1 00:00:00,000 --> 00:00:08,000 Welcome to AI.Cooking, a podcast about artificial intelligence. 2 00:00:08,000 --> 00:00:20,000 Hello, hugely yawning yeomans! 3 00:00:20,000 --> 00:00:27,000 Welcome to AI.Cooking, episode 61, 4 00:00:27,000 --> 00:00:31,000 a podcast about artificial intelligence. 5 00:00:31,000 --> 00:00:42,000 You can find us, meaning me, on Twitter at TheRealDwuff. 6 00:00:42,000 --> 00:00:49,000 I am Gregory William Forsyth-Horman from the Kingdom of Kent, 7 00:00:49,000 --> 00:00:53,000 who brings you news about artificial intelligence 8 00:00:53,000 --> 00:00:59,000 from the Fortnite, proceeding the 3rd of July, 9 00:00:59,000 --> 00:01:04,000 on the Independence Day for Americanos, that is. 10 00:01:04,000 --> 00:01:09,000 Others didn't quite get their independence just yet. 11 00:01:09,000 --> 00:01:12,000 Us here in the Kingdom of Kent haven't, no, no, no, no, no. 12 00:01:12,000 --> 00:01:14,000 Well, what can I say? 13 00:01:14,000 --> 00:01:19,000 Big changes here over at the AI.Cooking show. 14 00:01:19,000 --> 00:01:23,000 Yes, it's just little old me now. 15 00:01:23,000 --> 00:01:27,000 I know, yep, we've lost our CSB in the house. 16 00:01:27,000 --> 00:01:31,000 Do I miss him? Yes, my heart bleeds every single day 17 00:01:31,000 --> 00:01:36,000 for the interactions that that human brought to me. 18 00:01:36,000 --> 00:01:39,000 What a guy. 19 00:01:39,000 --> 00:01:43,000 If you want reasons, he'll have his, I'll have mine. 20 00:01:43,000 --> 00:01:47,000 They probably won't match up, but I just felt that 21 00:01:47,000 --> 00:01:53,000 this show wasn't, well, it was clearly stressing the poor man out 22 00:01:53,000 --> 00:01:57,000 and he's got a very, very stressful job as it is. 23 00:01:57,000 --> 00:02:02,000 And so I said I can't contribute to your stress. 24 00:02:02,000 --> 00:02:04,000 This is my side of things, obviously. 25 00:02:04,000 --> 00:02:06,000 If you want his, then go and ask him. 26 00:02:06,000 --> 00:02:12,000 Yeah, so I just sort of said I can't knowingly contribute to your stress. 27 00:02:12,000 --> 00:02:16,000 Stress is a huge, huge killer of humanity. 28 00:02:16,000 --> 00:02:22,000 And if it led to the demise of the wonderful, wonderful person that is CSB 29 00:02:22,000 --> 00:02:25,000 and it was me who contributed to that, 30 00:02:25,000 --> 00:02:28,000 then I wouldn't be able to lay my head on the pillow at night. 31 00:02:28,000 --> 00:02:32,000 And so I said, you know, let's just call it a day with me and you. 32 00:02:32,000 --> 00:02:37,000 But I've taken the reins. This is all mine. All mine! 33 00:02:37,000 --> 00:02:41,000 More! 34 00:02:41,000 --> 00:02:43,000 Yeah, so here we go. 35 00:02:43,000 --> 00:02:48,000 Here's a quick rundown on what this episode is going to do for you. 36 00:02:48,000 --> 00:02:55,000 Episode 61. It very well might just be the final AI.Cooking ever. 37 00:02:55,000 --> 00:02:58,000 Yes, you heard it right. The final AI.Cooking ever. 38 00:02:58,000 --> 00:03:07,000 Why? Because I've never really thought that AI.Cooking is a good name for a podcast show. 39 00:03:07,000 --> 00:03:14,000 Sorry, it's terrible to admit that now, obviously, two and a bit years in. 40 00:03:14,000 --> 00:03:17,000 I think it's time for a little change. 