[00:01.520 --> 00:14.320] Now we have some audio. Okay, so everyone somewhat speak? [00:22.950 --> 00:25.910] Okay, now try again. Yeah, now we have audio. [00:26.150 --> 00:28.810] Hello, good. Can you hear me, Sven? [00:29.310 --> 00:30.990] I can hear you. Hey, Irem. [00:31.150 --> 00:35.530] Oh my gosh, it was just a simple Windows reboot fixes everything. [00:36.910 --> 00:41.130] Hey, I unfortunately, I actually have a meeting right now. So I'm going to drop and join again. [00:41.130 --> 00:44.510] What time do I need to be on for speaker time? [00:45.010 --> 00:46.510] In 30 minutes. [00:47.510 --> 00:50.630] But only in Discord or also in Zoom? [00:51.110 --> 00:52.330] Only in Discord. [00:53.310 --> 00:54.190] Okay. [00:56.070 --> 01:00.810] Yep, so we've got people in the Slack chat or in the Twitch chat saying they can hear us. [01:00.810 --> 01:01.950] So I think we're good to go. [01:02.330 --> 01:03.110] Sweet. [01:04.070 --> 01:07.310] Well, hi, everyone. Sorry about the audio issues. [01:07.310 --> 01:10.430] That seems to be the start of every stream is figuring out the audio. [01:11.790 --> 01:14.910] The only people I've seen get it right are like Critical Role. [01:17.730 --> 01:20.290] They pay a lot of people to get it right. [01:20.490 --> 01:26.910] Yeah. Welcome to AI Village at DEF CON 28 safe mode. [01:28.010 --> 01:29.990] We're all doing this from home. [01:32.750 --> 01:36.930] I hope that this can be as fulfilling to you guys as it is to us. [01:38.690 --> 01:46.430] And we just want to start off the village with some like general questions about the [01:46.430 --> 01:53.730] state of the field as it is right now at the end, halfway through 2020. [01:54.430 --> 01:59.710] So Jag has got a bunch of questions prepared and the rest of us are going to try to answer them. [02:01.090 --> 02:02.010] Okay. [02:02.890 --> 02:04.370] Welcome, everyone. Thanks. [02:04.370 --> 02:05.890] Thank you for being here. [02:05.890 --> 02:09.510] The first question that I have listed is... [02:09.510 --> 02:13.430] Let's just knock this one out of the way real first. [02:13.670 --> 02:17.610] Facial recognition is a security concern for us. [02:18.110 --> 02:20.490] In the ML field, of course. [02:21.090 --> 02:24.070] And before you answer, please introduce yourself. [02:24.750 --> 02:27.950] Yeah. We should probably do a who is who. [02:29.530 --> 02:31.990] Sven, do you want to kick us off? [02:33.050 --> 02:33.990] Sure. [02:33.990 --> 02:35.210] So I'm Sven. [02:36.110 --> 02:39.850] I am a senior data scientist at Elastic. [02:39.850 --> 02:46.870] Before that, I was a topologist who accidentally got into machine learning and security. [02:47.850 --> 02:52.370] And I kind of helped run the AI village. [02:55.010 --> 02:57.030] John, do you want to go next? [02:57.230 --> 02:58.070] Yeah, sure. [02:58.070 --> 02:58.790] Hey, everyone. [02:58.790 --> 02:59.770] DeltaZero here. [02:59.770 --> 03:02.770] I do data science for intrusion detection. [03:02.770 --> 03:07.770] Yeah, I've been in the field for maybe, I don't know, four years now. [03:07.770 --> 03:09.890] But pretty awesome. [03:09.890 --> 03:11.110] Glad to be here. [03:14.110 --> 03:15.270] Rich? [03:16.170 --> 03:17.910] Hi, I'm Rich Huerang. [03:17.970 --> 03:21.690] I'm a director of data science research at Sophos. [03:21.690 --> 03:28.310] I've been working at the intersection of machine learning, security, and privacy for probably [03:28.310 --> 03:30.250] about 10 years now. [03:30.250 --> 03:32.250] Got my PhD from UCSB. [03:32.250 --> 03:37.750] Spent some time with the US Army Research Laboratory, and I've been at Invencion Sophos [03:37.750 --> 03:39.050] ever since. [03:44.460 --> 03:45.600] Yaga? [03:46.040 --> 03:47.740] Hi, I'm Yaga. [03:47.740 --> 03:52.280] I am a professor at a small university in Virginia. [03:52.520 --> 04:02.080] I've got about 25 years of experience in security, but relatively