, as he's doing it. 206 00:22:03.450 --> 00:22:10.049 David Bau: He's… he's trying to build this picture of what this is that he's found, and that keeps the discussion short again. 207 00:22:10.890 --> 00:22:19.010 David Bau: So it's, again, it's, like, different from, like, a science or nature paper, where the discussion section might be long and tries to describe the concepts. 208 00:22:19.640 --> 00:22:24.790 David Bau: in, like, Ganorb's paper, you try to describe the concepts along the way as you go, or maybe it's the introduction. 209 00:22:25.180 --> 00:22:26.610 David Bau: It talks about it. 210 00:22:27.350 --> 00:22:33.540 David Bau: Yeah, so, okay, so that's… so that's, that's, that's, that's just a couple papers to give a sense. 211 00:22:33.700 --> 00:22:35.590 David Bau: So, okay, so let's talk about… 212 00:22:38.290 --> 00:22:44.730 David Bau: The introduction. So, okay, introduction versus… what's the difference between an introduction And abstract. 213 00:22:47.460 --> 00:22:50.290 David Bau: What's the difference between an introduction and an abstract, you guys? 214 00:22:50.450 --> 00:23:08.580 David Bau: So, anybody? Yes? The amount of words you use. What'd you say? The amount of words you use. The amount of words he used? Okay, so which, which, then, which one would have more words? Introduction. Introduction has more words, so introduction seems to be longer. Yeah, usually introduction's longer, usually abstract shorter. It's probably not the right way to think of it. 215 00:23:08.690 --> 00:23:12.179 David Bau: I mean, you could have a very short introduction, it does its job. 216 00:23:12.550 --> 00:23:19.199 David Bau: Very briefly, and it could be very good, even if it's short in the abstract. Introduction could, 217 00:23:19.210 --> 00:23:33.300 David Bau: start with defining a problem, and go forward with… Yeah. Yeah, I think that's closer. So what's the purpose of an abstract, and what's the purpose of an introduction, maybe the way I should ask. 218 00:23:33.370 --> 00:23:35.570 David Bau: The abstract summarizes the whole paper. 219 00:23:35.710 --> 00:23:46.049 David Bau: That's right. And then what's the purpose of an introduction? My advisor always told me about, put your puzzle ended in the beginning. Put the puzzle name. That's exactly right. 220 00:23:46.220 --> 00:23:52.950 David Bau: That's exactly right. So, you know, people will say, so, okay, so the introduction is maybe the most important part of the paper. 221 00:23:53.080 --> 00:23:57.770 David Bau: But the funny thing about it is that in the introduction. 222 00:23:57.930 --> 00:24:01.249 David Bau: At least according to the standard form. 223 00:24:01.380 --> 00:24:04.339 David Bau: You're not supposed to answer the puzzle. 224 00:24:06.840 --> 00:24:08.950 David Bau: You know, you're supposed to present the puzzle. 225 00:24:09.810 --> 00:24:16.299 David Bau: And then, now, not every introduction is like that. Some introduction into the puzzle at the same time, right? 226 00:24:16.410 --> 00:24:27.000 David Bau: But the job of introduction is to, you know, make sure that the reader knows what is the puzzle, what is the question. 227 00:24:27.850 --> 00:24:29.139 David Bau: That is being asked. 228 00:24:29.430 --> 00:24:49.169 David Bau: And that's his job. And like Ted Ellison would say, and why you care about it. Like, okay, this is a puzzle, I don't care. So it should be, like, this motivational, thing to say, well, here's a grand puzzle on, but why, why you, your reader, should care. And now, the abstract is different. 229 00:24:49.290 --> 00:24:52.740 David Bau: The abstract is supposed to summarize the document, 230 00:24:53.320 --> 00:24:59.669 David Bau: For a bunch of different purposes, but the abstract is not necessarily supposed to be that. 231 00:25:00.410 --> 00:25:05.899 David Bau: motivational, well-written, right? The abstract is really just supposed to tell you Everything that's in the paper. 232 00:25:06.140 --> 00:25:09.239 David Bau: And so if you look at a standard abstract. 233 00:25:09.440 --> 00:25:13.839 David Bau: It'll literally just march through the sections of the paper, okay, so, you know. 234 00:25:13.900 --> 00:25:31.010 David Bau: we do this, you know, it's one sentence summary of the introduction, right? Then, you know, it's related to that, one sentence summary of the literature review, you know, to do this, we use this method, you know, one or two-sentence summary of, like, the methods. 235 00:25:31.110 --> 00:25:42.750 David Bau: And then we conduct this experiment, that experiment, the other experiment, the other experiment, and then, you know, our conclusion is that this. And that's, like, that's like an abstract, and it's almost like digesting every section of your paper down to, like, a sentence or two. 236 00:25:42.880 --> 00:25:45.900 David Bau: Right. And… and it's usually very boring. 237 00:25:46.010 --> 00:25:48.209 David Bau: But… but, like. 238 00:25:48.300 --> 00:25:58.439 David Bau: But what's the abstract good for? It's really important in the peer review process, because somebody is gonna have to, like, assign you one paper to be reviewed by somebody. 239 00:25:58.440 --> 00:26:09.369 David Bau: And they're like, okay, so, like, what's in this paper? Like, yeah, maybe the puzzle is something I could read your introduction and see it, but the method that you use is some… 240 00:26:09.380 --> 00:26:23.690 David Bau: weird diffusion model method or something like, oh, we need to have, like, a reader who knows that method, who would be a good reviewer for it. So you're… so you're abstract, and you're related to work, so there's a couple sections that actually kind of determine who's going to review your paper. 241 00:26:23.920 --> 00:26:26.550 David Bau: Even though it's, like, all these machine learning systems that are… 242 00:26:26.790 --> 00:26:37.529 David Bau: you know, sending your paper off to review now, it's the same thing. Like, one of the functions of a bunch of this front matter in the paper is to decide who it is that peer reviews it, so… 243 00:26:37.680 --> 00:26:39.889 David Bau: So something to… something to think about. 244 00:26:40.080 --> 00:26:44.190 David Bau: Okay, so, okay, so the introduction, but you're right. 245 00:26:44.900 --> 00:26:47.810 David Bau: The introduction must ask the main question. 246 00:26:49.030 --> 00:26:50.820 David Bau: That the paper addresses. 247 00:26:51.490 --> 00:26:54.319 David Bau: Oh, this is actually an interesting point. 248 00:26:55.020 --> 00:26:58.270 David Bau: Because the introduction has to ask the question. 249 00:26:58.680 --> 00:27:00.889 David Bau: But then, once you're done with your paper. 250 00:27:02.070 --> 00:27:03.929 David Bau: And it's easy to make this mistake. 251 00:27:04.180 --> 00:27:08.379 David Bau: You need to double-check that this question is actually answered by the paper, too. 252 00:27:12.650 --> 00:27:19.470 David Bau: Right? You know, so you want to make sure the question's really interesting and important, 253 00:27:19.590 --> 00:27:29.400 David Bau: But then if you make it too interesting and too important, then you might realize, oh, actually, that's too interesting, that's, like, beyond what our paper accurate answers. And so you need to scope the question. 254 00:27:29.520 --> 00:27:34.970 David Bau: Properly, so that when the reader gets to the end, they're like, just as promised! 255 00:27:35.100 --> 00:27:40.390 David Bau: You know, you promised that you were going to ask this question and investigate it, and just like you said you would. 256 00:27:40.810 --> 00:27:43.599 David Bau: Here it is. And that's, like, a satisfying pampering. 257 00:27:44.280 --> 00:27:45.210 David Bau: Make sense? 258 00:27:45.370 --> 00:27:51.050 David Bau: Okay, so… Here's a couple introductions, I guess. What I do. 259 00:27:52.930 --> 00:27:57.070 David Bau: Yeah, see? Here's me, here's one of my papers. 260 00:27:57.370 --> 00:27:59.040 David Bau: I'm, like, following this rule. 261 00:27:59.370 --> 00:28:02.279 David Bau: Where does a large language model store its facts? 262 00:28:02.460 --> 00:28:03.529 David Bau: Question mark. 263 00:28:03.820 --> 00:28:09.120 David Bau: It's like, if the first sentence of the introduction is a question, then you're probably following this rule. 264 00:28:09.300 --> 00:28:12.239 David Bau: You know, at least as long as it's a question that's answered. 265 00:28:12.520 --> 00:28:15.159 David Bau: By the paper. So that's a nice paper. 266 00:28:15.810 --> 00:28:19.350 David Bau: Yeah. 267 00:28:19.710 --> 00:28:22.349 David Bau: And so this is a function vector paper from Eric. 268 00:28:22.650 --> 00:28:27.759 David Bau: Since the study of the Lenos Capitalist. 1936, citation from 1936. 269 00:28:28.140 --> 00:28:33.829 David Bau: So, okay, maybe, maybe people have been interested in this for 100 years, right? Blah blah blah… 270 00:28:34.540 --> 00:28:48.470 David Bau: you know, in this paper, we report evidence that autoregressive transformers, not enlarge develop a rumored function references. So… so it's… it's a little different from posing a puzzle. It's a little bit answering it, right, right away, but it does… it does… 271 00:28:48.790 --> 00:28:58.600 David Bau: it does, you know, serve this function. It says… it says, you know, there's this idea of function references that people have been studying for a long time, here's why you should care about it. 272 00:28:58.910 --> 00:29:02.229 David Bau: And… and here, if you read this paper. 273 00:29:02.410 --> 00:29:06.839 David Bau: And we'll resolve the puzzle for you for, like, how function references might show up. 274 00:29:07.060 --> 00:29:09.200 David Bau: In, in big language models. 