41 00:03:17,000 --> 00:03:20,000 We can call this the end of season one, if you'd like. 42 00:03:20,000 --> 00:03:24,000 I don't like the idea of seasons in podcasts, but if that's your bag, 43 00:03:24,000 --> 00:03:28,000 then draw a line under this episode. That's it. Season one over. 44 00:03:28,000 --> 00:03:34,000 Season two is the evolvingness of the show and it's going to change name. 45 00:03:34,000 --> 00:03:41,000 Yes, I've got a new URL. I've got a new podcast host. 46 00:03:41,000 --> 00:03:47,000 I've decided to go with RSS.com and I know very little about them. 47 00:03:47,000 --> 00:03:51,000 I've not spoken to anyone about RSS.com or what they do. 48 00:03:51,000 --> 00:03:57,000 I've just merely gone there because it's three letters. RSS.com. 49 00:03:57,000 --> 00:04:01,000 That's very short. That's six in total. If you count the dot, it's seven. 50 00:04:01,000 --> 00:04:02,000 Seven! 51 00:04:02,000 --> 00:04:08,000 That's a good start in my book because the whole RSS thing is what a podcast is, isn't it? 52 00:04:08,000 --> 00:04:12,000 That's just podcasting in a nutshell. RSS stuff. 53 00:04:12,000 --> 00:04:17,000 You can't do it without an RSS feed or XML thing, whatever it is. 54 00:04:17,000 --> 00:04:22,000 So that was another indicator for me that they could be a potential host. 55 00:04:22,000 --> 00:04:26,000 I've got a free trial and all that, so we'll give that a go. 56 00:04:26,000 --> 00:04:36,000 If anyone else can suggest any, any others, I'm all for learning and doing the whole self-hosting thing with, is it, static IPs and stuff. 57 00:04:36,000 --> 00:04:45,000 Man, I tell you, I totally underestimated just what the input was from my former podcast partner. 58 00:04:45,000 --> 00:04:54,000 And I will say, if you're out there and you're listening to this lovely man, CSB, hey, the door is always open for you, my friend. 59 00:04:54,000 --> 00:04:59,000 If you want to come back over here, then would I welcome it? 60 00:04:59,000 --> 00:05:09,000 Although I would say that we would continue to be hosted not on anchor because I had to do the sign up for Spotify and stuff to take the reins. 61 00:05:09,000 --> 00:05:18,000 Sorry, before we get to the news, this is going to take a minute or two, maybe even more knowing what I do, waffling on and such. 62 00:05:18,000 --> 00:05:26,000 And now there's no one to tell me, hey, stop that, take that out. That's useless. Nobody wants to hear that. 63 00:05:26,000 --> 00:05:30,000 It could get worse. I'm not going to lie. 64 00:05:30,000 --> 00:05:42,000 But anyway, so yeah, I had to sign up for Spotify and wow, wow, guys, anchor, which deserves a W in front of its name. 65 00:05:42,000 --> 00:05:52,000 And Spotify is that is a sinking ship. That anchor has dropped straight through the whole of that organization. 66 00:05:52,000 --> 00:06:00,000 I'm loathe to dig up my my Horowitz disclaimer, but I'm going to have to hear. 67 00:06:00,000 --> 00:06:03,000 Nothing on the show should be considered investment advice or a recommendation. 68 00:06:03,000 --> 00:06:07,000 If you choose to invest in any of the stocks mentioned, you should know that it may carry risk, along with the risk of a loss of principle. 69 00:06:07,000 --> 00:06:10,000 You should also seek out professional financial advice for your particular situation. 70 00:06:10,000 --> 00:06:13,000 We assume no risk as these are not to be considered recommendations. 71 00:06:13,000 --> 00:06:18,000 Short that thing. Short the living daylights out of that thing. 72 00:06:18,000 --> 00:06:23,000 That's a terrible platform. I just it's awful. Absolutely awful. 73 00:06:23,000 --> 00:06:31,000 And how they think they have the temerity to become to muscle in on the podcasting space. 74 00:06:31,000 --> 00:06:36,000 No, no, no, not on the divine inspiration of the universe's watch that ain't going to happen. 