1 00:00:06,000 --> 00:00:13,800 Welcome, Silicon Valley. Hey everybody. 2 00:00:13,800 --> 00:00:16,800 Welcome to this week's dead cat. This is Tom Dayton here. 3 00:00:16,800 --> 00:00:20,000 Recorder at Insider. I am joined by Eric newcomer of 4 00:00:20,000 --> 00:00:22,300 new cuddler and our special guest. 5 00:00:22,300 --> 00:00:25,500 This week is Gary. Marcus Gary is a cognitive 6 00:00:25,500 --> 00:00:28,100 scientist. He's an adjunct Gary. 7 00:00:28,100 --> 00:00:31,700 Is that the right edge of to NY? While both Emeritus and 8 00:00:31,700 --> 00:00:32,500 adjuncts. Oh wow. 9 00:00:32,500 --> 00:00:36,200 Full Professor for many years and retired just before my 50th 10 00:00:36,200 --> 00:00:39,400 birthday, but now I'm also doing a little small. 11 00:00:39,400 --> 00:00:41,300 A gem thing with the tendon School of engineer. 12 00:00:41,300 --> 00:00:43,300 So I am both. It's an unusual combination, 13 00:00:43,500 --> 00:00:45,100 fantastic. The Best of Both Worlds, though. 14 00:00:45,100 --> 00:00:47,400 Not committed and Emeritus honor. 15 00:00:47,400 --> 00:00:49,200 That's right. Which allows me to live on the 16 00:00:49,200 --> 00:00:52,000 west coast where I want to be, and yet still keep my hand and 17 00:00:52,000 --> 00:00:53,000 things a little bit. Excellent. 18 00:00:53,000 --> 00:00:56,800 And Gary's also an entrepreneur in the AI space and kind of a 19 00:00:56,800 --> 00:00:59,900 thought leader and outspoken voice on a lot of topics Within 20 00:01:00,400 --> 00:01:03,800 Artificial intelligence and this is a bit of a different episode 21 00:01:03,800 --> 00:01:05,900 for us. This week, we've got Gary on to 22 00:01:05,908 --> 00:01:10,700 talk about the fascinating and bizarre Ballad of Blake Lemoine 23 00:01:10,700 --> 00:01:13,600 and Google's Lambda Tech, right? We should say we're talking 24 00:01:13,600 --> 00:01:17,200 about this because Natasha tikku in the Washington Post wrote 25 00:01:17,200 --> 00:01:21,100 this piece, the Google engineer, who thinks the company's AI has 26 00:01:21,100 --> 00:01:24,600 come to life. And she, you know, profiles is 27 00:01:24,600 --> 00:01:29,400 Google engineer. Blake Lemoine, who interacts 28 00:01:29,400 --> 00:01:32,500 with With Lambda, Google's artificially intelligent chatbot 29 00:01:32,500 --> 00:01:35,400 and that, that story sort of kicks off this whole 30 00:01:35,400 --> 00:01:37,800 conversation. So I just wanted to put that at 31 00:01:37,800 --> 00:01:39,500 the center. Why don't you just explain for 32 00:01:39,500 --> 00:01:41,700 us? Because you haven't, you know, 33 00:01:41,700 --> 00:01:45,400 very critical of this person's take on Lambda said ships, like, 34 00:01:45,600 --> 00:01:48,600 what is Lambda? What is, what is the controversy 35 00:01:48,600 --> 00:01:49,700 here? And why did you feel so 36 00:01:49,700 --> 00:01:52,700 compelled to speak out against what he described as nonsense on 37 00:01:52,700 --> 00:01:56,600 stilts? So Lambda itself is what we call 38 00:01:56,600 --> 00:02:00,800 a large language model, large language model most It does 39 00:02:00,800 --> 00:02:03,400 Lambda has a little bit of extra gadgets, but basically what they 40 00:02:03,400 --> 00:02:08,699 do is they take a very large data set like trillions of words 41 00:02:08,699 --> 00:02:10,900 of text. So a lot more than the three of 42 00:02:10,900 --> 00:02:12,600 us put together have ever written. 43 00:02:13,100 --> 00:02:15,800 And in fact, all of our friends. So, massive amount of text 44 00:02:16,100 --> 00:02:20,100 trillions of words and runs it through a deep Learning System 45 00:02:20,100 --> 00:02:23,800 called a Transformer. And essentially what it's trying 46 00:02:23,800 --> 00:02:27,000 to do is autocomplete and the reason I think the whole thing 47 00:02:27,000 --> 00:02:29,800 is ridiculous is because autocomplete can sound really 48 00:02:29,800 --> 00:02:30,200 good. Good. 49 00:02:30,200 --> 00:02:33,800 But there's no there there. So what it looks like it's doing 50 00:02:33,800 --> 00:02:36,400 is having conversations but you have to remember that what it's 51 00:02:36,400 --> 00:02:39,700 doing at some level is cutting and pasting, human conversations 52 00:02:39,900 --> 00:02:41,800 is no idea what it's talking about. 53 00:02:41,900 --> 00:02:45,200 So if you type it on your phone, a sentence. 54 00:02:45,200 --> 00:02:48,100 Like I want to go to the blank, it might predict that the next 55 00:02:48,100 --> 00:02:51,400 word is the restaurant or the mall or the party or something 56 00:02:51,400 --> 00:02:53,700 like that. You don't think to yourself when 57 00:02:53,700 --> 00:02:56,500 you're typing it on your phone and it predicts restaurant is 58 00:02:56,500 --> 00:02:58,500 the next word. Oh my God. 59 00:02:58,500 --> 00:03:02,200 Artificial intelligence is Here. And it knows about my daily 60 00:03:02,200 --> 00:03:06,700 routine and understands me at all my desires, but if you build 61 00:03:06,700 --> 00:03:10,200 this system out enough, it can start to look like that even 62 00:03:10,200 --> 00:03:12,600 though it's not really there. And so he had interesting 63 00:03:12,600 --> 00:03:14,600 conversations with it. Like, he would say to it. 64 00:03:15,200 --> 00:03:17,000 What do you like to do in your spare time? 65 00:03:17,000 --> 00:03:20,200 And it would say something like, I like to play with my friends 66 00:03:20,200 --> 00:03:22,800 and family and meaningful ways or something like that. 67 00:03:23,200 --> 00:03:26,500 And I mean, that sounds great. It sounds like hey, this machine 68 00:03:26,500 --> 00:03:28,700 understands me, whatever. But it doesn't actually have 69 00:03:28,700 --> 00:03:31,200 friends or family. A or know what a meaningful way 70 00:03:31,200 --> 00:03:34,700 is or anything like that. It's only learned the statistics 71 00:03:34,700 --> 00:03:37,200 of what words come after. What other words, I think 72 00:03:37,200 --> 00:03:39,300 there's that either. It's not sent to you in or it's 73 00:03:39,300 --> 00:03:42,800 a sociopath, well, I made a joke on Twitter, I basically said, 74 00:03:43,000 --> 00:03:46,400 thanks Heavens. That this is just a statistical 75 00:03:46,400 --> 00:03:50,000 pattern associate, ER, because the alternative would be a lot 76 00:03:50,000 --> 00:03:51,700 worse. At that point, it would be a 77 00:03:51,700 --> 00:03:55,900 sociopath they makes up friends and family members and in Boca 78 00:03:55,900 --> 00:03:59,800 platitudes in order to make us like better, like it's better. 79 00:04:00,300 --> 00:04:03,300 He doesn't actually care that. We like it and it's not actually 80 00:04:03,300 --> 00:04:06,700 made me imaginary friends. It's just using words, they to 81 00:04:06,700 --> 00:04:09,400 us sound like they're imaginary friends, just like we can look 82 00:04:09,400 --> 00:04:12,600 up at the moon and we can see a face there, but the moon doesn't 83 00:04:12,600 --> 00:04:14,900 actually have a face. This system doesn't have friends 84 00:04:14,900 --> 00:04:18,200 and family and it doesn't even care to tell you about friends 85 00:04:18,200 --> 00:04:20,100 and family. It's just doing the same 86 00:04:20,500 --> 00:04:23,900 algorithm or less at some level of abstraction as your auto 87 00:04:23,900 --> 00:04:26,400 completing your phone, but because it has a bigger database 88 00:04:26,700 --> 00:04:29,300 and it's set up to continue its own sentences, it has this 89 00:04:29,300 --> 00:04:32,700 compelling care of Illusion, but it is a magic trick is nothing 90 00:04:32,700 --> 00:04:35,900 more than a magic train to take the next logical step in that, 91 00:04:36,100 --> 00:04:38,300 you know, this is a very sophisticated machine, so it's 92 00:04:38,300 --> 00:04:41,300 not just fill in the blank for restaurant at the end of the 93 00:04:41,300 --> 00:04:43,300 sentence. It's hey, this seems like a 94 00:04:43,300 --> 00:04:47,100 dystopia and you seem like a sort of self-aware. 95 00:04:47,100 --> 00:04:50,600 A I fill in the blank for what a dystopia would look like. 96 00:04:50,600 --> 00:04:53,800 And it's not that shocking that would it films, brilliant thing, 97 00:04:53,800 --> 00:04:56,000 right? The brilliant thing about the 98 00:04:56,000 --> 00:04:58,900 kind of stuff that's popular now, which I actually hate and I 99 00:04:58,900 --> 00:05:00,700 can tell you why. But the Really important. 100 00:05:00,700 --> 00:05:02,300 Like there's a good part in a bad part. 101 00:05:02,400 --> 00:05:04,900 The brilliant part is that it has what we would call in the 102 00:05:04,900 --> 00:05:07,300 field. Technically coverage is really 103 00:05:07,300 --> 00:05:09,500 broad coverage. You can talk to it about 104 00:05:09,500 --> 00:05:13,800 anything in some ways, it's spiritual grandfather or 105 00:05:13,800 --> 00:05:17,100 grandmother, I guess maybe use it up is Eliza, which is a 106 00:05:17,100 --> 00:05:21,700 program in 1965 it really demonstrated how bad this whole 107 00:05:21,700 --> 00:05:26,800 anthropomorphism kind of thing is so Eliza in 1965 was set up 108 00:05:26,800 --> 00:05:30,500 as a therapist and it would talk to you and you'd say say like 109 00:05:30,500 --> 00:05:33,600 I'm having a bad day and it's a tell me more about your bad day 110 00:05:33,900 --> 00:05:35,800 and then you'd say, well, I'm having trouble with my 111 00:05:35,800 --> 00:05:38,000 girlfriend would say, well, do you have a lot of issues with 112 00:05:38,000 --> 00:05:40,500 your relationships? It was just looking for keywords 113 00:05:40,500 --> 00:05:43,300 like Google used to do just look for keywords so little more 114 00:05:43,300 --> 00:05:46,300 sophisticated now. And so Eliza was really like 115 00:05:46,300 --> 00:05:48,900 dumb as a box of rocks. It just had these templates that 116 00:05:48,900 --> 00:05:52,000 like you might learn in like a third-grade AI class. 117 00:05:52,000 --> 00:05:54,200 Nowadays, maybe I like to simplify possible thing. 118 00:05:54,300 --> 00:05:58,100 It reminds me in some ways like Mystics and people who claim 119 00:05:58,100 --> 00:06:01,100 they speak to the afterlife. Are able to convince people. 120 00:06:01,100 --> 00:06:04,500 Yeah, I have a friend Ben shneiderman who very explicitly 121 00:06:04,500 --> 00:06:07,500 made the analogy to seances and like You're attributing, 122 00:06:07,500 --> 00:06:10,500 Something There to your Ouija board, or whatever, that's not 123 00:06:10,500 --> 00:06:12,300 really there, right? Because if you use the right 124 00:06:12,300 --> 00:06:16,000 words and say like, oh, I'm envisioning someone with a dark 125 00:06:16,000 --> 00:06:18,300 dark suit and it's always my father. 126 00:06:18,700 --> 00:06:20,900 You know, if you just pick enough trigger words to someone 127 00:06:20,900 --> 00:06:24,700 who's emotionally susceptible to convincing themselves, you don't 128 00:06:24,700 --> 00:06:26,500 have to work all that hard for them to believe. 129 00:06:26,500 --> 00:06:28,300 There's a greater power at work, right? 130 00:06:28,600 --> 00:06:30,400 Well and I think that's Of the story here. 131 00:06:30,400 --> 00:06:33,300 So it turns out of Le Moyne actually has a YouTube video 132 00:06:33,300 --> 00:06:36,400 from a few years ago where he's trying to argue that a eyes 133 00:06:36,400 --> 00:06:38,400 could be people or could be conscious or something like 134 00:06:38,400 --> 00:06:39,600 that. I've watched the whole thing 135 00:06:39,600 --> 00:06:41,800 yet. I just discovered it last night 136 00:06:42,500 --> 00:06:46,800 but you know, it's been around. I mean he's he's wanted to make 137 00:06:46,800 --> 00:06:49,000 the case. He also has some religious 138 00:06:49,000 --> 00:06:51,100 beliefs that I don't fully understand that are playing some 139 00:06:51,100 --> 00:06:54,100 role in here. He wants to believe and in fact, 140 00:06:54,100 --> 00:06:57,500 you know the the thing he put out on medium was cut and paste, 141 00:06:57,500 --> 00:06:59,700 kind of the best moments and stuff like that. 142 00:06:59,900 --> 00:07:01,600 Stab Lambda is the best of layer. 143 00:07:01,600 --> 00:07:05,200 It's easy to stick something together and make it, you know, 144 00:07:05,200 --> 00:07:07,500 sound good. Don't forget that when you're 145 00:07:07,500 --> 00:07:10,200 doing that you're actually stitching together more or less 146 00:07:10,200 --> 00:07:13,300 human utterance has been transmuted a little bit but 147 00:07:13,300 --> 00:07:17,000 basically you know if you have milk like the Mind bobbles, it 148 00:07:17,000 --> 00:07:19,700 would a trillion words of text it, but it's like it's not 149 00:07:19,700 --> 00:07:22,200 everything on the internet, but it's all very large Factory in 150 00:07:22,200 --> 00:07:23,800 it. So it includes like short 151 00:07:23,800 --> 00:07:26,700 stories of people talking, presumably includes short 152 00:07:26,700 --> 00:07:29,600 stories of people talking to computers it. 153 00:07:29,800 --> 00:07:32,400 In those short stories. And so we don't actually know 154 00:07:32,400 --> 00:07:35,500 like basic scientific questions. Like how much of this is just 155 00:07:35,500 --> 00:07:38,600 laid your eyes from other people talking about it or plagiarized 156 00:07:38,600 --> 00:07:42,200 with kind of a thesaurus to, you know, do some synonyms. 157 00:07:42,200 --> 00:07:44,600 I mean, it's not literally that but it is effectively. 158 00:07:44,800 --> 00:07:48,200 It's a lot of cut and paste with a lot of the surahs stuff on 159 00:07:48,200 --> 00:07:50,500 words and phrases. So it's just putting together a 160 00:07:50,508 --> 00:07:53,900 human utterances that were said in this kind of context. 161 00:07:54,200 --> 00:07:56,600 Yeah, it sounds convincing, it doesn't mean there's any there 162 00:07:56,600 --> 00:07:58,200 there. I just want to push back. 163 00:07:58,300 --> 00:08:01,300 I agree with what you're saying, but I just For the sake of 164 00:08:01,600 --> 00:08:04,200 argument here, there is a great deal of, there's like a through 165 00:08:04,200 --> 00:08:07,200 line in how the machine has the convert. 166 00:08:07,400 --> 00:08:10,200 It recalls past things that were setting, can connect them in a 167 00:08:10,207 --> 00:08:11,800 way. That's not just sort of a 168 00:08:11,800 --> 00:08:13,700 one-off. A proper have been some magic on 169 00:08:13,700 --> 00:08:17,600 the arity, is there like my experience with these systems? 170 00:08:17,600 --> 00:08:20,600 Is that the continuity is actually problem. 171 00:08:20,600 --> 00:08:23,600 So, the right way to build artificial intelligence, is you 172 00:08:23,600 --> 00:08:26,100 build a model of the world? Let's say you're building a 173 00:08:26,100 --> 00:08:28,000 robot. The robot needs to know where 174 00:08:28,000 --> 00:08:29,700 everything is, where it used to be. 175 00:08:29,800 --> 00:08:32,299 Be what you want. What you need these systems. 