275 00:29:10.440 --> 00:29:15.440 David Bau: So… So, so more advice from different people, Jim Kaida. 276 00:29:20.200 --> 00:29:21.080 David Bau: It says. 277 00:29:21.350 --> 00:29:25.450 David Bau: You know, the most important thing about the paper from his point of view, is it's gotta be easy to read. 278 00:29:25.980 --> 00:29:44.240 David Bau: It's gotta be easy to read, you know, this poor academic stuff. So, who here has, like, been doing peer review for different things, like, you know… Peer review by ISNL, peer group, whatever, right? So, you know, it's one of the terrible duties of an academic to go, you know, do peer review for other people. 279 00:29:44.350 --> 00:29:48.699 David Bau: You know, you get a big pile of pen papers, you gotta read, and you gotta kinda grade each one. 280 00:29:48.920 --> 00:29:57.680 David Bau: And so, at least it's when you're going through the peer review process, but even afterwards, right, you have to sort of take pity on the readers. 281 00:29:57.910 --> 00:30:09.459 David Bau: you might have this incredibly dense amount of information that you want to convey, and you can write really, really complicated prose to, like, convey every single detail. But… but, like. 282 00:30:09.610 --> 00:30:14.679 David Bau: You know, the peer review is just trying to understand your paper and trying to figure out whether it's good or not. 283 00:30:15.020 --> 00:30:16.530 David Bau: Right? 284 00:30:16.780 --> 00:30:24.699 David Bau: And, and so, so… so you, so they're trying to figure out what is the problem you're solving. 285 00:30:24.890 --> 00:30:26.580 David Bau: Is it interesting? 286 00:30:26.700 --> 00:30:30.040 David Bau: Is it new? You know, is it cool? 287 00:30:30.160 --> 00:30:30.980 David Bau: Right? 288 00:30:31.230 --> 00:30:35.929 David Bau: And… and… and it's easy to make a mistake. 289 00:30:36.530 --> 00:30:38.050 David Bau: Of writing too much. 290 00:30:38.980 --> 00:30:41.340 David Bau: And losing track of these core things. 291 00:30:43.160 --> 00:30:58.309 David Bau: And so, where the reviewer or the… your future reader, past reviewer reads the paper, and can't figure out what it's about, whether it's new, you know, exactly what you're doing, because you have so many thoughts in your head, and you're getting them all out. 292 00:30:58.470 --> 00:31:03.780 David Bau: Right? You know? And so prioritize your thoughts, and look, this is a nice priority list. 293 00:31:04.020 --> 00:31:08.930 David Bau: All the stuff that you just have to… Make sure you're doing… Make sense? 294 00:31:09.130 --> 00:31:19.880 David Bau: There's, I mean, there's, I mean, you know, you guys are all brilliant. I, you know, reading your papers as people go through the PhD, and people have, like, brilliant things that they do, and their introductions. Oh, let me teach you a little bit of… 295 00:31:19.900 --> 00:31:35.159 David Bau: Ancient Greek history. Let me teach you a little… like, there's, like, there's brilliant things to do that can make a paper a lot of fun to read, but you have to be careful, because you need to, like, kind of do your work, right? You have to, like, you have to make it really clear, really easy to understand these things first. 296 00:31:35.480 --> 00:31:36.150 David Bau: Great. 297 00:31:37.470 --> 00:31:42.690 David Bau: And so… and so… and so his advice is, not only do you have to do these things well. 298 00:31:42.800 --> 00:31:51.430 David Bau: But it should be the first thing that you do. They should do it up front. In other words, like, you have to write a really, really good introduction, that's these things, and it should be the first thing that people read. 299 00:31:52.680 --> 00:31:58.889 David Bau: Okay, so that's… that's the connection. So, stating the problem and the concept, context is important. 300 00:31:59.040 --> 00:32:02.850 David Bau: Beyond that, Right. 301 00:32:03.280 --> 00:32:09.130 David Bau: So what… the common thing about, like, what makes it interesting. 302 00:32:09.320 --> 00:32:15.189 David Bau: is… That there are implications of what you're investigating. 303 00:32:15.570 --> 00:32:19.239 David Bau: And, and if you can clarify the implications. 304 00:32:19.440 --> 00:32:23.079 David Bau: Then it makes it clear to people why your paper is interesting. 305 00:32:24.630 --> 00:32:30.390 David Bau: So this is… this is, like, a key technique to use, especially, you guys are coming in with 306 00:32:30.630 --> 00:32:37.239 David Bau: concepts from left field that are going to be interdisciplinary. You say, hey, I'm going to study the concept of, you know, XYZ, 307 00:32:37.360 --> 00:32:42.729 David Bau: You know, computer science reader, you know, take a look at my paper. 308 00:32:42.850 --> 00:32:46.280 David Bau: And you might have to explain to them, like, why XYZ is… 309 00:32:46.390 --> 00:32:51.129 David Bau: important what the implications are, if you find one thing out or another. It'd be exclusive on it. 310 00:32:51.620 --> 00:32:57.920 David Bau: So, like, yes, the implications are obvious to you, right? 311 00:32:58.030 --> 00:33:01.380 David Bau: But if they're not obvious to your reader. 312 00:33:01.600 --> 00:33:04.320 David Bau: Then they're gonna be like, I don't know why somebody's done this. 313 00:33:04.940 --> 00:33:07.989 David Bau: Right? So they will have… they'll have empathy. 314 00:33:08.180 --> 00:33:11.380 David Bau: Or your point of view, but you have to explain your point of view. 315 00:33:16.590 --> 00:33:19.630 David Bau: Yeah, my students call this the sobot problem. 316 00:33:20.040 --> 00:33:23.080 David Bau: Yeah, so what? So, imagine the reviewer saying, so what? 317 00:33:23.280 --> 00:33:24.270 David Bau: So what? 318 00:33:24.410 --> 00:33:25.940 David Bau: Dude, I'll tell you, so what? 319 00:33:26.460 --> 00:33:27.140 David Bau: Right? 320 00:33:29.050 --> 00:33:33.140 David Bau: Oh, who knows who Donald Knuth is? Donald Knuth. 321 00:33:33.330 --> 00:33:35.549 David Bau: He always has a funny perspective. 322 00:33:35.780 --> 00:33:44.860 David Bau: And look, and here's Knoop with his advice, with his own custom font that doesn't work well, right? Knoop is famous for making custom everything. 323 00:33:45.140 --> 00:33:47.240 David Bau: And so, 324 00:33:47.780 --> 00:33:54.610 David Bau: But… but he says that not only should you pay attention to the first paragraph, the first section. 325 00:33:55.030 --> 00:34:03.869 David Bau: Of your paper, but the first paragraph is super important. And not only is the first paragraph super important, like, the first sentence is really important. 326 00:34:04.380 --> 00:34:12.420 David Bau: And so… so, Panuth would advise, that you should… really… 327 00:34:12.960 --> 00:34:17.179 David Bau: scrutinize that first sentence that you did. We'll do that exercise in a minute. 328 00:34:18.469 --> 00:34:21.420 David Bau: So here's some of his advice. He says. 329 00:34:21.550 --> 00:34:31.370 David Bau: You know, it should be an active sentence, it should get to the point, so he says, it's really bad for it to be past the sentence. An important method for internal sorting is quicksort. 330 00:34:31.530 --> 00:34:36.670 David Bau: And he says, you know, you should say quicksort, it's an important method for internal sorting, right? 331 00:34:36.969 --> 00:34:43.409 David Bau: So, so, like, don't write in a weak way, in the first sentence, like, You know, be direct. 332 00:34:43.699 --> 00:34:47.870 David Bau: You know, right off the bat, I think, you know, it's sort of his suggestion. 333 00:34:48.409 --> 00:34:54.810 David Bau: And so, okay, so… so you guys submitted things. I'm gonna pick on one, I'll pick on one soon. So Grace submitted this thing. 334 00:34:55.060 --> 00:34:58.010 David Bau: Okay? So, here's, like, an exercise. 335 00:34:59.110 --> 00:35:05.550 David Bau: So, Grape says that she did, she did, this, introduction 336 00:35:05.800 --> 00:35:14.840 David Bau: exercise by feeding her notes into Opus and pasting it in. And so, this is what Opus wrote for an introduction. 337 00:35:15.390 --> 00:35:26.950 David Bau: And the question is, which is… I think it's fine, but then you need to ask this question that I'm asking how to write, right? Which is, can we write better than Opus? 338 00:35:27.750 --> 00:35:31.540 David Bau: Think we can? I don't know if this is fixed or greater. What do you say? 339 00:35:31.850 --> 00:35:44.009 David Bau: For humanity's sake, I hope so. I hope so. I hope so. So, like, can we write better than all this? Yes, let's try. Let's try. So, the font here, I had to make the font small. 340 00:35:44.280 --> 00:35:46.739 David Bau: Because Opus is very tireless. 341 00:35:47.070 --> 00:35:49.330 David Bau: I'm happy to write lots and lots of words. 342 00:35:49.560 --> 00:35:51.350 David Bau: Right? 343 00:35:51.920 --> 00:36:00.169 David Bau: So I squeezed on and said, take a minute to read it, and think about this. And so I, I, I want… I want you to code to me 344 00:36:00.430 --> 00:36:02.879 David Bau: Let's do the Donald Knuth exercise. 345 00:36:04.870 --> 00:36:06.710 David Bau: What should be the first sentence here? 346 00:36:09.630 --> 00:36:11.330 David Bau: Would you be the first sentence? 347 00:36:21.690 --> 00:36:29.389 David Bau: I'll be very interested to see what you guys come up with. I have my things, but I, I wanna, I wanna, I wanna see what you guys think. 348 00:36:32.010 --> 00:36:33.989 David Bau: Would be a good question. 