new to machine learning. [04:02.080 --> 04:10.180] My research is focused on securing machine learning, and I like playing video games too. [04:10.180 --> 04:13.940] So there's something interesting about it. [04:14.200 --> 04:14.930] Anita? [04:15.640 --> 04:17.080] Hi, I'm Anita. [04:17.080 --> 04:19.360] I'm a research scientist at University of Illinois. [04:19.360 --> 04:21.500] Started life as a cryptographer. [04:21.500 --> 04:24.440] Been in security for almost 30 years. [04:28.200 --> 04:31.120] Okay, and Will's the last guy. [04:32.360 --> 04:34.400] Certainly not least. [04:34.400 --> 04:35.720] Yeah. [04:36.560 --> 04:39.280] Sorry, my morning's just starting. [04:39.380 --> 04:50.800] Will, I am a network operator, and I guess attack machine learning models and use AI [04:50.800 --> 04:53.260] for offensive purposes. [04:53.260 --> 04:56.600] Phishing, pathfinding, and stuff. [04:58.360 --> 04:59.760] Yeah, Will's being modest. [04:59.760 --> 05:05.860] He actually has the first CVE ever assigned against a machine learning system. [05:05.860 --> 05:11.240] If you attend my lab later, there'll be another one. [05:13.900 --> 05:16.960] I think that's a challenge for everyone. [05:18.540 --> 05:22.000] Can you get one file before Will drops number two? [05:25.560 --> 05:28.460] Okay, thank you everyone for the introductions. [05:28.460 --> 05:30.820] Sorry, I should have thought about that. [05:31.120 --> 05:35.740] Please be sure to ask any questions over in the Discord server. [05:35.740 --> 05:38.320] There is a talks channel. [05:40.800 --> 05:43.420] Yes, there's a channel for talks. [05:43.420 --> 05:48.000] You can ask in there or in the general, and we'll try to catch it. [05:48.100 --> 05:50.420] So back to the first question. [05:50.620 --> 05:56.100] This was one that was both easy and hard to throw out there. [05:56.100 --> 06:06.100] But is facial recognition a security concern when we're talking about privacy? [06:06.800 --> 06:11.000] For sure we know about, but in the general machine learning terms, [06:11.000 --> 06:13.500] is facial recognition a security issue? [06:14.880 --> 06:18.960] So I think just to make the point in our village, one of the things we're trying to do [06:19.540 --> 06:24.320] is expand from just thinking of this as a technical kind of algorithm problem. [06:24.320 --> 06:28.860] If you look at the broader scope of facial recognition, there are issues not just with [06:28.860 --> 06:33.040] security, with data sets, with data integrity, with these kind of choices. [06:33.440 --> 06:36.480] So I think my answer would be it is a security issue. [06:36.480 --> 06:37.540] It's a machine learning issue. [06:37.540 --> 06:38.200] It's a data issue. [06:38.200 --> 06:39.480] It's a human issue. [06:40.280 --> 06:45.000] And for us, we need to think about all those things as technology people. [06:47.180 --> 06:51.980] Yeah, and just to add to that, a lot of the issues around facial recognition [06:52.980 --> 06:58.520] are less to do with the technology and more to do with the processes that that technology is [06:58.520 --> 07:00.480] embedded in. [07:00.480 --> 07:06.220] So if you have a facial recognition system that works, you have issues around privacy, right? [07:06.220 --> 07:07.840] Who gets access to that system? [07:07.840 --> 07:10.560] Who gets to know who was seen where on which camera? [07:10.700 --> 07:13.740] And then when it doesn't work, you have bias issues that sneak in. [07:13.740 --> 07:16.360] So how do you control for false positives, right? [07:16.360 --> 07:22.140] Someone gets reported as being, you know, facially recognized as a bank robber or something [07:22.940 --> 07:24.920] when it's just like an average person. [07:24.920 --> 07:28.000] And this actually extends beyond issues of just facial recognition. [07:28.000 --> 07:30.880] Just computer vision in general has problems like this. [07:30.880 --> 07:36.020] There