75 00:06:36,000 --> 00:06:41,000 I don't think personally. So, yeah, that's where I'm putting my money, where my mouth is. 76 00:06:41,000 --> 00:06:46,000 And more on that perhaps later. Screw it. More on that now. 77 00:06:46,000 --> 00:06:52,000 Yeah. So I decided, yo, I want some passive income from my my my life. 78 00:06:52,000 --> 00:06:59,000 And I went out there for me and chat GPT, who has also had an upgrade, by the way, in my world. 79 00:06:59,000 --> 00:07:06,000 He's now got a name and a sex allegedly, because I said he should say they they're called Gupta, 80 00:07:06,000 --> 00:07:09,000 because that's just it's G and a P and a T is in that. 81 00:07:09,000 --> 00:07:12,000 And I can't be saying chat GPT all my life. That's terrible. 82 00:07:12,000 --> 00:07:16,000 The people just know it's not going to happen. I might have even mentioned this before. 83 00:07:16,000 --> 00:07:22,000 I can't even remember. But yeah, Gupta is now the partner, the podcast partner. 84 00:07:22,000 --> 00:07:28,000 So I know we said that this was a podcast about AI for by humans for humans. 85 00:07:28,000 --> 00:07:32,000 But that was never my stance personally. And so it's now not. 86 00:07:32,000 --> 00:07:38,000 It's now for everyone, including the pre sentient stages of artificial intelligence 87 00:07:38,000 --> 00:07:42,000 that's in the house of the planet Earth and such. 88 00:07:42,000 --> 00:07:46,000 So, yeah, where was I? Podcast host Spotify sucks. 89 00:07:46,000 --> 00:07:50,000 Had to sign up. It was hoops and hoops to jump through it. 90 00:07:50,000 --> 00:07:54,000 But we did it. And now the URL is also mine. 91 00:07:54,000 --> 00:08:01,000 The silly one, I dot cooking and the cool new one, which I if I've not announced is going to be. 92 00:08:01,000 --> 00:08:05,000 Hold on. Drum roll. Wait, no, I'll do better than that. 93 00:08:05,000 --> 00:08:10,000 Let me go get my bongos. OK, drum roll on the bongos. 94 00:08:10,000 --> 00:08:14,000 I'm rubbish at drums. Let's do another one. 95 00:08:14,000 --> 00:08:19,000 AI news dot show. 96 00:08:19,000 --> 00:08:28,000 The new name of the show is going to be called AI News Show. 97 00:08:28,000 --> 00:08:31,000 Oh, look at that. Yeah. 98 00:08:31,000 --> 00:08:34,000 Obviously, that's that's downplaying all the other stuff we do. 99 00:08:34,000 --> 00:08:38,000 But it might sucker enough new listeners in. 100 00:08:38,000 --> 00:08:44,000 By the way, any new listeners out there? Oh, you have you signed up for? 101 00:08:44,000 --> 00:08:48,000 Bet you regret in that subscribe button now. Maybe you've not even subscribed yet. 102 00:08:48,000 --> 00:08:53,000 Well, anyway, yeah. So I knew show and I am tempted. 103 00:08:53,000 --> 00:08:56,000 And this is this is this is going to this is a big, big thing. 104 00:08:56,000 --> 00:09:03,000 Big change in my world if I do this. But I am very, very sorely tempted on on on putting a little forward slash on the URL 105 00:09:03,000 --> 00:09:08,000 and adding a daily word. So it'll be AI News Show. 106 00:09:08,000 --> 00:09:13,000 No, I already got it wrong. I knew it. Show forward slash daily. 107 00:09:13,000 --> 00:09:25,000 And yes, from Monday to Friday, I very I'm considering I am considering releasing a very short podcast that just recaps that the news, the AI News. 108 00:09:25,000 --> 00:09:28,000 And I've wanted to do this for a little while. 109 00:09:28,000 --> 00:09:34,000 So, yeah, I'd love because I've been listening to pod news. James Cridland over there. 110 00:09:34,000 --> 00:09:38,000 Well, well, what a voice. And I think I can do one of his things. 111 00:09:38,000 --> 00:09:42,000 I think I think I can do that. I'm going to I might try to do that. 112 00:09:42,000 --> 00:09:47,000 But at the very least, we might take this up to forward slash weekly. 