WEBVTT 00:00.000 --> 00:09.500 This is dedicated to all the hackers and the crackers, to the hackers and the crackers. 00:09.980 --> 00:16.560 I see in binary, I speak source code, step on my toes, I'll post a million jpegs with you in a fag pose. 00:16.800 --> 00:21.800 In your digital stance, sitting firewalls hoes, cause I'm riding the net in my six foes. 00:21.980 --> 00:26.540 I was a dick with my 56, now with my cable I'm able to get that stable. 00:26.540 --> 00:32.560 On the out, my name is Ace and I'm a Leo. On the digital highway, my name is Neo and I'm a Nero. 00:32.760 --> 00:38.440 In a flash, I'll screw you on burning dreamtacks. You need some ISO? Let me do my hard drive rifle. 00:38.780 --> 00:43.600 Our exchange, you could never stifle. In a digital hood, you just caught the love bug. 00:43.920 --> 00:47.620 I bootlegged your CD, I caused the fight between Un and Jay-Z. 00:47.940 --> 00:52.320 Your CG, it's all gonna be free, whether we take it by force or we take it night free. 00:52.320 --> 00:56.340 You feel that rattle in your bones? When I tell you we just hacked out Jones. 00:56.540 --> 01:00.800 And NASDAQ leaves you flat on your back, cause I am he who loves to hack and crack. 01:01.160 --> 01:03.660 Cause I am he who loves to hack and crack. 01:04.080 --> 01:06.520 This is dedicated to the hackers and the crew. 01:07.120 --> 01:11.500 Serial pose, swords, toes, ISO, rad, fifth, fifth. 01:11.980 --> 01:13.840 This is dedicated to the hackers and the crew. 01:13.840 --> 01:18.480 Welcome to Binary Revolution Radio. This is episode 161. 01:18.480 --> 01:23.600 And we're gonna rename the show to Ben Verbradio, I think. 01:24.240 --> 01:28.080 Because Mr. Verbal, again, has come through and is the man about the show. 01:28.740 --> 01:30.920 And I'm Droops and Verbal's over somewhere. 01:31.880 --> 01:32.740 Hello, everyone. 01:33.140 --> 01:34.780 And we're recording on Asterisk. 01:34.980 --> 01:39.080 So, maybe the sound will be a little bit better than the last time Verbal was on. 01:39.080 --> 01:44.100 Um, admittedly, I have not listened to episode 159 yet. 01:44.520 --> 01:45.660 It's really hard to hear. 01:46.460 --> 01:48.840 I hear the sound quality is lacking. 01:49.140 --> 01:49.740 But that's okay. 01:50.620 --> 01:52.000 The content is there. 01:52.200 --> 01:53.140 And that's all that matters. 01:54.160 --> 01:54.680 Exactly. 01:55.360 --> 01:55.880 Exactly. 01:55.880 --> 01:56.000 Exactly. 01:58.000 --> 02:05.860 Well, I don't have any viewer mail other than Mr. Vaughn wrote in and said that Verbal had a sexy voice. 02:05.860 --> 02:09.260 Mr. Vaughn, give me a call. 02:09.740 --> 02:10.700 We're stepping out. 02:11.540 --> 02:14.300 I think he's over from the West Coast also. 02:14.480 --> 02:16.020 You might need to hook up. 02:17.420 --> 02:18.020 Oh, sweet. 02:18.120 --> 02:18.960 That's what I'm talking about. 02:19.020 --> 02:22.100 That's why I'm doing this in actuality. 02:23.260 --> 02:24.380 If you meet other men. 02:26.320 --> 02:26.800 Yes. 02:27.720 --> 02:28.280 All right. 02:28.360 --> 02:29.960 Actually, viewer mail. 02:30.380 --> 02:32.240 I did not get any viewer mail. 02:32.440 --> 02:33.780 I am pissed off. 02:33.780 --> 02:40.960 If you remember, in episode 159, I had talked about that mentoring program that I started. 02:41.640 --> 02:42.840 I mean, or I want to start. 02:43.460 --> 02:48.960 And I said, if you want to be mentored, or if you want to donate your time as a mentor, 02:49.840 --> 02:54.380 please email me at verbal at gmail dot com. 02:54.740 --> 02:57.000 And no one has effing emailed me. 02:57.000 --> 03:03.880 And, which is fine, but I see people asking questions on IRC, and people are still posting 03:03.880 --> 03:04.440 in the forums. 03:04.760 --> 03:10.600 So I know some of you out there want help, and you're not emailing me, emailing me. 03:10.960 --> 03:16.040 So after the show, you better fucking get on your computer and email me. 03:16.320 --> 03:17.780 Or email me right now, okay? 03:17.820 --> 03:19.120 You should be in front of your computer anyway. 03:19.120 --> 03:24.500 Anyway, so if in the next show I do, I get no more email, I'm going to come to your house 03:24.500 --> 03:25.400 and beat you. 03:26.520 --> 03:27.000 That's right. 03:27.100 --> 03:29.980 Even though he won't know who you are, because you didn't email him. 03:31.880 --> 03:32.280 Details. 03:32.820 --> 03:33.900 I will work that out. 03:36.040 --> 03:41.240 Well, you had quite a few topics for us to go through on this show. 03:41.440 --> 03:44.440 Do you want to do the personal hacking, or do you not like that section? 03:44.440 --> 03:48.400 What is the personal hacking? 03:48.660 --> 03:51.080 When you tell what you've been up to lately. 03:51.860 --> 03:52.740 No, I hate that. 03:53.200 --> 03:53.560 Awesome. 03:53.660 --> 03:54.580 I hate that segment. 03:55.100 --> 03:58.060 If there's something worth talking about, I will talk about it. 03:58.840 --> 04:01.180 Good deal, because I'm going to talk about what I've been working on. 04:01.520 --> 04:02.520 Yeah, no, go for it. 04:02.560 --> 04:03.000 Go for it. 04:03.120 --> 04:05.000 No, no, I'm going to include it into the show. 04:05.900 --> 04:06.260 Okay. 04:06.500 --> 04:07.360 You won't even know. 04:08.340 --> 04:10.660 Oh, I see how it is. 04:10.740 --> 04:12.160 You're going to whip it in there. 04:12.940 --> 04:13.300 Okay. 04:13.300 --> 04:15.920 I'm just going to whip it out, and you're not even going to notice. 04:16.800 --> 04:17.300 All right. 04:17.520 --> 04:19.060 Shut me up for the first segment, then. 04:20.300 --> 04:22.340 Well, why don't you start us off there, buddy? 04:23.260 --> 04:23.580 Yes. 04:25.380 --> 04:26.480 Oh, you want me to start? 04:26.640 --> 04:27.200 Oh, yeah. 04:27.980 --> 04:28.520 Oh, okay. 04:28.520 --> 04:28.860 I'm sorry. 04:29.780 --> 04:34.840 Okay, today we have actually two main topics, but I'm going to split them up. 04:35.300 --> 04:39.640 So, first we're going to talk about the AOL data that was released. 04:39.640 --> 04:42.480 There was a giant thread about it on the forums. 04:43.300 --> 04:45.960 And then we're going to have something a little different. 04:46.520 --> 04:59.900 We're going to have an actual live female call, and she's going to talk to us a little bit about kind of women in the hacking scene and all that stuff. 04:59.900 --> 05:01.400 So, more on that a little bit later. 05:01.400 --> 05:12.140 And then, finally, I'm going to go over something that some of you may not have heard about, but it is becoming a very important part of your hacking toolkit. 05:12.620 --> 05:17.180 And it's an area of computer science called machine learning and data mining. 05:17.640 --> 05:18.980 So, that should be fun. 05:19.080 --> 05:20.300 That will be the technical part. 05:20.900 --> 05:21.220 Excellent. 05:21.360 --> 05:23.220 I thought you said you were only talking about two things. 05:23.220 --> 05:26.260 No, but it's split up. 05:26.720 --> 05:27.120 Ah. 05:27.640 --> 05:31.980 So, I'm going to do the AOL data thing and do the non-technical part first. 05:32.400 --> 05:35.860 And then I'm going to split it up and do the technical stuff at the end. 05:36.240 --> 05:36.960 Good deal, dude. 05:38.620 --> 05:40.500 Do you even know what the show is about? 05:41.240 --> 05:42.200 I'm not even here. 05:42.320 --> 05:44.320 This is actually just the machine talking about me. 05:45.000 --> 05:45.920 Get out of here. 05:46.280 --> 05:46.460 Okay. 05:46.460 --> 05:47.440 It's an AGI script. 05:47.520 --> 05:48.280 I have an asterisk. 05:48.440 --> 05:49.320 I'm just wearing the numbers. 05:51.700 --> 05:52.380 Thank you. 05:52.560 --> 05:54.080 And come again. 05:55.260 --> 05:55.740 Alrighty. 05:56.680 --> 05:57.120 Ow. 05:57.280 --> 06:02.020 I just watched that family guy last night when Brian goes back to college. 06:02.440 --> 06:03.260 I love that one. 06:03.440 --> 06:05.760 There's that Stephen Hawking guy. 06:05.760 --> 06:09.760 And the sex scene is hilarious. 06:10.720 --> 06:12.100 The ow, ow, ow. 06:12.380 --> 06:13.400 Not so hard. 06:13.600 --> 06:15.100 You are hurting me. 06:15.100 --> 06:16.940 All right. 06:17.080 --> 06:18.340 Anyway, let's start. 06:18.460 --> 06:18.700 All right. 06:18.760 --> 06:19.460 We're going to have fucking. 06:20.060 --> 06:20.360 Okay. 06:21.280 --> 06:21.760 Okay. 06:23.340 --> 06:23.700 Woo. 06:24.000 --> 06:27.500 So, I don't really post on the phone that much. 06:27.620 --> 06:29.780 Or do I read it for whatever reason. 06:29.980 --> 06:33.740 But lately, I just had to go on because Stephen's giving me a hard time about it. 06:34.060 --> 06:39.080 So, I look and there's a thread that says that AOL has released research. 06:39.740 --> 06:44.020 I mean, has released search data that they want people to use for research. 06:44.020 --> 06:46.100 Troops, have you seen that thread? 06:46.340 --> 06:47.620 I have seen that thread. 06:48.900 --> 06:49.340 Okay. 06:49.480 --> 06:52.680 So, I don't know how much you went into it. 06:52.740 --> 06:56.380 But basically, they released about two gigs of data. 06:56.380 --> 07:00.400 And it is web queries from AOL users. 07:00.700 --> 07:02.740 And it doesn't give anyone's name. 07:03.260 --> 07:06.140 But each user is assigned a unique ID. 07:06.600 --> 07:13.940 So, you know, like who, or not who, but, you know, you know that this person executed all these queries. 07:14.040 --> 07:15.720 So, they're grouped per person. 07:15.720 --> 07:19.780 And now, you get to go and look through them and do whatever you want. 07:19.900 --> 07:26.420 And people have been having a field day with this because of all the weird, random shit that people are actually looking for. 07:26.980 --> 07:29.220 And, you know, there are a lot of examples of that on the forum. 07:29.740 --> 07:35.000 But the thing I want to talk about is the privacy issues surrounding this, right? 07:35.000 --> 07:43.380 Because, you know, this has really been, like, a hot topic lately, especially with Google and how, you know, they have, like, all of your data, right? 07:43.640 --> 07:45.760 So, it's like they know what you're searching for. 07:45.920 --> 07:47.280 You know, you do your email to them. 07:47.340 --> 07:48.680 And now there's Google Calendar. 07:48.680 --> 08:02.700 And, like, if you use Google Desktop, then you can, like, somehow speed up indexing or actually put your hard drive index onto their server so it's integrated when you search on the web and all this stuff. 08:03.100 --> 08:07.120 So, this privacy issue is kind of, you know, really important right now. 08:07.700 --> 08:18.500 And there has been a debate, actually, about this because people – so, AOL, obviously, is of the position that they are not violating any privacy. 08:18.880 --> 08:27.160 Laws or, you know, or whatever that they promise their users because, you know, they're not saying, okay, this guy executed these queries. 08:27.420 --> 08:28.740 But they have a unique ID. 08:29.140 --> 08:42.100 And, you know, if you're smart enough and clever enough and you go into the data set and you look at it, you know, you could figure out what some of these people – who some of these people are, at least, you know, where they are or something about them. 08:42.100 --> 08:48.480 You know, like someone from the forums that old people really love family genealogy. 08:49.020 --> 09:03.420 So, you know, if you see, like, one person who's, like, constantly searching for some last name and, like, variations of the first name with that last name, then, you know, chances are pretty good that, you know, that's their last name. 09:04.160 --> 09:05.520 And you're dealing with an old person. 09:06.420 --> 09:06.740 Right. 09:06.860 --> 09:07.720 Yeah, exactly. 09:08.340 --> 09:11.640 So, Troops, what do you think are the privacy implications of this? 09:11.720 --> 09:16.180 Like, is it wrong of AOL to do this and if they're violating any agreements? 09:18.660 --> 09:21.080 Well, I don't know. 09:21.100 --> 09:22.000 It's kind of a toss-up. 09:22.060 --> 09:24.920 I don't know the AOL agreements they have with their customers. 09:24.920 --> 09:31.060 I haven't been an AOL customer for years and years and years back when that was my only option. 