349 00:36:34.220 --> 00:36:41.679 David Bau: Okay, and… What might you say? What will happen when bank discrepancy… Is that kind of… 350 00:36:42.640 --> 00:36:45.120 David Bau: Improved phenomena, kind of thing. 351 00:36:45.690 --> 00:36:54.230 David Bau: What does it mean? Okay, think about it, think about it, yes. Yes, yes, yes. How about in the lines of, is your LLM lying to you? 352 00:36:54.570 --> 00:37:08.679 David Bau: Is your… so you could say, is your LLM aligned to you? Alying to you. Is your LLM lying to you? Yes, so I like the directness of that question, except that the main thing that I would… 353 00:37:08.860 --> 00:37:18.050 David Bau: caution me about, and you need, like, context to be able to criticize it in this way, is that LLM deception 354 00:37:18.710 --> 00:37:23.250 David Bau: is a huge… Question. It's, like, a huge area. 355 00:37:23.480 --> 00:37:28.430 David Bau: Of, like, oh, is your LLM lying to you? It includes a lot of other types of lying. 356 00:37:28.540 --> 00:37:36.839 David Bau: Besides this type. And so, if you… so you could start your paper off with that, but then you wanna… you wanna… 357 00:37:37.180 --> 00:37:44.439 David Bau: fit with that other advice I gave you, right, which is, it does need to ask an interesting question. That is a very interesting question. 358 00:37:44.790 --> 00:37:55.770 David Bau: But it has to be a question that your paper answers, right? And so… so if… if you… so is your lying to you is a little bit too broad for this paper, because it's studying a specific kind of lying. 359 00:37:56.210 --> 00:38:02.870 David Bau: Right? And, or, or… or something, a certain specific kind of behavior. So I would, I would probably… 360 00:38:03.310 --> 00:38:08.339 David Bau: Not use that sentence. Sometimes the way you do a news report would cover your paper. 361 00:38:08.600 --> 00:38:18.790 David Bau: Is your Alan lying to you? Yeah, that's more of a news… that's more like a headline. That's something that, like, Rupert Murdoch would put on the top of your paper, and then… and then you read, you know, you open the Wall Street Journal, and you're like. 362 00:38:19.310 --> 00:38:30.130 David Bau: He's like, the headline lied to me. It's not what the article is about. But Rupert would be like, well, you bought the paper, didn't you? It's like, I guess I did. 363 00:38:30.300 --> 00:38:32.780 David Bau: And so, 364 00:38:32.990 --> 00:38:42.249 David Bau: So there's… so there's… so there's a little bit of this going on. You want to avoid that, because you have to go through two rounds. You're not just getting somebody to read your paper. 365 00:38:42.450 --> 00:38:44.859 David Bau: You have to get through the peer reviewers first. 366 00:38:45.040 --> 00:38:47.869 David Bau: And the peer reviewers are like, this is false advertising. 367 00:38:48.230 --> 00:38:58.009 David Bau: You know, and they will really say this. This is actually one of the most common ways of getting a paper rejected. It says, where people say, you know, you didn't support the claims in your paper. 368 00:38:58.410 --> 00:39:00.779 David Bau: Oh, you claim to be asking 369 00:39:00.950 --> 00:39:05.300 David Bau: If you're answering the question, did your LM lie to you? But you didn't actually do that. 370 00:39:08.330 --> 00:39:09.619 David Bau: Any other ideas? 371 00:39:12.010 --> 00:39:16.480 David Bau: Yeah. At least you're not allowed to say anything, because it's your, you're writing it, and… 372 00:39:16.630 --> 00:39:28.839 David Bau: This is your fault. Yeah, it was my fault. To clarify, it's much shorter than the list of bullet points that I fed in. Oh, okay. 373 00:39:29.120 --> 00:39:32.300 David Bau: Yeah. It's interesting, alright, yeah. Yeah. 374 00:39:32.620 --> 00:39:38.150 David Bau: Any other, any other ideas? A combination of what they argued, the third… 375 00:39:39.480 --> 00:39:44.749 David Bau: So the third sentence… so can you read the… tell me which words? Yeah, we argue. We argue… 376 00:39:45.080 --> 00:39:49.600 David Bau: That sycophancy has not been rigorously characterized. 377 00:39:49.840 --> 00:39:55.980 David Bau: Or something like that, but, like, do you like this sentence? Yeah, but the person doesn't know, maybe they don't know much to… 378 00:39:56.190 --> 00:40:02.060 David Bau: Yeah, they maybe don't know what it is. It's hard. It's a hard job you have on that first sentence. What the heck can you do? 379 00:40:02.380 --> 00:40:09.730 David Bau: Right, because they… they need to understand what it is, and… and but… but I agree with you, this is a, like… 380 00:40:10.100 --> 00:40:17.380 David Bau: Like, this sentence is appealing, Because it describes what you're trying… to do. 381 00:40:18.020 --> 00:40:27.490 David Bau: But then it has the problem that it doesn't explain what it is. Yeah, I feel like you could do the, you know, clickbait first thing, where you're like. 382 00:40:27.510 --> 00:40:39.839 David Bau: Yes? Is your LLM lying to you? Please you? Please you, yes, okay, yes, yes. Sicovince, past work on Sykvins, he would suggest. Okay. Yes. Okay, so you could immediately follow it up with… 383 00:40:39.980 --> 00:40:45.370 David Bau: But it didn't… they didn't differentiate… 384 00:40:45.690 --> 00:40:48.959 David Bau: The user level kind of confess in its own. 385 00:40:52.360 --> 00:40:54.089 David Bau: No repeat for me. 386 00:40:55.630 --> 00:41:05.290 David Bau: Yeah, but I think that's… that's right. You can have a pair of sentences, so you're not… that first sentence doesn't stand alone, so if you come in right away with your caveats. 387 00:41:05.370 --> 00:41:13.990 David Bau: you know, as long as those caveats are interesting to read, then you can rescue the first sentence before the… yeah. Probably what I want to do, I was… 388 00:41:13.990 --> 00:41:26.640 David Bau: I would probably try to get a, like, quote from one of the past, like, popular anthropic papers about pregnancy, where they suggest, like. 389 00:41:26.690 --> 00:41:34.870 David Bau: It's trying to blog with your private, like, it'll, like, have some sentence where, like, this is the… Both the thing. But, like, is this an Android? Like, like… Yes. 390 00:41:35.250 --> 00:41:45.189 David Bau: Yeah. Yeah. And, anthropic depreciation. And throw all your good stuff on. Yeah. 391 00:41:46.790 --> 00:41:56.689 David Bau: Yeah, so you could use a quote to give context, and that helps make it so that, well, I'm not saying the crazy things, and it said the crazy thing. 392 00:41:57.130 --> 00:42:02.029 David Bau: Yeah, any other ideas? Like, okay, okay, you guys, how would you… what would you pick for the first sentence? 393 00:42:03.060 --> 00:42:06.889 Jasmine C.: What about, like, a historical anecdote? So something like, 394 00:42:07.000 --> 00:42:13.760 Jasmine C.: During the Great Leap Forward, Mao's officials told him exactly what he wanted to hear, and, like, 30 million people died for it. 395 00:42:15.600 --> 00:42:23.550 Jasmine C.: Well, I'm just saying, like, some… sometimes it's, like, it's, like, illustrative, you're like, look, like, yes men bring down countries, like, I don't know. 396 00:42:24.340 --> 00:42:36.270 David Bau: You're better than Rupert Murdoch! This is… this is gonna sell some papers. No, I think this is… I like it. I think that's great because it communicates 397 00:42:36.380 --> 00:42:44.739 David Bau: why you're interested in it. But then I think that you, right after you do that, you do have to do the thing that Grace talks about. You have to sort of rescue this. 398 00:42:44.740 --> 00:42:46.479 Jasmine C.: Yeah, of course, of course. 399 00:42:47.020 --> 00:42:51.900 David Bau: Does that make sense? Yes. I think that maybe when we could put it as well as, like. 400 00:42:52.220 --> 00:42:54.350 David Bau: Basically, like, the way… 401 00:42:55.360 --> 00:43:13.049 David Bau: the way that you deal with problematic behavior depends on the root cause of it. Yes, yes. And so, like, if someone is lying to you because they're, you know, a 4-year-old that wants more candy, or if they're lying to you because they're, you know, a ruthless sociopath, you know, these are different, and so… Absolutely. 402 00:43:13.050 --> 00:43:16.840 David Bau: We can say that, like, people have documented Awesome. 403 00:43:17.100 --> 00:43:29.940 David Bau: behavior, but have not actually tried to look at the psychology of… So, as a writing exercise, right, you see a few different approaches here. So, you know, one of the approaches is 404 00:43:30.180 --> 00:43:34.879 David Bau: if the job is to ask a question, you just ask the question. 405 00:43:35.300 --> 00:43:45.800 David Bau: Up front. And then you can… and then while we're asking questions, you can sex it up. You can be like, you know, so, you want to kill half a million people? Right? You know, kind of thing. 406 00:43:46.030 --> 00:43:47.420 David Bau: Right? 407 00:43:47.680 --> 00:43:59.200 David Bau: So, you know, you can ask the question, but then there's a second job that you're trying to do in the introduction, which is make sure that the reader understands your scope. 408 00:43:59.360 --> 00:44:01.889 David Bau: They understand, like, what… 409 00:44:02.010 --> 00:44:08.630 David Bau: Like, so, like, there's this tension between the broad, expansive question that you want to ask. 410 00:44:08.790 --> 00:44:13.760 David Bau: And the narrow scope of, like, what you're actually going to be able to do in these 8 pages. 411 00:44:14.100 --> 00:44:18.160 David Bau: And we're navigating between them, and you see that there's this… 412 00:44:18.360 --> 00:44:21.009 David Bau: Two ways that introductions are often written. 