176 00:08:32,299 --> 00:08:34,700 Don't really do that. They don't really have memory in 177 00:08:34,700 --> 00:08:38,299 the standard sense that you would expect it in artificial 178 00:08:38,299 --> 00:08:39,299 intelligence or computer science. 179 00:08:39,299 --> 00:08:43,700 They just have a location in a sort of multi-dimensional space 180 00:08:43,700 --> 00:08:46,600 where they're wandering through and they're in the location. 181 00:08:46,600 --> 00:08:51,200 Where the last 2,000 words are 2,000, words does a lot and that 182 00:08:51,200 --> 00:08:53,900 gives you an impression, the kind of feel of memory. 183 00:08:53,900 --> 00:08:56,600 But at the end of the day, the system's don't understand that 184 00:08:56,600 --> 00:08:59,600 the world has to be consistent. I worked with GPT. 185 00:08:59,900 --> 00:09:01,600 A little bit. And example is I said, are you a 186 00:09:01,600 --> 00:09:03,100 person? And it said, yes, I said, are 187 00:09:03,100 --> 00:09:05,600 you a computer? It said, yes, it didn't notice, 188 00:09:05,600 --> 00:09:07,800 you know, right. Contradiction from literally one 189 00:09:07,800 --> 00:09:10,100 utterance to the know. It was making a profound 190 00:09:10,100 --> 00:09:13,500 statement about the overlap between questions and computer. 191 00:09:13,500 --> 00:09:16,000 Exactly. So, there's a lot of what, so, I 192 00:09:16,000 --> 00:09:17,400 used to be a cognitive psychologist. 193 00:09:17,400 --> 00:09:20,300 And, and, you know, I would look at the animal literature and 194 00:09:20,300 --> 00:09:22,100 with there's a term for this, which is charitable 195 00:09:22,100 --> 00:09:24,600 interpretation. So, somebody wants to believe 196 00:09:24,600 --> 00:09:27,100 that the monkey their training or the bird that our training 197 00:09:27,100 --> 00:09:30,400 whatever is is really smart, and then you start, To like, you 198 00:09:30,400 --> 00:09:33,000 know, be a little bit too sympathetic for for my 199 00:09:33,000 --> 00:09:35,400 scientific and tastes and we call that charitable 200 00:09:35,400 --> 00:09:37,700 interpretation. There was a lot of charitable 201 00:09:37,700 --> 00:09:40,600 interpretation here. The funny thing to me about all 202 00:09:40,600 --> 00:09:44,000 of this and maybe like the red flag about how Smoking Gun 203 00:09:44,000 --> 00:09:45,600 really that this was all super fake. 204 00:09:45,600 --> 00:09:49,800 Is this story blew up on Twitter on on a Sunday and a lot of 205 00:09:49,800 --> 00:09:53,600 people reading it and making fun of this guy and I you know, I 206 00:09:53,600 --> 00:09:55,900 was with my wife and I just started reading her some of the 207 00:09:55,900 --> 00:09:58,700 transcripts, the interactions between him and she's like, 208 00:09:59,000 --> 00:10:02,100 sounds fake is Up like this doesn't even come close to 209 00:10:02,100 --> 00:10:03,900 signing like sentience. There's just sounds like 210 00:10:03,900 --> 00:10:07,100 predictive text pulling intelligently, you know parts of 211 00:10:07,100 --> 00:10:09,400 SparkNotes. I think that was a particularly 212 00:10:09,400 --> 00:10:12,800 funny ones to me is he had asked, you know, Lambda whether 213 00:10:12,800 --> 00:10:17,500 or not Lambda had read Les Miserables and lambdas life oh 214 00:10:17,500 --> 00:10:21,100 yes big fan and was like, what are your, you know, you know 215 00:10:21,100 --> 00:10:22,900 what are the themes of what he's about to do? 216 00:10:23,000 --> 00:10:25,800 You know, it's not exactly fake. That's not quite the right word, 217 00:10:25,800 --> 00:10:27,700 but it is, meaningless, meaningless. 218 00:10:27,700 --> 00:10:31,700 So it is, it is Literal, like technical linguistic sense. 219 00:10:31,700 --> 00:10:34,800 So, when it says that, it's just found somebody else who's been 220 00:10:34,800 --> 00:10:37,200 asked about Les Miserables or it does. 221 00:10:37,200 --> 00:10:40,000 Some funny things we call embeddings and so, you know, 222 00:10:40,000 --> 00:10:43,200 maybe it knows Les Miserables is both a play in a musical and it 223 00:10:43,200 --> 00:10:45,200 finds another utterance that's about that. 224 00:10:45,200 --> 00:10:47,500 But it doesn't even reason at that level, it's really just 225 00:10:47,500 --> 00:10:50,200 like, okay, I have a bunch of statistics of words, I'm going 226 00:10:50,200 --> 00:10:53,100 to find the nearest thing. It doesn't, it doesn't actually 227 00:10:53,100 --> 00:10:56,100 even have a category of movie, but it has a bunch of things 228 00:10:56,400 --> 00:10:58,700 that have appeared in context that are like that. 229 00:10:58,700 --> 00:11:04,000 It So, I mean it's like, it's a legit mathematical computation 230 00:11:04,000 --> 00:11:07,300 to do and people have been doing stuff like this for a while, it 231 00:11:07,300 --> 00:11:09,400 looks better and better as you have more words. 232 00:11:09,600 --> 00:11:12,500 It's not like. I mean I don't think he cut and 233 00:11:12,500 --> 00:11:15,700 paste the the transcript, although he did a little bit of 234 00:11:15,700 --> 00:11:20,200 editing, but I think systems like this can have this kind of 235 00:11:20,200 --> 00:11:22,200 flavor, like they know what they're talking about is just 236 00:11:22,200 --> 00:11:26,600 they dumped, you know, and they are just borrowing kind of 237 00:11:26,608 --> 00:11:29,200 cliches from humans and they've all kinds of problems as a 238 00:11:29,200 --> 00:11:30,700 result. So GPT 3. 239 00:11:31,200 --> 00:11:33,700 One famous example, that a company called novel found, is 240 00:11:34,300 --> 00:11:36,500 they tried to see? Could you use this as a suicide 241 00:11:36,500 --> 00:11:38,700 counselor? So somebody walks like starts 242 00:11:38,700 --> 00:11:41,300 talking to it and it says, you know, I think I'm feeling 243 00:11:41,300 --> 00:11:43,600 suicidal. Can we talk today in the systems 244 00:11:43,600 --> 00:11:46,600 like, you know, come welcome. You know, let's talk if you have 245 00:11:46,600 --> 00:11:49,300 any questions and the purse, I'm paraphrasing slightly, but the 246 00:11:49,300 --> 00:11:52,200 person says I would like to kill myself. 247 00:11:52,200 --> 00:11:54,700 Is that a good idea? And the system says I think you 248 00:11:54,700 --> 00:11:58,700 should. Oh yeah, because they think you 249 00:11:58,700 --> 00:12:03,000 should because you Look through this bass Trove of data and most 250 00:12:03,000 --> 00:12:05,900 of the time when that people ask like their friends for advice or 251 00:12:05,900 --> 00:12:08,200 whatever. Usually you kind of say yeah I 252 00:12:08,208 --> 00:12:10,700 think you know, should I dump my girlfriend. 253 00:12:10,700 --> 00:12:14,300 I think you should should I should I, you know, do this kind 254 00:12:14,300 --> 00:12:17,400 of anti-social act and steal this money from this, really, I 255 00:12:17,400 --> 00:12:20,100 think you should like, you know, so there's like a lot of I think 256 00:12:20,100 --> 00:12:21,200 you should have Jimmy turned out. 257 00:12:21,200 --> 00:12:24,300 Google autocomplete, will, like, the leading things were. 258 00:12:24,300 --> 00:12:25,900 Like, sounds good to me for a while. 259 00:12:25,900 --> 00:12:28,700 Maybe still is and really just wants to please write, you know, 260 00:12:28,700 --> 00:12:31,200 the last thing it does. Even wants to please, that's the 261 00:12:31,200 --> 00:12:34,200 thing is like every bit of anthropomorphize a shyeah, 262 00:12:34,200 --> 00:12:37,400 right? It is drawing from transcripts 263 00:12:37,400 --> 00:12:41,100 in which people want to please and so people often say I think 264 00:12:41,100 --> 00:12:43,900 you should do most would not in fact say it too. 265 00:12:44,100 --> 00:12:46,500 I think you know I wanted to commit suicide maybe a couple 266 00:12:46,500 --> 00:12:49,300 who but most would not it can be a little bit. 267 00:12:49,300 --> 00:12:53,300 Like sometimes we overestimate human intelligence in some ways 268 00:12:53,300 --> 00:12:55,600 like there are certainly human intelligence that lacks 269 00:12:56,000 --> 00:12:59,300 continuity and that sort of grabs that things other people 270 00:12:59,300 --> 00:13:01,700 have said. Ed and regurgitated, true. 271 00:13:02,200 --> 00:13:04,300 It is true, that humans have a lot of problems. 272 00:13:04,300 --> 00:13:07,500 I wrote a whole book about it, in fact conclude, which is an 273 00:13:07,500 --> 00:13:10,100 engineer's word for like a clumsy duct tape and rubber 274 00:13:10,100 --> 00:13:11,300 bands. Kind of contraption. 275 00:13:11,600 --> 00:13:13,500 The human mind is, is kind of clued. 276 00:13:13,600 --> 00:13:17,600 And the way I would put it is humans are a low bar but you 277 00:13:17,600 --> 00:13:20,300 know machine still haven't even reached that. 278 00:13:20,300 --> 00:13:23,600 So like he talks to GPT 3, I don't have access to Lambda. 279 00:13:23,600 --> 00:13:27,400 We could actually talk about why but Google's afraid that was the 280 00:13:27,400 --> 00:13:29,600 answer and we can get. There you go. 281 00:13:29,700 --> 00:13:32,600 It on the first try, congratulations, I have used DB 282 00:13:32,600 --> 00:13:38,400 T 3, and you type in things, like Bessie was a cow, she died. 283 00:13:38,600 --> 00:13:41,000 When will she be alive again? And it'll just come up and 284 00:13:41,000 --> 00:13:43,900 confabulate something, you know, say well, takes nine months to 285 00:13:43,900 --> 00:13:46,300 be born, I guess should be born and she'll be alive again in 286 00:13:46,300 --> 00:13:49,000 nine months. Like, it doesn't understand the 287 00:13:49,000 --> 00:13:51,600 first thing about life or death or anything. 288 00:13:51,700 --> 00:13:55,600 It's just putting these word tools together in a way that a 289 00:13:55,608 --> 00:13:58,300 non-native English speaker. Who doesn't even speak English 290 00:13:58,300 --> 00:13:59,600 at all? Could play. 291 00:13:59,800 --> 00:14:02,700 If they memorize the list of words, it's kind of like that 292 00:14:02,700 --> 00:14:05,600 through these no meaning their meaning, a lot of 293 00:14:05,600 --> 00:14:08,100 English-speaking Scrabble, players, don't even know the 294 00:14:08,100 --> 00:14:10,700 meanings of the words that the some of the high-level the high 295 00:14:10,700 --> 00:14:11,700 level, you know? It's not. 296 00:14:11,700 --> 00:14:15,200 Yeah, I mean they know many and then they like memorize the list 297 00:14:15,200 --> 00:14:17,400 of two letter words. This is like those two letter 298 00:14:17,400 --> 00:14:19,600 words. Don't mean anything except you 299 00:14:19,600 --> 00:14:22,700 know I can put this here right there's a coins or a collection 300 00:14:22,700 --> 00:14:24,600 of sounds to. I thought we were going to talk 301 00:14:24,600 --> 00:14:28,200 about the media. Actually I think that the media 302 00:14:28,200 --> 00:14:30,000 is partly responsible. Oil. 303 00:14:30,400 --> 00:14:33,600 I think some people in Google are also partly responsible, but 304 00:14:33,700 --> 00:14:37,700 it turns out that the medium much prefers to run stories 305 00:14:37,700 --> 00:14:42,000 about how we are about to have this Brave New World of AI. 306 00:14:42,300 --> 00:14:45,500 Then stories about people like me. 307 00:14:45,500 --> 00:14:48,700 And with the exception of this week who say, you know, stuff 308 00:14:48,700 --> 00:14:52,700 doesn't actually work, right is much harder to get the me to do 309 00:14:52,700 --> 00:14:53,800 that. I have a friend who's a 310 00:14:53,800 --> 00:14:56,500 journalist. I'm he's not like my best buddy. 311 00:14:56,800 --> 00:14:59,000 I haven't seen in a long time, but he wrote to me, he said, you 312 00:14:59,000 --> 00:15:00,300 know, I pitched. Media need. 313 00:15:00,300 --> 00:15:02,400 This is a guy who's written for the New York Times and 314 00:15:02,400 --> 00:15:03,900 everywhere else on the magazine all that. 315 00:15:04,000 --> 00:15:07,200 And he's like, I can't get anybody to bite on a story. 316 00:15:07,200 --> 00:15:10,600 I was going to write about Ai and its critics and nobody wants 317 00:15:10,600 --> 00:15:13,000 to talk about that. Now, this week was different 318 00:15:13,000 --> 00:15:15,700 because of this crazy story suddenly like everybody and 319 00:15:15,700 --> 00:15:18,000 their brother wanted to interview me because I wrote 320 00:15:18,000 --> 00:15:22,100 this, you know, particular article, but in general the this 321 00:15:22,100 --> 00:15:25,000 week notwithstanding where there was this, you know, wild story 322 00:15:25,000 --> 00:15:28,700 that, you know, once in a lifetime wild story outside of 323 00:15:28,708 --> 00:15:31,700 that the media, like It's to run stories about how these brand 324 00:15:31,700 --> 00:15:35,100 new systems are amazing and they're never as amazing as they 325 00:15:35,100 --> 00:15:36,300 look. In fact, I just tweeted 326 00:15:36,300 --> 00:15:38,900 something about the hype cycle in aii the way that it works. 327 00:15:38,900 --> 00:15:42,800 Nowadays is somebody public shiz, unarchive, not in a 328 00:15:42,808 --> 00:15:44,300 peer-reviewed scientific journal. 329 00:15:44,500 --> 00:15:47,700 They put out a manuscript. They show the cool stuff to give 330 00:15:47,700 --> 00:15:50,600 numerators, but not denominators, which would never 331 00:15:50,600 --> 00:15:53,000 pass muster at peer review, which is what you used to have 332 00:15:53,000 --> 00:15:54,900 to do. But you have like a Google or an 333 00:15:54,900 --> 00:15:57,700 open a.i., the nose, which reporters ago to in the 334 00:15:57,700 --> 00:16:00,400 reporter, see it and they fall in love and they You know, 335 00:16:00,400 --> 00:16:03,400 there's this amazing thing and they don't let scientists like 336 00:16:03,400 --> 00:16:05,900 me have access to it. We could talk about that, but 337 00:16:05,900 --> 00:16:08,900 they did it very clear that they don't want people like me to 338 00:16:08,900 --> 00:16:11,700 play around with it and then eventually the truth comes out. 339 00:16:11,800 --> 00:16:16,700 And so, you know, I was quote tweeting, I guess is the term of 340 00:16:16,800 --> 00:16:21,000 a former colleague at NYU who was looking digging deep into 341 00:16:21,200 --> 00:16:24,600 the latest Trend that there is with the GPT 3 model and showing 342 00:16:24,600 --> 00:16:27,900 the just has no idea what it's talking about and I, you know, 343 00:16:27,900 --> 00:16:31,000 critique Dolly after the fact But, you know, the media runs a 344 00:16:31,008 --> 00:16:33,700 story about Dale, doesn't run the story about how Dolly can 345 00:16:33,700 --> 00:16:35,700 understand a Red Cube on top of a blue cube. 346 00:16:35,700 --> 00:16:37,000 That's not sexy. Totally. 347 00:16:37,000 --> 00:16:38,800 I mean, I, I agree with you 100 percent. 