09:31.060 --> 09:39.040 But, like, if I go to, like, Yahoo.com, I'm not paying them any money and I use their service. 09:39.160 --> 09:46.060 I would expect them to retain that data of me doing my searches and try to make their money off of it. 09:46.100 --> 09:56.800 Like, if I call a 1-800 number, I expect them to keep my phone number, you know, my caller ID, and use it for whatever purposes they want because they're paying for it. 09:56.860 --> 09:59.300 You know, I'm getting a free service off of it and nothing's free. 09:59.300 --> 10:03.500 But with AOL, now, this is their actual users. 10:03.680 --> 10:06.880 Like, you have to have the AOL application, I believe, whatever it's called. 10:07.560 --> 10:08.280 Yeah, yeah, yeah. 10:08.340 --> 10:09.160 And do those searches. 10:09.240 --> 10:10.480 It's a gateway to the Internet. 10:11.080 --> 10:11.400 Yes. 10:11.580 --> 10:15.420 It's not like people that are just going to AOL.com and using that for search, right? 10:18.200 --> 10:18.560 Right. 10:18.900 --> 10:19.220 Okay. 10:19.400 --> 10:19.920 Just making sure. 10:20.020 --> 10:21.840 I wasn't 100% on that. 10:22.140 --> 10:23.140 I wasn't any percent on that. 10:23.160 --> 10:23.860 I was just making it up. 10:23.860 --> 10:28.560 But, like, with AOL, you're paying them money for a service. 10:29.700 --> 10:41.860 And you would kind of expect, even though you're not paying Yahoo or not paying Google when you are paying AOL, that if they're going to try to make their money off of that, they're not going to release pertinent information about you. 10:41.860 --> 10:43.660 You know, like... 10:43.660 --> 10:45.320 Well, hold on. 10:45.500 --> 10:46.440 Let me hear what I'm saying. 10:46.520 --> 10:47.180 I don't know. 10:47.180 --> 10:50.720 Well, I mean, the thing is, it really depends on the user agreement. 10:51.160 --> 11:00.020 Like, even if you're paying for it, if the user agreement says, you know, all the data we collect are ours and we can sell it, then they can do that, even if you're paying for it. 11:00.020 --> 11:05.840 But, I think, in most cases, I mean, like, I'm no lawyer and I don't read the U.S. 11:05.840 --> 11:08.300 Well, you, of course, expect to have that privacy. 11:08.380 --> 11:08.720 Right. 11:08.940 --> 11:09.440 Right, right, right. 11:09.440 --> 11:10.020 It's a common person. 11:10.840 --> 11:11.260 Right, right. 11:11.360 --> 11:23.720 So, I think what happens is they say, okay, you're paying for this, you know, we retain the rights to, like, so-and-so, but we're not going to, you know, invade your privacy. 11:23.720 --> 11:27.520 We're not going to sell your, you know, fucking credit cards to the bank. 11:27.800 --> 11:28.560 I mean, that's the bank. 11:29.000 --> 11:29.140 Yeah. 11:29.140 --> 11:35.200 Yeah, we're not going to sell your credit cards to, like, Russian mobs and we're not going to do this and we're not going to do that and you're going to be safe. 11:35.780 --> 11:50.780 Now, I think the issue is that they're not technically breaking any laws by releasing this data, but because, you know, of the way the searches are and what people do, you know, you can figure it out. 11:50.780 --> 11:54.340 And it's kind of, you know, it's basically a PR nightmare. 11:55.300 --> 11:55.460 Yeah. 11:55.460 --> 12:02.160 So, you know, it's like, should companies do that even if they're not breaking any rules? 12:02.160 --> 12:08.820 Well, I personally feel it's a corporation or a business's response. 12:09.080 --> 12:10.920 Like, they're responsible to be ethical. 12:12.020 --> 12:17.420 You know, like, they shouldn't just be going around shitting on people, even though maybe they're allowed to. 12:18.460 --> 12:18.760 Right. 12:18.760 --> 12:22.720 And in this modern age, of course, you know, that idea is out the window. 12:23.820 --> 12:23.920 Right. 12:23.920 --> 12:24.960 Do whatever the fuck you want. 12:25.900 --> 12:27.540 And, I don't know. 12:28.520 --> 12:32.240 I'm kind of at a toss-up with just because AOL, I haven't seen their agreement. 12:33.520 --> 12:34.440 Right, right, right. 12:34.440 --> 12:36.880 But you aren't paying them for a service. 12:36.960 --> 12:37.740 You wouldn't expect them to do that. 12:37.780 --> 12:43.040 Now, we built, like, a search engine little part of the application we're building for work. 12:43.740 --> 12:53.380 And we keep all those searches, obviously, in the database so that I can go in there later and look and see what screw it up or who's mistyping what. 12:53.380 --> 13:02.600 But, you know, like, if they keep mistyping my last name, searching for me, I can, you know, point it to me a lot easier and see how that is. 13:02.900 --> 13:05.660 Right, but, I mean, that's, right, but that's, right. 13:05.660 --> 13:06.940 But I'm not exactly calling you. 13:07.680 --> 13:08.300 Right, right. 13:08.440 --> 13:10.040 I mean, right, right, right. 13:10.100 --> 13:14.100 Because, I mean, obviously, everyone at Google knows what I'm searching for. 13:14.360 --> 13:18.620 But what I care about is they don't tell, like, the rest of the world. 13:18.860 --> 13:19.180 Yeah. 13:20.580 --> 13:21.980 I'm not publishing that data. 13:22.740 --> 13:23.360 Right, right. 13:23.380 --> 13:26.740 So, I mean, like, you know, like, Google tries a lot. 13:26.820 --> 13:35.340 Like, I have a ton of friends who work at Google, and they tell me that they try very, very hard to not be evil and to not be creepy. 13:35.780 --> 13:37.180 I mean, sometimes it's very hard. 13:37.300 --> 13:44.840 Like, for example, if you work there, then, like, they do so much more stuff for you because you work for them. 13:44.980 --> 13:50.600 So, they will do all the creepy stuff and the weird stuff that they won't do for public users. 13:50.600 --> 14:01.240 Like, they'll track exactly where you are and, like, where you go if you use any of their services that do that, you know, that gather that information, and they'll track what you're searching for. 14:01.560 --> 14:05.820 You know, they'll track your email and, like, because everyone at Google uses Gmail. 14:05.820 --> 14:14.680 So, they'll track all of that, and they'll, like, actually look at it and, like, try to find out stuff to improve on, you know, their current products. 14:15.000 --> 14:17.120 But they won't do that for public people. 14:17.620 --> 14:18.060 Yeah. 14:18.060 --> 14:29.020 But, you know, for AOL to release this, I mean, personally, I don't know what the motivation is because it's like, okay, we're releasing this because we're good people 14:29.020 --> 14:39.940 and we want to, like, you know, give a data set to whoever, you know, researching these kind of things and, like, to help further academic studies or whatever. 14:40.080 --> 14:40.540 That's fine. 14:40.540 --> 14:51.400 And, see, that's definitely a good thing about all this is that now we do have a data set that we can go into, you know, and do data mining on it or whatever we need. 14:52.680 --> 14:53.080 Right. 14:53.080 --> 15:06.120 But then you get, like, these people who, like, figure out, you know, who is user, like, you know, FUBAR, and they call them and, like, go to the house and say stuff like, oh, hey, what's up? 15:06.160 --> 15:11.280 I found all your info and what you're searching for from the data set that AOL released. 15:11.860 --> 15:13.400 And the AOL user's like, what? 15:13.680 --> 15:13.920 Huh? 15:14.300 --> 15:21.860 Yeah, you know, like, wait, like, so on the forums, like, some people have done this, like, Rikos does. 15:21.860 --> 15:25.080 And, you know, he's not, like, I know him, and he's a great guy. 15:25.220 --> 15:35.860 So I know he's not going to, like, stalk someone or, like, you know, find people who are into, like, pyro necrobeciality or something, you know, 15:35.980 --> 15:43.320 and then, like, you know, find where they live and, like, have flaming cat sex with them, flaming dead cat sex with them. 15:43.600 --> 15:45.420 But, like, other people could do this. 15:45.420 --> 15:52.700 And, you know, AOL has to be aware of that because, you know, it's like they don't just randomly pick two gigs of data and release it. 15:52.860 --> 15:55.560 Or maybe they did, and if they did, that's really stupid of them. 15:55.840 --> 16:04.540 But they should know what the data set contains, and they should know that it is possible to find certain people from this data set. 16:04.540 --> 16:07.700 So I don't know, like, what they're thinking. 16:08.280 --> 16:09.740 I mean, that's just weird to me. 16:12.540 --> 16:24.120 Well, so it's a bad thing for the AOL users, but it's a good thing for people that need a good viable data set that's not just randomly made up crap. 16:24.120 --> 16:30.220 Right, like, there are tons of data sets out there for training your spam filters. 16:30.740 --> 16:39.960 You know, like, people just collect spam, and they have them bundled, and you can just download it, and, you know, you can use it for, like, if you're writing a spam folder. 16:40.460 --> 16:41.560 And that's fine. 16:41.680 --> 16:44.400 I mean, you know, that's not really hurting anyone because it's just spam. 16:44.780 --> 16:47.280 But, like, other more sensitive data, like, you know, I think... 16:47.280 --> 16:48.440 Well, it's hurting the spam guys. 16:49.780 --> 16:51.540 Yeah, that's true. 16:51.540 --> 16:51.620 True. 16:54.120 --> 16:56.480 I mean, I guess... 16:56.480 --> 16:58.960 I mean, criminals have rights, too, I guess. 16:59.200 --> 16:59.980 Yeah, they do. 17:00.540 --> 17:01.680 Yeah, they do. 17:02.320 --> 17:03.880 But, all right, let's move on, then. 17:04.880 --> 17:06.660 So, I was looking... 17:06.660 --> 17:13.640 Well, the lady, the chief information officer at AOL, I forgot her name, but she resigned. 17:15.060 --> 17:16.340 Oh, oh, really? 17:16.720 --> 17:17.300 Oh, really? 17:18.840 --> 17:19.980 Because of this? 17:20.480 --> 17:22.560 It was right after this, like, a couple days. 17:22.560 --> 17:25.740 Right, but, like, she resigned because of this incident? 17:26.620 --> 17:28.660 Well, it didn't say why she resigned. 17:29.140 --> 17:30.040 Oh, right, okay. 17:30.240 --> 17:32.040 But that's what one would assume. 17:33.580 --> 17:39.060 That's ridiculous, because you would think that she would, like, make someone lower rank than her take the fall. 17:39.860 --> 17:42.980 Well, see, I'm a firm believer in the... 17:42.980 --> 17:46.500 If you make a mistake and you learn from it, then you're better off. 17:47.540 --> 17:47.980 Right. 17:48.360 --> 17:56.720 So, like, I would rather have an employee that has made a thousand mistakes, and we've already, you know, gone through it and resolved the issues, 17:56.720 --> 18:01.760 and have that employee that have to fire him and hire a new guy that's going to make the same thousand mistakes. 18:02.660 --> 18:03.780 Oh, I mean, of course. 18:03.780 --> 18:09.380 But, I mean, the whole reason they do that is because for PR reasons, you know. 18:09.480 --> 18:14.940 It's like, they can't keep her around because she's a person who did X and it was bad. 18:15.480 --> 18:15.600 Yeah. 18:15.600 --> 18:17.040 You know, like, it's really not. 18:17.160 --> 18:19.320 You know, like, because people are stupid. 18:19.680 --> 18:22.920 Basically, it comes down to people are stupid, and that's the end of that. 18:23.600 --> 18:24.320 So, okay. 18:24.600 --> 18:26.440 So, let's change this topic a little bit. 18:27.460 --> 18:34.360 So, I looked at the forums, and basically, I want to see what people are doing with this data, right? 18:34.480 --> 18:42.480 I mean, it's there, and, you know, hopefully, you can extract some cool information from it and learn something about the data set 18:42.480 --> 18:47.940 or, you know, find interesting, like, clumps of information from these things. 18:47.940 --> 18:54.300 And it's like, you know, I'm looking, and people seem to be grepping for weird shit. 18:54.960 --> 18:57.500 And that is pretty much all that's going on. 18:57.640 --> 19:03.920 Like, I don't know if you play with it, but what kind of things would you want to figure out? 19:04.080 --> 19:06.840 Like, you know, like, what would you do to the data set? 19:06.940 --> 19:08.480 Like, you have this data, right? 