413 00:44:21.530 --> 00:44:32.770 David Bau: Which is doing one, and then the other, or the other, and then the one. And they're both… they're both fine, right? You know, it's a style choice for the specific paper of what you might… 414 00:44:33.040 --> 00:44:42.960 David Bau: It's like, oh, let me tell you the big question, oh, and, you know, here's my scope, or let me introduce a little bit of scope, and then this, you know, once you have the scope, here's… 415 00:44:43.140 --> 00:44:50.609 David Bau: the question that you might want to know. And so, the problem that I have with what Opus wrote here, like, personally. 416 00:44:50.850 --> 00:44:52.600 David Bau: is that… 417 00:44:53.650 --> 00:44:58.580 David Bau: It, like, mixed up scope and con… you know, it mixed up, you know, the scope and the question. 418 00:44:58.690 --> 00:45:01.079 David Bau: In this big paragraph, which is mostly scope. 419 00:45:01.880 --> 00:45:12.220 David Bau: It's mostly about scope, but also, like, talks a little bit about the question, you know, hints at the question in there, but not totally clear, right? And then, and before, like, so, so, so see these three words? 420 00:45:12.790 --> 00:45:16.319 David Bau: In this work, blah, blah, blah. So as a reviewer. 421 00:45:16.460 --> 00:45:27.679 David Bau: When you're, like, reading a paper that's kind of wordy and not that well written, right? You, like, skip, skip, skip, skip, and then you're flipping pages, and, oh, look, on page 3, finally, like, in this work. 422 00:45:27.830 --> 00:45:32.200 David Bau: Right? And then you're like, ha! Now I can read, like, what the heck this paper's about. 423 00:45:32.320 --> 00:45:33.840 David Bau: Right? That makes sense? 424 00:45:34.330 --> 00:45:51.769 David Bau: because they're just, like, they have too much they want to get out of their head. And so Opus here has too much, it's trying to get a… you know, Grace… Grace gave Opus, you know, 53 bullet points, and, like, Opus is like, okay, I better, like, talk about all that stuff, like, my head is full of this stuff. And then finally it says, in this work is what's going on. But… but, like, you know, this… so, like… 425 00:45:51.860 --> 00:45:58.229 David Bau: Like, in concept, whatever job this sentence is doing, or this paragraph is doing, like, that could be the first sentence. 426 00:45:58.360 --> 00:46:05.920 David Bau: Except that… except that, like, this… the way that Opus has constructed it here, Hence, like, totally dependent on 427 00:46:06.090 --> 00:46:11.010 David Bau: this, this paragraph. So, but you could imagine, like, writing this paragraph in some way that, you know. 428 00:46:11.440 --> 00:46:22.750 David Bau: you know, it could stand alone, and that's a good first sentence. Or it could be a good sentence, second sentence. And that's sort of the way that I… I think of it. So you want to see what I wrote? 429 00:46:26.800 --> 00:46:28.509 David Bau: What would you say, you guys? 430 00:46:29.260 --> 00:46:31.169 David Bau: What would your first sentence be? 431 00:46:33.790 --> 00:46:35.890 David Bau: Okay, we'll do this in a minute. 432 00:46:36.530 --> 00:46:39.960 David Bau: So I, so I have, I have a couple attempts here. 433 00:46:40.810 --> 00:46:51.609 David Bau: But I did, I did the thing that you guys suggested. I said, oh, let's ask a question up front. We'll see if that works. So I said, are LLMs being deceptive when they act as sacrifice? 434 00:46:51.790 --> 00:46:53.680 David Bau: Or are they just getting confused? 435 00:46:54.220 --> 00:46:55.000 David Bau: Right. 436 00:46:55.290 --> 00:46:57.640 David Bau: So I, so I tried to… 437 00:46:58.310 --> 00:47:02.649 David Bau: I tried to do two things in that same sentence, I don't know if it's successful. 438 00:47:02.820 --> 00:47:06.629 David Bau: Right? Which is, I'm asking the question that you're asking. 439 00:47:06.950 --> 00:47:11.710 David Bau: Is your LLM lying to you? But I'm trying to scope it down at the same time. 440 00:47:11.950 --> 00:47:22.020 David Bau: I'm trying to scope it down. I'm trying to say, okay, I'll take this word sucripantic and stick it in there. It wasn't quite the right word form for the sentence I was constructing, so I said it a different way. 441 00:47:22.130 --> 00:47:26.080 David Bau: Right, but maybe close enough that people know that I'm talking about succinctly. 442 00:47:26.430 --> 00:47:32.330 David Bau: And… and then… and then I… and then, you know, when I digest what was going on in the… 443 00:47:32.490 --> 00:47:35.210 David Bau: First paragraph about what the contrast is. 444 00:47:35.590 --> 00:47:45.139 David Bau: of what's doing, I say, oh, actually, it's like an either-or. It's like, are they being semantic and kind of deceptive to you, or is it just, like, they're just messing up, they're just getting confused? 445 00:47:45.250 --> 00:47:49.869 David Bau: Right? And so I feel like if that's the main argument, you can kind of put it up front. 446 00:47:50.330 --> 00:47:53.709 David Bau: And say, this is what we're going to try and distinguish between, and then you have to… 447 00:47:53.950 --> 00:48:00.989 David Bau: And then you have to, like, say something. Oh, look, and then this is the thing that I… that… this is… this is the thing that Grace says that you should do. 448 00:48:01.290 --> 00:48:14.049 David Bau: You know, the appearance-specific behavior has been widely observed with quote, quote, quote, quote, quote, quote, I don't know what quotes, but, like, I could, you know, you could do a little literature review and say, oh, Anthropic says this and that, right? And then you could cite somebody and say, this is why you care. 449 00:48:14.200 --> 00:48:21.290 David Bau: Right? And it actually leads users to assume this, and so this is… this is why Anthropic cares. Maybe this is why you care. 450 00:48:21.690 --> 00:48:28.399 David Bau: Right? So, like, we're gonna do this, this is why you should care, everybody cares about it, so don't accuse me of… 451 00:48:28.770 --> 00:48:30.939 David Bau: Doing a piece of research, nobody cares about 452 00:48:31.230 --> 00:48:38.649 David Bau: I've proved to you twice in the first paragraph somebody cares about it, and then… and then… and then I do the thing that, like, I didn't like. 453 00:48:39.060 --> 00:48:52.049 David Bau: Which is, I have this in this paper thing, and a little too late, in the second paragraph, here we go. In this paper, we present with our… based on this, and we do this, right? So… but… but this is the job. So this is… so this… in this paper sentence. 454 00:48:52.160 --> 00:48:54.890 David Bau: It's the one that's doing the job of telling you, like, what the… 455 00:48:55.110 --> 00:49:02.589 David Bau: what the paper's about, what the core puzzle is that you're solving, or something like that. But this is my… this is my first attempt to rewrite. It's fewer words. 456 00:49:02.980 --> 00:49:05.580 David Bau: Right? You can kinda, kinda see, right? 457 00:49:05.760 --> 00:49:09.599 David Bau: And, and then after I wrote this, and I was like. 458 00:49:09.970 --> 00:49:13.609 David Bau: And then you edit it again, right? So here's my second attempt, you can see. 459 00:49:14.110 --> 00:49:22.509 David Bau: And I just deleted all the stuff in between this, and I said, okay, you know, you should just get to end this paper. So I take the end paper sentence, and just shove it up as far as I can. 460 00:49:22.760 --> 00:49:29.479 David Bau: And so the first… I don't know, I thought the first sentence was okay, I might have removed a couple words, but, like, okay, our elements means that there's… 461 00:49:29.790 --> 00:49:36.390 David Bau: And then… and then we say, in this paper, we present evidence that AI sufficiency might not arise from this, but from that, or something like that. 462 00:49:36.530 --> 00:49:39.210 David Bau: And then now, this is, like. 463 00:49:39.900 --> 00:49:42.450 David Bau: You know, okay, I'm gonna prove this. 464 00:49:43.070 --> 00:49:46.990 David Bau: Right? Just, like, as quickly as you can, you just, like, get to the point. 465 00:49:47.230 --> 00:49:53.920 David Bau: you know, you might not agree with me, but I, you know, in the next 8 pages, I'm gonna argue, you know. 466 00:49:54.120 --> 00:49:55.310 David Bau: Or something like this. 467 00:49:55.480 --> 00:49:57.189 David Bau: Right? That make sense? 468 00:49:57.690 --> 00:50:00.639 David Bau: Yeah, so you see? See how much I deleted? 469 00:50:01.120 --> 00:50:01.810 David Bau: Yeah. 470 00:50:03.480 --> 00:50:14.230 David Bau: And then… and then… and then you say, okay, and then… and then introduction, and then you might be like, okay, then I can actually scope it more. I can say, in this… to say this, we begin with this, I'm gonna define how we're going to ask the question. 471 00:50:14.590 --> 00:50:22.470 David Bau: Right? So this is a little bit of giving away the game, but it's an analysis paper, so the… you know, you have this question. 472 00:50:22.930 --> 00:50:24.610 David Bau: So in a methods paper. 473 00:50:24.740 --> 00:50:29.759 David Bau: you probably didn't… don't want to start with this. Like, in a methods paper, you don't want to say. 474 00:50:29.890 --> 00:50:36.790 David Bau: In this paper, we're going to prove that our method is the best method, since all methods have ever been invented. 475 00:50:37.160 --> 00:50:40.609 David Bau: Like, we're the best. Like, you wouldn't say that in the introduction. 