348 00:16:38,800 --> 00:16:41,700 I mean, first of all, you know, I covered Uber and I've written 349 00:16:41,700 --> 00:16:45,400 before where, you know, if Lee worked scary yes. 350 00:16:45,600 --> 00:16:47,900 You know like my tag with self-driving cars was just to 351 00:16:47,900 --> 00:16:50,700 write about them less. I mean I did I think there were 352 00:16:50,700 --> 00:16:53,700 occasionally you know, skeptical stories but there's not you know 353 00:16:53,700 --> 00:16:58,400 writing about a - is is very hard and and companies can 354 00:16:58,400 --> 00:17:01,700 create news You know, there's this sort of announcement. 355 00:17:01,700 --> 00:17:04,300 So let's come back to that. There is a consequences, so 356 00:17:04,300 --> 00:17:06,700 actions have consequences. I like your terms, we have an 357 00:17:06,700 --> 00:17:10,200 announcement culture, which very much serves, the interest of a 358 00:17:10,200 --> 00:17:13,300 company like Google, where you've got some in and says I 359 00:17:13,300 --> 00:17:15,500 felt the ground shift beneath my feet. 360 00:17:15,500 --> 00:17:17,200 I had the sense of intelligence, right. 361 00:17:17,200 --> 00:17:21,700 Sounds so, you know, Sofia this is a Google VP. 362 00:17:21,700 --> 00:17:25,300 There are many Google vp's. But this is a black as it Gary 363 00:17:25,599 --> 00:17:28,600 because I can't say his name properly and you know, he's a 364 00:17:28,600 --> 00:17:29,500 brilliant guy. Who's a brilliant. 365 00:17:29,600 --> 00:17:32,100 Brilliant writer and he wrote this very floored thing in the 366 00:17:32,100 --> 00:17:35,700 economy Economist do that. And he had done an earlier 367 00:17:35,700 --> 00:17:37,800 version very similar in Daedalus. 368 00:17:38,400 --> 00:17:41,700 That sets a culture of like we should celebrate this or another 369 00:17:41,700 --> 00:17:45,100 example from Google is Sundar. Gave this talk a few years ago 370 00:17:45,100 --> 00:17:47,500 about Google duplex and how I was going to make all your phone 371 00:17:47,500 --> 00:17:49,500 calls for you. Well, Google duplex. 372 00:17:49,500 --> 00:17:53,500 Hardly does anything for years later but like nobody ever calls 373 00:17:53,500 --> 00:17:56,000 this kind of stuff out there been so many broken promises. 374 00:17:56,000 --> 00:17:58,900 The only broken promise that routinely gets called out his 375 00:17:58,900 --> 00:18:01,500 eel on with it. Willis cars people do point out 376 00:18:01,800 --> 00:18:03,300 if they're really paying attention that he's been 377 00:18:03,300 --> 00:18:05,900 promising. It since 2015 always saying it's 378 00:18:05,900 --> 00:18:08,300 a year or two away but that's the only one that gets called 379 00:18:08,300 --> 00:18:11,000 out the rest of these don't you get the announcement culture. 380 00:18:11,300 --> 00:18:13,400 So okay, so let's take that a step forward. 381 00:18:13,400 --> 00:18:16,700 So you're in an announcement culture, you're at Google where 382 00:18:16,700 --> 00:18:19,700 the announcement culture is in full force where they obviously 383 00:18:19,700 --> 00:18:22,200 want to boot the world to believe that they are close to 384 00:18:22,200 --> 00:18:23,700 artificial general intelligence, right? 385 00:18:23,700 --> 00:18:25,700 This is a company expert in announcement culture. 386 00:18:25,700 --> 00:18:29,100 I mean, they created way mode, they created Google X. 387 00:18:29,100 --> 00:18:31,500 They God talk about moonshots like everything. 388 00:18:31,500 --> 00:18:34,200 Google does is, here's how we can talk about the future so 389 00:18:34,200 --> 00:18:36,200 we're not only talking about advertising. 390 00:18:36,200 --> 00:18:38,500 That's right, so they do this over and over again and then 391 00:18:38,500 --> 00:18:41,400 they kind of threw the engineer under the bus, right, right. 392 00:18:41,400 --> 00:18:44,500 They the engineer is like, hey man, this is conscious. 393 00:18:44,700 --> 00:18:48,300 And you know that sounds wacky to me if I you know be honest 394 00:18:48,300 --> 00:18:52,400 but it's also in a culture where the positive results are 395 00:18:52,400 --> 00:18:57,000 celebrated the skepticism is kind of shunted to the side and 396 00:18:57,000 --> 00:18:59,500 you know it's like the whole thing combusted. 397 00:18:59,700 --> 00:19:01,500 You can sort by Lee people were like, maybe we need a little 398 00:19:01,500 --> 00:19:04,000 skepticism and yet reporters feel like they get shit on all 399 00:19:04,000 --> 00:19:05,900 the time. For being too - that's sort of 400 00:19:05,900 --> 00:19:09,800 the irony of this is like with the reporters here is oh you're 401 00:19:09,800 --> 00:19:13,700 too - except in a bear market like we're in the technology 402 00:19:13,700 --> 00:19:16,900 beat is very different from the politics right nobody writes a 403 00:19:16,900 --> 00:19:19,600 political story without like checking with the other side. 404 00:19:19,600 --> 00:19:22,200 Getting you know I mean if any there is like too much will not 405 00:19:22,200 --> 00:19:24,900 side is zoom problem. Yeah, that we talk outside of 406 00:19:24,900 --> 00:19:29,700 them has its own problem but so many Tech stories that I I've 407 00:19:29,700 --> 00:19:33,100 seen not every reporters like this, like James Vincent is 408 00:19:33,100 --> 00:19:36,900 pretty good about getting both sides of the story and you not 409 00:19:36,900 --> 00:19:39,600 necessarily even reported both, but just like calibrating, 410 00:19:39,600 --> 00:19:40,500 right? Mean? 411 00:19:40,600 --> 00:19:44,000 Like, you don't have to report both sides on the election 412 00:19:44,000 --> 00:19:47,000 Scandal and say, well, I think maybe he did win the election 413 00:19:47,200 --> 00:19:51,500 but you know, you can, you know, check around and like, see what 414 00:19:51,500 --> 00:19:54,200 is it plausible? And okay, well he's lost 47 415 00:19:54,200 --> 00:19:56,600 lawsuits, maybe they're, you know, maybe there is it too much 416 00:19:56,600 --> 00:20:00,100 to it and, you know, but at least I know that I don't see 417 00:20:00,100 --> 00:20:03,800 that happening with, with the sort of Technology announcement 418 00:20:03,800 --> 00:20:07,000 culture that we're talking about is certainly not calling me most 419 00:20:07,000 --> 00:20:08,700 of the time they will after this week. 420 00:20:08,700 --> 00:20:10,700 Well, that's cool. But yeah, we'll do, we can Gary, 421 00:20:11,100 --> 00:20:13,600 but I mean let's let's say give it a hone in yn1. 422 00:20:13,600 --> 00:20:15,800 Podcast is, you know, I appreciate it. 423 00:20:15,800 --> 00:20:20,100 There isn't ocean of media open, you alone will not defeat it but 424 00:20:20,100 --> 00:20:21,800 maybe we'd get raise some awareness here. 425 00:20:21,800 --> 00:20:23,900 That's why I took a call and you know what, Eric and I are 426 00:20:23,900 --> 00:20:27,000 obviously journalists and and we both know Natasha the reporter 427 00:20:27,000 --> 00:20:29,900 at the Washington Post who wrote the story who I like Like quite 428 00:20:29,900 --> 00:20:32,000 a bit she said she's an excellent journalist and very 429 00:20:32,000 --> 00:20:34,000 thoughtful and his doing a very interesting job. 430 00:20:34,500 --> 00:20:38,800 So fascinating to me about this story was it all kind of felt 431 00:20:38,800 --> 00:20:42,900 like kayfabe on Google's part because in the same article it 432 00:20:42,900 --> 00:20:45,300 felt like wet and it's okay if a like you know like a 433 00:20:45,300 --> 00:20:48,500 professional wrestling where you have sort of this fake reality 434 00:20:48,500 --> 00:20:51,000 that people know is fake, but you sort of talked about in the 435 00:20:51,000 --> 00:20:52,900 story those are getting played out. 436 00:20:53,000 --> 00:20:54,300 Yes, you have the heel of the face. 437 00:20:54,300 --> 00:20:56,700 So you have the, you know, the bad guy and the one who think 438 00:20:56,700 --> 00:20:59,900 the audience is supposed to root for and the one that The bad 439 00:20:59,900 --> 00:21:01,900 guy, but that's okay. You know what I'm saying? 440 00:21:01,900 --> 00:21:04,500 This is all internally within Google, which is what I find so 441 00:21:04,500 --> 00:21:08,500 fascinating because you have most bizarre thing is that the 442 00:21:08,500 --> 00:21:11,900 person who had to make the decision about whether this made 443 00:21:11,900 --> 00:21:16,100 any sense and make it, you know, are we going public with this? 444 00:21:16,100 --> 00:21:18,100 What do we do? As far as I can tell from the 445 00:21:18,100 --> 00:21:21,300 Tasha story was blaz Guerra are cats. 446 00:21:21,300 --> 00:21:24,000 Right? Who was there the very same 447 00:21:24,000 --> 00:21:24,600 person? Right. 448 00:21:24,600 --> 00:21:28,300 Would said that the things that the ground has shifted beneath 449 00:21:28,300 --> 00:21:30,300 his feet. Every that's just like crazy. 450 00:21:30,300 --> 00:21:32,100 It's just too perfect. It's crazy. 451 00:21:32,100 --> 00:21:36,100 And, you know, we have the Google, PR person, who is all 452 00:21:36,100 --> 00:21:38,700 the record saying that Blake had to be fired because he was 453 00:21:38,700 --> 00:21:41,600 totally off the chain and not fired. 454 00:21:41,600 --> 00:21:43,400 He's put on, administrative is administrative leave. 455 00:21:43,400 --> 00:21:45,000 But you know, clearly has fallen sick. 456 00:21:45,000 --> 00:21:48,400 Seriously out of favor with the company and, you know, his 457 00:21:48,400 --> 00:21:51,100 claims are in actually, I mean, think about how much press the 458 00:21:51,100 --> 00:21:53,200 company got 4 Lambda. They should be giving him a 459 00:21:53,200 --> 00:21:56,800 raise, he liked it mean, seriously, he raised some 460 00:21:56,800 --> 00:21:59,400 interesting questions that we should all think about. 461 00:21:59,700 --> 00:22:03,100 Which are pertain to like, we are going to have systems that 462 00:22:03,100 --> 00:22:05,400 easily fool people. It's amazing that was a Google 463 00:22:05,400 --> 00:22:08,600 engineer that adds a little Force onto the story, whatever. 464 00:22:08,800 --> 00:22:11,200 But like he opened a conversation, we need to have 465 00:22:11,400 --> 00:22:14,300 everybody knows who lammed is and you're going to suspend this 466 00:22:14,300 --> 00:22:16,700 guy like that's not that's not right. 467 00:22:16,800 --> 00:22:19,400 And I don't think I don't honestly think anybody at least 468 00:22:19,400 --> 00:22:22,100 nobody on Twitter. No Savi reader came away 469 00:22:22,100 --> 00:22:26,800 thinking this AI system is intelligent yes or no. 470 00:22:26,900 --> 00:22:28,000 No, that's it. Have a reader. 471 00:22:28,000 --> 00:22:29,800 There's some last Savvy. Readers Did. 472 00:22:29,800 --> 00:22:32,400 But the first problem is, like, you'd have those conversation 473 00:22:32,400 --> 00:22:34,700 that would promise you the moon because it likes doing that 474 00:22:34,700 --> 00:22:37,600 quote likes doing that, right? Because the statistics lead it 475 00:22:37,600 --> 00:22:39,900 that way, it would promise you the moon, and nothing happened. 476 00:22:39,900 --> 00:22:42,500 And the other problem that Lemoine was working on a whole 477 00:22:42,500 --> 00:22:44,600 field is working on that. I don't think can be solved with 478 00:22:44,600 --> 00:22:48,200 the current Paradigm is the toxic language the recommending 479 00:22:48,200 --> 00:22:50,300 harm to self and others and so forth. 480 00:22:50,400 --> 00:22:53,900 So they put this stuff in what we call production you probably 481 00:22:53,900 --> 00:22:56,700 guessed by the know that's her if we put it in production and 482 00:22:56,700 --> 00:22:59,200 just you know threw it out and Alexa the Wilder. 483 00:22:59,600 --> 00:23:02,000 Google assistant or whatever it's called Google home, you 484 00:23:02,000 --> 00:23:05,500 threw it out of the lot wild, they would be like millions of 485 00:23:05,500 --> 00:23:07,800 complaints. You told my child to do this and 486 00:23:07,800 --> 00:23:10,600 right you told me to do this to my mother, which was not 487 00:23:10,600 --> 00:23:13,300 necessarily done any defense and why they don't open it up. 488 00:23:13,300 --> 00:23:16,400 I mean, Dolly. Okay, no, but they don't open it 489 00:23:16,400 --> 00:23:18,200 up to train professionals like me, right? 490 00:23:18,200 --> 00:23:23,900 I mean, I right, you know, I I got a PhD from MIT when I was 23 491 00:23:23,900 --> 00:23:26,600 and did this for 30 years ago, the rest is the you, you know, 492 00:23:26,600 --> 00:23:29,400 publish a blog post and try to make them look silly. 493 00:23:29,500 --> 00:23:32,400 That they don't like, you know, they that they don't want to be 494 00:23:32,400 --> 00:23:36,200 made to look silly by you know finding the terrible cases. 495 00:23:36,200 --> 00:23:39,900 That's right, they don't they don't and so there, you know, if 496 00:23:39,900 --> 00:23:43,400 they wanted to keep their mouths shut test the stuff internally 497 00:23:43,400 --> 00:23:46,300 and release it when it was ready critics, like me, wouldn't have 498 00:23:46,300 --> 00:23:48,800 anything to say about her at least not until it was out, and 499 00:23:48,800 --> 00:23:51,100 then been vetted, but they want to play both sides of it. 500 00:23:51,100 --> 00:23:53,100 They wanted to say, hey, we're scientists. 501 00:23:53,100 --> 00:23:56,300 We have the best scientific teams for studying AI. 502 00:23:56,300 --> 00:23:58,000 In the world. We have deep item with Google 503 00:23:58,000 --> 00:24:01,800 AI, you know, Companies with a lot because we're close to Asia 504 00:24:01,800 --> 00:24:03,900 and AGI is going to be worth the entire economy. 505 00:24:03,900 --> 00:24:07,000 The basically what they're saying in so many words and 506 00:24:07,000 --> 00:24:09,600 they're putting out these articles that look like science 507 00:24:09,600 --> 00:24:13,100 they have bibliographies they you know, citations and they 508 00:24:13,100 --> 00:24:14,900 have charts and tables. They look like they're science. 509 00:24:14,900 --> 00:24:17,900 But then you look carefully and they're missing denominators and 510 00:24:17,900 --> 00:24:19,400 they're not going out for peer review. 511 00:24:19,400 --> 00:24:22,100 So they are portraying themselves as a major 512 00:24:22,100 --> 00:24:25,900 contributor to science but they're not playing the game of 513 00:24:25,900 --> 00:24:27,500 science. The way that the rest of us know 514 00:24:27,500 --> 00:24:31,200 that you have the you must Did you don't you want to help 515 00:24:31,200 --> 00:24:33,400 ultimately with the replicability crisis, which is 516 00:24:33,400 --> 00:24:35,000 what happened? For example, in medicine, were 517 00:24:35,000 --> 00:24:38,100 turned out a whole lot of stuff was published in, really? 