19:08.600 --> 19:11.020 Like, how are people going to use it? 19:13.160 --> 19:14.300 Are you asking me? 19:15.380 --> 19:16.980 Oh, I'm putting you on the spot. 19:16.980 --> 19:26.920 Oh, I'll tell you what I would do is I would first search for porno passwords to see if anybody had found any good sites with good porno passwords. 19:28.200 --> 19:30.700 You're glad Mrs. Joops is not on the show tonight. 19:32.360 --> 19:33.780 Yeah, she went to bed early. 19:35.280 --> 19:37.540 And that's actually a really dumb joke. 19:37.540 --> 19:42.900 Well, I kind of am interested in how sick fucking people are. 19:43.880 --> 19:45.920 And I would probably search for just sick things. 19:46.460 --> 19:53.120 And then when I found a user ID that this guy was into something weird, I would just grab for his user ID and see what else he was into. 19:53.480 --> 20:00.120 So to see if he's, like, into, like, stock car racing and then, like, other weird shit. 20:00.120 --> 20:03.760 People have definitely done that. 20:04.180 --> 20:11.820 And, you know, there are a lot of – I mean, like, the thing that surprised me most is that people search for, like, questions. 20:12.160 --> 20:16.620 Like, do you remember the website I think called Dogpile? 20:17.440 --> 20:17.680 Yeah. 20:17.680 --> 20:23.240 People – yeah, like, you're supposed to ask it questions, you know, so it's more user-friendly. 20:23.900 --> 20:31.320 But, like, you know, people still do that with AOL, and there's these, like, long, long strings of questions and other weird crap. 20:31.780 --> 20:33.520 I like how people will search for Google. 20:34.720 --> 20:35.120 Yeah. 20:35.600 --> 20:41.120 Or they'll search for, like, you know, like, something.com instead of just typing it into the address bar. 20:42.200 --> 20:42.680 Yeah. 20:42.680 --> 20:49.140 I mean, I mean, there are definitely some really weird shit in there. 20:49.280 --> 20:58.680 Like, someone actually made a search – a web front end for the search, and, like, search what people has queried for, 20:58.780 --> 21:02.280 if you put in, say, like, a text string or the user ID. 21:02.820 --> 21:06.100 And, like, there's some really, really weird stuff here. 21:06.200 --> 21:08.180 Let me give – okay. 21:08.180 --> 21:16.540 This is a user that was actually posted to the forum, and just to give you an idea of this weird stuff – okay. 21:17.380 --> 21:23.300 This is a search for how to destroy demons that live in apartment above. 21:24.200 --> 21:24.580 Okay. 21:26.420 --> 21:30.000 Is hip-hop and rap music a form of Satanism? 21:31.340 --> 21:34.720 Are niggers Satan or demons or gremlins? 21:35.160 --> 21:36.460 Like, like, questions. 21:36.460 --> 21:36.860 Gremlins. 21:36.860 --> 21:38.580 Oh, that wasn't a question. 21:38.680 --> 21:39.060 I'm sorry. 21:40.500 --> 21:44.280 You know, do niggers have x-ray vision? 21:46.340 --> 21:51.040 I mean, there is just fucking weird shit. 21:51.600 --> 21:55.840 Like, I don't even – I mean – but I digressed. 21:56.580 --> 21:59.980 And then there's something normal, like, jobs going bad. 22:00.620 --> 22:03.260 You know, I don't understand. 22:03.260 --> 22:11.740 And my favorite, this guy searched for this about 20 times, old, rushed nuns for sex. 22:13.120 --> 22:13.640 Sweet. 22:14.360 --> 22:16.440 I don't know what to say about that. 22:16.580 --> 22:19.380 See if anyone searched for, like, A-plus exam. 22:20.660 --> 22:21.020 Yeah. 22:22.020 --> 22:22.500 Yeah. 22:22.500 --> 22:31.320 So, later in the show, actually, we're going to change gears again a little bit. 22:31.860 --> 22:44.520 Later in the show, I'll get back to talking about machine learning and data mining and basically how these techniques can help us maybe extract useful information from data theft. 22:44.520 --> 22:55.400 Because the point is, you're going to end up with a sea of crap, just, you know, like, gigs and gigs and gigs of text that's data from anything. 22:55.500 --> 22:56.100 It doesn't matter. 22:56.660 --> 23:01.580 And you're not going to be, you know – you can only grep for cool shit for so long. 23:01.580 --> 23:10.420 But if you want to find useful, like, information in it and, like, how things relate to each other, then you're going to need to do something a little smarter than that. 23:10.820 --> 23:19.260 So, later in the show, I'm going to talk about some of the ways you can do that and software that's out there, you know, so you can easily implement this kind of stuff. 23:19.260 --> 23:28.700 But right now, we're going to do a segment, which is unnamed, and we're going to invite a guest, actually. 23:28.980 --> 23:38.880 A friend of mine, whose name is Nolani, is going to come on, and I'm just going to ask her some questions, and we'll see what she says, okay? 23:39.760 --> 23:40.620 Hi, Nolani. 23:40.740 --> 23:42.240 Thanks for calling into the show. 23:42.240 --> 23:52.340 So, we have Nolani, who's a friend of mine, and she has been married to a hacker for how long, for four years or so? 23:52.980 --> 23:55.440 Yeah, four years, but we've been together for about seven. 23:56.440 --> 23:56.920 Okay. 23:57.180 --> 23:58.180 Okay, so that's great. 23:58.180 --> 24:04.820 So, she knows what it's like to have a relationship with a hacker and just a geek in general. 24:06.280 --> 24:09.820 So, I understand that you were at DEF CON recently, right? 24:10.000 --> 24:10.200 I was. 24:10.200 --> 24:11.100 And it was your first time? 24:11.560 --> 24:12.160 Yes, it was. 24:13.200 --> 24:20.880 Um, so, okay, what I was saying, actually, in episode 159, when we're doing the DEF CON wrap-up, 24:21.060 --> 24:32.400 I was very surprised at the amount of female attendance at the CON, because I was there, like, you know, three years ago, 24:32.780 --> 24:37.820 and, you know, there was, like, no women there at all, you know, 24:37.820 --> 24:45.340 and this year we have women presenting with people in contests and everything, and just generally attending. 24:46.060 --> 24:51.840 And so, I was, you know, so I want to ask your perspective of what you thought, you know, 24:51.840 --> 24:59.200 about kind of, like, the role of women in hacking and, you know, attending the CON and getting involved in the community. 24:59.200 --> 25:08.300 Well, my husband, the first year that he went was, I think, DEF CON 7, and he said there weren't many females there. 25:08.300 --> 25:12.600 And he said the turnout, just like you, has gone up greatly. 25:13.020 --> 25:19.200 And I think, just like any profession, whether it's, you know, a physician or a hacker, 25:19.200 --> 25:24.000 you're always going to start out with a low number of females, because, you know, the man's always keeping us down. 25:24.780 --> 25:26.300 And then we slowly... 25:26.300 --> 25:27.820 I don't know, right. 25:28.960 --> 25:31.360 What exactly do you mean by that? 25:31.760 --> 25:36.280 Can you elaborate on the man putting us down or keeping us down? 25:37.140 --> 25:46.080 Well, I think that, you know, a common misconception is that women can't do a lot of things as well as men can, 25:46.080 --> 25:50.100 when, in reality, we can do it sometimes just as good or better. 25:51.100 --> 25:51.820 All right, right on. 25:51.980 --> 25:52.220 Okay. 25:54.540 --> 25:58.960 So, I think, you know, the representation, of course, is going to go up year by year. 25:59.640 --> 26:02.920 My husband and I are growing our two hackers. 26:03.320 --> 26:04.700 We have a lockpicker in the family. 26:05.800 --> 26:11.080 So, hopefully, we'll still grow up to be good hackers, too. 26:12.520 --> 26:13.500 Okay, yeah. 26:13.500 --> 26:16.900 Yeah, I mean, that's something I think we're going to see. 26:17.500 --> 26:25.400 But, so, you were mentioning that, you know, the female attendants in, you know, the tech field, 26:25.540 --> 26:31.720 not just the hacker field, but, you know, people being programmers or being electrical engineers or whatever, 26:31.720 --> 26:38.440 just generally in the tech field are lower for, you know, for females just because of, you know, 26:38.440 --> 26:46.120 not necessarily of any actual, like, ability, you know, differentiation or whatever, but, you know, 26:46.240 --> 26:48.220 something like a social thing, right? 26:48.720 --> 26:49.200 Yeah, it is. 26:49.360 --> 26:49.680 Yeah. 26:49.680 --> 26:58.420 So, what do you suggest, you know, maybe, like, someone who is interested in getting into the field 26:58.420 --> 27:04.080 but kind of has, you know, feel that invisible, like, social pressure that, like, okay, 27:04.140 --> 27:09.860 maybe girls shouldn't be doing math and, like, they shouldn't be, like, good at science, et cetera, et cetera. 27:09.860 --> 27:16.280 And, you know, like, if someone really went against that field, like, kind of, not can't, but it's hard for them to, 27:16.340 --> 27:18.140 like, what do you recommend they do? 27:18.140 --> 27:30.780 I think one of the things that they can do is just really attach on to a group of guys that are in the field. 27:31.160 --> 27:35.460 And, you know, a lot of the guys that I know really are willing to take someone in, 27:35.640 --> 27:41.620 regardless of how you look as a female, to really teach them the tricks of the trade and teach things. 27:41.940 --> 27:45.820 One of the misconceptions, I think, of hackers is that they're not really friendly. 27:45.820 --> 27:52.540 And all the people that I met at DEF CON were excessively friendly to me and, you know, and my older daughter. 27:53.020 --> 27:55.460 And we didn't have any problems with people being rude or anything. 27:56.180 --> 28:00.780 So, I think just stepping into the arena would be good and going from there. 28:02.200 --> 28:02.720 Okay. 28:03.080 --> 28:03.380 Okay. 28:03.380 --> 28:06.520 But I think you actually mentioned something that's, like, a stickler for me. 28:07.020 --> 28:13.320 You know, girls in general, there's this whole thing that happens in middle school 28:13.320 --> 28:18.280 where math and science doesn't become important to them because their hormone levels start to go up. 28:19.000 --> 28:19.040 Right, right. 28:19.040 --> 28:24.120 And it's been taught that, you know, in order to get a good guy, that you've got to be a student. 28:24.120 --> 28:24.480 Right. 28:25.300 --> 28:30.400 So, I mean, like, one of the things I'm trying to get at is, you know, 28:30.520 --> 28:34.600 if you're in that position, and you're kind of in middle school or early high school, 28:34.600 --> 28:37.120 like, how do you fight against that, you know? 28:37.200 --> 28:39.480 Like, you know, like, now, I'm okay. 28:39.600 --> 28:44.680 So, since we're all older, we know, like, okay, it doesn't really matter, you know, 28:44.720 --> 28:50.600 if, like, people don't get you or, like, they shun you for, like, being in stuff you're not supposed to. 28:51.060 --> 28:54.140 But from the point of view of someone in that position, you know, 28:54.160 --> 28:57.320 having friends and being popular is actually, you know, really important. 28:58.160 --> 29:04.100 So, you know, like, you know, to them, how would you tell them to kind of, you know, do both? 29:05.020 --> 29:09.380 Yeah, I think, well, I mean, and this is from my parents' point of view. 29:09.480 --> 29:13.240 It really starts with, you know, building the good foundation when they're smaller. 29:13.380 --> 29:18.020 But, you know, just because you're smart doesn't mean that it's a bad thing 29:18.020 --> 29:23.080 and you're not going to be able to hang out with people or get a boyfriend or a girlfriend or whatever. 29:23.080 --> 29:27.520 And, you know, a lot of the schools, at least in our area, are focusing on that 29:27.520 --> 29:33.080 and really getting teenage girls involved in these programs where they are exposed to, you know, 29:33.460 --> 29:38.740 geeky women in their field being successful and having a happy life. 29:39.020 --> 29:41.960 And I think that schools are starting to recognize this whole thing. 29:43.080 --> 29:46.460 And letting people know, you know, at least with our kids, we say, 29:46.840 --> 29:51.080 it doesn't really matter what happens now. 29:51.080 --> 29:56.460 Because all those people who make fun of you for being smart are going to be fat and ugly and unsuccessful. 