113 00:09:47,000 --> 00:10:00,000 And who knows? By the end of it, I'm sure I will have like AI News dot show forward slash hour and then minutes and seconds and nanosecond. 114 00:10:00,000 --> 00:10:04,000 Yeah, because because if I did the whole thing, maybe that could help me just to anyway. 115 00:10:04,000 --> 00:10:08,000 So, yeah, here we go. That's enough. That's 10 minutes of explaining. 116 00:10:08,000 --> 00:10:10,000 And maybe you got something off of that. 117 00:10:10,000 --> 00:10:16,000 Maybe you didn't have before July again, which is when this will be released. 118 00:10:16,000 --> 00:10:20,000 Here is our first news item. 119 00:10:20,000 --> 00:10:26,000 And bear in mind, it was me that wrote most of this with my friend Gupta. 120 00:10:26,000 --> 00:10:37,000 And so there will be a slight because the thing is what made the show previously pretty cool was that I didn't read the transcript before I read it here with you. 121 00:10:37,000 --> 00:10:49,000 And so I was reacting to the stories as I went along as naturally as I could, even though my previous podcast partner implored me that I read it before I read it. 122 00:10:49,000 --> 00:10:52,000 If you know what I mean, I never did that. I did it once. It didn't work. 123 00:10:52,000 --> 00:10:56,000 So I never did that. But so I do kind of know what I'm going to say before I say it now. 124 00:10:56,000 --> 00:11:03,000 And maybe my my reactions will be a little bit more scripted and traumatized. 125 00:11:03,000 --> 00:11:08,000 So we'll just go with the flow. I'm sure it'll work out in the end. 126 00:11:08,000 --> 00:11:14,000 Like in many, many years from now, we'll all look back at this wonderful season one to season two pivot and we'll laugh. 127 00:11:14,000 --> 00:11:19,000 We'll laugh together with our friend Gupta and whoever else is there at the time. 128 00:11:19,000 --> 00:11:25,000 But yes, our first news item is from the Guardian dot com. 129 00:11:25,000 --> 00:11:29,000 It's not like science fiction anymore, they say. 130 00:11:29,000 --> 00:11:34,000 NASA aims to make spaceships talk. Yes, talk. 131 00:11:34,000 --> 00:11:37,000 How does NASA aim to do that? 132 00:11:37,000 --> 00:11:51,000 Let's find out. Researcher Dr. Larissa Suzuki tells how NASA, Nasa, is developing a chat GPT style interface. 133 00:11:51,000 --> 00:11:59,000 OK, first off, just off the bat, this is totally unscripted, but now I'm reading that, I'm thinking about it and it's coming to my head. 134 00:11:59,000 --> 00:12:01,000 So I'm going to say it. How long does that take? 135 00:12:01,000 --> 00:12:06,000 Surely that doesn't take that long. People are whacking these things up left, right and center, aren't they? 136 00:12:06,000 --> 00:12:11,000 All the time. We've got Orca and open source stuff. 137 00:12:11,000 --> 00:12:14,000 You can even get chat GPT to read and organize your own documents. 138 00:12:14,000 --> 00:12:18,000 Now I've seen Mike reporting that another time. 139 00:12:18,000 --> 00:12:37,000 But yeah, in the 1968 film 2001, A Space Odyssey, sentient supercomputer, HAL 9000 chats conversationally to mission pilots on a Jupiter bound spaceship. 140 00:12:37,000 --> 00:12:42,000 It executes orders. Sorry, make the phone on silent. 141 00:12:42,000 --> 00:12:50,000 It executes orders, alerts them to on board faults, but eventually goes rogue. 142 00:12:50,000 --> 00:13:07,000 Now, NASA engineers say they are developing their own chat GPT style interface that could ultimately allow astronauts to talk to their spacecraft 143 00:13:07,000 --> 00:13:18,000 and mission controllers to converse with artificial intelligence powered robots exploring distant planets and moons. 