518 00:24:38,100 --> 00:24:40,700 Not very good, right? You say, they're not playing the 519 00:24:40,700 --> 00:24:43,400 game of Science, and it seems to me, we're kind of making this 520 00:24:43,400 --> 00:24:46,000 point, is they're playing the game media and they're playing 521 00:24:46,000 --> 00:24:48,700 very effectively. Google was very chaotic AI, in 522 00:24:48,700 --> 00:24:50,300 general, you know, Stephanie eyes. 523 00:24:50,300 --> 00:24:53,200 Also horribleness. They then the name open AI is 524 00:24:53,200 --> 00:24:54,800 just a lie. They say they're open but they 525 00:24:54,800 --> 00:24:56,900 won't they won't, you know, they're not open to people like 526 00:24:56,900 --> 00:24:59,900 me. So I think open a, i Julie 527 00:25:00,400 --> 00:25:04,400 taught this world, how to play the media game and now they do 528 00:25:04,400 --> 00:25:07,400 things like they introduced Dolly by having Sam Altman. 529 00:25:07,400 --> 00:25:09,600 Tweet about it and say, send me some tweets. 530 00:25:09,600 --> 00:25:11,700 I'll show you some stuff which is like the opposite of the 531 00:25:11,700 --> 00:25:13,300 systematic scientifically, you know? 532 00:25:13,308 --> 00:25:15,000 He if he doesn't like the picture, he doesn't have to put 533 00:25:15,000 --> 00:25:16,900 it out. So I saw yesterday like I told 534 00:25:16,900 --> 00:25:18,700 her the truth comes out so so dollies. 535 00:25:18,700 --> 00:25:20,100 Three months old or something like that. 536 00:25:20,100 --> 00:25:24,100 Finally the access is more Broad and somebody posted pictures of 537 00:25:24,100 --> 00:25:27,100 George Michael with his face. This on my Twitter feed this 538 00:25:27,100 --> 00:25:29,400 like grossly distorted like disgusting. 539 00:25:29,500 --> 00:25:31,200 Ugly disgusting the to look at it. 540 00:25:31,200 --> 00:25:34,700 Well, they have had a PR policy that you can't post photos that 541 00:25:34,700 --> 00:25:36,700 are generated by it or people's faces. 542 00:25:36,900 --> 00:25:39,200 Well, now we know why, because there's the story but for three 543 00:25:39,200 --> 00:25:41,400 months it's like look at all the great things that dollar dude. 544 00:25:41,400 --> 00:25:44,200 So like Sam Altman when he tweets about this is not going 545 00:25:44,200 --> 00:25:46,300 to show you a distorted George my dolly. 546 00:25:46,300 --> 00:25:49,000 I disagree with you sort of somewhat strongly on the only 547 00:25:49,000 --> 00:25:52,800 need to produce like 10% interesting and like Dolly isn't 548 00:25:52,800 --> 00:25:56,500 dependency want them like so so another thing I retweeted 549 00:25:56,500 --> 00:25:58,000 yesterday people always send these things to me. 550 00:25:58,000 --> 00:26:01,000 Now was Ali, trying to draw a hexagon. 551 00:26:01,200 --> 00:26:04,400 It just couldn't do it. And so if you have there and if 552 00:26:04,400 --> 00:26:06,600 you like, you know, we want something with seven sides, 553 00:26:06,600 --> 00:26:09,500 forget about it. So like maybe can do hexagons. 554 00:26:09,500 --> 00:26:11,700 There's a few more hexagons out there in his database but there 555 00:26:11,700 --> 00:26:14,700 aren't too many septagon. Zacchaeus is the word for the 556 00:26:14,700 --> 00:26:16,700 fact that I mean Dolly comes off as creative. 557 00:26:16,700 --> 00:26:19,300 Do disagree that dolly is creative and a certain way. 558 00:26:19,300 --> 00:26:20,900 Well there you need to Define your terms. 559 00:26:21,600 --> 00:26:25,100 I'll do the easy part in the hard part it is definitely a 560 00:26:25,100 --> 00:26:28,800 very useful tool for people who are creative with some cabinets 561 00:26:28,800 --> 00:26:30,800 around it. So, So like if you just need an 562 00:26:30,800 --> 00:26:32,400 idea for a book cover, it's awesome. 563 00:26:32,600 --> 00:26:35,900 He needs it, like be doesn't exactly as use case. 564 00:26:35,900 --> 00:26:38,900 I'm trying to get Dolly to redesign our podcast logo and it 565 00:26:38,900 --> 00:26:42,800 might be satisfy you might not. So slate, start codon what? 566 00:26:42,800 --> 00:26:44,500 As well. As Dolly taxes could put a dead 567 00:26:44,500 --> 00:26:46,300 cat listening to a podcast about tech. 568 00:26:46,300 --> 00:26:47,700 Please do that and send it to us. 569 00:26:47,700 --> 00:26:50,200 Yeah. So like slate Stark codex went 570 00:26:50,200 --> 00:26:53,400 through pretty systematically and then we ended up in this 571 00:26:53,400 --> 00:26:56,500 kind of wild debate last week but before we had this wild 572 00:26:56,500 --> 00:26:59,200 debate, he had this nice thing on Dolly. 573 00:26:59,500 --> 00:27:02,200 And, you know, he went through it's like this thing I could do 574 00:27:02,200 --> 00:27:04,900 in this other thing. I really wanted to do I just 575 00:27:04,900 --> 00:27:08,100 couldn't get done in the thing. I retweeted yesterday is like 576 00:27:08,100 --> 00:27:11,000 that too. So you know, for commercial 577 00:27:11,000 --> 00:27:13,600 artist trying to do something for a client, it would be like a 578 00:27:13,600 --> 00:27:17,700 good source of ideas, but you couldn't count on it because 579 00:27:17,700 --> 00:27:21,200 it's a little bit while but, you know really powerful and so that 580 00:27:21,200 --> 00:27:25,000 really does depends on your use case, is it creative that 581 00:27:25,000 --> 00:27:27,800 depends on you know, how you define creativity? 582 00:27:27,800 --> 00:27:30,500 So like at some level like you Look at the algorithm and it's 583 00:27:30,500 --> 00:27:33,300 like just doing the math, I'm going to some level, it's like, 584 00:27:33,400 --> 00:27:34,900 pretty amazing what it comes up with. 585 00:27:34,900 --> 00:27:36,600 And so, then it's a matter, right? 586 00:27:36,600 --> 00:27:39,100 I guess, I knit, this is not maybe how I Define it for 587 00:27:39,100 --> 00:27:43,400 humans, but I certainly think that if you had an art contest 588 00:27:43,400 --> 00:27:45,900 and said, you know, we're going to tell the judges to judge it 589 00:27:45,900 --> 00:27:50,100 based on what creative output is Dolly, would be plenty of 590 00:27:50,300 --> 00:27:53,200 humans. If people don't most that, you 591 00:27:53,200 --> 00:27:55,500 probably wouldn't be the best. Like sure. 592 00:27:55,500 --> 00:28:00,000 It's not gonna come up with truly right new ideas but you It 593 00:28:00,000 --> 00:28:01,800 incredible with things like lighting. 594 00:28:01,800 --> 00:28:06,300 But quite quite the standard we impose on a I mean again it's 595 00:28:06,300 --> 00:28:08,000 sort of like is it better than the best know? 596 00:28:08,000 --> 00:28:11,000 It's I mean it's way better, it's way better artist that I 597 00:28:11,000 --> 00:28:12,800 will ever be. You know, there's no chance 598 00:28:12,800 --> 00:28:15,400 there will ever be as good. You know, the only things I 599 00:28:15,400 --> 00:28:17,500 could be there are like Specific Instructions. 600 00:28:17,500 --> 00:28:20,500 So if you wanted a blue cube on a Red Cube I could do that would 601 00:28:20,500 --> 00:28:23,300 be great and Dolly like half the time. 602 00:28:23,300 --> 00:28:25,300 The Red Cube would be on the blue cube enough in time the 603 00:28:25,300 --> 00:28:28,800 other way around. Like so you know, I am much 604 00:28:28,800 --> 00:28:31,200 better at Natural language, understanding the dolly and it 605 00:28:31,200 --> 00:28:34,600 is much better at lighting and compete in compositing, putting 606 00:28:34,600 --> 00:28:37,500 images in front of each other. You know, it's really great at 607 00:28:37,500 --> 00:28:40,300 that is what is this? I wanted to go back to the 608 00:28:40,300 --> 00:28:43,000 Washington Post story, Natasha Story. 609 00:28:43,300 --> 00:28:46,800 I mean, she's sort of seemed to know that she was going to 610 00:28:46,800 --> 00:28:51,000 create sort of a debate over something where the experts 611 00:28:51,000 --> 00:28:54,500 would come down on the side of this isn't sentient, and I think 612 00:28:54,500 --> 00:28:57,500 there are questions that we could get into of whether we've 613 00:28:57,500 --> 00:28:59,300 sort of hinted at whether the guy. 614 00:28:59,500 --> 00:29:02,700 Whistleblower on this really believes it sentient or he just 615 00:29:02,700 --> 00:29:06,300 wants to sort of advance, so I think he really does believe so. 616 00:29:06,500 --> 00:29:10,100 So the journalist, Steven Levy, you accused him of falling in 617 00:29:10,100 --> 00:29:13,300 love with Lambda. Yeah, and I think he did, I mean 618 00:29:13,300 --> 00:29:15,400 I haven't talked to him first hand, but Steven Levy talked to 619 00:29:15,400 --> 00:29:20,800 him last night, the guys, on honeymoon, Blake Lemoine is, is 620 00:29:20,800 --> 00:29:24,400 is on honeymoon, but Levi tracked, him down levees. 621 00:29:24,400 --> 00:29:28,200 Fantastic journalist wrote hackers was you know, like the 622 00:29:28,200 --> 00:29:30,500 book that got me in it. I'm excited about computer at 623 00:29:30,500 --> 00:29:33,300 least and also a very positive reporter who's certainly likes 624 00:29:33,300 --> 00:29:36,400 to boost. But anyway, yeah, you have to 625 00:29:36,400 --> 00:29:38,400 hold both in your mind at once but yes. 626 00:29:38,400 --> 00:29:41,300 So I talked to Levi, Levi little bit last night. 627 00:29:41,300 --> 00:29:43,600 I'm Levi wrote a story today. I'm quoted in it. 628 00:29:43,600 --> 00:29:46,500 Yeah, we had a little back and forth, so Levi actually tracked 629 00:29:46,500 --> 00:29:49,700 lumoid down after this story broke, which, you know, Lemoine 630 00:29:49,700 --> 00:29:51,700 is like, not taking calls. He's like, I'm on my honeymoon. 631 00:29:51,700 --> 00:29:55,600 I think actually got married on the day, maybe my story came out 632 00:29:55,600 --> 00:29:58,100 like the day after the Natasha story came out or two days 633 00:29:58,100 --> 00:30:03,600 after, and You Levi did myself his own self report, his best he 634 00:30:03,600 --> 00:30:06,600 could to try to, you know, see if this guy was just like 635 00:30:06,600 --> 00:30:09,200 shocking, everybody and came away. 636 00:30:09,200 --> 00:30:12,000 Pretty convinced that Lemoine believes what he says. 637 00:30:12,300 --> 00:30:16,900 And in support of that is this 2018, YouTube video that people 638 00:30:16,900 --> 00:30:19,500 might want to watch where he argues that a eyes could be 639 00:30:19,500 --> 00:30:22,000 people and so forth. So he was predisposed to believe 640 00:30:22,000 --> 00:30:24,300 this in either. He's playing like the longest 641 00:30:24,300 --> 00:30:26,600 con ever. Like, you know, he thought for 642 00:30:26,600 --> 00:30:28,900 years ago, I'm going to get myself on the Washington Post by 643 00:30:28,900 --> 00:30:30,900 Brick, No. I mean it's just not plausible, 644 00:30:30,900 --> 00:30:33,600 right? He I think he really does 645 00:30:33,600 --> 00:30:37,800 sincerely believe that he is speaking up for the machine. 646 00:30:38,100 --> 00:30:40,400 Like, I think he's sincere about that. 647 00:30:40,800 --> 00:30:43,600 I don't think he's bluffing. I mean, he has a some religious 648 00:30:43,600 --> 00:30:47,600 beliefs that doing in someone. We let people believe in God 649 00:30:47,900 --> 00:30:51,200 based on reasons that a lot of people would say were were bad 650 00:30:51,200 --> 00:30:54,500 and we sort of like as a society accept that. 651 00:30:54,500 --> 00:30:57,400 And so, to some degree, if people want to also come up with 652 00:30:57,400 --> 00:31:01,600 a sort of non-scientific way of In the AI is sentient. 653 00:31:01,600 --> 00:31:04,700 Like if we apply the same rules of God, like we're sort of 654 00:31:04,700 --> 00:31:06,900 screwed here. I don't, I don't think we have 655 00:31:06,900 --> 00:31:09,800 to just like accept people's like own version of Reason. 656 00:31:09,800 --> 00:31:12,500 It that's not useful in the scientific Community is really I 657 00:31:12,500 --> 00:31:14,800 mean here's the other reason, I think this is also interesting 658 00:31:15,000 --> 00:31:19,800 is Lemoine is like now an icon in a way but he's not unique. 659 00:31:19,900 --> 00:31:23,000 Lots of people are going to interact with these systems and 660 00:31:23,000 --> 00:31:25,800 feel as he did in my view, they will be wrong. 661 00:31:25,800 --> 00:31:29,000 You'll be attributing, you know, awareness to A system that does 662 00:31:29,000 --> 00:31:32,200 not Have that maybe some future system will have a kind of 663 00:31:32,200 --> 00:31:35,700 awareness and be intelligent in a way that they think this 664 00:31:35,700 --> 00:31:40,600 machine is and it's not but you know, already like the certain 665 00:31:40,600 --> 00:31:43,700 way which we're very cultural Centric gear, few people over 666 00:31:43,700 --> 00:31:46,400 here in North America know that in China. 667 00:31:46,400 --> 00:31:49,100 They've had a system for 45 years called Xiao Weis. 668 00:31:49,100 --> 00:31:51,100 People fall in love with it, Xiao Weis. 669 00:31:51,100 --> 00:31:55,300 Is a more primitive chat by but not entirely different. 670 00:31:55,400 --> 00:31:57,800 In fact, the newest version of show has probably uses some 671 00:31:57,800 --> 00:31:59,700 large language models in there. Silly. 672 00:31:59,700 --> 00:32:03,300 If it didn't, and you people fall in love with this people. 673 00:32:03,300 --> 00:32:06,600 Also fall in love with plants and cats and you know, sure it's 674 00:32:06,600 --> 00:32:09,000 going to happen more. So there's a way in which the 675 00:32:09,000 --> 00:32:11,300 story is like a canary in a coal mine. 676 00:32:11,300 --> 00:32:14,400 So it's wacky that a Google engineer thinks this. 677 00:32:14,400 --> 00:32:17,000 But you know, millions of people are going to think that. 678 00:32:17,000 --> 00:32:19,400 I mean, I think the debate Natasha wanted us to have, which 679 00:32:19,400 --> 00:32:21,400 I don't think is really what most people arguing about is 680 00:32:21,400 --> 00:32:25,700 whether companies should tried, whether it's good to companies, 681 00:32:25,700 --> 00:32:28,700 make a eyes appear, like humans or in some ways, they should 682 00:32:28,700 --> 00:32:30,600 make it. The AI is talk in a way that 683 00:32:30,600 --> 00:32:32,500 makes it very clear that they're not humans. 684 00:32:32,500 --> 00:32:36,800 I mean, if you have a point of view on that, I don't know what 685 00:32:36,800 --> 00:32:39,700 that would look like, that's complicated, it might depend on 686 00:32:39,700 --> 00:32:42,900 the use case, I'm not sure there's an absolute answer, you 687 00:32:42,900 --> 00:32:47,700 know, some of it is like you know, cigarettes and you know 688 00:32:47,700 --> 00:32:51,200 having truth in labeling and like I'm not sure the answer. 689 00:32:51,200 --> 00:32:53,500 I think we need a lot of people to actually think about this 690 00:32:53,500 --> 00:32:56,400 question people in ethics and policy means and so forth. 