29:57.640 --> 29:59.040 I know, right? 29:59.640 --> 30:00.200 Totally. 30:00.600 --> 30:02.240 Well, we see the unsuccessful part. 30:02.920 --> 30:03.280 Yes. 30:05.160 --> 30:13.080 Yeah, I mean, yeah, but, you know, it's so hard to get that point across to, you know, people in that situation. 30:13.540 --> 30:17.080 Because I was in that situation not too long ago. 30:17.080 --> 30:23.920 And, you know, it's really hard to see that because, you know, you're living at that point in time 30:23.920 --> 30:26.200 and not, like, in five years in the future. 30:26.600 --> 30:29.000 But, I mean, it's something to definitely have to learn. 30:30.040 --> 30:30.480 Yes. 30:30.480 --> 30:30.720 Okay. 30:31.360 --> 30:38.760 So another thing I wanted to ask you, actually, is what is it like being married to one of the biggest geeks I know? 30:38.760 --> 30:47.440 I would say that the thing that definitely attracted me to him first was his passion and something, 30:48.460 --> 30:50.440 just his pure passion into something. 30:50.780 --> 30:53.960 And, you know, it's definitely something that I've had to get used to. 30:54.120 --> 31:00.880 I know that he goes through periods of time where he just needs to focus and be very productive. 31:00.880 --> 31:04.140 And I need to step back and work on my own things. 31:04.740 --> 31:06.440 And I'm totally okay with that. 31:06.660 --> 31:11.100 So it's just a whole adjustment period and learning, just like any relationship. 31:12.040 --> 31:12.940 Right, right. 31:12.960 --> 31:13.900 About a person and whatnot. 31:13.900 --> 31:21.180 But I think geeks are definitely a special type of person where they're just, 31:21.420 --> 31:26.360 they get so focused on one thing or another and you have to learn not to take it personally. 31:26.920 --> 31:30.540 But it's not that they love you any less or think you're less attractive. 31:30.780 --> 31:33.900 It's that, you know, this is it. 31:35.060 --> 31:35.500 Right. 31:35.820 --> 31:38.880 Now, has he had to make any adjustments for you? 31:38.880 --> 31:43.800 I mean, you know, so you're coming from, you know, what you're doing for him and all this stuff. 31:44.020 --> 31:49.180 But, like, have you guys kind of made, you know, a conscious effort to kind of make a compromise somewhere? 31:49.360 --> 31:58.200 You know, let's say, like, maybe he'll, like, set aside specific chunks of time that he's not staring at his computer for 16 hours a day. 31:59.680 --> 32:03.620 You know, he really has his priorities set. 32:03.620 --> 32:09.380 And we go through periods of time where, you know, he is staring at his computer for 16 hours a day. 32:09.540 --> 32:15.480 And I'm okay with that because I'm actually accomplishing my own things, like working on my photography or working on movies or whatnot. 32:15.700 --> 32:19.260 And it's times that I can really focus on things that I'm working on. 32:20.200 --> 32:25.060 And, you know, I'm just not one to sit around waiting for him to finish what he needs to. 32:25.060 --> 32:41.440 So, I mean, do you think it takes, like, a certain type of women to date geeks and hackers who are, you know, prone to prolonged periods of isolation and, you know, being immersed in what they're doing? 32:42.260 --> 32:44.120 You know, like, you have your own thing going on. 32:44.200 --> 32:45.860 So, you guys are both working on something. 32:46.120 --> 32:48.460 So, it's cool that he's off doing whatever. 32:48.460 --> 32:58.020 Yeah, I think it definitely, I don't think it takes a special type of person, per se, but I think it takes somebody who's definitely patient and willing to work in a relationship. 32:59.060 --> 33:01.560 You know, just like any relationship, it doesn't come naturally. 33:01.880 --> 33:10.060 So, the first time he sat down to work at his computer for 16 hours, I was like, you know, hey, what's going on? 33:12.360 --> 33:13.880 Come on, I want to do it. 33:14.580 --> 33:14.980 Exactly. 33:14.980 --> 33:18.620 True, what about you? 33:21.260 --> 33:23.340 I'm just enjoying listening to the show. 33:24.580 --> 33:26.060 You're hosting the show. 33:26.200 --> 33:27.460 Stop listening to it. 33:28.880 --> 33:43.000 My wife and I have similar problems that I, I don't know how you do it, but when I'm working on my computer, or if I'm playing on the computer, which we tend to do that sometimes also, 33:43.000 --> 33:45.720 it's always just playing on the computer. 33:47.520 --> 33:53.640 Like, I think she's getting it now, but in the beginning, she would look at me working on the computer as me just playing. 33:54.640 --> 33:55.580 I see, yeah. 33:56.340 --> 33:58.880 And that would aggravate the ever-living hell out of her. 33:58.880 --> 34:06.440 Yeah, I've never thought that, because I think for me, it was something that we talked about when we first got together. 34:06.800 --> 34:08.940 And, you know, the first time I was like, what the heck? 34:09.480 --> 34:10.380 Why are you on your computer? 34:10.480 --> 34:13.460 He explained to me in detail what he's done. 34:13.460 --> 34:25.780 And I actually see the things as tangible, and he's sharing, even though I don't understand, like, 50, probably 70% of what he's saying to me when he's talking about his coding or this really cool thing you made at work, 34:25.780 --> 34:32.760 I can see the passion in his eyes and understand that there's something tangible that I'm seeing that he's done. 34:33.620 --> 34:34.480 That's hot, right? 34:36.100 --> 34:40.880 Well, Ms. Troops is very supportive when I do something that's cool to me. 34:41.640 --> 34:43.660 And granted, she doesn't get a flip about it. 34:44.340 --> 34:47.820 But, you know, I'll get all excited, and I'll show it to her like I'm a little kid. 34:47.820 --> 34:51.860 And, you know, she pays attention to it, and she's like, oh, that's really cool. 34:52.720 --> 34:59.900 You know, doesn't understand what I'm doing, but she's not a nerd, so she doesn't have to. 35:02.320 --> 35:06.140 All right, so that sounds good. 35:06.240 --> 35:07.560 So there's hope for me, yes. 35:08.200 --> 35:08.860 There is hope. 35:08.980 --> 35:10.940 And she's also into photography. 35:11.860 --> 35:14.540 Were you into photography before the nerd? 35:15.540 --> 35:21.760 I was into photography before, but I got more into it because my nerd is also a hot skater boy. 35:22.540 --> 35:22.980 Awesome. 35:23.460 --> 35:28.620 So I get to take pictures of a hot skater boy and other hot skater boys at the skate park. 35:28.800 --> 35:32.320 So I edit and do all kinds of fun things with those. 35:32.740 --> 35:36.700 It wasn't just the fact that he was on his computer for 16 hours and you had to entertain yourself? 35:37.240 --> 35:38.280 No, it wasn't. 35:38.820 --> 35:39.120 Okay. 35:39.120 --> 35:44.460 I mean, because he was on for 16 hours, I learned how to do like CSS and make my own web page 35:44.460 --> 35:47.480 and I started blogging and all that kind of stuff, but. 35:48.720 --> 35:49.420 Good deal. 35:49.780 --> 35:50.220 It wasn't. 35:50.860 --> 35:51.280 Okay. 35:51.840 --> 35:55.060 So now this is the part where you have to help me. 35:56.100 --> 35:56.440 Okay. 35:56.960 --> 36:06.700 So if I'm looking for someone to go out with and I, bear with me. 36:06.700 --> 36:19.220 So, okay, so from what I understand, normal people will go to like bars and clubs to meet people to like do or something. 36:19.700 --> 36:30.940 But if I'm looking for someone to go out with that is, let's say, of a certain disposition, where would I go to find those people? 36:30.940 --> 36:36.440 I mean, you know, it's like, I don't know where they are. 36:37.380 --> 36:45.820 I would first move to New Zealand because the women to men ratio is like, what is it, like three to one. 36:45.840 --> 36:46.560 Do you say New Zealand? 36:47.200 --> 36:47.680 Yes. 36:48.240 --> 36:50.500 You would have a good chance there. 36:50.500 --> 36:57.420 I mean, you can go to like certain computer stores. 36:59.000 --> 37:08.240 See, that's kind of creepy because like if I go to Borders or like Fry's and I'm just standing around, you know, 37:08.500 --> 37:12.400 it's like analogous to picking up women at the supermarket. 37:13.080 --> 37:14.040 Who does that? 37:14.480 --> 37:15.240 Who does that? 37:15.240 --> 37:22.420 I know, but like if you're going, you know, say to the Apple store to buy like a MacBook Pro or something. 37:23.400 --> 37:24.160 Or a MacBook. 37:24.160 --> 37:25.140 There's hot chicks. 37:25.380 --> 37:26.340 No, MacBook Pro. 37:27.460 --> 37:31.700 There's hot chicks there standing around. 37:31.820 --> 37:40.680 And you could like, you know, go over to the software section where they are and be like, oh, I was looking for that SimCity video game too. 37:42.900 --> 37:43.940 Okay, okay, okay. 37:43.940 --> 37:50.620 So now let's say I found her at, you know, your local friendly Apple store. 37:51.260 --> 38:00.680 Now, what do I do to like, but I don't want to say pick up mine because I think personally that's really stupid. 38:01.200 --> 38:06.680 But how do I start a conversation and like engage her without scaring her away? 38:07.700 --> 38:09.680 Because I'm not going to say like. 38:09.680 --> 38:11.420 Ask her how she likes her tech product. 38:11.420 --> 38:13.140 Be like, how do you like your MacBook Pro? 38:13.940 --> 38:17.640 Because then they'll be like, oh, I love you, I love you, this, this, blah, blah, blah, blah. 38:17.800 --> 38:19.260 And then just pretend like you're interested. 38:20.300 --> 38:20.540 Okay. 38:20.920 --> 38:27.000 Take me through the conversation from when I see her to when I end up with a date. 38:28.060 --> 38:28.560 And go. 38:28.560 --> 38:34.700 So, all right, so you're standing there and she's looking at Final Cut. 38:35.780 --> 38:36.140 Okay. 38:36.660 --> 38:37.880 She's looking at Final Cut. 38:38.260 --> 38:44.940 And you could say, oh, you know, I was really interested in buying that software, but, you know, I'm not sure how user friendly it is. 38:44.940 --> 38:53.360 And then you can go on with the conversation and she can tell you about it and you guys can discuss the different things that you do in Final Cut. 38:54.120 --> 38:57.960 And then you could always say at the end, you know, you're a really good source of information. 38:58.100 --> 38:59.920 Do you mind if I get like your email address? 39:00.040 --> 39:06.120 Because I'm going to buy this and I really need a good source of information to, you know, ask questions about. 39:06.120 --> 39:12.740 Okay, so, so, so the first thing is trick her into giving her your contact information. 39:13.220 --> 39:13.620 Okay. 39:13.860 --> 39:14.520 Okay, I got it. 39:14.580 --> 39:15.100 Well, no, no, no. 39:15.120 --> 39:18.420 The first thing is ask them a question and see how they talk. 39:18.500 --> 39:21.540 And see if they annoy the shit out of you, like the first minute they're talking. 39:22.220 --> 39:23.920 You just want to make an excuse and walk away. 39:24.600 --> 39:25.000 Exactly. 39:25.560 --> 39:26.760 Okay, okay, okay. 39:26.760 --> 39:36.360 See, because in movies, what happens is you see him go up and say, hi, how are you doing? 39:36.640 --> 39:38.080 And the next thing is there's sex. 39:38.580 --> 39:40.060 I don't know what goes on in the middle. 39:41.280 --> 39:48.020 I mean, that whole conversation period is undocumented. 39:48.800 --> 39:55.300 Well, you know, even though it's not documented, stuff that happens in the movies does not really happen. 39:56.760 --> 39:57.900 It's just not true. 39:57.900 --> 39:58.700 Are you sure? 39:59.440 --> 40:06.460 Okay, my husband did not come riding up to me on a white horse with roses to ask me to marry him. 40:07.600 --> 40:11.100 So I'm pretty sure that the stuff that happens in movies doesn't happen. 40:12.380 --> 40:22.620 All right, so you would say the, so basically get a conversation started and then express interest by, like, you know, 40:22.620 --> 40:29.500 asking her for her e-mail or her number to, like, talk about some mutual interest that you guys have. 40:29.880 --> 40:30.180 Yeah, something like that. 40:30.180 --> 40:31.100 Is that a good first step? 40:31.540 --> 40:34.820 Yeah, something like that definitely is, you know, non-threatening. 