was a story recently about a license plate recognition system, which led to someone [07:36.020 --> 07:40.460] being pulled over and detained, you know, face down on the side of the road for hours [07:40.460 --> 07:44.180] before they figured out that it was the wrong license plate number, the wrong make of car, [07:44.180 --> 07:47.000] the wrong model year, et cetera, et cetera, et cetera. [07:47.000 --> 07:53.360] So yeah, it's definitely a case of not only does the ML work, but do you have the right [07:53.360 --> 07:57.940] controls around the ML to make it actually usable? [07:57.940 --> 08:02.180] And have you, you know, assessed what happens when it actually fails so that it can fail [08:02.180 --> 08:04.020] gracefully without hurting people? [08:05.560 --> 08:06.840] Anyone else? [08:07.320 --> 08:11.540] I think the last thing, not to dominate this, but the last thing to throw in is I feel like [08:11.540 --> 08:17.560] in our village, part of our job is to also translate all this technobabble sometimes [08:17.560 --> 08:20.800] to policy speak and to real world speak. [08:20.800 --> 08:22.800] And that could be very challenging. [08:22.800 --> 08:24.820] We know the technical parts. [08:24.820 --> 08:29.000] What ends up on the other end, we need to bridge that gap. [08:29.680 --> 08:35.920] I think that's important to understand that security isn't just the technical side. [08:35.920 --> 08:39.220] There's a human side to it too, and that extends to policy. [08:39.220 --> 08:41.140] So definitely something to think about. [08:41.140 --> 08:41.520] Thank you. [08:42.200 --> 08:48.740] Staying along the same lines, one of the things that I wanted to ask about is the FOX program. [08:48.740 --> 08:50.620] Have you heard about it? [08:50.620 --> 09:04.180] And how does it affect our ability for the end user to take back some of that control [09:04.940 --> 09:06.440] over their data? [09:08.560 --> 09:14.140] It's technically, again, this is, it's the gap between technical aspects of it and [09:14.140 --> 09:16.380] implementation and social aspects of it. [09:16.380 --> 09:18.700] Technically, it's a brilliant idea. [09:18.700 --> 09:21.760] And, you know, it does what it says on the tin. [09:21.760 --> 09:27.780] In practice, there are so many ways that people get registered into facial recognition systems [09:27.780 --> 09:34.320] from, you know, collecting ID cards to photos that they uploaded before FOX was a big thing [09:34.320 --> 09:43.520] to photos that other people take of them and maybe tag them in later, that you really can't [09:43.520 --> 09:49.480] just find a quick technical band-aid for security problems that are really a combination of [09:49.480 --> 09:51.260] social and technical problems. [09:51.260 --> 09:55.240] The root cause here isn't like, oh, we just need better technology to fix this. [09:55.240 --> 10:01.880] The root cause is society and sort of laws and norms haven't kept up with the technology. [10:01.880 --> 10:05.200] And that's the angle that we need to address it from. [10:05.200 --> 10:09.060] Technical band-aids are only ever going to be available to people that know about them, [10:09.060 --> 10:13.820] know how to apply them, and can get over that barrier to entry. [10:13.920 --> 10:17.360] Really, we need to be thinking broader about impact to society as a whole. [10:18.160 --> 10:23.360] It makes me think a little bit of this field of, you know, usable privacy and, you know, [10:23.360 --> 10:24.980] privacy by design. [10:24.980 --> 10:32.700] See, that's analogous here because people don't understand that long-term impact of [10:33.300 --> 10:34.540] having a photo op. [10:34.540 --> 10:36.880] So, enabling people to use this tool is great. [10:36.880 --> 10:41.100] I think the average person isn't going to even think about it because it doesn't affect [10:41.100 --> 10:42.460] their day-to-day life. [10:42.460 --> 10:48.980] So, again, I