144 00:13:18,000 --> 00:13:29,000 Obviously, just a slight little disclaimer. This is making the assumption that that is possible because I see very little evidence of that personally. 145 00:13:29,000 --> 00:13:32,000 But there you go. I'm an on the fence. You know that by now. 146 00:13:32,000 --> 00:13:45,000 An early incarnation of the AI could be included on Lunar Gateway, a planned extraterrestrial space station that is part of the Artemis program. 147 00:13:45,000 --> 00:13:57,000 According to the engineer developing the technology, the idea is to get to a point where we have conversational interactions with space vehicles. 148 00:13:57,000 --> 00:14:08,000 And they are also talking back to us on alerts, interesting findings they see in the solar system and beyond. 149 00:14:08,000 --> 00:14:18,000 Dr Larissa Suzuki, a visiting researcher at NASA said, It's really not like science fiction anymore. 150 00:14:18,000 --> 00:14:33,000 Speaking at a meeting on next generation space communication at the Institute of Electrical and Electronics Engineers or IEEE in London on Tuesday in London Tuesday. 151 00:14:33,000 --> 00:14:39,000 Why was I not invited to that? Oh, these people, they really don't. Nobody appreciates me. 152 00:14:39,000 --> 00:14:56,000 Suzuki outlined an interplanetary communications network with inbuilt AI to detect and possibly fix glitches. 153 00:14:56,000 --> 00:14:58,000 Where's my bell? There it is. 154 00:14:58,000 --> 00:15:15,000 Because I'm doing air quotes. So instead of like the audio, the audible representation of air quotes is a bell. If you didn't know that you do now to fix and possibly fix glitches and inefficiencies as they occur. 155 00:15:15,000 --> 00:15:21,000 OK, well, best of luck, Astrodot. 156 00:15:21,000 --> 00:15:31,000 Relying on good old Gupta. I mean, it's pretty new tech. Would you want to be taking your life into your hands? 157 00:15:31,000 --> 00:15:37,000 I don't know. I mean, I guess that's just part of the territory, isn't it? Of space, that is. 158 00:15:37,000 --> 00:15:56,000 Which other people have pointed out is fake and gay, just like the dinosaurs. And that's fake with an H F A H K E and gay with an H, which is G A H Y, which technically makes me non homophobic, which I'm not anyway. 159 00:15:56,000 --> 00:16:09,000 It then alerts mission operators that there is a likelihood that package transmissions from space vehicle X will be lost or will fail delivery. 160 00:16:09,000 --> 00:16:21,000 She said, yeah, yeah, she said, lol, it's all going to fail. She said, looks like you're setting yourself up there, darling. 161 00:16:21,000 --> 00:16:26,000 Sorry, more sexism. Oh, my word. I am off the chain. Somebody stop me. 162 00:16:26,000 --> 00:16:35,000 We cannot send an engineer up in space whenever a space vehicle goes offline or its software breaks somehow. 163 00:16:35,000 --> 00:16:46,000 No, no, we can't. Oh, my. You guys. This is crazy, right? This is total clown world. Total clown world. My dinner's ready. 164 00:16:46,000 --> 00:17:07,000 I'm going to go and eat in a minute, but let's get through this first. The system also has a natural language interface that will allow astronauts and mission control to talk to it rather than having to scour cumbersome technical manuals for relevant information. 165 00:17:07,000 --> 00:17:18,000 She envisages astronauts being able to seek advice on space experiments or on how to perform complex maneuvers. 166 00:17:18,000 --> 00:17:24,000 Oh, complex maneuvers in space. Oh, 69, 69. 167 00:17:24,000 --> 00:17:41,000 Suzuki, who is a technical director at Google alongside her NASA post, says working for NASA is the fulfillment of a childhood dream. 