476 00:50:40.800 --> 00:50:53.739 David Bau: Right? Because that's a little bit of a puzzle. In this paper, you know, we ask, is there a better way of doing this method? You know, there's some indication that it should be a good way, so there's this nice thing that you can do. 477 00:50:53.830 --> 00:51:08.149 David Bau: we bring this idea and that idea together, to bring our method together. And then later on, you know, we ask, you know, is this any better? Right? And then you can ask… answer that and the rest of the story. But in an analysis paper, which I think this is an analysis paper. 478 00:51:08.580 --> 00:51:11.500 David Bau: then I think that the, 479 00:51:11.770 --> 00:51:17.260 David Bau: the answer to your puzzle is the analysis itself. So, you state your hypothesis. 480 00:51:17.460 --> 00:51:23.220 David Bau: You say, hey, I'm gonna say this controversial thing that you might not even think is measurable. 481 00:51:23.710 --> 00:51:28.960 David Bau: Like, does… does an AI really believe what it says? 482 00:51:29.440 --> 00:51:47.670 David Bau: Right? And then you're like, that's not even a scientific question, right? How do you do it? And then you're like, we think we can answer that. We're gonna do this, we're gonna probe this, we're gonna look at that, we're gonna change this. I think that we're gonna try to get a signal on whether Alan believes what it says, or something, right? 483 00:51:48.710 --> 00:51:56.150 David Bau: And then how that all rolls out is, like, the interesting meat in the paper. And so you can kind of think, like, what is… 484 00:51:56.430 --> 00:51:58.960 David Bau: Like, what's… what's the… 485 00:51:59.290 --> 00:52:04.239 David Bau: what's the interesting thing that you're answering? It might not literally be the answer to the question. 486 00:52:05.510 --> 00:52:13.390 David Bau: Makes sense. Okay. And then you'd hold that back. Okay, so your turn. So let's do this. Okay, everybody, gonna do this. 487 00:52:13.580 --> 00:52:17.330 David Bau: So, so this is the URL, so if you go to the course website. 488 00:52:18.500 --> 00:52:23.110 David Bau: click on the link that says… I forget what it's called, like, interactive. 489 00:52:23.790 --> 00:52:29.320 David Bau: You know, editing, And, and do these 5 steps. Will you do these 5 steps? 490 00:52:30.210 --> 00:52:32.160 David Bau: So, it's a spreadsheet. 491 00:52:32.780 --> 00:52:34.409 David Bau: Put your name in column 1. 492 00:52:35.230 --> 00:52:37.470 David Bau: To claim a random piece of writing. 493 00:52:37.760 --> 00:52:39.440 David Bau: Don't put your name next to your own. 494 00:52:42.260 --> 00:52:45.190 David Bau: Until… until all the… until all the rows are full. 495 00:52:46.710 --> 00:52:47.380 David Bau: Right. 496 00:52:47.790 --> 00:52:53.660 David Bau: If you didn't submit one, it's okay, you can put your name next to one anyway, but just make sure all the rows are full. If the rows… 497 00:52:53.900 --> 00:52:55.040 David Bau: If there's a… 498 00:52:55.220 --> 00:52:59.580 David Bau: And this empty road payment. And if all the roads are already full, then take a break. 499 00:53:00.050 --> 00:53:02.179 David Bau: We can put your name, you can put a second name next to one. 500 00:53:02.930 --> 00:53:08.620 David Bau: And then… and then take the text out and copy it out to a Google Doc, so that you just write, like. 501 00:53:08.760 --> 00:53:10.680 David Bau: You know, in a regular Google Doc. 502 00:53:11.080 --> 00:53:16.390 David Bau: And then think about, what is the first sentence? What's the best first sentence? 503 00:53:16.410 --> 00:53:32.739 David Bau: your comment might be that this is already a brilliant first sentence, I want to change it, and it's okay. But, like, really think about this question. What do you think is the best first sentence, the first two or three sentences of the thing, and then make a shot at editing the document to just write what you think would be the… 504 00:53:33.860 --> 00:53:38.600 David Bau: The best, like, the most banger first sentence for the paragraph that you picked. 505 00:53:38.950 --> 00:53:39.700 David Bau: Right. 506 00:53:41.400 --> 00:53:46.629 David Bau: Or the first… it could be a few sentences. And then… and then once you're done, so just think about that for… 507 00:53:47.070 --> 00:53:48.250 David Bau: 10 minutes? 508 00:53:48.800 --> 00:53:50.950 David Bau: Once you're done, paste your suggestion back. 509 00:53:51.890 --> 00:53:58.629 David Bau: And alcohol time at… I'll call time in 10 minutes. So, relax, guys, and work on it. 510 00:53:58.800 --> 00:54:02.230 David Bau: And then, and then, and then we will reveal… 511 00:54:02.890 --> 00:54:06.730 David Bau: Whose thing you edited, and go find who edited your thing. 512 00:54:07.060 --> 00:54:07.760 David Bau: I can't 513 00:55:34.240 --> 00:55:36.609 David Bau: These are intros, right, so these are not abstracts. 514 00:55:39.980 --> 00:55:42.770 David Bau: Now that you know the difference between an intro and abject. 515 01:04:39.360 --> 01:04:40.909 David Bau: Hey guys, so, 516 01:04:41.020 --> 01:04:46.380 David Bau: If you ever put… pasted your thing, your suggestion into the spreadsheet, take a minute to do that now. 517 01:04:46.560 --> 01:04:54.250 David Bau: You can take the sentence that you've written, or on your notes, and paste them into the third column. 518 01:04:56.050 --> 01:05:00.529 David Bau: And, give me a minute to do that. And once you're done with that. 519 01:05:01.780 --> 01:05:07.800 David Bau: I want you to find the piece of writing that you wrote, submitted last night, or the other night. 520 01:05:08.130 --> 01:05:11.749 David Bau: And find who it is who's been editing. 521 01:05:12.500 --> 01:05:13.520 David Bau: And go find them. 522 01:05:14.340 --> 01:05:19.789 David Bau: And ask them, why God would have you done to my writing it? 523 01:05:20.920 --> 01:05:22.440 David Bau: Alright, do you do that? 524 01:05:23.570 --> 01:05:39.330 David Bau: Yeah, and go and just chat, and then chat a little bit about the sentence and about thoughts. Yeah, like, actually stand up, everybody stand up. Are people gonna be going to, like… Oh, yes, okay, yes, so it's a little bit chaotic, so, try to find… 525 01:05:40.020 --> 01:05:44.580 David Bau: And then if… if they're already talking to somebody, then wait. 526 01:05:44.840 --> 01:05:48.339 David Bau: Until somebody wants to chat with you? Sorry, it's… 527 01:05:48.470 --> 01:05:56.210 David Bau: there's probably, multiple cycles of people, but… but I think that we could probably do it about 2 rounds. I think 10 minutes… 528 01:05:56.360 --> 01:06:06.309 David Bau: There's no easy way to define it. There's no easy way to define it. So, so, yeah, so, because we did it in this chaotic way, 529 01:06:06.530 --> 01:06:09.929 David Bau: But, I wouldn't go down the… we could go down the list. 530 01:06:10.030 --> 01:06:16.179 David Bau: So who, who was the… who… who had their thing edited by, you know, Avery? Who was, like, the first on the list? 531 01:06:16.330 --> 01:06:18.489 David Bau: I don't know who's that, who's that was. 532 01:06:18.880 --> 01:06:24.330 David Bau: So yeah, so come back to Avery. And who was the second on the list? 533 01:06:25.410 --> 01:06:27.950 David Bau: supposed to say. Oh, just, just, just. 534 01:06:29.350 --> 01:06:31.330 Jesseba Fernando: Second on the list is me. 535 01:06:31.720 --> 01:06:40.300 David Bau: Oh, it was you? Oh, okay, sorry, you're… so you're… you're online. So, what you can do is you can log into the Zoom and chat, chat. Who, who was, 536 01:06:40.440 --> 01:06:44.269 David Bau: So you were the… so Emery edited yours, right? So… 537 01:06:44.270 --> 01:06:45.140 Jesseba Fernando: Maybe. Yep. 538 01:06:45.670 --> 01:06:49.899 David Bau: Maybe Emery can, like, log into the Zoom and then chat with you a little bit, but I can mute the Zoom. 539 01:06:50.110 --> 01:06:55.640 David Bau: So that you're not talking out loud, or what you can do is you can put in the Zoom chat. 540 01:06:56.530 --> 01:07:02.180 David Bau: Just have a, maybe your own Zoom that you can use to chat with, Amy. 541 01:07:02.180 --> 01:07:03.569 Jesseba Fernando: Oh, yeah, I can do that. 542 01:07:04.190 --> 01:07:05.299 David Bau: That's great, thanks. 543 01:07:06.410 --> 01:07:13.139 David Bau: And then, Okay, and then, and then down, and down the list. So, let's… 544 01:07:13.540 --> 01:07:16.470 David Bau: Basically, all I wanted to do is just talk about, like, 545 01:07:16.600 --> 01:07:20.880 David Bau: you know, what, 546 01:07:21.100 --> 01:07:23.819 David Bau: You know, why you chose a different first sentence. 547 01:07:24.250 --> 01:07:26.840 David Bau: You know, what might be missing, that type of thing. 548 01:07:27.030 --> 01:07:30.479 David Bau: You know, does this really capture what's needed? 549 01:07:30.730 --> 01:07:32.360 David Bau: Just, just for 5 minutes. 550 01:07:36.150 --> 01:07:37.590 David Bau: Nobody's talking about. 551 01:07:37.760 --> 01:07:45.010 David Bau: Oh, okay, should we just keep on going down the list? Who, who, who, who is… Yu Chen added one. Where is Yu-chan? 552 01:07:45.160 --> 01:07:50.619 David Bau: Oh, here you. Yeah, yeah, so, so, yeah, go, go talk to whoever, whoever's… 553 01:07:51.120 --> 01:07:59.449 David Bau: you know, whoever your… yours is from. And then Claire edited one, but you're already talking, so you… Oh, that's… 554 01:07:59.460 --> 01:08:13.229 David Bau: Grayson, Grace said the next one, so, I noticed the broken one, and Grace had any people taking this starter. Or maybe what we should do is we should, like, you could reveal who you were that I deleted. 555 01:08:13.280 --> 01:08:14.440 David Bau: Should I find that? 556 01:08:15.320 --> 01:08:16.859 David Bau: I can find it, I have it in this 557 01:08:17.950 --> 01:08:24.369 David Bau: Should I just put whoever it was so that you guys continue to check that out? Alright. Let me go do that. I'm sorry. 