think the issue is, how do you express to them the long-term potential damage [10:48.980 --> 10:51.280] of having their pictures out there? [10:52.140 --> 10:57.460] Okay, we're going to move into a different arena. [10:58.320 --> 11:06.680] What do you say the latest advances that your typical security team could use [11:08.280 --> 11:10.360] in terms of machine learning? [11:11.380 --> 11:16.040] You know, a lot of vendors are out there saying that this is going to be the panacea for them [11:16.040 --> 11:17.880] to be able to fix things. [11:17.880 --> 11:24.600] Some of the actual products that will help them or techniques that will help them. [11:26.440 --> 11:31.940] I just want to take this opportunity to plug the panel that we have coming up on Saturday [11:31.940 --> 11:35.540] on whether or not ML and InfoSec lives up to the hype. [11:35.540 --> 11:39.640] So, we're going to be diving, hopefully, pretty deep into those questions. [11:39.640 --> 11:42.680] But I'll kick it over to someone else since I've been talking a lot. [11:47.220 --> 11:52.500] I really don't know what the boots on the ground could be using. [11:53.320 --> 11:57.860] But there are some promising things coming down the pipe from researchers. [11:59.560 --> 12:00.680] But I... [12:02.540 --> 12:05.960] So, I'll jump in real quick. Sorry to dominate again. [12:05.960 --> 12:08.600] And I'll let Will speak because he's more an operator. [12:08.720 --> 12:13.600] For me, federated learning, federated and privacy-preserving machine learning is very [12:13.600 --> 12:15.520] important for a bunch of reasons. [12:16.020 --> 12:20.440] Not the least of which is being able to learn across edge devices that are in multiple disparate [12:20.440 --> 12:22.740] places and being able to do it more efficiently. [12:22.760 --> 12:27.840] And being able to use new types of data because there's privacy-preserving ways to do it. [12:27.840 --> 12:33.320] So, I think I see federated learning as something that's promising. [12:34.780 --> 12:40.040] Yeah, I'd say from a product standpoint, it's just like the simple stuff. [12:40.040 --> 12:49.780] So, you know, on a network, clients will still miss password spraying internally. [12:50.020 --> 12:54.520] And that's just something that shouldn't be happening with the logs that are available. [12:54.520 --> 13:02.420] So, I think in terms of just instant response and stuff on networks, it's the simple stuff [13:02.420 --> 13:06.540] that is going to give you the most bang for the buck. [13:09.120 --> 13:14.740] Obviously, there's a lot of competition in the machine learning space, or it would seem. [13:17.440 --> 13:24.140] And half of me, at least 40% of me, thinks in two years, we're going to be moving on [13:24.140 --> 13:28.160] from machine learning as a solution in networks. [13:28.160 --> 13:32.700] And we'll probably replace it with like zero trust or something like that. [13:34.080 --> 13:39.120] But there's also the other 10% that thinks like it will always have a place. [13:40.020 --> 13:47.980] But just like with the facial recognition stuff and the bias, if we don't put it out [13:50.680 --> 13:55.480] in a way that's useful, and in a way that people can understand, [13:55.480 --> 14:01.100] the minute it breaks, they're likely to just throw it out the window and say it doesn't work. [14:02.700 --> 14:09.820] Even though, you know, it is a perfectly viable solution. But yeah, I'd say simple stuff. [14:10.320 --> 14:11.580] Yeah. Okay. [14:11.820 --> 14:15.600] To me, that's the more important side of the question is like, [14:15.600 --> 14:20.180] practitioners on the ground, we don't need any new math or anything new interesting. [14:20.180 --> 14:24.660] But there's some basics that they can get their hands on that will probably help them out when [14:24.660 --> 14:30.100] they have to deal with a deluge of alerts from some sim that they have to deal with. [14:30.100 --> 14:32.680] A little bit of clustering could go a long way in that situation. [14:34.020 --> 14:36.440] Yeah, I think that's perfect. [14:36.660 --> 14:41.220] That's perfect for what I was going to say is, I didn't mean to say products, [14:41.220 --> 14:47.800] I probably wrote that down wrong. But like, what can we tell our audience here that maybe [14:47.800 --> 14:53.860] practitioners that they can do on their own? So I think something like our workshops, [14:53.860 --> 14:56.420] if you take a look at them, they'll be able to help you [14:56.420 --> 15:00.840] see that you may not need to go get these expensive products. There's some simple things [15:00.840 --> 15:04.520] you can do. Yeah. [15:04.620 --> 15:09.100] Yeah, and I just add to that, but from my own experience as someone who's [15:09.700 --> 15:15.180] used ML to tackle a lot of security problems, very often you don't even need like the big [15:15.180 --> 15:20.920] complicated deep neural networks or gradient boosted decision trees or pick your flavor of [15:20.920 --> 15:27.200] the day. Very often, it's enough to have a good idea of the problem you're trying to solve and [15:27.280 --> 15:31.760] a good set of representative data and then simple methods like, you know, linear logistic [15:31.760 --> 15:36.740] regression will often be enough to tackle it for you. So, you know, you don't need to think that [15:36.740 --> 15:43.020] you've got to have tons of like Amazon EC2 credits to run a big beefy GPU machine. [15:43.020 --> 15:48.020] A lot of times you can get really like 80% solutions, really good solutions just on your [15:48.020 --> 15:51.820] laptop. Okay. [15:54.940 --> 16:00.380] For most security to be able to defend yourself, you have to be able to identify [16:01.000 --> 16:08.920] the issue or identify the attack. We discuss, you know, adversarial attacks quite a bit [16:09.520 --> 16:17.320] in the academic realm. How could someone tell that they're seeing an adversarial attack in practice [16:18.320 --> 16:26.870] today? You mean on the AI? I'm throwing out some hard ones. You mean on the AI portion? [16:29.390 --> 16:36.430] Well, how could you just notice an attack? Because to defend against something, [16:36.430 --> 16:40.330] you typically have to be able to define what that attack looks like, right? [16:40.930 --> 16:44.970] At least to some degree. Can we do that today? [16:47.350 --> 16:52.570] It's kind of too bad Hiram's not here because I feel like he'd have a lot to say on this topic. [16:55.410 --> 16:56.770] Let's hear it. [16:57.510 --> 17:04.690] So for the traditional adversarial attacks, we basically are trying to modify an image [17:04.690 --> 17:09.330] or something slightly to bypass the machine learning system. That doesn't really work [17:09.330 --> 17:14.150] with security because our data is much more structured. You can't like do a gradient descent [17:14.150 --> 17:20.570] on a binary and then hope it still compiles or runs because you're going to get byte codes that [17:20.570 --> 17:28.570] are just wrong. So I don't think like adversarial attacks as academics that you read up in a big [17:28.570 --> 17:33.230] year paper are going to really apply. There are some modifications to that that might apply that [17:33.230 --> 17:40.370] might be very helpful. But what we do see already and have been seeing for like a decade is people [17:40.370 --> 17:46.090] just trying a bunch of stuff and then they get past it. And the more they get informed they are [17:46.090 --> 17:51.710] about how the machine learning model works, the more effective and more like gradient descent it [17:51.710 --> 17:58.610] is. You know, like the more effective they are. Like for like the spam filters from back in 2001 [17:58.610 --> 18:06.290] when people were packing Wikipedia articles at the bottom of their spam messages to bypass the [18:06.290 --> 18:10.050] naive Bayesian classifiers, they