40:35.460 --> 40:37.680 And, you know, it's not like, oh, Final Cut Pro. 40:38.540 --> 40:39.620 Hey, that's really cool. 40:39.680 --> 40:40.340 How does that work? 40:40.420 --> 40:41.820 And then, can I have your e-mail address? 40:43.100 --> 40:43.500 Okay. 40:43.980 --> 40:46.520 So you have to have some conversation skills. 40:46.920 --> 40:47.240 Yes. 40:47.840 --> 40:50.080 Hmm, maybe I'll just stick with MySpace. 40:51.040 --> 40:52.760 Just keep asking them questions. 40:53.240 --> 40:54.100 That's all you have to do. 40:54.180 --> 40:58.460 Like, you just have to listen for keywords and ask them a question. 40:58.740 --> 41:02.340 If she's like, well, the filters on Final Cut are really cool, you can do this. 41:02.820 --> 41:04.320 What kind of filters are there? 41:05.580 --> 41:06.020 Okay. 41:06.020 --> 41:13.180 Okay, so I'll write a script to do that, and then we'll see how it works. 41:13.580 --> 41:13.960 Okay. 41:14.500 --> 41:14.860 Gotcha. 41:15.620 --> 41:15.920 Okay. 41:16.340 --> 41:18.800 I would help with this market research, but I can't. 41:20.120 --> 41:20.660 You can't. 41:20.860 --> 41:25.340 Tell Mrs. Droops that this is the educational purpose. 41:25.460 --> 41:26.840 It's for science. 41:27.220 --> 41:27.840 It's for science. 41:27.840 --> 41:30.500 She's going to be like, it's for your stupid computer friends, isn't it? 41:30.500 --> 41:37.400 All right, well, I think that's all the time we have for this segment right now. 41:38.080 --> 41:39.400 No, Alani, thank you for coming. 41:39.480 --> 41:41.200 I'm sure we'll be talking to you again. 41:41.660 --> 41:42.020 All right. 41:42.080 --> 41:43.160 Thanks for having me. 41:43.300 --> 41:44.080 Thank you very much. 41:45.300 --> 41:46.100 No problem. 41:46.360 --> 41:47.660 You guys enjoy the rest of your show. 41:48.400 --> 41:49.240 We will. 41:49.900 --> 41:53.480 Now we can talk, go into our technical topics for the evening. 41:53.920 --> 41:57.880 And as I said before, we're going to talk about machine learning and data marketing. 41:58.740 --> 42:04.520 So, Droops, have you heard about this machine learning, data mining thing? 42:05.200 --> 42:07.280 Only from what I've heard from you today. 42:09.320 --> 42:09.800 Okay. 42:10.100 --> 42:10.360 Okay. 42:10.480 --> 42:11.220 That works. 42:11.900 --> 42:18.040 Now, machine learning is pretty much artificial intelligence. 42:19.320 --> 42:22.720 Now, when you think of AI, what do you think of? 42:24.320 --> 42:25.620 That Will Smith movie. 42:27.460 --> 42:27.740 Okay. 42:28.780 --> 42:36.360 Now, artificial intelligence is actually much less awesome than that. 42:37.980 --> 42:40.460 Machine learning is a little option of AI. 42:40.920 --> 42:49.080 And it's pretty much all of the algorithms and all the theorems and all the stuff that people can actually use. 42:49.080 --> 42:53.700 So, if you think machine learning, think practical AI. 42:55.100 --> 42:59.500 Now, for machine learning, there are a lot of ways to do it. 43:00.300 --> 43:05.200 You know, there are what they call supervised learning methods and unsupervised learning methods. 43:05.200 --> 43:12.000 But all of that stuff basically is trying to do one thing. 43:12.000 --> 43:29.240 You have the assumption that all types of data, like all types of things, whether it is, you know, where Droops walks in his house throughout the day to, you know, who searches what on the Internet. 43:29.240 --> 43:34.600 All of those things can be modeled by a mathematical function. 43:35.160 --> 43:41.740 Now, that function can be really complex and, you know, some crazy shit that no one can solve. 43:41.740 --> 43:43.180 But it exists. 43:44.120 --> 43:51.600 So, what that means is, you know, you can predict what's going to happen if you know how something works. 43:51.600 --> 43:58.940 I mean, that's a really general idea, but, you know, I mean, it's always been there, like, with physics and all that stuff. 43:59.480 --> 44:01.260 You know, it's basically what science is about. 44:01.980 --> 44:05.160 So, let me get into some detail. 44:06.140 --> 44:12.920 Now, there are two main categories of machine learning algorithms, supervised and unsupervised. 44:13.280 --> 44:18.980 Now, you always start with a data set, like the old data set they gave you for research. 44:18.980 --> 44:33.200 Now, for supervised, what that means is while you're running the algorithm and getting, you know, your software to learn what that function is that you're trying to have your machine learn, 44:33.980 --> 44:41.000 you have outputs that you're, you know, trying to match against. 44:41.220 --> 44:44.000 So, you know what it should be. 44:44.000 --> 44:49.220 So, you know, it's like, think about it as a function with inputs and outputs. 44:49.640 --> 44:54.420 You have a data set that's representative of the correct one. 44:54.940 --> 45:05.240 Now, you take those and you're trying to predict future occurrences of that, of items from that data set, right? 45:05.680 --> 45:08.380 When supervised, you know what the output is. 45:08.380 --> 45:12.820 So, let's say, you know, you have a mystery function and you don't know what it is. 45:13.220 --> 45:17.000 So, you plug in one and three and then you get out 206, right? 45:17.660 --> 45:27.860 Having the answer 206 means it's a supervised, I mean, the data set can be run with a supervised learning algorithm. 45:28.860 --> 45:32.360 Now, if we're unsupervised, that's just random data. 45:32.500 --> 45:34.900 That's just random crap that's out there. 45:34.900 --> 45:40.220 You know, so something like that would be the web search. 45:40.680 --> 45:42.740 Like, there's really no output. 45:43.180 --> 45:47.960 Like, you know, you have to figure it out yourself or it just isn't there. 45:48.060 --> 45:50.020 Or, you know, it's not that kind of data. 45:50.660 --> 45:55.640 So, an example of supervised that's actually pretty pertinent to this is spam, right? 45:55.640 --> 46:06.300 So, you have all these attributes for a spam email, like the link and, you know, like what's in the header and what kind of words are in the message, who it's from, et cetera, et cetera. 46:06.920 --> 46:10.340 And your output is whether it's a spam email or not. 46:10.880 --> 46:17.840 So, you know, there's some classification that you already know, but not everything is like that. 46:17.840 --> 46:32.620 Now, I'm going to go over four prominent algorithms, and those are agentry, key nearest neighbors, artificial neural nets, and clustering. 46:33.080 --> 46:39.400 Now, the first three are supervised algorithms, and clustering is the only unsupervised algorithm I'm going to go over. 46:39.400 --> 46:49.100 Now, at the end of the show, well, actually, in the show notes, I'm going to put in a link to a framework. 46:49.280 --> 46:56.260 Well, it's really a software package called WECA, W-E-K-A, and that has all this stuff implemented. 46:56.840 --> 47:07.320 So, what you can do is go and just run these algorithms on whatever data set you want, and then it'll spit back results, and all you have to know is how to interpret it. 47:07.320 --> 47:13.980 So, stuff I'm talking about now won't be a complete waste, and you don't have to implement it yourself. 47:15.200 --> 47:17.500 Now, okay, so let's start with decision trees. 47:18.120 --> 47:22.520 Now, decision trees is one of the simplest, simplest things you can do. 47:24.560 --> 47:27.120 Okay, actually, let's talk about the data set a little bit. 47:27.120 --> 47:39.520 So, you have – so, if you looked at the AOL data set, for example, they say, okay, each line, it's either a, you know, a query that someone did. 47:39.640 --> 47:42.900 So, you'll have the UID, and then I'll have the query string. 47:42.900 --> 47:52.420 And then another one is UID, query string, and some other variables. 47:52.780 --> 48:00.480 Now, those variables mean that they actually clicked on a result that was in, you know, what they queried for. 48:01.060 --> 48:03.960 So, there are additional attributes for that. 48:04.640 --> 48:09.900 Now, each attribute is one dimension of the data set. 48:09.900 --> 48:17.300 Now, one of the most important things is to figure out which attributes you want and which attributes you're going to use. 48:17.940 --> 48:24.480 Now, you may get a data set, and you don't want to use every single attribute because some of them maybe, you know, don't care, 48:25.020 --> 48:29.620 or they're not pertinent to what you're doing, or you may not have all the values for it. 48:29.620 --> 48:33.540 Now, there are techniques to deal with that, but you don't have to worry about that right now. 48:33.540 --> 48:42.360 So, once you have your data set, and since we're doing a supervised learning algorithm, you need to have an output. 48:43.420 --> 48:51.620 So, I don't, you know, so the AOL data set probably wouldn't work too well unless you categorize them yourself first. 48:53.380 --> 48:58.900 And what we're trying to do here is get, build like a software model. 48:58.900 --> 49:03.720 I mean, you know, you can even say some sort of, like, brain, right? 49:04.460 --> 49:06.420 What that can do is you're going to train it. 49:06.760 --> 49:10.320 So, you're going to take the data set and feed it into this algorithm, 49:10.680 --> 49:17.480 and it's going to give you a configuration for some kind of, you know, brain. 49:17.680 --> 49:22.340 I have no better word for it, but it'll probably come later to me. 49:22.340 --> 49:28.080 But you have some configuration, then what you want to do is predict future values. 49:28.080 --> 49:35.880 So, in the context of spam, you train your whatever learning algorithm you have with all the spam you have, right? 49:36.220 --> 49:43.360 And then, hopefully, the new spam will be close enough in certain characteristics with the old spam. 49:43.620 --> 49:48.720 It can recognize it as a spam message, and a real email is not a spam message. 49:49.600 --> 49:51.500 So, decision tree. 49:52.140 --> 49:54.220 What happens is this. 49:54.500 --> 49:57.540 You have a bunch of inputs. 49:57.540 --> 50:00.360 So, let's say you have an input that has a bunch of attributes. 50:01.280 --> 50:06.580 So, each attribute is going to end up being a split in the tree. 50:07.060 --> 50:12.820 So, you create this tree where you say, like, where you end up following it down. 50:13.040 --> 50:20.580 So, you say, okay, let's say I have, you know, a data set where each input has ten attributes, 50:20.580 --> 50:25.760 and the attributes are, you know, just some simple greater than or less than thing, right? 50:25.760 --> 50:36.800 So, let's say I put it in the input, and the first decision I make is, okay, is attribute A less than or greater than ten? 50:36.800 --> 50:45.280 And if it is less than ten, then I go one way, and if it's greater than ten, then I go the other way in the decision tree. 50:45.280 --> 50:52.740 Now, you keep doing that, and then you end up using all your ten attributes. 50:53.500 --> 50:59.500 And at the very end of the attribute tree, you know, so you follow some path, right? 50:59.640 --> 51:02.160 And at the very end, you come up with a classification. 51:02.160 --> 51:14.240 You say, oh, okay, I'm at the end, and from what I, you know, from the decision tree, I classify this as, you know, something or something else, 51:14.280 --> 51:15.540 and you defined that before. 51:16.580 --> 51:23.780 So, what you do is you end up training your decision tree on all of these example data, 51:24.360 --> 51:30.060 and then you kind of get some sort of notion of how well you do, right, because you're not always right. 51:30.060 --> 51:36.760 So, since you, I mean, you know, you would be always right if you had all the data possible from that function that you're trying to deduce, 51:37.020 --> 51:39.860 but you don't, so you're not going to be 100% correct. 51:40.220 --> 51:44.220 So, at some point, you're going to be wrong, so that's going to give you, like, a percentage of how well you're doing. 51:45.000 --> 51:49.260 So, that is what a decision tree does. 51:50.000 --> 51:52.980 Now, the interesting thing is how you make one. 51:52.980 --> 51:57.820 Because when you start, right, you don't know what your decision tree is. 51:57.820 --> 52:02.000 Like, when you take attribute A and say, you know, is A greater than 10? 