558 01:08:26.109 --> 01:08:30.109 David Bau: I, I shall review. 559 01:08:30.830 --> 01:08:31.660 David Bau: bubbled. 560 01:08:32.410 --> 01:08:43.719 David Bau: Improving a formula, like… We show that. Maybe it does have answered this question, or at least helps to… 561 01:08:44.060 --> 01:08:44.840 David Bau: Oops. 562 01:08:45.120 --> 01:08:46.439 David Bau: Okay, oh. 563 01:08:47.240 --> 01:08:48.769 David Bau: No problem, I think. 564 01:08:50.960 --> 01:08:53.019 David Bau: Okay, now you know who you should talk to. 565 01:08:55.240 --> 01:08:57.219 David Bau: Original, right? Oh, there it is! 566 01:09:00.830 --> 01:09:01.569 David Bau: Chill. 567 01:09:02.529 --> 01:09:10.050 David Bau: Yes, I would. Yeah. The first two sentences is fine. 568 01:09:10.660 --> 01:09:17.180 David Bau: That's what I did, too. And then I just started writing, so… No, I don't… 569 01:09:17.410 --> 01:09:41.850 David Bau: Oh, that's okay. I'm good, I'm good. And then you can… so, if your student is online, then… Trying to make rounds in Canada. 570 01:09:41.850 --> 01:09:45.999 David Bau: They are. They are at very similar cases a little bit. 571 01:09:46.040 --> 01:10:05.820 David Bau: However, it's not a canon. 572 01:10:06.770 --> 01:10:13.780 David Bau: Senator Diaz. I don't think… 573 01:10:13.780 --> 01:10:38.170 David Bau: I wonder if that… burning is going to definitely have to change based on the rest of the space, because, well, I guess what I'm trying to say is, I mean, I tried to pull out the things I thought you were in pieces, but I, you know. 574 01:10:38.170 --> 01:10:45.140 David Bau: Especially as it gets further down. But I put this question in my beginning, this just came out. 575 01:10:45.850 --> 01:10:57.510 David Bau: That's okay. Now, I realize, like, after this class, I need to definitely… 576 01:10:57.610 --> 01:11:13.790 David Bau: And also, like, we have not… I'm not understanding. Oh, okay. 577 01:11:13.820 --> 01:12:01.040 David Bau: Whether it's something that… Yeah, I'm not sure. I don't know, I find the introduction… 578 01:12:01.040 --> 01:12:22.439 David Bau: My advisor always seems to write the introduction box. Like, write everything else, and then once you have realized that there was that sufficient, I can't go back. 579 01:12:22.440 --> 01:12:25.300 David Bau: outrage my work. 580 01:12:25.460 --> 01:12:33.470 David Bau: No, it's okay, Bob, it's like a lot short time. 581 01:12:33.470 --> 01:12:57.790 David Bau: Oh, right. Yeah, so, no, it's… 582 01:12:57.790 --> 01:13:14.090 David Bau: I'm going to just change this to discriminations. 583 01:13:14.090 --> 01:13:38.349 David Bau: No, I called the word that, investors look at a company's currently called them, I seamlessly, I call them decisions. 584 01:13:38.350 --> 01:13:52.960 David Bau: I, 585 01:13:59.140 --> 01:14:08.449 David Bau: But the most important. 586 01:14:47.790 --> 01:14:48.870 David Bau: I don't know that. 587 01:15:12.590 --> 01:16:00.180 David Bau: So, so now… It makes it easier. 588 01:16:00.180 --> 01:16:14.150 David Bau: What is it? Right. Actually, I know, it's not, it feels, it feels, it feels like… 589 01:16:20.640 --> 01:16:21.580 Kai Nylund: That's amazing. 590 01:16:31.560 --> 01:16:35.470 David Bau: No, why… why do they always do that? The idea that they can't do that. 591 01:16:37.580 --> 01:16:40.460 David Bau: bed… I'm gonna, pick up the background. 592 01:16:41.290 --> 01:16:46.280 David Bau: And continue lecturing for a bit now that everybody's chatted for a couple minutes. Is that okay? Everybody's cool? 593 01:16:46.390 --> 01:16:54.929 David Bau: I have a few more things to say, and just a few more things to relate, okay? And we can chat more. We'll have another writing exercise, so thank you guys for kind of warming up. 594 01:16:55.050 --> 01:17:06.380 David Bau: with this warm-ups exercise for a few minutes, we're gonna do more of it on the Thursday, so now that you get the idea, you know, on Thursday, so who here thinks that 595 01:17:07.840 --> 01:17:10.000 David Bau: It needs to be a little more organized than what I do. 596 01:17:11.030 --> 01:17:13.869 David Bau: Okay, so I will organize it a little bit better. 597 01:17:14.490 --> 01:17:17.199 David Bau: I'll, like, maybe make a spreadsheet of, like. 598 01:17:18.050 --> 01:17:36.010 David Bau: of, like, rounds, of, like, who talks to who on which round, and stuff like that, so it's a little bit clearer. But to give me some ingredients, we're gonna… look, I want to try it on paper, so that it's sort of physical, so come, come, you know, come if you can, in person to class. 599 01:17:37.140 --> 01:17:43.680 David Bau: If you're stuck on Zoom, then, you know, put them on Google Docs, it's fine, right? But, like, I'd like to see if we can… we'll just try it on paper. 600 01:17:43.800 --> 01:17:49.579 David Bau: Just for one round. We'll probably go back to Google Docs after this, right? But I think it's a nice exercise. 601 01:17:49.990 --> 01:17:52.639 David Bau: Okay. So, alright. 602 01:17:52.850 --> 01:18:04.509 David Bau: So, that's introduction, and… and… and you get the idea, right? So, okay, so let me talk about the rest of the sections of the paper, because when you write your thing for Thursday, I want you to write… 603 01:18:05.100 --> 01:18:17.689 David Bau: of stuff as well, right? So, so after you talk about introduction and actually talk about related work, you, you talk about the… you talk about the background and method. So this is where we teach. 604 01:18:17.820 --> 01:18:24.409 David Bau: So, one of the things that I always tell my students is that research He's a teaching. 605 01:18:25.550 --> 01:18:30.240 David Bau: Research is teaching. Research is not… just discovery. 606 01:18:31.000 --> 01:18:31.700 David Bau: Right? 607 01:18:31.970 --> 01:18:33.440 David Bau: Research YouTube. 608 01:18:33.970 --> 01:18:34.930 David Bau: Teaching. 609 01:18:35.080 --> 01:18:41.949 David Bau: There's, you know, it's not just that Columbus might fit find Latin America. 610 01:18:42.060 --> 01:18:43.120 David Bau: Or something. 611 01:18:43.220 --> 01:18:46.810 David Bau: That he came back and wrote all these books. 612 01:18:47.090 --> 01:18:50.720 David Bau: about it. That's why it's so amazing. 613 01:18:51.360 --> 01:18:54.320 David Bau: You know, if he had just gone there and 614 01:18:54.760 --> 01:18:58.629 David Bau: Hung out there and retired, then it, like, wouldn't have been as amazing. 615 01:18:59.120 --> 01:19:16.420 David Bau: And so… so don't just go and do your discovery and go and hang out and rehire. You're gonna teach it. And so, this is where the teaching happens, in the background method section. It's, like, where the technical details are taught. And so, so why is this thing covering my, my slide? But… but it's, 616 01:19:17.270 --> 01:19:21.309 David Bau: But the idea here is… Where's… where's mine? 617 01:19:22.330 --> 01:19:23.240 David Bau: Mouse. 618 01:19:24.400 --> 01:19:25.349 David Bau: Hmm, not here. 619 01:19:26.500 --> 01:19:29.039 David Bau: Love these custom computer systems. 620 01:19:29.470 --> 01:19:32.430 David Bau: Set up professional audio. Great thing, it's like an advertisement. 621 01:19:33.120 --> 01:19:41.460 David Bau: Okay, so… Okay. 622 01:19:42.440 --> 01:19:43.730 David Bau: So this is, like. 623 01:19:46.800 --> 01:19:54.040 David Bau: This is, like, a typical… section, like, paragraph in a method section, blah blah blah blah blah, right? So… 624 01:19:55.080 --> 01:19:58.380 David Bau: So the… the thing… the thing that you want to do 625 01:19:58.950 --> 01:20:10.279 David Bau: like, if you have, like, if you have an equation of any technical material, one, one Greek letter, right, one, one formula, anything, it doesn't matter, like, you, you, you have some numbers, right? 626 01:20:11.050 --> 01:20:14.100 David Bau: You know, to have pity on your reader. 627 01:20:14.580 --> 01:20:24.479 David Bau: take some time and make a figure that explains the technical idea. Like, if you were trying to teach your idea to an undergrad, who's like, hey, come and join my project. 628 01:20:24.800 --> 01:20:43.360 David Bau: And the undergrad's like, I don't get it, I don't get it, I don't get it, right? Then you would probably take your whiteboard marker, you'd probably draw something on the whiteboard and help them understand it. So draw that. Draw that for your reader, right? Take some time and draw a nice figure that explains what's going on. And so there's some secrets for making a nice figure. 629 01:20:43.370 --> 01:20:54.340 David Bau: Right? But if you… like, so there's a lot of different figures you can always make for a paper, there's a lot of different ones, and so… but a really good figure for, like, a figure 1 serves as a conceptual map. 630 01:20:54.360 --> 01:20:58.519 David Bau: the paper. And there's a lot of different things that you can do, and a really good 631 01:20:58.650 --> 01:21:12.509 David Bau: bigger to help out a paper, right? Like, like, you have to… you have to have empathy with your reader and think about all the weird jargon that you're using that… that, like, is natural to you, but to a new reader, they don't… they won't, like, understand what that word is. 632 01:21:12.510 --> 01:21:28.239 David Bau: So if you… if there's any words like that, it's like, my… my… my blah blah blah, sophisticated, awesome data set, you know, my… my R3, you know, user, you know, group, my, my, my whatever, right? Like, if there's any jargon. 