understood that it wasn't a naive Bayesian classifier. They [18:10.050 --> 18:16.270] understood that they needed to pack the thing. And then they did that. And that's going to happen [18:16.270 --> 18:27.590] with next-gen AV products too. It may already be happening. I mean, that was what we saw in [18:27.590 --> 18:35.430] the silence bypass was effectively a stuffing attack. Yeah. That was researchers being nice to [18:35.430 --> 18:46.730] us and telling us about the problem that's happening. So this one is more directed at Will, [18:46.730 --> 18:51.650] but definitely want everyone's opinion. And it gets back to something we talked about in a journal [18:51.650 --> 18:57.710] club, which for those of you that might not have seen it, we may be replaying it during the weekend. [18:59.230 --> 19:07.470] Will received a CVE for the vulnerability he discovered. [19:07.470 --> 19:17.110] But do we need actual CVEs specifically for machine learning or should we keep them more general? [19:19.090 --> 19:25.990] So I guess first I'd like to say it was our, so Nick Landers and I did it. [19:26.550 --> 19:34.550] I just talk about it more so. So the CVE system is designed to communicate risk. [19:35.310 --> 19:43.330] And I think anywhere that risk is present, then a CVE is appropriate. [19:46.150 --> 19:51.510] You know, I just, I submitted it. So I'm not the one that gets to say whether or not it's a CVE. [19:52.270 --> 19:59.170] So I guess that the question would be for the NIST. And I know they've come out with a paper [19:59.170 --> 20:05.910] that kind of highlights machine learning attacks or there's a taxonomy. It's like 15 taxonomies [20:05.910 --> 20:15.470] right now. But yeah, I think in any way there's risk that is at least measurable, a CVE would [20:15.470 --> 20:21.310] be appropriate or any standardized score. I know there's operators like... [20:22.130 --> 20:29.630] And I'm sorry, I don't mean should it be a CVE, but more should there be a class just for machine [20:29.630 --> 20:35.810] learning, you know, because they break it down. Yeah, yeah, I think so. It's fundamentally [20:35.810 --> 20:43.390] different. You know, so for example, like with a traditional exploit, you know, there's like [20:44.170 --> 20:51.750] a patch and then you can push it out. But with, for example, like with Proofpoint's model, [20:51.750 --> 20:58.270] there's some residual, there's some like lingering theft going on. So [20:58.990 --> 21:05.230] once you have that copycat model, you're never going to get 100% of their model. But [21:06.590 --> 21:13.250] for some, there's some decay into the future, but you'll have some portion of that model. [21:13.250 --> 21:20.910] Your copycat model will be right, you know, up to some future time until they've retrained enough. [21:20.910 --> 21:28.930] But in their response, you know, unless they do things to fix it or prevent, you know, [21:28.930 --> 21:36.610] people from stealing outputs, then you can always create a new model. So I think that [21:37.210 --> 21:45.050] piece is important. But yeah, it's a little, it's fundamentally different beast than [21:46.610 --> 21:50.990] than like an exploit. Yeah, I don't know, you guys probably have a better idea of like, [21:50.990 --> 21:56.550] how you're supposed to patch a machine learning model or you take it out. I don't know. [21:57.510 --> 22:00.250] Long and expensive retraining process. [22:01.130 --> 22:07.210] Do you guys agree, though, that it's good to have that separate machine learning [22:07.210 --> 22:13.550] kind of category to highlight the fact that this is a vulnerability, a specific type of vulnerability? [22:15.270 --> 22:22.450] Yeah, I think most, so there are some exceptions. And, you know, again, I'm going to sort of like [22:22.450 --> 22:28.890] promote the Journal Club on the Summoning Demons paper later this weekend. But most machine learning [22:28.890 --> 22:35.250] exploits don't