52:02.080 --> 52:09.800 Like, you don't know how to do that, because A is just some value that all the input has, and, you know, that's it. 52:10.180 --> 52:15.580 So, you have to figure out what to make the decision tree. 52:15.820 --> 52:21.660 And the most important thing to do that is a concept called information gain. 52:21.660 --> 52:25.200 Now, so picture it this way. 52:25.300 --> 52:31.320 You have all your data, and now you have, let's say, let's take attribute A, for example. 52:32.320 --> 52:40.820 If you put in a decision that says, okay, let's split all the data, the input data, 52:41.500 --> 52:45.520 on if attribute A is less than 10 or greater than 10. 52:45.520 --> 52:51.280 Now, you do that, and then you figure out how that splits the data. 52:52.280 --> 52:58.200 Now, if it only splits out one input out of, you know, like, the thousands you have, 52:58.480 --> 53:02.320 then that's not very good at, you know, sorting out the info you have, right? 53:02.660 --> 53:10.960 So, what you want to do is find attributes and, you know, those kind of decision splitting points 53:10.960 --> 53:13.600 that's going to give you the most split. 53:13.600 --> 53:19.040 Now, for example, one of the best things to do, you know, at least in the beginning, 53:19.280 --> 53:23.540 is find one that is going to give you an even split. 53:23.940 --> 53:28.980 So, that's going to kind of narrow your chances the most. 53:30.260 --> 53:35.680 Now, so you do this in a greedy fashion, and so what that basically means is, okay, 53:35.840 --> 53:38.740 you used up attribute A, you split the data set in half. 53:38.740 --> 53:45.880 Now, the next decision you make is you do it with, you know, attribute B, right, the next one, 53:46.020 --> 53:51.080 and then you keep doing this until you end up with a whole tree that uses all your attributes. 53:51.080 --> 54:00.700 And since you have the results from the input, at the very end, you know, you categorize all your data, 54:01.020 --> 54:05.780 and you're going to get clumps of them, right, because you have, like, thousands and thousands of input, 54:05.980 --> 54:09.060 and you only have, say, like, a fraction of attributes. 54:09.500 --> 54:16.080 So, when you get down to the end of the tree, you're going to end up having, like, 200 things that are categorized by, 54:16.080 --> 54:19.880 that have the same attributes, or, you know, are categorized by the same attributes. 54:20.900 --> 54:29.420 So, when you get to the very end, you're going to look at the results or, you know, the output is for all those sample data. 54:29.800 --> 54:34.300 And, you know, they're not all going to agree, but you take the ones that's the majority. 54:34.300 --> 54:43.100 And then you say, okay, if I see an input in the future that takes me down this tree, down to that point, 54:43.620 --> 54:49.660 then I'm going to say the output, I'm going to predict the output is going to be, you know, 54:50.020 --> 54:55.580 this thing where it's the majority of the output that I've seen in the past. 54:55.580 --> 55:06.740 And then once you have that tree, then you can start, you know, making decisions and predicting output for future input. 55:07.540 --> 55:13.580 The next supervised algorithm that I'm going to talk about is something called K-nearest neighbor. 55:14.520 --> 55:23.020 Now, all this is, is you plot your, okay, so forget about the tree, you know, forget about, you know, 55:23.020 --> 55:26.580 all that crap right now, take your data set. 55:27.260 --> 55:30.660 Now, it has, let's say, the same 10 attributes, right? 55:31.620 --> 55:37.360 Then what you want to do now is to make a 10-dimensional graph. 55:37.920 --> 55:43.160 Just, just visual, not, you know, not visually, but plot, you know, 55:43.200 --> 55:46.960 pretend like you're plotting all the inputs on this 10-dimensional graph. 55:47.340 --> 55:53.000 So, you have this 10-dimensional space and all your data points are in there, right? 55:53.020 --> 55:54.500 And they're just floating around. 55:55.100 --> 55:58.180 So, so it's just this big blob of shit. 55:59.180 --> 56:03.560 Now, what you want to do is find things that are alike, you know? 56:03.620 --> 56:10.000 Like, that is kind of the, the, the overall strategy of most of these things. 56:10.120 --> 56:17.660 You want to find things that are alike so you know what, you know, some future input is going to be. 56:17.660 --> 56:30.020 So, so it's called K-nearest neighbors because you kind of group things and you make these little groups of K dots, right? 56:30.780 --> 56:35.680 Now, let's say you have 2,000 input points, right? 56:35.680 --> 56:38.900 Now, you get to choose what K is. 56:39.180 --> 56:42.300 So, what that means is, let's say I make K 10. 56:42.780 --> 56:48.780 So, from, you know, 1,000 of these points, I will make 10 groups. 56:49.440 --> 56:57.080 Now, you know, you can do K to whatever you want, but the point, the key is to find something interesting, 56:57.080 --> 57:05.500 like an interesting K, so it actually, you know, puts a circle around the dots that, that makes sense. 57:05.740 --> 57:12.880 Because ultimately, you want to say, okay, this group right here that I drew a circle around means, you know, 57:13.340 --> 57:18.360 it's people searching for grandmothers who want to have sex. 57:18.360 --> 57:25.620 And, you know, this other group are people who want to learn how to swallow swords, you know, or whatever. 57:26.460 --> 57:29.260 See, like, let's say you make too big. 57:29.420 --> 57:30.980 Let's say you make K 2,000. 57:31.420 --> 57:34.160 So, every single dot is its own group. 57:34.620 --> 57:35.780 That doesn't really tell you that. 57:36.560 --> 57:39.740 Now, we can make K something ridiculous, like 2,000. 57:40.320 --> 57:44.040 Then, then each, each input would be its own group. 57:44.040 --> 57:48.480 Now, you can say something meaningful about each one of those groups, 57:48.620 --> 57:53.520 but that's not going to really help you classify future data into, you know, 57:53.640 --> 57:58.180 into people like, who like to swallow swords or people who want to have sex with grandmothers. 57:58.960 --> 58:03.440 Now, so, you know, because not every single person is going to end up being the same. 58:03.520 --> 58:05.280 So, the key is to find a K that works. 58:06.100 --> 58:10.520 Now, what you're going to do is not worry about it, 58:10.520 --> 58:13.820 because the algorithm will pick K for you. 58:14.700 --> 58:21.980 See, because you end up kind of defining a distance that, you know, some distance where, okay, 58:22.040 --> 58:28.440 if two points are sufficiently far apart, then we don't want to put them in the same cluster. 58:28.820 --> 58:32.420 So, we'll, you know, put K to be like K plus one or something. 58:32.760 --> 58:36.140 So, there are ways to find K, but I'm not going to go into them here. 58:36.140 --> 58:43.980 Now, to illustrate how this works, there are several methods for doing K nearest neighbor. 58:44.380 --> 58:48.040 But, for example, this is one method to do it. 58:48.620 --> 58:52.680 You start with K being, let's say, 2,000. 58:52.800 --> 58:54.020 So, everyone's in their own group. 58:54.020 --> 59:02.720 Now, what you do is take two of the points that are the closest, then you group them together, 59:03.120 --> 59:05.580 and then make K one less. 59:06.460 --> 59:12.740 So, you know, if two points are really, really close, then you say, okay, those people are essentially the same. 59:12.740 --> 59:18.860 So, you know, at some point, we can extract some meaningful data out of that group. 59:19.200 --> 59:25.600 Then you keep doing that until, you know, so you can keep doing that until there's only one group. 59:26.440 --> 59:29.340 But what you end up with is like this tree again. 59:29.700 --> 59:34.900 Now, it's not like the decision tree, but it's a tree of kind of the order you group people in. 59:34.900 --> 59:43.180 So, you know, so this is kind of like a Deitcher's algorithm thing where you take two points that are the closest, 59:43.520 --> 59:48.660 and then you say, okay, these are one group, and then the next closest. 59:48.860 --> 59:55.200 Now, they might be two separate points completely, but you don't really care because eventually they're going to hook up with something else. 59:56.040 --> 01:00:00.680 So, when you make this tree of, you know, so you start grouping things together, 01:00:00.680 --> 01:00:08.900 you have this tree, and you can look yourself at like, okay, let me look at when K is equal to 7, you know. 01:00:09.100 --> 01:00:16.740 Like, what does that graph look like, you know, when you put the dots all in, you know, your 10-dimensional graph? 01:00:17.100 --> 01:00:18.480 You know, like, how does that look? 01:00:18.560 --> 01:00:22.580 Like, what if I go to K equals 6? 01:00:23.100 --> 01:00:27.140 Then am I going to put two really large groups together? 01:00:27.140 --> 01:00:33.860 Because if I am, then maybe those two groups have something interesting that, you know, 01:00:34.060 --> 01:00:39.440 that there's, you know, have some individuality that I can extract from it. 01:00:39.480 --> 01:00:41.180 So, maybe I don't want to put them together. 01:00:42.320 --> 01:00:49.720 Now, so you see what I mean when I say that K is kind of dynamic, depending on what you're trying to find out? 01:00:49.720 --> 01:00:55.940 So, you do that, and then when you get to the K you want, you end up with, okay, 01:00:56.040 --> 01:00:58.680 so let's say what makes sense for me is K equals 7. 01:00:59.220 --> 01:01:00.460 So, I have seven groups. 01:01:00.900 --> 01:01:02.740 So, I've grouped my data into seven things. 01:01:03.300 --> 01:01:08.000 Now, then I have to go and say, okay, what do these things have in common? 01:01:08.920 --> 01:01:11.260 But the hard part is done for you. 01:01:11.360 --> 01:01:13.060 You know they have something in common. 01:01:13.700 --> 01:01:16.760 I mean, you know, so basically it's trying to tell you something, 01:01:17.140 --> 01:01:19.360 and you need to figure out what it is. 01:01:19.360 --> 01:01:23.080 That becomes your job after you run the algorithm, you know. 01:01:23.260 --> 01:01:28.500 You look at the data, and you're like, oh, okay, these people look like they're into, you know, 01:01:29.300 --> 01:01:33.560 interspecies erotica, so maybe that's what this is telling me, you know. 01:01:34.040 --> 01:01:37.100 Like, all these people are into that, and that's cool, 01:01:37.940 --> 01:01:43.660 but they're different from this other group of people who might be into whatever else. 01:01:43.660 --> 01:01:52.720 So, that is a supervised thing, because what happens is, you know, so this is supervised for them. 01:01:52.800 --> 01:01:59.740 So, you end up having kind of the, not the answer, but you have the output. 01:01:59.740 --> 01:02:07.740 So, you know what, you know, these clusters of things are. 01:02:07.840 --> 01:02:11.980 Well, actually, cluster is not the right word, but you know what this group of dots mean. 01:02:12.400 --> 01:02:16.500 Because, you know, you, once again, you can do what you do in the decision tree, 01:02:16.500 --> 01:02:24.800 and you can say, like, okay, all, you know, out of these 367 groups, you know, of input, 01:02:26.120 --> 01:02:30.020 you know, like 300 of them have, you know, this for an output. 01:02:30.320 --> 01:02:39.540 So, we can safely say that, you know, if at some other point we get a dot that falls into that area, 01:02:39.540 --> 01:02:46.120 the output will most likely be whatever the output is for those 300 people or inputs that we have. 01:02:47.440 --> 01:02:52.080 So, you know, that's what, you know, that's kind of the same way you would use it. 01:02:52.200 --> 01:02:56.520 So, you know, let's run this, and you figure out what each of those groups mean. 01:02:56.660 --> 01:03:02.720 You can save that, and if you later get more data points from that same source, 01:03:03.220 --> 01:03:08.660 then, you know, then you can know where they belong, and you don't, you know, like, 01:03:08.660 --> 01:03:11.840 if, like, let's say, once again, you're predicting spam, right? 