633 01:21:28.370 --> 01:21:48.230 David Bau: it should be on the figure, like, with the jargon on the figure, like, next to, like, the arrow or the box that that jargon, like, indicates. Like, don't… don't, like, use jargon in the paper and then not put it in the figure. The whole point of the figure is to make it easier to understand your jargon, so the jargon should be here, too. So, like, so, like, all these crazy Greek letters, like. 634 01:21:48.670 --> 01:21:54.260 David Bau: This is a good figure, because all the Greek letters are on the paper… on the figure. Like, they're in here. Like, you might still not understand them. 635 01:21:54.570 --> 01:22:08.499 David Bau: But at least they're in the figure, so you can point at them, and if you see them in the paper and you don't understand what's going on, and you're puzzling over it, at least when you're puzzling, you can stare at this figure and wonder, like, why is this… why is this Greek letter here? 636 01:22:08.610 --> 01:22:09.410 David Bau: Right? 637 01:22:09.510 --> 01:22:12.310 David Bau: And it's next to some arrow and whatever. Does that make sense? 638 01:22:12.620 --> 01:22:24.949 David Bau: And so… so that's, that's, that's sort of that thing. And then there's, like, little keys, like, this Greek letter means this, this Greek letter, those little pictures, right? So, like, all the jargon is supposed to be here. Some of the jargon is in English. 639 01:22:25.090 --> 01:22:33.369 David Bau: oh, this is… this method is called the ESD method, this method is called the whatever, you know, you know, so there's English in here, right? 640 01:22:33.650 --> 01:22:36.590 David Bau: Okay, so, so that's, that's my suggestion. 641 01:22:36.980 --> 01:22:47.469 David Bau: Is… is… and… and… and you can also think of it as… so, like, every major concept, ideally, would be in a good figure, and… and then you can also think of it as sometimes 642 01:22:47.680 --> 01:23:00.510 David Bau: a figure is, like, actually a map of the paper. If you have, like, three major things you're doing in the paper, sometimes a figure will have, like, three sections in it that correspond to the expediency data. You know, so… but think of it as, like, a map. 643 01:23:01.010 --> 01:23:03.330 David Bau: Not just as, you know, a picture. 644 01:23:03.850 --> 01:23:14.559 David Bau: And, okay, so, Bill Freeman says that when you're teaching something difficult technically, what he likes to do is he likes to start with a simplified setup. 645 01:23:14.740 --> 01:23:23.379 David Bau: And it's especially important. If you've worked on your method for a whole year, and then by the end of it, your method is, like, really good, that it's a little complicated. 646 01:23:23.650 --> 01:23:41.119 David Bau: And so, if you just start off by, like, hitting somebody with, like, a complicated form of your method, then it'd be like, oh my god, what is this? It's like, you know, you have to… you have to deal with all these rounding error problems, right? Like, this whole method is about rounding errors. No, what if it's not, right? What if the rounding errors are just to get the last 5%? 647 01:23:41.290 --> 01:23:46.440 David Bau: of performance, and actually the core of the method is something else. So Bill says, you know. 648 01:23:46.660 --> 01:24:02.770 David Bau: you can take some time in your method section to introduce the simple version of your thing. Like, maybe even a toy problem. Maybe even, like, don't solve the real problem yet, like, consider what happens in this toy thing, look at how this simple thing works. 649 01:24:02.980 --> 01:24:13.539 David Bau: If you do your writing well, you can sometimes, you know, describe the simplified toy question that you're asking in a couple paragraphs, and then you can ask the question 650 01:24:13.700 --> 01:24:26.370 David Bau: you know, what happens if you take this toy thing and then, you know, do it on a scale experiment. So that's, like, one technique. And so these are all examples of papers that have been published recently 651 01:24:26.490 --> 01:24:31.889 David Bau: That used FOI examples to start off with, which… which I liked a lot. 652 01:24:32.200 --> 01:24:45.479 David Bau: And so, okay, so, so those, so those are the couple things that I would suggest about your methods thing. And then, and then separate your methods from experiments. So, you're, you know, basically, 653 01:24:45.970 --> 01:24:51.010 David Bau: Your… your methods section is the section where you're teaching all the concepts. 654 01:24:51.230 --> 01:24:54.960 David Bau: So that if a reader is like, I don't get it. 655 01:24:55.070 --> 01:24:56.929 David Bau: They're gonna go back to that section. 656 01:24:58.490 --> 01:25:04.250 David Bau: And then your experiment section is a little bit of the section for readers who are like, I don't believe it. 657 01:25:04.640 --> 01:25:07.770 David Bau: Right? That makes sense. 658 01:25:07.940 --> 01:25:09.860 David Bau: And there's where you're defending 659 01:25:10.110 --> 01:25:29.270 David Bau: your paper, with your experimental evidence, and then there… so there, you're a little bit nerdier, you're like, okay, this is exactly how we measured it, and exactly what we did, exactly what we saw. You're very neutral, you're like, I'm not trying to hide anything from you, you know, I just tell you all the facts, right? So this is a classic science paper section of it. 660 01:25:29.270 --> 01:25:34.759 David Bau: And, you know, it's important, like, you have to… you have to have convincing evidence, and you have to be very forthright. 661 01:25:34.760 --> 01:25:43.670 David Bau: About what's going on there, and there's a little bit of teaching going on as you go through there, but, you know, most of your conceptual teaching is done by the time you're done with your meditation. 662 01:25:43.910 --> 01:25:50.850 David Bau: And so, and so, the key thing to think about when you're doing experiments is triangulation. 663 01:25:51.000 --> 01:26:00.329 David Bau: So, you're… the… What you want to avoid is, like, an experiment section That has 5 different experiments. 664 01:26:00.440 --> 01:26:02.120 David Bau: And each experiment 665 01:26:02.300 --> 01:26:09.069 David Bau: Like, wanders to, like, a different part of the continent, and, like, let's take a look at this experiment, looks over here… 666 01:26:09.410 --> 01:26:20.650 David Bau: And then, oh, there's this other interesting thing, this experiment looks over here, and there's this other interesting thing, and this experiment looks over here. So, what you would rather have as a field for your experiments is 667 01:26:21.350 --> 01:26:27.509 David Bau: Here's the thing that we're interested in. Right here, this is the spot in the introduction, it says we're interested in this question. 668 01:26:27.830 --> 01:26:29.519 David Bau: And we have 5 experiments. 669 01:26:29.710 --> 01:26:32.030 David Bau: One experiment looks at it like this way. 670 01:26:33.200 --> 01:26:53.180 David Bau: And then the other experiment's, like, over here, and we look at the same spot, like, over here this way. You can see a different thing. And then here's our third experiment. Our third experiment, like, is over here. And then once you get, like, to the third experiment, and you look at the same phenomenon, like, the third way, people will start to get convinced. They're like, well, I don't know, I didn't know if I believed the first experiment, but the second one's kind of… 671 01:26:53.310 --> 01:27:10.820 David Bau: surprising, because you looked at the same thing. The third one, I can't, like, it's hard to deny, there's something going on here, and then… and then you hit them, and you're like, no, no, there's a fourth way. You could look at the fourth thing, there's a fifth thing, and then… and then if you do that, and then you take… then you step in there, and you do that, and you throw it into the air, and it does something else. 672 01:27:10.820 --> 01:27:14.449 David Bau: Then it also happened, and people were like, wow, that's crazy. 673 01:27:14.700 --> 01:27:20.349 David Bau: Right? And then they're convinced. That's, like, what a good experiment section feels like. Does that make sense? 674 01:27:20.490 --> 01:27:23.809 David Bau: So it's not just, like, a tour of the British Museum. 675 01:27:24.200 --> 01:27:25.810 David Bau: Right? Does that make sense? 676 01:27:26.180 --> 01:27:28.030 David Bau: And so, 677 01:27:28.220 --> 01:27:38.239 David Bau: And then after you've done that, hopefully you've wowed your reader, they're like, oh, there's really something here, and you just write a very brief, humble conclusion section. 678 01:27:38.360 --> 01:27:43.000 David Bau: That says, well, this paper proposed that you could figure this thing out. 679 01:27:43.160 --> 01:27:49.679 David Bau: We introduced this method that you use to figure this out. We did some experiments, it turns out that you can figure it out, it works. 680 01:27:49.870 --> 01:27:52.819 David Bau: You know, this, this, this probably has some implications. 681 01:27:52.950 --> 01:27:54.550 David Bau: for future sciences. 682 01:27:54.960 --> 01:27:55.839 David Bau: And that's it. 683 01:27:56.100 --> 01:27:57.950 David Bau: That's your introduction, that's your conclusion. 684 01:27:58.310 --> 01:27:59.260 David Bau: Make sense? 685 01:27:59.890 --> 01:28:05.449 David Bau: And that's sort of the social form here. So, you know, you might point out, 686 01:28:05.580 --> 01:28:08.499 David Bau: What the implications are of your work. 687 01:28:08.670 --> 01:28:10.220 David Bau: But just very briefly. 688 01:28:10.920 --> 01:28:12.730 David Bau: Don't make a big deal at 5. 