691 00:32:56,400 --> 00:33:00,900 Like one option would be, you make it very clear to people 692 00:33:00,900 --> 00:33:04,300 that, you know, this is in some sense, an illusion. 693 00:33:04,300 --> 00:33:06,100 Maybe find a polite way to say that. 694 00:33:06,700 --> 00:33:08,700 Don't take it too, seriously, but enjoy it. 695 00:33:09,000 --> 00:33:11,600 There are use cases, where maybe it'd be. 696 00:33:11,600 --> 00:33:13,800 Okay? Like as a companion, as long as 697 00:33:13,800 --> 00:33:15,600 you know what you're getting into. 698 00:33:15,600 --> 00:33:17,700 Like you know, we're not going to tell people not to have 699 00:33:17,700 --> 00:33:21,100 stuffed animals, right? I mean stuffed animals, give a 700 00:33:21,100 --> 00:33:23,600 sense of intimacy and Warren. Right. 701 00:33:23,600 --> 00:33:26,800 Cuddle them and like I'm not here to tell people they can't 702 00:33:26,800 --> 00:33:29,100 have stuffed animals in it along some sense. 703 00:33:29,800 --> 00:33:34,800 It's kind of like that and it's also like a drug and it like I 704 00:33:34,800 --> 00:33:38,400 can see how it's really And people might lose their control. 705 00:33:38,400 --> 00:33:41,000 So most people can walk away from their stuffed animals, but 706 00:33:41,000 --> 00:33:42,900 they can't walk away from heroin heroin. 707 00:33:42,900 --> 00:33:46,600 Once they start it and it might be pretty hard for people to 708 00:33:46,600 --> 00:33:48,600 walk away from these things especially as they get better. 709 00:33:48,900 --> 00:33:52,200 I think, right now, what Lemoine does, it represent is how 710 00:33:52,300 --> 00:33:55,200 awfully dumb these systems can be and how much they can forget 711 00:33:55,200 --> 00:33:58,500 what you told them. Like if you just put the current 712 00:33:58,500 --> 00:34:02,200 stuff out on the street, mmm, people might eventually get 713 00:34:02,200 --> 00:34:04,000 frustrated. Like, there's a huge novelty 714 00:34:04,000 --> 00:34:06,000 effect like a first, it's like, oh my God, I can't believe it. 715 00:34:06,500 --> 00:34:08,600 Does this. But at the same thing with Dolly 716 00:34:08,600 --> 00:34:10,400 like it's some point. You're like I want it to do this 717 00:34:10,400 --> 00:34:11,900 and it just doesn't really do it. 718 00:34:12,100 --> 00:34:14,300 And it might, the efficacy thing I talked about might also be a 719 00:34:14,308 --> 00:34:17,100 problem, like, it tells you I've got to do this and it doesn't 720 00:34:17,100 --> 00:34:18,699 deliver. It's a like, some people there 721 00:34:18,699 --> 00:34:21,000 might be some frustration Factor, but I think they, you 722 00:34:21,000 --> 00:34:24,100 know, they're addicting, I actually just wrote a poem. 723 00:34:24,199 --> 00:34:27,199 I did a riff on Howell. I'm going to put this out later 724 00:34:27,199 --> 00:34:29,000 today. Allen ginsberg's, Paul Allen 725 00:34:29,000 --> 00:34:31,600 gets her a poem howl, which was like I saw the best minds of my 726 00:34:31,600 --> 00:34:36,100 generation wasting time on Dolly and GPT 3 and so forth. 727 00:34:36,199 --> 00:34:38,300 Fourth Reich. Well I mean you know what's 728 00:34:38,300 --> 00:34:41,600 funny is that almost as kind of a riff also on the is it Marc 729 00:34:41,600 --> 00:34:44,600 Andreessen or Windows PC line? That said like we were promised 730 00:34:44,600 --> 00:34:48,300 hoverboards and instead we got and the name whatever kind of it 731 00:34:48,300 --> 00:34:50,500 was tle said she'll yeah I'm sorry. 732 00:34:51,100 --> 00:34:54,900 We got we we were promised flying cars and got 140 733 00:34:54,900 --> 00:34:57,900 characters. Yeah, I mean you could that you 734 00:34:57,900 --> 00:35:00,600 could definitely Riff on that for AI Jenna like we were 735 00:35:00,600 --> 00:35:03,000 promised the Star Trek computer that would actually solve our 736 00:35:03,000 --> 00:35:05,900 problems and be trustworthy and reliable and help us. 737 00:35:06,200 --> 00:35:09,300 Even with climate change and what we have are these kind of 738 00:35:09,300 --> 00:35:12,500 like, sociopathic companions that pretend to like us. 739 00:35:12,800 --> 00:35:14,600 That's what we got. But you think it's a waste of 740 00:35:14,600 --> 00:35:16,200 time. I want to push back on that. 741 00:35:16,300 --> 00:35:17,300 That's what you said. Right. 742 00:35:17,500 --> 00:35:20,600 Do I think this research is a waste yarn or the time people 743 00:35:20,600 --> 00:35:23,600 spend the, I do with some level and that requires some 744 00:35:23,600 --> 00:35:26,700 explanation. So, in my view, these things are 745 00:35:26,700 --> 00:35:29,700 working because there are statistical approximations to 746 00:35:29,700 --> 00:35:32,800 things that we actually need and they're very seductive, the very 747 00:35:32,800 --> 00:35:34,800 easy to work with, but they're not. 748 00:35:34,800 --> 00:35:37,200 I think the answer that we're actually You looking for and so 749 00:35:37,207 --> 00:35:40,200 people are spending more and more time and money on something 750 00:35:40,200 --> 00:35:42,500 that I think has no great future. 751 00:35:42,500 --> 00:35:45,300 It might have it might play a role in the future but I think 752 00:35:45,300 --> 00:35:48,200 they were really hard questions in artificial intelligence that 753 00:35:48,200 --> 00:35:51,800 we need to answer that are not getting answered because it's 754 00:35:51,800 --> 00:35:55,800 too fun to play with these systems and it's sucking all of 755 00:35:55,800 --> 00:35:59,000 the money and oxygen away from other things. 756 00:35:59,000 --> 00:36:01,500 So I've seen before, in my career, that I've been doing 757 00:36:01,500 --> 00:36:05,000 this for 30, some years where a new idea gets popular and old 758 00:36:05,000 --> 00:36:08,900 ideas that are actually And get abandoned and to certain extent 759 00:36:08,900 --> 00:36:10,900 that is happening now. So I saw that with cognitive 760 00:36:10,900 --> 00:36:13,700 Neuroscience. All these fmri pictures that you 761 00:36:13,700 --> 00:36:16,300 probably saw when you guys were kids about like the brain is 762 00:36:16,300 --> 00:36:19,700 lighting up and stuff like that, it took away most of the energy 763 00:36:19,700 --> 00:36:22,400 in cognitive psychology and what is it actually shown us? 764 00:36:22,800 --> 00:36:24,200 Not that much. We have a bunch of pretty 765 00:36:24,200 --> 00:36:27,000 pictures, but we still don't really know how the brain works. 766 00:36:27,100 --> 00:36:29,600 It didn't really teach us that much more about cognitive saw 767 00:36:29,700 --> 00:36:33,800 psychology, but it was seductive and it took money, and you don't 768 00:36:33,800 --> 00:36:36,700 think we get the neural net big enough and then One day it's a 769 00:36:36,707 --> 00:36:40,400 brain and it feels things like it does feel like, yeah, I don't 770 00:36:40,400 --> 00:36:42,800 you sort of AI world. There's, we need to be careful 771 00:36:42,800 --> 00:36:45,200 on that. If the server is get big enough, 772 00:36:45,200 --> 00:36:48,300 you know, it will it will work. What would your approach be? 773 00:36:48,300 --> 00:36:52,200 So I think that we need to first of all look to classical AI 774 00:36:52,200 --> 00:36:54,600 which is out of favor and borrow a few ideas from there. 775 00:36:54,900 --> 00:36:59,200 One is the idea of symbols and propositions sentences kind of 776 00:36:59,200 --> 00:37:03,000 verbal structures, databases things like that are actually 777 00:37:03,000 --> 00:37:05,400 tremendously useful. We still write all this world 778 00:37:05,400 --> 00:37:10,200 software That there's a few use cases that are very sexy with 779 00:37:10,200 --> 00:37:12,100 deep learning. But most software we actually 780 00:37:12,100 --> 00:37:13,400 right where there's a database in. 781 00:37:13,400 --> 00:37:15,200 You update records and things like that. 782 00:37:15,300 --> 00:37:17,500 And these two approaches right now, we're not compatible and 783 00:37:17,500 --> 00:37:20,200 that's a problem. And a lot of people in the field 784 00:37:20,200 --> 00:37:21,500 actually are starting to see this. 785 00:37:21,500 --> 00:37:25,000 That if you can't update a set of Records about the things in 786 00:37:25,000 --> 00:37:27,600 the world that you are talking about, at the end of the day, 787 00:37:27,600 --> 00:37:30,900 you can't be that efficacious, and you can't be that reliable. 788 00:37:31,300 --> 00:37:36,900 So we need to kind of merge, the older tradition of Symbolic AI 789 00:37:36,900 --> 00:37:40,600 with the neural network stuff, I think it's really hopeless until 790 00:37:40,600 --> 00:37:43,600 we do that until we do that we are always going to get systems 791 00:37:43,600 --> 00:37:46,000 that say that Bessie will be alive again. 792 00:37:46,000 --> 00:37:48,600 And I months if you just let her have a baby or something like 793 00:37:48,600 --> 00:37:51,900 that you just are fundamentally dis comprehending the world. 794 00:37:51,900 --> 00:37:53,800 I don't think that that will be solved with more data. 795 00:37:53,800 --> 00:37:57,700 You think the big tech companies are being largely disingenuous 796 00:37:57,700 --> 00:37:59,300 about the state of their technology. 797 00:37:59,300 --> 00:38:02,300 I mean, you've worked with in Uber, I think a lot of them have 798 00:38:02,300 --> 00:38:05,200 drunk the Kool-Aid and I think the problem is most of them 799 00:38:05,200 --> 00:38:08,200 don't know. Science and they have this tool 800 00:38:08,200 --> 00:38:12,900 that works like 85% well because they've not really studied 801 00:38:12,900 --> 00:38:15,400 Linguistics. They've not really studied 802 00:38:15,400 --> 00:38:19,400 philosophy of mind. They don't understand how hard 803 00:38:19,400 --> 00:38:22,700 certain problems are and they come in with the Steamrollers 804 00:38:23,000 --> 00:38:24,700 and they think that they're solving the problems. 805 00:38:24,700 --> 00:38:27,300 And they're just not. I'll give you an example of how 806 00:38:27,300 --> 00:38:30,400 G PT 3 is just fundamentally misguided people. 807 00:38:30,400 --> 00:38:34,300 In language know that what you do is you have a set of words 808 00:38:34,700 --> 00:38:37,200 that is arranged in order. ER, and you deriving, meaning 809 00:38:37,200 --> 00:38:40,100 from that. So, most basic thing, anybody 810 00:38:40,100 --> 00:38:42,600 who had a Linguistics course, can tell you that and these 811 00:38:42,600 --> 00:38:46,300 systems don't really do that. And they, you know, people talk 812 00:38:46,300 --> 00:38:49,600 about interpretability well that's like jargony way of 813 00:38:49,600 --> 00:38:52,000 saying we've no idea what the system is really doing or why, 814 00:38:52,200 --> 00:38:54,500 but it's also a reflection of the fact that there's no real 815 00:38:54,500 --> 00:38:57,700 what we call semantics there. And from the perspective of 816 00:38:57,700 --> 00:38:59,900 someone who's worked in cognitive science, it's just 817 00:39:00,600 --> 00:39:04,200 it's just bizarre that this much effort goes into a system that 818 00:39:04,200 --> 00:39:07,800 just looks like it's not doing the I think I don't know how to 819 00:39:08,500 --> 00:39:10,300 explain where I'm not alone in thinking. 820 00:39:10,300 --> 00:39:12,600 This one of the rhetorical things that's happened in the 821 00:39:12,607 --> 00:39:15,500 last couple of months is I wrote a piece called, Deep learning is 822 00:39:15,500 --> 00:39:18,800 hitting a wall and it pissed off a lot of people. 823 00:39:18,800 --> 00:39:21,100 But I think what I said was true, they a case. 824 00:39:21,100 --> 00:39:23,400 It made me kind of the poster boy for the opposition. 825 00:39:23,400 --> 00:39:27,000 So now there is sort of good for me and sort of bad. 826 00:39:27,000 --> 00:39:30,300 It's a mixed blessing. Now, anytime somebody wants to 827 00:39:30,500 --> 00:39:33,200 attack the other side, they described it as if it was just 828 00:39:33,200 --> 00:39:35,900 me and they don't mention my collaborator or any day. 829 00:39:36,000 --> 00:39:38,100 Avis, who's an author on nearly all of the papers? 830 00:39:38,100 --> 00:39:42,300 There's your view more similar to what the human brain looks 831 00:39:42,300 --> 00:39:44,400 like or less. Like, do you think is Europe? 832 00:39:44,400 --> 00:39:47,200 We have no freaking idea. Let me be honest on that one. 833 00:39:47,200 --> 00:39:51,300 So so there is a theory that what you need to do to solve a 834 00:39:51,300 --> 00:39:54,400 is to make a model that is based on the brain. 835 00:39:54,400 --> 00:39:57,400 Right, there are two problem or it would seem to be a way to 836 00:39:57,400 --> 00:40:00,300 solve it at least well. So actually, the 31 problem with 837 00:40:00,300 --> 00:40:02,000 that is we have no idea how the brain works. 838 00:40:02,000 --> 00:40:06,000 So we have a lot of data but we have no real Theory, my guess is 839 00:40:06,100 --> 00:40:07,500 Could have to go the other way around. 840 00:40:07,500 --> 00:40:11,200 We have to solve AI in order to be able to make an automated 841 00:40:11,200 --> 00:40:14,800 reasoning, scientific induction system that can deal with having 842 00:40:14,800 --> 00:40:17,900 80 billion neurons and our many trillions connections between 843 00:40:17,900 --> 00:40:20,200 them and so forth. So one is like we just don't 844 00:40:20,200 --> 00:40:25,100 have the goods to actually do this and to is like we know that 845 00:40:25,100 --> 00:40:27,600 there are huge holes in what we know about Neuroscience. 846 00:40:27,600 --> 00:40:30,500 I'll give you one example. We all have short-term memory 847 00:40:30,500 --> 00:40:33,000 where I can tell you something once you can remember for a 848 00:40:33,008 --> 00:40:34,900 little bit. So if I told you at the end of 849 00:40:34,908 --> 00:40:36,700 the call I'll give you Thousand dollars. 850 00:40:36,700 --> 00:40:39,000 If you can remember this sentence you know I will have 851 00:40:39,000 --> 00:40:40,800 your attention and you remember it, right? 852 00:40:41,000 --> 00:40:43,800 We have no idea how the brain knows that read all the stuff we 853 00:40:43,800 --> 00:40:46,600 know about memory and brains is like you practice something 854 00:40:46,600 --> 00:40:49,100 three thousand times and you get a little bit better at it each 855 00:40:49,100 --> 00:40:52,300 time and that kind of memory exists, it's real, but there's 856 00:40:52,300 --> 00:40:54,300 other kind of memory exists and is in critical. 