01:03:12.200 --> 01:03:18.180 Then, if you get an email in, and that dot falls into that group for whatever attribute, 01:03:18.700 --> 01:03:25.020 then you know that it's probably spam or not spam based on the, you know, 500 dots surrounding it. 01:03:25.700 --> 01:03:28.040 So, that's how you would use something like that. 01:03:29.540 --> 01:03:36.620 Now, the next supervised and the last supervised thing I'm going to talk about is artificial neural nets. 01:03:36.620 --> 01:03:44.100 Now, you might be fairly, you know, if you haven't heard of anything I've talked about yet, 01:03:44.300 --> 01:03:47.400 you probably may have heard about this. 01:03:47.900 --> 01:03:55.800 Now, neural nets is, you know, was invented in, what, like, the freaking 70s, probably? 01:03:56.400 --> 01:03:58.340 And, you know, it's been around forever. 01:03:58.340 --> 01:04:05.680 And the premise is you were trying to model what the brain does. 01:04:06.760 --> 01:04:12.500 So, you have, you know, so back, you know, when they were doing this, the biologists had figured out, like, 01:04:12.560 --> 01:04:18.820 okay, you know, you have these neurons in your brain, and you have inputs and outputs. 01:04:18.820 --> 01:04:25.020 Now, what happens is you get electrical signals, and they, you know, they go in the inputs, 01:04:25.500 --> 01:04:32.860 and then depending on how much of a voltage you get, like, how strong this signal is from your various inputs, 01:04:33.380 --> 01:04:36.260 then your output will fire, or it will not. 01:04:37.440 --> 01:04:42.720 So, you have one of these things, and then you hook them up, and your entire brain is made of this. 01:04:42.720 --> 01:04:47.120 And then, you know, so you have inputs, connect, or outputs, so everything is networked together, 01:04:47.500 --> 01:04:56.000 and that somehow, you know, leads to an eventual output that says, you know, like, do this, or, you know, don't do that. 01:04:56.820 --> 01:05:03.500 So, the AI, you know, the computer scientists and the AI people wanted to kind of utilize that, 01:05:03.840 --> 01:05:11.960 because we want to, you know, the goal, I guess, is to make, you know, an actual intelligence that can make decisions for you. 01:05:12.720 --> 01:05:19.140 But, as it turns out, 20, 30 years later, the brain is much more complicated than that. 01:05:19.720 --> 01:05:23.820 But, we still manage to find ways to use it. 01:05:24.760 --> 01:05:30.300 Now, the way a neural net works is you have this concept of a perceptron. 01:05:30.860 --> 01:05:40.200 A perceptron is exactly the same thing, or logically the same thing, as what they thought a neuron was. 01:05:40.200 --> 01:05:45.920 So, you end up having these inputs in your perceptron. 01:05:46.500 --> 01:05:48.260 Let's say you have three inputs. 01:05:49.000 --> 01:05:55.060 Now, the input can either be, well, it can be anything between zero and one. 01:05:55.980 --> 01:05:58.260 So, any real number between zero and one. 01:05:59.000 --> 01:06:00.360 Now, you have an output. 01:06:00.840 --> 01:06:03.620 The output can only be zero or one. 01:06:03.860 --> 01:06:06.700 So, a perceptron either fires or it doesn't. 01:06:06.700 --> 01:06:13.060 Now, inside the perceptron, not inside, but the perceptron has this notion of a threshold. 01:06:14.060 --> 01:06:20.820 So, let's say my threshold is, I don't know, two and a half. 01:06:20.820 --> 01:06:31.700 So, if every input I have, you add up the input values, they're from zero to one. 01:06:32.360 --> 01:06:41.060 And, if that value is greater than your threshold, which is two and a half, then your perceptron will fire, and it'll send out a one. 01:06:41.060 --> 01:06:43.560 So, that's the basic building block. 01:06:44.060 --> 01:06:56.580 Now, you hook all these things up into some sort of network, and you have this, you know, this is where neural nets come from, like the terminology. 01:06:57.040 --> 01:07:01.620 So, you have this thing, and now, we go back to the inputs that you have. 01:07:01.620 --> 01:07:04.280 You have 10 attributes, okay. 01:07:04.800 --> 01:07:13.120 So, we have 10 perceptron, and you hook them up in some fashion, and then you get one output. 01:07:13.760 --> 01:07:24.180 So, going back to our example with spam, you get, you know, so you get all these attributes from the email, 01:07:24.800 --> 01:07:29.000 and it'll have one output, and either say it's spam or not. 01:07:29.000 --> 01:07:37.220 So, I'm going to skip, kind of, you know, how you determine the middle right now, and I'll go back to that. 01:07:37.780 --> 01:07:42.860 But, before I do, so let's say you, you know, so, think back to the decision tree. 01:07:43.560 --> 01:07:54.500 Let's say you have an attribute A, and, you know, you say, okay, if attribute A is less than 10, then put in a zero. 01:07:54.920 --> 01:07:56.920 Or, if it's greater than 10, put in a one. 01:07:56.920 --> 01:08:04.520 Now, that's good and all, but you also get a range of values as the input. 01:08:05.040 --> 01:08:11.840 So, what you can do is normalize your attribute into, from zero to one. 01:08:12.180 --> 01:08:17.720 So, let's say if I have a range for my attribute that's like, I don't know, 20. 01:08:17.720 --> 01:08:27.080 Then, you know, then I can normalize it to zero and one, and based on every single value of attribute A, 01:08:27.580 --> 01:08:33.400 I can put into the first perceptron some value between zero and one. 01:08:33.400 --> 01:08:39.360 So, you know, the closer you're at 20, the higher, the closer you're going to be at one for the input. 01:08:39.940 --> 01:08:47.960 So, you do that for all the attributes, and you end up with, you know, some perceptrons firing and some not firing. 01:08:48.700 --> 01:08:53.900 So, the natural question, then, is, say, okay, how do you set thresholds? 01:08:53.900 --> 01:08:55.560 Because you don't know what to put them. 01:08:56.920 --> 01:09:08.600 So, there's an algorithm for neural nets that lets you train a neural net that kind of starts at a default factory setting state. 01:09:08.600 --> 01:09:15.640 So, it starts at some, you know, stupid thing where, like, all the thresholds are, like, one or something, 01:09:16.120 --> 01:09:20.220 and then you keep putting data in and you train it. 01:09:20.820 --> 01:09:25.800 And in your neural net, you can, most times, you can have a feedback loop. 01:09:26.580 --> 01:09:35.760 So, what happens is, since in your output, you know, in your input set, you know what the output is for that training data, right? 01:09:35.760 --> 01:09:39.760 So, what you can do is put one in, see what happens. 01:09:40.360 --> 01:09:45.840 You know, if, say, the end result fires, then, you know, if the end result is a one, 01:09:46.180 --> 01:09:51.940 then you can look at what the output actually is for your training data. 01:09:52.500 --> 01:09:56.980 And then, if it's right, then you say, okay, guys, good job, you know, moving on. 01:09:57.260 --> 01:10:01.540 But if it's wrong, then you say, okay, why the hell is this wrong? 01:10:01.540 --> 01:10:06.620 So, you have a feedback loop that goes back to the beginning and says, okay, you guys are wrong. 01:10:07.100 --> 01:10:13.000 So, someone screwed up, and someone's threshold is not correct, so you need to change it. 01:10:13.260 --> 01:10:16.340 And we'll see, you know, how well you do. 01:10:17.960 --> 01:10:20.260 So, I'm not going to go into details on how that works, 01:10:20.620 --> 01:10:25.240 but there's a process whereby you put in your input one after one, 01:10:25.480 --> 01:10:29.200 and then you keep adjusting the thresholds based on what you see. 01:10:29.200 --> 01:10:36.260 So, this is kind of, you know, like a steady, I guess what they call a gradient descent 01:10:36.260 --> 01:10:40.260 towards the approximation function that you want. 01:10:41.180 --> 01:10:49.080 So, you little by little get to the place where, you know, it would categorize each thing correctly. 01:10:49.580 --> 01:10:55.640 So, this works totally, completely differently from the decision tree and the k-nearest neighbors algorithm. 01:10:55.640 --> 01:11:00.620 Now, this is, like, because the decision tree and the k-nearest neighbor kind of make sense 01:11:00.620 --> 01:11:04.640 and kind of relate to each other in that you're, like, grouping how close things are, 01:11:04.960 --> 01:11:08.380 and it's kind of like a discrete value sort of thing. 01:11:09.240 --> 01:11:12.840 But for artificial neural nets, it's a totally different thing. 01:11:13.600 --> 01:11:18.140 So, you're training your neural net, and at the end of that, you end up with this network 01:11:18.140 --> 01:11:25.740 that has different weights in all of the nodes or the perceptron. 01:11:26.480 --> 01:11:34.780 Then you can use that, like, you use the other two methods to try to classify new things that come in. 01:11:36.100 --> 01:11:44.340 Now, before, I didn't talk about how you kind of construct that network, right? 01:11:44.340 --> 01:11:51.720 Because you know you need to have, say, ten inputs, because that's, ten inputs into your neural net, 01:11:51.980 --> 01:11:53.980 because that's the number of attributes you have. 01:11:54.460 --> 01:12:01.000 But, and you know you need one output for the actual classification of whatever you put in the neural net. 01:12:01.380 --> 01:12:04.240 But all that stuff in the middle, it's kind of vague. 01:12:04.640 --> 01:12:10.680 The reason that it's vague is because no one really knows what's the best thing to put. 01:12:10.680 --> 01:12:13.700 So, you can do anything you want. 01:12:14.180 --> 01:12:18.560 You can have, you know, you can have 70 layers of neural nets, 01:12:18.640 --> 01:12:20.840 and you can just keep hooking them up to each other, 01:12:21.240 --> 01:12:25.940 and, you know, just have a full mesh of layers and layers of perceptrons. 01:12:26.060 --> 01:12:30.620 And each time you go through the training data, you know, those weights will change. 01:12:31.580 --> 01:12:36.580 Now, or you can have just one row where you have a perceptron for each input, 01:12:36.580 --> 01:12:41.000 and all of them are just going to hook up to one perceptron for the output. 01:12:41.080 --> 01:12:41.700 You can do that. 01:12:42.460 --> 01:12:48.400 Now, depending on what you do in the network, your outcome is going to be different. 01:12:48.500 --> 01:12:52.380 Like, your neural net is going to be, you know, it's going to perform differently. 01:12:52.380 --> 01:12:58.920 So, what you have to do is you have to tell it how many layers to have 01:12:58.920 --> 01:13:03.940 and also how many, how wide each layer is, 01:13:03.980 --> 01:13:10.020 like how much input aggregation you do as you add each new layer, 01:13:10.140 --> 01:13:13.460 because you eventually have to get from the number of attributes down to one. 01:13:13.460 --> 01:13:20.440 Now, what I have found works the best is to have two layers. 01:13:20.920 --> 01:13:23.720 Now, sometimes, not always, so you have to try this out. 01:13:24.460 --> 01:13:28.260 If, you know, sometimes for something, having one layer works really well. 01:13:28.540 --> 01:13:32.820 And if you add a little layer to it, either it doesn't do anything or it fucks out completely. 01:13:33.280 --> 01:13:35.000 And at other times, it's vice versa. 01:13:35.000 --> 01:13:44.380 But in my experience, one or two layers is enough to capture the complexity of a function. 01:13:45.080 --> 01:13:48.020 Because that's, I mean, that's what adding layers means. 01:13:48.340 --> 01:13:55.660 You know, you add a level of indirection so you have room to tweak any intermediate values that you want. 01:13:55.820 --> 01:13:58.860 So it's not like, oh, if this is on, then that's on, right? 01:13:58.860 --> 01:14:06.620 So you let, you allow for the possibility of other, you know, of things affecting each other 01:14:06.620 --> 01:14:09.120 because they're going to be in a full mesh, right? 01:14:10.460 --> 01:14:15.880 So, you know, so what I found out is try one or try two layers with the software. 01:14:16.320 --> 01:14:20.980 But if neither of them works, then neural nets are probably not the best thing to classify that data set. 01:14:23.760 --> 01:14:24.320 Pause. 01:14:25.320 --> 01:14:26.280 What's up, Droofs? 01:14:26.280 --> 01:14:26.680 You're typing. 