689 01:28:13.060 --> 01:28:19.760 David Bau: So Bill says that… This is very common for, like, first-year students. 690 01:28:20.200 --> 01:28:21.439 David Bau: Don't do it. 691 01:28:22.690 --> 01:28:24.040 David Bau: Future work. 692 01:28:24.510 --> 01:28:28.629 David Bau: So, you know, in your conclusions, you're like, okay, here, here's, like, what we did. 693 01:28:29.510 --> 01:28:34.080 David Bau: Future work. Here is, like, future stuff that we're also going to do. 694 01:28:34.570 --> 01:28:35.570 David Bau: Right? 695 01:28:36.660 --> 01:28:40.569 David Bau: Never put future work in your paper. Never write it down. 696 01:28:40.730 --> 01:28:46.339 David Bau: There's, like, so many problems with future work. So, one, like, is to, like, if you're in a competitive area. 697 01:28:46.540 --> 01:28:51.269 David Bau: Like, everybody's trying to, like, make a cool new method. Then your future work reads like this. 698 01:28:51.510 --> 01:28:58.800 David Bau: It reads, oh, so it reads, okay, here is a list of all the things that we wish we could do, but we didn't have time, we, like, ran out of time. 699 01:28:59.140 --> 01:28:59.940 David Bau: break? 700 01:29:00.280 --> 01:29:05.069 David Bau: And so the reviewers were like, huh, Why didn't you do that? 701 01:29:05.550 --> 01:29:12.599 David Bau: kind of reject it, because you didn't do that thing. It's just like a list for all the reasons a reviewer should reject you. It's like, why are you listing this out to the reviewer? Right? 702 01:29:12.710 --> 01:29:31.109 David Bau: Or, it's like, oh, here's a… like, we did all the good stuff, but there's, like, there's even more ambitious stuff you should do if you're reading it. Here's a list of good ideas that you should go and steal from us now, and you do it before we get a chance to do it. It's like, why… why are you doing this to your research competitors? It's like, please do this for us, right? 703 01:29:31.250 --> 01:29:37.059 David Bau: You know, it's like… like, there's a couple, you know, functions that, you know, future work 704 01:29:37.550 --> 01:29:43.719 David Bau: You know, serves. And neither one of them is really in your interest, and neither one of them is really of interest to most readers. 705 01:29:43.980 --> 01:29:54.899 David Bau: Right? Like, the readers don't want to be told what they should be doing in the future. They're going to figure that out for themselves. Unless you have a really great idea, and then you can steal it from them. Yeah, I don't know. Don't write future work. 706 01:29:55.340 --> 01:29:56.250 David Bau: Right? 707 01:29:56.580 --> 01:30:03.199 David Bau: If you, if you read the paper, so write a paper to be read, Like, 10 years from now. 708 01:30:03.460 --> 01:30:07.370 David Bau: It's really weird to go… like, if you wrote an important paper. 709 01:30:07.690 --> 01:30:23.699 David Bau: And then somebody went to read it 10 years later, and there's, like, a future work section. It's no longer future work, right? Like, maybe that happened, or maybe it didn't. I don't know, it's weird, right? So no matter what happened, it's like, it's weird, like, you wouldn't read the paper to read the thing that happened in the future, you'd read the future paper, right? And you also… 710 01:30:23.800 --> 01:30:27.919 David Bau: But then also, you know, maybe the future didn't happen the way that you imagined. 711 01:30:28.160 --> 01:30:36.380 David Bau: And then it's this weird anachronism. Future work? This is, like, what they thought was gonna happen in the future didn't work out that way at all. It's, like, very strange. Like, don't write this. 712 01:30:36.660 --> 01:30:37.580 David Bau: Make sense? 713 01:30:38.120 --> 01:30:39.020 David Bau: Okay. 714 01:30:39.380 --> 01:30:44.840 David Bau: So, just to end with the conclusion or summary, just say a few words, don't write a lot in your 715 01:30:45.810 --> 01:30:47.690 David Bau: That's, that's my rule of thumb. 716 01:30:48.200 --> 01:30:49.799 David Bau: Don't worry too much about it. 717 01:30:50.480 --> 01:30:54.330 David Bau: Okay, so, okay, so as you write your things for Thursday, a couple general tips. 718 01:30:54.560 --> 01:30:56.780 David Bau: Couldn't hear a response again. 719 01:30:58.040 --> 01:31:02.760 David Bau: Perhaps the most important principle of good writing is to keep the reader uppermost in mind. 720 01:31:04.220 --> 01:31:11.150 David Bau: Yeah, what does the reader know so far? Okay, so actually, you know, I… I… so… so there's… there's a couple things about this introduction I want to… 721 01:31:11.250 --> 01:31:18.469 David Bau: to cover. Because there's a thing that's my own personal beef that's not really here on one of these slides. 722 01:31:18.650 --> 01:31:21.979 David Bau: So, what people are saying here is they're saying. 723 01:31:22.450 --> 01:31:28.560 David Bau: you know, be kind to your reader. So the hardest thing about writing, I can work on a technical project. 724 01:31:28.700 --> 01:31:29.929 David Bau: For a few months. 725 01:31:30.170 --> 01:31:36.970 David Bau: Is… is to have empathy, To the reader who hasn't thought about it as long as you have. 726 01:31:37.470 --> 01:31:41.240 David Bau: Right? And so, it's like a guest coming into your house. 727 01:31:42.930 --> 01:31:48.230 David Bau: You know, like, for example, in my house, I've got this toilet that doesn't quite work right. 728 01:31:48.630 --> 01:31:51.060 David Bau: And I know that it doesn't work right. 729 01:31:51.220 --> 01:31:54.070 David Bau: But if I don't have empathy for my guests. 730 01:31:54.280 --> 01:32:06.390 David Bau: then when they come and use that toilet, then they'll be like, oh no, I'm here in the bathroom, I can't figure out how to use the toilet, whatever, right? So I realized, oh, this is one of the things that I have to explain. And your paper's kind of the same way. 731 01:32:06.640 --> 01:32:11.310 David Bau: Right? You know, you know there's all these things about your technology. 732 01:32:11.770 --> 01:32:14.440 David Bau: And if a reader needs to understand it, then he should 733 01:32:14.680 --> 01:32:16.560 David Bau: Try to have empathy and explain it. 734 01:32:16.750 --> 01:32:23.469 David Bau: Up front. So it's like, oh, would you like the panels? And, oh, by the way, let me show you how to get this toilet to work. 735 01:32:23.960 --> 01:32:27.079 David Bau: And, and then here's the sheets for your bed. 736 01:32:27.220 --> 01:32:29.459 David Bau: And breakfast is served at 6. 737 01:32:29.840 --> 01:32:31.460 David Bau: You know, whatever. 738 01:32:32.130 --> 01:32:33.140 David Bau: That make sense? 739 01:32:33.630 --> 01:32:38.860 David Bau: And, Right? Fewer words. 740 01:32:39.780 --> 01:32:54.619 David Bau: Oh, no, what does drunken White saying? They said, omit needless words. Oh, they're so much more elegant. But I say just write fewer words. Yes. Okay, yes, we're almost done. So, so we already did the writing exercise, but here's the suggestion. 741 01:32:54.930 --> 01:33:04.660 David Bau: as I was thinking about all the introductions you guys wrote, there's a really common way of writing the introductions, and like, I would recommend trying to be bold and not writing this way. 742 01:33:04.810 --> 01:33:10.730 David Bau: So, in our field, we write these funnel introductions that start like warranties. 743 01:33:11.340 --> 01:33:14.770 David Bau: But it was the best of times. It was the worst of times. 744 01:33:15.040 --> 01:33:19.720 David Bau: Right? It's like… you know, In recent decades. 745 01:33:19.720 --> 01:33:43.639 David Bau: artificial intelligence has grown in importance until it's, like, taken over the entire world, like, it's such an important thing. You know, AI is, like, this big, big thing. It's, like, like, important for this, it's important for that. AI is everywhere, there's no avoiding AI, and then you take this funnel, and it's like… and then in our town, AI is used everywhere, and in my neighborhood, AI is used everywhere, and then in my… on my street. 746 01:33:43.640 --> 01:33:44.689 David Bau: at my house. 747 01:33:44.690 --> 01:33:53.989 David Bau: Right? On my countertop, right here on this table, right here is AI! Right, and then you're, like, a page into your introduction, and then you say, so, since AI is right here, we're gonna study this. 748 01:33:54.260 --> 01:34:03.760 David Bau: Right? And it's like, you don't need that section. You can just say, what, the reader? Look at this whole thing here, we're gonna do this, right there on this table. 749 01:34:03.940 --> 01:34:07.700 David Bau: And you can do that in, like, the first sentence, but it feels very bold. 750 01:34:07.980 --> 01:34:12.509 David Bau: It's like, why would anybody care about my funny little thing on my table? Right? 751 01:34:12.620 --> 01:34:17.040 David Bau: That make sense? But, like, but actually, it's refreshing. 752 01:34:17.230 --> 01:34:23.279 David Bau: to read it. It's easier to read. It's actually… if you write it well, it's kind of cool. 753 01:34:23.400 --> 01:34:35.179 David Bau: So that's what I would suggest, like, write your words, get straight to the point, don't write these big funnel things, that make sense? And have empathy for your reader. Okay, so I'll have more to say later, but… 754 01:34:35.380 --> 01:34:40.930 David Bau: Write a couple sections of your paper with your team, bring 5 copies of it on paper on Thursday. 755 01:34:41.150 --> 01:34:41.969 David Bau: more this. 756 01:34:42.210 --> 01:34:44.429 David Bau: Thanks, everybody. See you later today. 757 01:34:51.280 --> 01:35:00.080 David Bau: You've been pretty special, but…