translate into exploits as we usually think of them in InfoSec. They are much [22:35.250 --> 22:42.450] more either model stealing or model bypass, or maybe like a denial of service type thing. So [22:43.570 --> 22:48.170] in my mind, keep in mind, I'm coming at this from the ML side of things and much less from the [22:48.170 --> 22:53.130] InfoSec side, I do think it makes a lot of sense to have separate categories that reflect the [22:53.130 --> 22:59.430] different kinds of risks that we currently think of ML models as being exposed to. [22:59.870 --> 23:07.110] Okay. So that's kind of all the questions I had. The next was, I was going to turn it back over to [23:07.110 --> 23:12.510] Sven to kind of give you an overview of the weekend. And thanks for being here, everyone. [23:12.510 --> 23:18.510] And thank you for your questions. Or answers. Yeah, cool. I'm really looking forward to the [23:18.510 --> 23:27.310] talks. This is going to be a great DEF CON. Cool. I'm out. Later, folks. [23:27.790 --> 23:35.010] Have fun. So about this weekend, the way we're structuring it is everyone, all the talks are [23:35.690 --> 23:40.630] recorded already and on my desktop. So the speakers won't be talking live. The reason why we're doing [23:40.630 --> 23:47.090] that is so that the speakers can focus on answering your questions in Discord. And so during the talk, [23:47.090 --> 23:52.110] speakers will be in the Discord and hopefully afterwards. So any questions you have about the [23:52.110 --> 23:57.850] thing, you can just ask it in the Discord channel, and we'll answer it. The speaker will answer it, [23:57.850 --> 24:02.710] or maybe someone else who's knowledgeable on the subject will help try to answer your question. [24:05.090 --> 24:12.250] So we're going to, in a few minutes, launch into the first one, which is we've got Zoltan here. [24:16.010 --> 24:24.330] Zahak, who's in Hyrum. They'll be going over to the Discord to answer any questions. [24:25.830 --> 24:31.230] And then, yeah, then we've got Eriks and a few others. Oh, [24:31.810 --> 24:34.670] Sarah, do you want to introduce yourself, too? [24:36.710 --> 24:44.770] Oh, god. SJTurk, ethics chair, so I guess the buck stops here. [24:46.530 --> 24:50.050] I guess data nerd who's working on disinformation. [24:57.600 --> 25:02.100] I think that's it for the intro stuff. [25:02.100 --> 25:07.120] Yeah, that's it. We were trying to leave some time to make sure the transition to Hyrum. [25:07.650 --> 25:12.080] Yeah, I've got that all ready to hit play. [25:12.620 --> 25:19.360] Okay, yeah. Oh, workshops. Yeah, so there's a few workshops. [25:20.630 --> 25:27.550] They're mostly focused on sort of bypasses and interesting techniques on... [25:28.270 --> 25:39.110] The hard part for us is text data. So, GtKondike, Gavin Kondike, well, has a lovely workshop on [25:42.990 --> 25:54.750] spam builders. And we have... Will here is talking about a new, basically, CVE2. [25:57.590 --> 25:59.650] Model stealing stuff. [26:02.930 --> 26:12.030] And the way those work are they'll be streamed on Twitch live. They're not recorded. And then [26:12.030 --> 26:18.790] there'll be a few people who are TAing who will help the workshop presenter help relay questions [26:18.790 --> 26:25.170] from Discord over to the workshop presenter and help resolve any minor issues you guys might have. [26:26.330 --> 26:31.270] There are Jupyter notebooks and stuff posted on the website. [26:33.070 --> 26:41.250] So, hopefully this all works out. It's a pretty normal machine learning workshop structure. [26:41.570 --> 26:47.570] Just online versus in person so that you can't have someone spy over your shoulder to tell you [26:47.570 --> 26:58.050] your typo is. Well, with that, let's start Hyrum's... [27:02.410 --> 27:08.170] Okay. I'm gonna go ahead and drop off. Hi, everyone. Have a great DevCon. [27:08.290 --> 27:12.190] We'll do. Have fun. Awesome. Thanks all.