01:14:26.680 --> 01:14:30.560 No, that was low-tech coming through the door. 01:14:31.700 --> 01:14:32.100 Sorry. 01:14:34.140 --> 01:14:34.540 Hello? 01:14:34.820 --> 01:14:35.060 Sorry? 01:14:35.320 --> 01:14:35.640 Hey. 01:14:36.540 --> 01:14:37.380 Hey, what's up? 01:14:37.740 --> 01:14:39.440 That was low-tech coming through the door. 01:14:40.380 --> 01:14:41.300 Oh, cool, cool, cool. 01:14:41.840 --> 01:14:44.260 Okay, so I just finished talking on neural nets. 01:14:45.900 --> 01:14:50.380 I'm going to type with, like, the last thing and then close it. 01:14:52.820 --> 01:14:53.260 Okay. 01:14:54.120 --> 01:14:54.520 Okay? 01:14:55.200 --> 01:14:55.840 Sounds good. 01:14:56.680 --> 01:14:57.200 Okay. 01:14:57.580 --> 01:14:58.700 So still no questions? 01:14:59.760 --> 01:15:01.760 No, I haven't come up with any questions yet. 01:15:01.760 --> 01:15:02.140 I'm sorry. 01:15:02.820 --> 01:15:03.520 That's all right. 01:15:03.600 --> 01:15:04.640 I'll just keep talking. 01:15:05.420 --> 01:15:06.460 All right, start. 01:15:07.760 --> 01:15:11.500 You know, I've mentioned three supervised algorithms. 01:15:11.500 --> 01:15:20.980 Now, those tend to be a little more tame when you're doing machine learning because you know, you know, you know what the output is. 01:15:21.020 --> 01:15:22.480 So you have a set to train on. 01:15:23.060 --> 01:15:27.840 You know, that's the most important thing because all of these things have, like, configuration parameters, 01:15:27.840 --> 01:15:40.140 whether it is deciding which decisions to have in the branch or, you know, how to group your input data or, you know, what to make your thresholds for your neural net. 01:15:40.140 --> 01:15:43.760 You know, all of that requires some sort of training. 01:15:43.760 --> 01:15:56.280 And ultimately, you know, you're never – sometimes you'll get a data set that has no clear, you know, has no clear output. 01:15:56.280 --> 01:15:59.200 So it's just going to be, like, data of people doing stuff. 01:15:59.600 --> 01:16:02.460 And you're not going to know, you know, what it means. 01:16:02.780 --> 01:16:12.640 So, you know, let's say I take, like, traffic patterns, you know, or like how people, you know, call each other and all that stuff. 01:16:12.640 --> 01:16:18.440 Like, you're not – you don't know, you know, what the end goal is. 01:16:18.820 --> 01:16:25.200 You just know that you have this massive collection of data and you know that, you know, 01:16:25.260 --> 01:16:34.080 there's probably some sort of interesting information in there and you don't want to just rep it and find weird shit. 01:16:35.020 --> 01:16:38.500 So you need a way to extract that information. 01:16:38.500 --> 01:16:46.180 And there are a bunch of unsupervised learning algorithms, machine learning algorithms. 01:16:46.640 --> 01:16:51.640 But this one, I think, is going to be the easiest to kind of grasp. 01:16:51.780 --> 01:16:54.280 And I think it works the best, actually. 01:16:55.320 --> 01:16:57.260 So this is called clustering. 01:16:58.180 --> 01:17:02.900 Now, clustering is going to sound very similar to King Nearest Neighbor, 01:17:03.800 --> 01:17:07.760 but it's slightly different in the mathematical sense. 01:17:07.760 --> 01:17:09.600 But you don't care about that. 01:17:10.200 --> 01:17:14.080 What's going to happen is – I mean, this is the easiest thing to use. 01:17:14.520 --> 01:17:16.500 You take the software. 01:17:17.260 --> 01:17:22.040 Now, you feed it the data, and then you run a clustering algorithm on it. 01:17:22.120 --> 01:17:23.360 Now, there's a bunch of them. 01:17:23.740 --> 01:17:27.680 But what it basically does is what King Nearest Neighbor does. 01:17:27.680 --> 01:17:37.060 It looks at your data set, and it tells you what circles to draw around which dots. 01:17:37.560 --> 01:17:43.720 So it's going to cluster all the data, and then it's going to say, like, hey, I found, you know, 01:17:44.200 --> 01:17:46.920 like one or two clusters of interesting things. 01:17:47.000 --> 01:17:49.540 Like, these things over here are the same somehow. 01:17:49.540 --> 01:17:52.100 And those things over there are the same somehow. 01:17:53.020 --> 01:17:57.120 And we don't really know why, but that's what you need to figure out. 01:17:58.020 --> 01:18:07.740 Now, when you have something like this, the hardest part is to, you know, figure out – to interpret the results. 01:18:07.740 --> 01:18:16.400 But you can do that by looking at the data and, you know, doing whatever to it. 01:18:18.240 --> 01:18:29.420 So you don't have – okay, so when I was mentioning a function, like you have an actual function that you're trying to get, you know, 01:18:29.540 --> 01:18:30.740 trying to model, right? 01:18:31.380 --> 01:18:35.640 There's no notion of that in an unsupervised learning environment. 01:18:35.640 --> 01:18:38.400 But you end up having, like, a generator. 01:18:38.600 --> 01:18:44.700 Like, you have the concept of a generator that generates the actual model, right? 01:18:45.440 --> 01:18:49.900 And then you have sample data that's not linked. 01:18:50.720 --> 01:18:56.400 So you don't know really what you're learning, and you don't really know what your goal is. 01:18:57.020 --> 01:19:03.560 But the good thing is, clustering is going to tell you what you need to know. 01:19:03.560 --> 01:19:08.280 Now, there are a lot of ways to cluster. 01:19:08.800 --> 01:19:16.820 And I guess for every single way that you cluster, you can apply it to the K-Nurse and everything. 01:19:17.260 --> 01:19:26.340 So you can do, like, the mean distance between clusters, and you can, you know, calculate things like the minimum distance between clusters, 01:19:26.340 --> 01:19:31.960 clusters, and the mean internal distancing cluster, and all this stuff. 01:19:32.700 --> 01:19:36.940 And what you end up with is just groups of things. 01:19:38.060 --> 01:19:44.880 So you can tell a lot by grouping things, to my surprise when I first learned about this. 01:19:44.880 --> 01:19:51.180 So what you're going to end up having to do is, you know, okay, so once you have this data, 01:19:51.240 --> 01:19:56.480 you can find things like, you know, where noise is and where an interesting segment of data is. 01:19:56.680 --> 01:20:02.600 So you don't have to waste your time looking, you know, combing through gigs and gigs of stuff. 01:20:02.600 --> 01:20:11.660 And you can just concentrate on the meaningful centers of, you know, the nuggets of interesting things. 01:20:11.980 --> 01:20:14.640 And you can find out how they relate to each other. 01:20:15.320 --> 01:20:21.160 Now, a lot of this stuff is, you know, so I'm not talking, I'm not really talking about how the clustering works. 01:20:21.160 --> 01:20:31.420 But once you get it, get the clusters, you know, you need to find a lot of visual way to represent the data that, you know, that you have isolated. 01:20:32.320 --> 01:20:39.940 And you have to, you know, kind of like, from then on, it's kind of like create a process of trying to figure out, you know, 01:20:40.060 --> 01:20:46.720 what is important from, you know, whatever the algorithm tells you. 01:20:46.720 --> 01:20:55.000 So those are the predominant machine learning algorithms. 01:20:55.360 --> 01:20:57.540 And I really suggest you guys go. 01:20:57.920 --> 01:21:00.220 Just fucking download Wicca. 01:21:00.500 --> 01:21:02.080 Download It's in Java. 01:21:02.280 --> 01:21:02.900 Download it. 01:21:03.360 --> 01:21:03.840 Run it. 01:21:03.940 --> 01:21:06.220 Just feed AOL data into it, you know. 01:21:06.460 --> 01:21:15.280 Let your machine, you know, do something with the idle cycle and just feed the data into it and see what you get, you know. 01:21:15.280 --> 01:21:20.660 So it could be nothing, but you may find something interesting, you know. 01:21:20.740 --> 01:21:23.000 And it's all in how you interpret it. 01:21:23.900 --> 01:21:27.160 Well, that's all the time that we have here for Binary Revolution. 01:21:27.480 --> 01:21:29.180 And as we say every week. 01:21:29.400 --> 01:21:30.620 We'll see you again next week. 01:21:30.760 --> 01:21:31.660 Same hack time. 01:21:32.200 --> 01:21:33.340 Same hack channel. 01:21:33.340 --> 01:21:35.060 It's 3 a.m. 01:21:35.060 --> 01:21:37.040 And I want to go to bed. 01:21:37.500 --> 01:21:40.340 I got a lady running through my head. 01:21:40.940 --> 01:21:43.960 Ran out of money looking for the night shift. 01:21:43.960 --> 01:21:45.340 It's 3 a.m. 01:21:45.340 --> 01:21:47.720 And I want to go to bed. 01:21:47.780 --> 01:21:50.660 I know a lady way down in my country. 01:21:51.200 --> 01:21:54.560 She was so pretty that my eyes threw the sizes at me. 01:21:54.720 --> 01:21:57.560 Now we will sit and we'll wonder about our future. 01:21:58.180 --> 01:22:01.560 But now I'm thinking that today is down front of me. 01:22:01.700 --> 01:22:04.340 Well, I've been working five days full time. 01:22:05.080 --> 01:22:08.540 Ain't got no money, but everything is going fine. 01:22:08.540 --> 01:22:11.160 Well, I've been tired in my head. 01:22:11.760 --> 01:22:14.900 Well, I've been tired in my head. 01:22:15.240 --> 01:22:16.600 It's 3 a.m. 01:22:16.600 --> 01:22:18.260 And I want to go to bed. 01:22:19.040 --> 01:22:22.040 I got a lady running through my head. 01:22:22.460 --> 01:22:25.520 Run out of money looking for the night shift. 01:22:25.920 --> 01:22:26.920 It's 3 a.m. 01:22:26.920 --> 01:22:29.280 And I want to go to bed. 01:22:29.280 --> 01:22:32.740 I know a lady whose eyes fly right to me. 01:22:32.740 --> 01:22:36.200 And she will sit and stare directly at me. 01:22:36.340 --> 01:22:39.240 And that laugh will take me to my future. 01:22:39.820 --> 01:22:43.080 Throughout my past, there is nothing left of me. 01:22:43.220 --> 01:22:46.460 But I've been working five days full time. 01:22:46.660 --> 01:22:50.100 Ain't got no money, but everything is going fine. 01:22:50.200 --> 01:22:52.740 Well, I've been tired in my head. 01:22:52.740 --> 01:22:56.480 Said, I've been tired in my head. 01:22:57.080 --> 01:22:58.120 It's 3 a.m. 01:22:58.120 --> 01:22:59.720 And I want to go to bed. 01:23:00.560 --> 01:23:03.540 I got a lady running through my head. 01:23:04.000 --> 01:23:07.020 Run out of money looking for the night shift. 01:23:07.420 --> 01:23:08.420 It's 3 a.m. 01:23:08.420 --> 01:23:10.820 And I want to go to bed. 01:23:10.820 --> 01:23:26.580 I feel my lady late night. 01:23:26.580 --> 01:23:29.160 She comes to me and baids my mind. 01:23:29.420 --> 01:23:31.460 Reminds me of where we should be. 01:23:31.580 --> 01:23:34.500 So we just sit and we dream about our future. 01:23:35.060 --> 01:23:38.360 Throughout my past, there is nothing left for me. 01:23:38.500 --> 01:23:40.440 But I've been thinking five days. 01:23:40.440 --> 01:23:41.700 Full time. 01:23:41.900 --> 01:23:45.360 Ain't got no money, but everything is going fine. 01:23:45.440 --> 01:23:47.820 But I've been tired in my head. 01:23:48.840 --> 01:23:51.640 Well, I'm tired in my head. 01:23:52.280 --> 01:23:53.300 It's 3 a.m. 01:23:53.300 --> 01:23:55.000 And I want to go to bed. 01:23:55.760 --> 01:23:58.780 I got a lady running through my head. 01:23:59.220 --> 01:24:02.260 Run out of money looking for the night shift. 01:24:02.640 --> 01:24:03.640 It's 3 a.m. 01:24:03.640 --> 01:24:06.560 And I want to go to bed. 01:24:06.560 --> 01:24:10.280 My lady won't leave my head. 01:24:10.440 --> 01:24:14.260 I just want to rest my head through bed. 01:24:14.660 --> 01:24:16.660 She won't let me go. 01:24:17.820 --> 01:24:19.840 I don't want to go. 01:24:19.840 --> 01:24:19.900 I don't want to go. 01:24:19.900 --> 01:24:26.520 I don't want to go. 01:24:32.660 --> 01:24:33.080 Good night. 01:24:33.080 --> 01:24:33.680 I don't want to go. 01:24:33.680 --> 01:24:34.000 I don't want to go. 01:24:34.000 --> 01:24:35.120 Give it to her yes. 01:24:35.120 --> 01:24:35.420 I don't want to go. 01:24:35.420 --> 01:24:35.780 I don't want to go. 01:24:35.900 --> 01:24:36.280 You know what it is. 01:24:36.280 --> 01:24:37.180 I don't want to go. 01:24:37.180 --> 01:24:37.260 I don't want to go. 01:24:37.260 --> 01:24:39.160 I don't want to go. 01:24:39.160 --> 01:24:39.580 I don't want to go. 01:24:39.580 --> 01:24:41.580 You