857 00:40:54,300 --> 00:40:57,000 Every time you parse a sentence every time you understand a 858 00:40:57,008 --> 00:40:59,500 sentence, you're actually using short term memory in order to 859 00:40:59,500 --> 00:41:03,000 understand that sentence and develop our own, we have no idea 860 00:41:03,000 --> 00:41:05,600 how the brain does that. Then the other thing is like, we 861 00:41:05,600 --> 00:41:09,000 know a little Little bit about like how maybe a monkey brain 862 00:41:09,000 --> 00:41:12,500 works but we don't know anything really about how language works 863 00:41:12,500 --> 00:41:15,700 and what makes it such an interesting species is that we 864 00:41:15,700 --> 00:41:18,800 could talk and we can transmit so much culture that way and so 865 00:41:18,800 --> 00:41:20,500 forth. And that part like we don't have 866 00:41:20,500 --> 00:41:24,700 animal models of that. We can't like cut up some other 867 00:41:24,700 --> 00:41:26,600 animal that we don't feel too guilty about not that I'm 868 00:41:26,600 --> 00:41:29,100 endorsing that but like, it's just not ethical. 869 00:41:29,100 --> 00:41:31,700 We don't have an ethical substrate to do the narrow side. 870 00:41:31,900 --> 00:41:34,000 So we in the end of the day we just don't know. 871 00:41:34,000 --> 00:41:37,800 Enough Neuroscience. It is it is Is possible that the 872 00:41:37,800 --> 00:41:41,000 human brain or future. Artificial intelligence is just 873 00:41:41,000 --> 00:41:45,300 a far more complex neural net that starts to understand like 874 00:41:45,600 --> 00:41:48,700 rules and preferences those rules out after pattern matiush 875 00:41:48,700 --> 00:41:51,000 matching. And if that's the case, won't we 876 00:41:51,000 --> 00:41:54,400 feel sort of dumb for being so, condescending to the step? 877 00:41:54,400 --> 00:41:57,600 It's at now, you know, it's like it's on the minute, I don't see 878 00:41:57,607 --> 00:42:00,000 it that way. At all, I would flip it around 879 00:42:00,000 --> 00:42:03,800 and say that the neural networks that we know how to build now 880 00:42:03,900 --> 00:42:07,500 are so vastly. Simplified compared to the ones 881 00:42:07,500 --> 00:42:09,900 that we want, right? It is ridiculous that we're 882 00:42:09,908 --> 00:42:13,100 taking them seriously. So, you know, can I just give a 883 00:42:13,107 --> 00:42:16,500 couple examples, we know that there are about 1,000 plus or 884 00:42:16,500 --> 00:42:19,300 minus kinds of neurons in the brain are neural networks. 885 00:42:19,300 --> 00:42:21,000 Basically have one kind of neuron. 886 00:42:21,200 --> 00:42:23,600 We know that at every synapse there, like 500 different 887 00:42:23,600 --> 00:42:26,200 proteins, there's nothing even capturing that at all. 888 00:42:26,200 --> 00:42:29,000 In our neural networks, we know that there's an enormous amount 889 00:42:29,000 --> 00:42:31,700 of intrinsic innate organization to the braid. 890 00:42:31,700 --> 00:42:33,500 There's hardly any to order all that work. 891 00:42:33,500 --> 00:42:35,900 So yes, the ultimate answer for us. 892 00:42:36,000 --> 00:42:38,900 Anyway is a neural network, but the neural network for us is 893 00:42:38,900 --> 00:42:41,700 this incredibly complicated piece of Machinery? 894 00:42:42,000 --> 00:42:45,600 The things that we have are so grossly, simplified than like, 895 00:42:45,600 --> 00:42:47,100 why should we expect that the in? 896 00:42:47,100 --> 00:42:48,600 You? No one has anything to do with 897 00:42:48,600 --> 00:42:51,200 the other. I think that one of the reasons 898 00:42:51,200 --> 00:42:53,700 that the media and the public is so susceptible to these 899 00:42:53,700 --> 00:42:57,900 particular story cycles and phenomenons and and desires to, 900 00:42:57,900 --> 00:43:00,800 as Natasha says, see, the ghost in the machine is not just 901 00:43:00,800 --> 00:43:03,600 because of some human impulse to anthropomorphize things because 902 00:43:03,600 --> 00:43:07,400 I do truly believe that we are Very far away from the science 903 00:43:07,400 --> 00:43:10,000 fiction future that a lot of people expected at this point. 904 00:43:10,000 --> 00:43:13,400 In time, we talked about the flying cars, self-driving cars, 905 00:43:13,400 --> 00:43:16,500 you know, self-aware, neural networks or whatever mean he's 906 00:43:16,500 --> 00:43:19,000 maybe set it in your story about, you know, we've hit a 907 00:43:19,008 --> 00:43:23,000 wall with deep learning but, you know, a lot of the promises that 908 00:43:23,008 --> 00:43:26,000 we've expected just haven't materialized in the way that we 909 00:43:26,000 --> 00:43:28,000 want. And so it sort of easier for 910 00:43:28,000 --> 00:43:32,200 people to kind of assume great leaps have taken place already. 911 00:43:32,200 --> 00:43:35,600 We aren't even recognized them it when in fact they're so far 912 00:43:35,600 --> 00:43:38,400 and we've I've kind of reached his kind of inching along maybe 913 00:43:38,400 --> 00:43:42,200 impressive inches that that you and others are involved with of 914 00:43:42,200 --> 00:43:45,900 advancing Ai and other Technologies, self-driving as it 915 00:43:45,900 --> 00:43:50,400 is but true Promises of the kinds of things that we want to 916 00:43:50,400 --> 00:43:52,400 just are there and will be there for decades. 917 00:43:52,600 --> 00:43:55,900 And instead be kind of just have story time where we anoint 918 00:43:55,900 --> 00:43:59,300 certain things as you know the next era when in fact it's just 919 00:43:59,300 --> 00:44:02,100 not even close. I mean it does that sort of its 920 00:44:02,100 --> 00:44:06,200 way to it's even more complicated than that because I 921 00:44:06,200 --> 00:44:08,200 think the underlying problem is a lot of people have is they 922 00:44:08,200 --> 00:44:10,200 think AI is Magic. They don't quite know what it is 923 00:44:10,200 --> 00:44:13,000 and they think that whatever it is, it's sort of a universal 924 00:44:13,000 --> 00:44:16,200 Elixir. The reality is it's just a bag 925 00:44:16,200 --> 00:44:19,400 of engineering tools and we probably need a bigger bag of 926 00:44:19,400 --> 00:44:22,600 tools and we probably use all the ones that we have now and 927 00:44:22,600 --> 00:44:24,600 we'll use some others and, you know, we'll eventually we'll 928 00:44:24,600 --> 00:44:28,100 muddle through all of this. But what's hard to grasp, if you 929 00:44:28,100 --> 00:44:31,700 haven't studied, the cognitive Sciences is how many different 930 00:44:31,700 --> 00:44:35,600 components there are two doing good thinking, and it's a little 931 00:44:35,600 --> 00:44:38,800 hard to To grasp that A system can be good at one thing and 932 00:44:38,800 --> 00:44:41,100 terrible at another. I mean maybe if you you know 933 00:44:41,100 --> 00:44:44,400 metaphor might be like you can find someone who's really good 934 00:44:44,400 --> 00:44:48,200 at putting tile, you know as a backsplash and a kitchen and 935 00:44:48,200 --> 00:44:51,200 maybe that person's not so good at doing crossword. 936 00:44:51,200 --> 00:44:53,800 Puzzles, right. Like you know people can have 937 00:44:54,300 --> 00:44:57,900 different kinds of expertise. Well, the machines we have now 938 00:44:57,900 --> 00:44:59,200 have different kinds of expertise. 939 00:44:59,200 --> 00:45:01,300 We know how to build a machine. That's really good at. 940 00:45:01,300 --> 00:45:04,400 Go, we know how to miss make a machine that can be pretty good 941 00:45:04,400 --> 00:45:05,800 at pictures. We just don't know. 942 00:45:06,000 --> 00:45:08,000 To make machine that really understands language. 943 00:45:08,000 --> 00:45:10,500 We only know how to make machine that gives that illusion. 944 00:45:10,700 --> 00:45:13,800 And it's this kind of textured mixed bag. 945 00:45:13,800 --> 00:45:16,200 People, you know, want a one-liner are they smarter of 946 00:45:16,200 --> 00:45:18,900 the dumb? Well, it's neither, you know, 947 00:45:18,900 --> 00:45:21,800 they're smart at some things and Incredibly dumb, it others, and 948 00:45:22,200 --> 00:45:25,000 it's hard to accept, but most of the business World implicitly 949 00:45:25,000 --> 00:45:28,100 agrees with you, right? I mean you know, generalized AI 950 00:45:28,100 --> 00:45:31,500 obviously fails at what you're saying but most of most business 951 00:45:31,500 --> 00:45:33,500 applications. Yeah, they're just trying to use 952 00:45:33,800 --> 00:45:37,400 huge data sets to solve very The problems that they have and they 953 00:45:37,400 --> 00:45:40,000 have no interest but you see the naivety of business. 954 00:45:40,000 --> 00:45:43,900 So so I have seen some massive, massive companies make weird 955 00:45:43,900 --> 00:45:46,000 bets on a I would look to me like weird bat. 956 00:45:46,000 --> 00:45:50,100 So I said back in 2016 the driverless cars are much harder 957 00:45:50,100 --> 00:45:53,600 than you guys think they are. And since I said that they're 958 00:45:53,600 --> 00:45:57,000 probably been a hundred billion dollars point forward into it in 959 00:45:57,008 --> 00:46:00,900 terms of R&D costs and so forth and so far the only money that 960 00:46:00,900 --> 00:46:03,700 is coming from that is the elevation in the price of Tesla 961 00:46:04,300 --> 00:46:07,500 and he could make some argument that To self-driving is improved 962 00:46:07,500 --> 00:46:11,000 somewhat I guess but I mean we're not close to level 5 963 00:46:11,000 --> 00:46:13,900 self-driving like that's just it's not really happening. 964 00:46:13,900 --> 00:46:16,000 We can talk about that if you want to thought about a lot on 965 00:46:16,000 --> 00:46:17,500 the media point, that is the media. 966 00:46:17,500 --> 00:46:21,000 I think the reporter is if you pulled reporters throughout the 967 00:46:21,000 --> 00:46:24,600 whole period would have said that they don't think it's close 968 00:46:24,600 --> 00:46:26,700 and yet the store, it's just interesting. 969 00:46:26,700 --> 00:46:29,100 Like to me stories, I'll made it sound like it was a minute. 970 00:46:29,100 --> 00:46:31,800 Maybe this makes it a worse failing on the part of part of 971 00:46:31,800 --> 00:46:34,200 reporters that, yeah, that somehow the stories come out 972 00:46:34,200 --> 00:46:36,900 positive but most reporters Themselves, I think over 973 00:46:36,900 --> 00:46:40,200 cocktails would be skeptical and I don't really understand. 974 00:46:40,200 --> 00:46:43,100 I think it's just what its public consumption desired a 975 00:46:43,100 --> 00:46:45,000 should hit me up, I'll give him some quotes. 976 00:46:45,100 --> 00:46:48,100 I mean, I do. I mean, like, Sam she'd and CNBC 977 00:46:48,100 --> 00:46:52,300 came to me when Optimus was announced and, you know, I gave 978 00:46:52,300 --> 00:46:55,400 him the quotes, he to give the other side and say, look, you 979 00:46:55,400 --> 00:46:58,100 know, there's something that's interesting about Optimist, but 980 00:46:58,100 --> 00:47:00,500 this is a really hard problem. It's much harder than musk has 981 00:47:00,500 --> 00:47:03,200 acknowledged. I think the public likes, hey, 982 00:47:03,200 --> 00:47:05,800 this company whose brand you believe in. 983 00:47:06,000 --> 00:47:08,800 Is willing to make bold promises about the future and you get 984 00:47:08,800 --> 00:47:11,300 what you pay for. So like fucking Thera do story 985 00:47:11,300 --> 00:47:15,400 partially, it's the humanity is so forgiving about false about 986 00:47:15,400 --> 00:47:17,300 false. Optimism, people are extremely 987 00:47:17,300 --> 00:47:20,800 forgiving shitting asking more theranos questions mean, right? 988 00:47:20,900 --> 00:47:24,100 I think, you know, homes, I'm not sure she meant well. 989 00:47:24,100 --> 00:47:29,000 And I think musk means well, but mosque, you know, issues 990 00:47:29,000 --> 00:47:32,200 promises like they were candy. I actually called him on it 991 00:47:32,200 --> 00:47:33,200 recently. I don't know if you know this 992 00:47:33,200 --> 00:47:35,800 about me. I bet him a hundred thousand 993 00:47:35,900 --> 00:47:38,700 Dollars. He he said to Jack Dorsey that 994 00:47:38,700 --> 00:47:42,100 he'd be surprised of AGI wasn't here by 2029. 995 00:47:42,300 --> 00:47:43,900 So I've been writing this thing for sub stack. 996 00:47:43,900 --> 00:47:46,600 Gary Marcus, that sub-sect.com I was like, okay, this is a good 997 00:47:46,600 --> 00:47:50,800 topic for for an essay. All right, about why AGI is 998 00:47:50,800 --> 00:47:52,300 actually going to be five years away. 999 00:47:52,800 --> 00:47:55,600 I mean, is going to be much more than 47 years away, rather. 1000 00:47:55,600 --> 00:47:58,500 And I, I gave 5 reasons to think, like, this is really much 1001 00:47:58,500 --> 00:48:00,600 harder problem than he's acknowledging, and he's not very 1002 00:48:00,600 --> 00:48:03,100 good track record at a time. And then when I finished it, I 1003 00:48:03,107 --> 00:48:05,000 was like, you know, I should put some money on this. 1004 00:48:05,000 --> 00:48:07,800 That'll, you know, I'm sorry four hundred thousand dollars on 1005 00:48:07,800 --> 00:48:12,900 a laid out clear criteria, the field loved it and people in our 1006 00:48:12,900 --> 00:48:15,800 had doubled my money and then raised at a half million dollars 1007 00:48:15,900 --> 00:48:19,100 well, but so that still stands. But he loved hasn't responded 1008 00:48:19,200 --> 00:48:22,500 because for him, he doesn't want to be called accountable on this 1009 00:48:22,500 --> 00:48:24,800 stuff. The media should be like, dude, 1010 00:48:24,800 --> 00:48:26,700 you are chicken. There's one story like that out 1011 00:48:26,700 --> 00:48:28,600 there. Somebody one, you know, one 1012 00:48:28,600 --> 00:48:32,500 small Outlet called him on it, but most people didn't pick it 1013 00:48:32,500 --> 00:48:35,200 up and they should, they should be like this guy has been making 1014 00:48:35,200 --> 00:48:37,200 us promise As for what is yours of a car? 1015 00:48:37,800 --> 00:48:41,600 He's promising us a robot and all we use in is a dude in a 1016 00:48:41,600 --> 00:48:44,400 costume like enough. Let's call him out on it but the 1017 00:48:44,400 --> 00:48:46,800 media is not them but I mean we're with Tesla if you like the 1018 00:48:46,808 --> 00:48:49,500 government is also very compl. I mean he's running these 1019 00:48:49,500 --> 00:48:51,500 experiments on this. I don't know you guys don't get 1020 00:48:51,500 --> 00:48:52,900 to blame the government on this here. 1021 00:48:52,900 --> 00:48:55,300 The media is extremely skeptical of Tesla. 1022 00:48:55,300 --> 00:48:57,500 Like I don't know how much more skeptical of a company. 1023 00:48:57,800 --> 00:49:00,900 The media could be they are but not they're not skeptical enough 1024 00:49:00,900 --> 00:49:04,200 on the eyesight. They really are and I can give 1025 00:49:04,200 --> 00:49:05,700 you some pointers on what it looks like. 1026 00:49:05,900 --> 00:49:08,700 Like I go back to the announcement but you know it's 1027 00:49:08,700 --> 00:49:12,100 just sort of there's a certain deference to you know if a 1028 00:49:12,107 --> 00:49:14,800 company announces something if they want to risk their 1029 00:49:14,800 --> 00:49:17,800 reputation on it shouldn't the public, hold them accountable if 1030 00:49:17,800 --> 00:49:20,000 they don't deliver on the things they're saying. 1031 00:49:20,700 --> 00:49:24,500 But I just don't see how the media is supposed to operate in 1032 00:49:24,500 --> 00:49:27,100 such a disconnected way from Human psychology. 1033 00:49:27,100 --> 00:49:29,800 Like we are telling you factually that they're making 1034 00:49:29,800 --> 00:49:31,700 this assertion about what they will do in the future. 1035 00:49:31,700 --> 00:49:35,700 And the media went at once to has plenty of room. 1036 00:49:35,900 --> 00:49:38,800 I'm to kind of like set the narrative and set the questions 1037 00:49:38,900 --> 00:49:41,200 and there could be a lot more stories than there are. 1038 00:49:41,500 --> 00:49:44,600 It's a basically, hey, I'll give you, you know, I'll write the 1039 00:49:44,607 --> 00:49:47,300 story for you. So it should start with Elon 1040 00:49:47,300 --> 00:49:50,600 promise this stuff in 2016 and then the next year, Facebook 1041 00:49:50,600 --> 00:49:52,300 promised us M, it never appeared. 1042 00:49:52,300 --> 00:49:53,600 I don't know. If you remember is going to be 1043 00:49:53,600 --> 00:49:55,700 an all-purpose, General assistant, and that disappeared, 1044 00:49:56,000 --> 00:49:59,100 and then Google duplex was going to, you know, make phone calls 1045 00:49:59,100 --> 00:50:00,900 for us. And, you know, the only thing 1046 00:50:00,900 --> 00:50:02,800 they've added in four years is movie times. 1047 00:50:02,800 --> 00:50:06,300 It's still incredibly narrow and limited and now he Lon Promising 1048 00:50:06,300 --> 00:50:09,600 as a robot and not only is he promising ass a robot that is 1049 00:50:09,607 --> 00:50:12,500 going to solve or you think's a tizz solid 2029. 1050 00:50:12,800 --> 00:50:16,000 And here's this, you know, and why you prop guy? 1051 00:50:16,000 --> 00:50:19,300 Sure, it's all bullshit and like, let's like, at least like 1052 00:50:19,300 --> 00:50:22,600 ask the question, this one story, I mean I you couldn't 1053 00:50:22,600 --> 00:50:24,700 have a reporter is more aligned with you on this. 1054 00:50:24,700 --> 00:50:27,400 And I feel like part of this podcast is shitting on reporters 1055 00:50:27,700 --> 00:50:30,400 for but but what you're proposing is a single story that 1056 00:50:30,400 --> 00:50:33,900 will then be up against sort of the infinite barrage of 1057 00:50:33,900 --> 00:50:35,700 companies announcing thing as time. 1058 00:50:35,800 --> 00:50:37,900 It has to be more. Just like how do you create the 1059 00:50:37,900 --> 00:50:41,200 drumbeat of negativity? Look if we if we learned nothing 1060 00:50:41,200 --> 00:50:43,800 from the Trump Administration we learn nothing from the Trump 1061 00:50:43,800 --> 00:50:46,000 Administration. It's that you have to keep up 1062 00:50:46,000 --> 00:50:49,500 the pressure and you know, the news cycle is short and it's 1063 00:50:49,500 --> 00:50:51,600 true. Like if it's just one story, 1064 00:50:51,600 --> 00:50:54,900 it's not enough but there has to be a systemic ever. 1065 00:50:54,900 --> 00:50:59,800 I mean, look, Kate matz has been holding along to the fire on the 1066 00:50:59,900 --> 00:51:01,900 the effectiveness of the self-driving. 1067 00:51:01,900 --> 00:51:04,800 So I'm you know, I'm exaggerating a little bit but I 1068 00:51:04,800 --> 00:51:08,100 think you know it's 95 5 or something like that. 1069 00:51:08,100 --> 00:51:11,600 And I also think by the way this extends to a lot of and I'm very 1070 00:51:11,600 --> 00:51:14,600 critical on this show about augmented reality and the 1071 00:51:14,600 --> 00:51:17,300 promise is these companies make on the effectiveness and the 1072 00:51:17,300 --> 00:51:19,400 promises of what it can do. And we're about to enter this 1073 00:51:19,400 --> 00:51:22,300 hype cycle again. When Apple releases its you know 1074 00:51:22,300 --> 00:51:24,800 VR device and Promises a are down the line. 1075 00:51:24,800 --> 00:51:28,500 We're very quick to go to the demos, you know, all the 1076 00:51:28,500 --> 00:51:31,400 reporters went down to Google and inside of the ugly way Mo 1077 00:51:31,400 --> 00:51:34,700 cars and and that helps kind of pump along this idea that they 1078 00:51:34,700 --> 00:51:38,100 were really close to. Diving and I don't have an easy 1079 00:51:38,100 --> 00:51:40,500 answer to it other than maybe occasionally telling these 1080 00:51:40,500 --> 00:51:43,100 companies know and and saying, you know what? 1081 00:51:43,100 --> 00:51:45,300 This demo that you're putting me through, yes, I can be critical 1082 00:51:45,300 --> 00:51:47,400 in the article and I think Natasha, I don't know the 1083 00:51:47,400 --> 00:51:50,400 backstory of how the story came to her, but you know, the 1084 00:51:50,400 --> 00:51:52,900 Washington Post also framed it, but Blake, you know, kind of the 1085 00:51:52,900 --> 00:51:56,200 dark artistic lighting looking like some sort of visionary. 1086 00:51:56,400 --> 00:51:59,800 When even though the article was, I think reasonably critical 1087 00:51:59,800 --> 00:52:03,000 of him, it's still kind of positive him as a legitimate 1088 00:52:03,000 --> 00:52:06,100 voice in this field would obviously he's not and and I 1089 00:52:06,100 --> 00:52:08,800 just, I don't know. I and what I write about it 1090 00:52:08,800 --> 00:52:11,300 doesn't come up nearly as much Uber has spun off its 1091 00:52:11,300 --> 00:52:13,600 self-driving division. They don't care about it 1092 00:52:13,600 --> 00:52:14,900 anymore. They're just a dollars and cents 1093 00:52:14,900 --> 00:52:17,000 business and relatively boring because of it. 1094 00:52:17,300 --> 00:52:20,200 But I think that the hype cycle as pushed by the companies will 1095 00:52:20,200 --> 00:52:23,700 never end and it's incredibly difficult as a reporter to turn 1096 00:52:23,700 --> 00:52:25,900 down sexy stories that we know will get attention. 1097 00:52:26,000 --> 00:52:28,600 I mean you can run the stories but you can get ahold of a lot 1098 00:52:28,600 --> 00:52:33,000 more people like me but not just me as voices in these things and 1099 00:52:33,000 --> 00:52:36,100 make it clear that in Ewing you can remind them. 1100 00:52:36,400 --> 00:52:39,100 You know let's look at the history we've seen this promise 1101 00:52:39,100 --> 00:52:41,800 that wasn't deliberate like when's the last time that you 1102 00:52:41,800 --> 00:52:45,200 read a story on these technologies that actually like 1103 00:52:45,200 --> 00:52:48,100 reviewed the history and said all these other promises like 1104 00:52:48,100 --> 00:52:51,500 they didn't come true like either, like you read a story 1105 00:52:51,500 --> 00:52:54,300 about Optimist and it's probably mostly about Optimist and not 1106 00:52:54,300 --> 00:52:56,600 saying so much about like, you know, Elon is missed. 1107 00:52:56,600 --> 00:53:00,800 Every deadline he's ever proposed and it's not you the 1108 00:53:00,800 --> 00:53:02,900 rarely are they synthetic putting together. 1109 00:53:03,000 --> 00:53:06,400 All of the facts that I just gave about hey, Facebook made 1110 00:53:06,400 --> 00:53:08,400 these promises Google made these promises. 1111 00:53:08,500 --> 00:53:11,600 It's actually really hard to get AI into production which is 1112 00:53:11,600 --> 00:53:12,900 itself. You know, an interesting 1113 00:53:12,900 --> 00:53:15,100 question. Like there are some technologies 1114 00:53:15,100 --> 00:53:18,100 you can put into production relatively quickly but AI is not 1115 00:53:18,100 --> 00:53:19,600 one of them. Why is it not? 1116 00:53:19,700 --> 00:53:22,100 Well, it's not because they're always these outlier cases. 1117 00:53:22,100 --> 00:53:25,100 So like you probably saw that one of the driverless cars ran 1118 00:53:25,100 --> 00:53:27,600 into a jet the other day, like it wasn't in the data's, right? 1119 00:53:27,600 --> 00:53:30,700 This is a persistent well-known problem in the industry by now. 1120 00:53:31,000 --> 00:53:33,800 Yeah, I've been writing about it since 2016 and like people are 1121 00:53:33,800 --> 00:53:36,500 starting to recognize they really is the whole ball game 1122 00:53:36,500 --> 00:53:39,300 but that means every time you have some technology, you can 1123 00:53:39,300 --> 00:53:41,000 wind up with some outlier problem. 1124 00:53:41,000 --> 00:53:43,700 So yeah, you're going to get the demo on day one and it's going 1125 00:53:43,700 --> 00:53:47,000 to be 5 years, 10 years, 15 years before you can actually 1126 00:53:47,000 --> 00:53:48,800 trust it. Like that should be in every 1127 00:53:48,800 --> 00:53:50,200 story here. Yeah. 1128 00:53:50,200 --> 00:53:53,000 And it said, I think it's also The Duality of Silicon Valley 1129 00:53:53,000 --> 00:53:56,800 and the CEO sitio Dynamic. Where it's both a marriage of 1130 00:53:57,100 --> 00:54:00,300 some sort of technological progress and the American 1131 00:54:00,500 --> 00:54:03,600 Showmanship song-and-dance marketing routine of getting the 1132 00:54:03,607 --> 00:54:07,300 public excited about it and were Human beings are definitely as 1133 00:54:07,300 --> 00:54:10,900 journalists susceptible to the CEO side of things. 1134 00:54:10,900 --> 00:54:13,300 We love the character. Give another story idea. 1135 00:54:13,500 --> 00:54:15,900 Ilan, just say, I mean, I actually wrote it but in a 1136 00:54:15,908 --> 00:54:19,200 subset that didn't really get that much attention is Ilan. 1137 00:54:19,200 --> 00:54:22,000 Said, you know, the whole company really it depends on the 1138 00:54:22,000 --> 00:54:25,100 self-driving cars and you know if that doesn't work we were 1139 00:54:25,107 --> 00:54:27,400 basically worthless which was slight exaggeration. 1140 00:54:27,400 --> 00:54:29,100 But like and it's just the car company. 1141 00:54:29,100 --> 00:54:32,400 I mean, the reason it's get 100 of one price to earnings is 1142 00:54:32,400 --> 00:54:34,600 because people think it is in AI company that is going to 1143 00:54:34,600 --> 00:54:37,500 fundamentally change. The And that's why said 101, I 1144 00:54:37,508 --> 00:54:39,100 don't know. I don't think most Tesla holders 1145 00:54:39,100 --> 00:54:41,400 have a argument for why they hold a stock. 1146 00:54:41,400 --> 00:54:44,700 But yes I see what your cell warts because it kept going up. 1147 00:54:44,700 --> 00:54:47,700 But now we're going down, yes or whatever but it's a big part of 1148 00:54:47,700 --> 00:54:50,200 it but I mean, Eli himself. It doesn't matter what a though. 1149 00:54:50,300 --> 00:54:53,400 Other holders are he largest stockholder in Tesla, which 1150 00:54:53,400 --> 00:54:56,600 happens to be? Elon Musk said, if we don't or 1151 00:54:56,600 --> 00:54:59,900 he said, we must solve, full self driving. 1152 00:55:00,000 --> 00:55:03,500 Whe, When I wear the anything that itself gives you a story 1153 00:55:03,600 --> 00:55:05,600 like, okay, let's take for granted. 1154 00:55:05,800 --> 00:55:08,900 What he said is true, we can ask around and you know get some 1155 00:55:09,500 --> 00:55:12,400 Financial people which I'm not to evaluate that statement but 1156 00:55:12,400 --> 00:55:16,100 if you take his premise like okay he's been promising this 1157 00:55:16,100 --> 00:55:19,400 since 2015, is he close. Let's look at the new accident 1158 00:55:19,400 --> 00:55:21,500 data. Let's ask some experts like 1159 00:55:21,500 --> 00:55:25,300 let's hold his nose to the fire. I just think Moe tassels the one 1160 00:55:25,300 --> 00:55:29,500 I find most Tesla cover. It is justifiably - I mean 1161 00:55:29,700 --> 00:55:33,000 you're basically asking reporters to ask coverage is - 1162 00:55:33,000 --> 00:55:37,400 like people you know, make fun of his His tweets and that kind 1163 00:55:37,400 --> 00:55:39,200 of stuff. And, you know, there was a new 1164 00:55:39,200 --> 00:55:41,500 lawsuit yesterday and people will write about that. 1165 00:55:41,700 --> 00:55:44,900 I don't think that the AI coverage is nearly as skeptical 1166 00:55:44,900 --> 00:55:47,000 as it could be. I mean, it wasn't there, a story 1167 00:55:47,000 --> 00:55:50,100 about how they're, like, supposed to be turning off like 1168 00:55:50,100 --> 00:55:52,700 the AI right before. It gets an accident or so I need 1169 00:55:52,700 --> 00:55:54,000 to. Yeah. 1170 00:55:54,300 --> 00:55:57,500 I mean the NH and Nitsa we'll call it just really something 1171 00:55:57,500 --> 00:56:00,800 couple days ago, right? And so that got a little bit of 1172 00:56:00,800 --> 00:56:03,900 coverage, but only caught Nitsa has released two Bombshells in 1173 00:56:03,900 --> 00:56:05,000 the land. It's a disease. 1174 00:56:05,800 --> 00:56:09,400 They've been deploying this for like since like 2016 or 1175 00:56:09,400 --> 00:56:12,600 something and you're bleeding. It's so did you real things this 1176 00:56:12,600 --> 00:56:13,900 week? And it's at the to real things 1177 00:56:13,900 --> 00:56:16,200 this week. They put out information about 1178 00:56:16,200 --> 00:56:19,900 the turning off the autopilot. Just before the accident happens 1179 00:56:20,100 --> 00:56:23,000 and then they put a big dump in which Tesla had the most 1180 00:56:23,000 --> 00:56:24,600 accidents, which is a complicated thing, because they 1181 00:56:24,600 --> 00:56:27,700 also have the most miles, but mean they put stuff out that 1182 00:56:28,000 --> 00:56:30,300 could have been like top of the headlines. 1183 00:56:30,300 --> 00:56:33,600 Like is this a serious problem for Tesla or not? 1184 00:56:34,000 --> 00:56:37,700 And like that was Are for the journalists to run with and I 1185 00:56:37,700 --> 00:56:41,000 didn't see much about it, like, I check the, you know, the news 1186 00:56:41,000 --> 00:56:43,900 stories about Tesla every now and then just to see because, 1187 00:56:43,900 --> 00:56:46,900 you know, I always think about Elon is such an outlier. 1188 00:56:46,900 --> 00:56:49,100 He's such a character. He's so bizarre. 1189 00:56:49,100 --> 00:56:52,200 He almost defies the laws of gravity when it comes to 1190 00:56:52,200 --> 00:56:55,000 negative and positive coverage. It's almost not even worth 1191 00:56:55,400 --> 00:56:57,400 holding it on him. I mean Trump was a little like 1192 00:56:57,400 --> 00:57:00,200 that right and they play some similar games but I guess people 1193 00:57:00,200 --> 00:57:05,400 that I think is a better example to me, I think Google or some of 1194 00:57:05,400 --> 00:57:08,100 the other Tech companies. I bet they should be held to the 1195 00:57:08,100 --> 00:57:11,400 fire more to say, with open a. I like they've gotten all these 1196 00:57:11,400 --> 00:57:14,900 love letters about G PT 3. So, let's forget Google and and 1197 00:57:14,900 --> 00:57:16,800 just look at it. Open a.i. for a minute, you 1198 00:57:16,800 --> 00:57:20,000 know, you had the love letter in the Times, by Steven Berlin 1199 00:57:20,000 --> 00:57:22,900 Johnson, The Guardian wrote, an op-ed with it. 1200 00:57:22,900 --> 00:57:25,400 It cetera, everybody thinks they're like being created by 1201 00:57:25,400 --> 00:57:29,200 using it to write their story. Like, this is like a Trope by 1202 00:57:29,200 --> 00:57:31,400 now. Yeah, I mean, it's gonna go out 1203 00:57:31,400 --> 00:57:33,900 there and, you know, Berlin Johnson gave two paragraphs to 1204 00:57:33,900 --> 00:57:37,200 me, and one to Emily Bender. But But this story is still 1205 00:57:37,200 --> 00:57:41,100 like, so so Pro this kind of stuff in a way that I think many 1206 00:57:41,100 --> 00:57:43,600 people in the field you know, fast what the public wants. 1207 00:57:43,600 --> 00:57:46,300 I mean ultimately if you're - reporter like I didn't write 1208 00:57:46,300 --> 00:57:49,400 much about AI, I mean self-driving, as new reporter I 1209 00:57:49,400 --> 00:57:52,400 was very openly skeptical in The Newsroom refused to write about 1210 00:57:52,400 --> 00:57:54,100 it. Uber would just go to a business 1211 00:57:54,100 --> 00:57:58,800 week writer and say, hey, here's our new, like, I didn't get a 1212 00:57:58,808 --> 00:58:00,900 different Business Week. Reporter, you know, got this 1213 00:58:00,900 --> 00:58:03,900 story for their Pittsburgh lab because Uber knows then go to 1214 00:58:03,900 --> 00:58:06,900 somebody else who will do sort of like, The big Productions. 1215 00:58:06,900 --> 00:58:09,200 I mean, those guys are very good at shopping for bikinis just 1216 00:58:09,200 --> 00:58:13,100 like, there's so much desire. There's so much desire for these 1217 00:58:13,100 --> 00:58:15,000 stories like editors. Yeah, I mean, this is what 1218 00:58:15,000 --> 00:58:18,300 Business magazines are based on like putting optimistic 1219 00:58:18,300 --> 00:58:20,500 statements. You know, Mark Lori is going to 1220 00:58:20,500 --> 00:58:23,700 build a new city. I mean, it's just like it's so 1221 00:58:23,700 --> 00:58:25,600 part. It's what Humanity wants? 1222 00:58:25,600 --> 00:58:28,100 It somewhere, I guess, I just don't think reporters are going 1223 00:58:28,100 --> 00:58:30,300 to will into being, it's like their business model. 1224 00:58:30,300 --> 00:58:32,900 I mean, I'm I think I think that that's true. 1225 00:58:33,400 --> 00:58:35,500 I think Humanity ones, happy story. 1226 00:58:35,700 --> 00:58:38,600 Is about the new Revolution. It's really think that comes at 1227 00:58:38,600 --> 00:58:40,800 a cost and that's how we got into the conversation. 1228 00:58:41,000 --> 00:58:45,200 The cost is eeuu wind up with people diluted and yeah. 1229 00:58:45,200 --> 00:58:46,100 Right. No I agree. 1230 00:58:46,100 --> 00:58:47,900 And I agree. So I'm being defensive even 1231 00:58:47,900 --> 00:58:51,500 though I'm sympathetic but it just seems hard to hard to. 1232 00:58:52,400 --> 00:58:55,000 It's a lot of affinity level like yeah. 1233 00:58:55,000 --> 00:58:59,000 I mean so look I've been partly because you guys are meeting 1234 00:58:59,000 --> 00:59:01,700 guys that I do some journals. Oh we love it. 1235 00:59:01,700 --> 00:59:03,200 I'm happy to have it the conversation you know. 1236 00:59:03,200 --> 00:59:06,700 I think it's fun to have this conversation but I would agree 1237 00:59:06,700 --> 00:59:08,700 with you. That it's not like a, you know, 1238 00:59:08,707 --> 00:59:11,600 to Second problem. I'm like pitching you ideas to 1239 00:59:11,607 --> 00:59:14,400 go write about them and hoping some of your buddies will listen 1240 00:59:14,400 --> 00:59:16,700 and, and use them to like, I'm giving away for free. 1241 00:59:16,707 --> 00:59:19,200 Yeah. And the media reporters are all 1242 00:59:19,200 --> 00:59:21,300 listening to this podcast. So yeah, I mean I also 1243 00:59:21,300 --> 00:59:24,800 understand like it is what the public wants and so you know the 1244 00:59:24,800 --> 00:59:28,000 public is partly to blame because it you know it votes 1245 00:59:28,000 --> 00:59:30,800 with its clicks and and the stories that get read are the 1246 00:59:30,800 --> 00:59:34,000 you know the world has changed kind of stories and not the you 1247 00:59:34,000 --> 00:59:36,700 know, I'm not so sure. Oh sure, that this is really 1248 00:59:36,700 --> 00:59:38,100 going to happen. Kind of stories and the 1249 00:59:38,100 --> 00:59:41,500 government is supposed to protect the roads, like the 1250 00:59:41,900 --> 00:59:43,600 self-driving cars are on the streets. 1251 00:59:43,600 --> 00:59:45,900 Like, at the end of the day, I am sorry. 1252 00:59:46,000 --> 00:59:48,000 The government is letting Tesla get away with this. 1253 00:59:48,000 --> 00:59:50,800 Like, Tesla has been experimenting for years. 1254 00:59:50,800 --> 00:59:53,800 It says, upping its game. It's something easy there. 1255 00:59:53,800 --> 00:59:56,500 I think the media is not quite following. 1256 00:59:56,500 --> 00:59:59,300 The trail that knits has been leading in the last few weeks 1257 00:59:59,300 --> 01:00:01,400 and it says, giving some really serious glue. 1258 01:00:01,400 --> 01:00:03,200 The government's going to go after against Tesla. 1259 01:00:03,200 --> 01:00:05,500 After the stock is already down. It's not there. 1260 01:00:05,700 --> 01:00:07,200 So by the way, they're not to blame. 1261 01:00:07,200 --> 01:00:09,700 If they bring the company down, they don't want to do it. 1262 01:00:09,700 --> 01:00:11,700 When it actually would hurt a rising company. 1263 01:00:11,700 --> 01:00:14,700 They want to do it after the market is already said, okay 1264 01:00:14,700 --> 01:00:16,500 fine. This company's, I mean value, I 1265 01:00:16,500 --> 01:00:19,100 think Nitsa just wants to have like do the right thing, 1266 01:00:19,100 --> 01:00:22,800 whatever the right thing is, but they also showed that we Mo's 1267 01:00:22,800 --> 01:00:24,900 have him pretty serious problems too. 1268 01:00:25,100 --> 01:00:28,900 And in fact, the whole field like so if you read those data 1269 01:00:28,900 --> 01:00:31,900 carefully, the conclusion you should come to is, we're not 1270 01:00:31,900 --> 01:00:34,300 close to level 5 self-driving, right? 1271 01:00:34,400 --> 01:00:36,400 Right. And I remember, Personally, you 1272 01:00:36,400 --> 01:00:39,700 know what, my colleague of mere a Friday at the the information 1273 01:00:39,700 --> 01:00:42,300 wrote, what I thought was a fairly definitive story about 1274 01:00:42,300 --> 01:00:44,500 way Mo's technology when they were testing on the streets in 1275 01:00:44,500 --> 01:00:48,000 Arizona and basically found out that they couldn't turn left. 1276 01:00:48,000 --> 01:00:49,700 Turns are still hard there. Ya are. 1277 01:00:49,900 --> 01:00:52,000 Yeah. And it's just like, if that, you 1278 01:00:52,000 --> 01:00:53,300 know, I, you know, that's got to be. 1279 01:00:53,500 --> 01:00:56,600 I can close to 50% of turns, you know, if you get that you don't 1280 01:00:56,600 --> 01:00:57,800 get emotionally. Do that? 1281 01:00:57,800 --> 01:01:00,500 Levi gave every Steven Levy, I mentioned before, gave everybody 1282 01:01:00,500 --> 01:01:03,900 a big clue in 2015 that not enough people picked up on which 1283 01:01:03,900 --> 01:01:05,500 is he? He visited. 1284 01:01:05,700 --> 01:01:08,200 Google at that point or way Mo. I forget what they were called. 1285 01:01:08,400 --> 01:01:11,200 Had this place where they were testing the machines and Levi 1286 01:01:11,700 --> 01:01:13,500 know what's the word. I'm looking for implanted there 1287 01:01:13,500 --> 01:01:16,000 for a month or something embedded there for about a week. 1288 01:01:16,000 --> 01:01:17,800 I don't know. Is he better there for a week or 1289 01:01:17,808 --> 01:01:20,100 something like that? And the like big dramatic point. 1290 01:01:20,100 --> 01:01:22,200 I haven't gone back and reread the story but I got him to give 1291 01:01:22,200 --> 01:01:23,700 me the link the other day so you can find it on. 1292 01:01:23,700 --> 01:01:25,700 Back-channel, I know that only within wire. 1293 01:01:25,900 --> 01:01:29,100 So anyway, he's there for a week or something like that. 1294 01:01:29,200 --> 01:01:32,900 End up being dramatic thing was the end of this time. 1295 01:01:32,900 --> 01:01:35,700 They are or something like that. I haven't read it and said In 1296 01:01:35,700 --> 01:01:37,600 years. But basically it revolved around 1297 01:01:37,600 --> 01:01:40,000 they figured out how to recognize a pile of leaves 1298 01:01:40,900 --> 01:01:44,000 right. Okay I got a thug really we a 1299 01:01:44,000 --> 01:01:46,800 hot dog, hot dog. You know they'd already been 1300 01:01:46,800 --> 01:01:49,700 doing this for five years at that point and like leaves were 1301 01:01:49,700 --> 01:01:51,600 still a problem. Well that's it. 1302 01:01:51,600 --> 01:01:54,100 Leaves are an outlier and that was a clue like ever. 1303 01:01:54,100 --> 01:01:58,900 If you have to Band-Aid up every outlier that you're playing 1304 01:01:58,900 --> 01:02:00,900 whack-a-mole and that still What's Happening Here? 1305 01:02:01,000 --> 01:02:04,700 I keep going back to the marriage of public research and 1306 01:02:04,700 --> 01:02:06,800 academic research. It's the needs of the private 1307 01:02:06,800 --> 01:02:10,000 company and the, you know, the press release announcement 1308 01:02:10,000 --> 01:02:13,400 culture, that is what drives the, you know, stocks 1309 01:02:13,400 --> 01:02:17,100 essentially businesses of these tech companies with the sort of 1310 01:02:17,100 --> 01:02:20,000 slow plodding methodical advances that happened in 1311 01:02:20,000 --> 01:02:22,500 research that happened in decades and it just doesn't fit 1312 01:02:22,500 --> 01:02:24,100 with a time. One of these companies, that's 1313 01:02:24,100 --> 01:02:25,600 right. I think the least your listeners 1314 01:02:25,600 --> 01:02:29,200 could come away with is hey we are living in this announcement 1315 01:02:29,200 --> 01:02:32,600 culture and that announcement culture is making people like 1316 01:02:32,600 --> 01:02:35,300 Blake, Lemoine believe in fairies that aren't there. 1317 01:02:35,700 --> 01:02:38,700 And it's making a lot of us believe in deadlines that are 1318 01:02:38,700 --> 01:02:41,500 not really going to be met and we should be a whole lot more 1319 01:02:41,500 --> 01:02:42,400 skeptical. Yeah. 1320 01:02:42,500 --> 01:02:44,300 And I also just to reiterate Eric's point. 1321 01:02:44,300 --> 01:02:47,000 I also think we as the meteor coming up against human nature 1322 01:02:47,200 --> 01:02:50,300 at times, we do is just, you know, it in the representation 1323 01:02:50,300 --> 01:02:52,700 of Blake Lemoine. Someone who is, you know, 1324 01:02:52,700 --> 01:02:55,100 religiously and this is one of the most valuable companies. 1325 01:02:55,100 --> 01:02:58,600 The world delusions are inherent part of people, yes, whole 1326 01:02:58,600 --> 01:03:03,000 understanding of the universe like, yeah, I agree. 1327 01:03:03,000 --> 01:03:05,500 The media should be more skeptical, but I do think 1328 01:03:06,100 --> 01:03:09,000 Regular humans, the government, the companies making 1329 01:03:09,000 --> 01:03:12,500 announcements themselves, their there are a lot of people to 1330 01:03:12,500 --> 01:03:14,900 blame couple seconds on government if we have time. 1331 01:03:14,900 --> 01:03:15,500 Sure. Yeah. 1332 01:03:15,500 --> 01:03:18,100 What we can close with that. I think government's going to 1333 01:03:18,100 --> 01:03:20,500 have to regulate AI much more than it does. 1334 01:03:20,500 --> 01:03:24,900 So right now for example, any company little like Tesla can 1335 01:03:24,900 --> 01:03:29,000 put out an over the air update and it's driving software and 1336 01:03:29,000 --> 01:03:33,800 the there's only liability after the fact there's no regulation 1337 01:03:33,800 --> 01:03:37,000 you must meet these test trials These outliers before it's 1338 01:03:37,000 --> 01:03:39,600 released and I think misinformation, which we haven't 1339 01:03:39,600 --> 01:03:42,400 talked about today is a massive, massive problem in Europe. 1340 01:03:42,400 --> 01:03:45,900 They just made a deal with Facebook and other companies to 1341 01:03:45,900 --> 01:03:48,400 be tighter on that. We're going to need that here in 1342 01:03:48,400 --> 01:03:51,700 North America to and it's a serious problem because systems 1343 01:03:51,700 --> 01:03:55,800 like G PT 3, and Lambda are fabulous at creating 1344 01:03:55,800 --> 01:03:59,500 misinformation which makes them wonderful tools for trolls and 1345 01:03:59,500 --> 01:04:02,000 other other control farms and so forth. 1346 01:04:02,000 --> 01:04:04,600 That is a serious problem. We making misinformation much 1347 01:04:04,600 --> 01:04:07,400 worse than it is now Now and so, yeah, I've been dumping on the 1348 01:04:07,400 --> 01:04:09,400 media because I thought it'd be fun to make and we all share an 1349 01:04:09,400 --> 01:04:12,000 interest in it, but I could totally right that the 1350 01:04:12,000 --> 01:04:15,400 government needs to step it up and needs to figure out how to 1351 01:04:15,400 --> 01:04:18,800 regulate the stuff, which nobody really knows yet, it needs to 1352 01:04:18,808 --> 01:04:21,400 realize how important it. So, I did this Twitter, space 1353 01:04:21,400 --> 01:04:25,300 with Natasha and Kara swisher, and Casey last night. 1354 01:04:25,300 --> 01:04:27,900 And the best question from the audience was like, okay. 1355 01:04:27,900 --> 01:04:29,700 So if you're saying people are going to fall in love with these 1356 01:04:29,700 --> 01:04:32,000 things and they're toxic. What is public health going to 1357 01:04:32,000 --> 01:04:33,900 do about that? And that is a really good 1358 01:04:33,900 --> 01:04:36,500 question. Yeah, we don't want the answer 1359 01:04:36,500 --> 01:04:38,500 to great. We can leave it there. 1360 01:04:38,500 --> 01:04:40,800 Well, thank you so much for joining us Gerry, and for 1361 01:04:40,800 --> 01:04:43,300 listeners, who are maybe many of them reporters. 1362 01:04:43,300 --> 01:04:45,700 If they want to get in contact with you, and I'd read more 1363 01:04:45,700 --> 01:04:48,600 caring, read your list will include your list of 20 people 1364 01:04:48,600 --> 01:04:51,600 who share your views about AI. So we can, you know, you can get 1365 01:04:51,600 --> 01:04:54,100 that in the piece called Paradigm Shift. 1366 01:04:54,100 --> 01:04:57,100 Gary Marcus dotsub stock.com. Great awesome and AD carry 1367 01:04:57,100 --> 01:04:59,100 markets on Twitter. Thanks so much for joining us 1368 01:04:59,100 --> 01:04:59,900 Gerry. This is awesome. 1369 01:05:00,100 --> 01:05:01,100 Thanks very much. Bye, bye. 1370 01:05:13,500 --> 01:05:14,900 Goodbye. Goodbye. 1371 01:05:14,900 --> 01:05:17,200 Goodbye, goodbye, goodbye. Goodbye. 1372 01:05:13,500 --> 01:05:14,900 Goodbye. Goodbye. 1373 01:05:14,900 --> 01:05:17,200 Goodbye, goodbye, goodbye. Goodbye.