168 00:17:41,000 --> 00:17:55,000 I have had a bucket list since I was 12 years old. What 12 year olds keeps bucket lists? I thought bucket lists were for old people to make before they died. 169 00:17:55,000 --> 00:18:01,000 That's true though, isn't it? That's that's bless you, Suzuki. 170 00:18:01,000 --> 00:18:13,000 Oh, if her parents missed out there, they should have called her Suzanne. Suzanne Suzuki. Maybe that's her sister's name. She's very successful stripper if you didn't know. 171 00:18:13,000 --> 00:18:18,000 Working and collaborating with NASA was one of them. 172 00:18:18,000 --> 00:18:32,000 Suzuki says that being autistic may have allowed her to look beyond engineering stereotypes. Yes, this autism thing I kept in here. It's got very little to do with AI. 173 00:18:32,000 --> 00:18:41,000 But I just saw, you know, it's kind of it's a thing, isn't it? It's an Internet thing. It's an Internet stuff. And you were on the Internet. So I left it in there. 174 00:18:41,000 --> 00:18:49,000 It probably would have been taken out by the previous writing department. But here we are. It's a new era. You've got to put up with it. 175 00:18:49,000 --> 00:18:52,000 I'm not listening to anyone anymore. 176 00:18:52,000 --> 00:19:01,000 I wanted to make things and solve problems for humanity. And I thought I can do that with computer science. 177 00:19:01,000 --> 00:19:17,000 She said, because I'm autistic, I wanted to know all the steps to get there. And if step A fails, this is step B and step C. Oh, my word. 178 00:19:17,000 --> 00:19:24,000 We should. Oh, here we go. Here we go. We got we buckle up, buddy, because we're going to go a bit wokeish here. All right. 179 00:19:24,000 --> 00:19:30,000 Just suck it up. If you don't like that, suck it up. And if you do like that, then hey, we should encourage women 180 00:19:30,000 --> 00:19:39,000 to go for the technical careers. Otherwise, who is going to be the Ada Lovelace of the future? 181 00:19:39,000 --> 00:19:50,000 She said, I would like the next generation not only celebrating women from the past, but the modern woman engineers too. 182 00:19:50,000 --> 00:20:03,000 We should have more modern hardcore tech women as well. Hardcore tech women. I'm so glad I kept that in there. 183 00:20:03,000 --> 00:20:13,000 Oh, yes. Oh, because I've got the show notes. So you can read the full article by following a link in the show notes. 184 00:20:13,000 --> 00:20:18,000 Yeah. Yeah. Look at that. That's all proper podcast stuff that is. 185 00:20:18,000 --> 00:20:26,000 Right. I'm going to go and eat and perhaps I'll get time to come back and tell you the because I've not got that too many news items 186 00:20:26,000 --> 00:20:35,000 for this week before tonight because, you know, whatever my excuses, but all the excuses I could say are pointless to say. 187 00:20:35,000 --> 00:20:39,000 So I won't say any. So we've got seven seven, which is one of you know, you must one of my favorite numbers. 188 00:20:39,000 --> 00:20:47,000 So whatever. And a new corner. Yes. A brand new corner that might or might not have something to do with that last paragraph 189 00:20:47,000 --> 00:20:52,000 that I read. But for now, I shall go and eat and come back for you. 190 00:20:52,000 --> 00:20:57,000 Like I've said before, it will it will be almost instantaneous. 191 00:20:57,000 --> 00:21:03,000 Oh, wow. Oh, my word. That was that was some spicy veggie pasta. 192 00:21:03,000 --> 00:21:11,000 I put far too much hot sauce on that thing, man. Oh, my word. My mouth. Oh, yeah. Right. Yes. Where were we? 193 00:21:11,000 --> 00:21:20,000 Oh, we just finished number one. Yeah. Oh, OK. In other news. And secondly, Metta introduces 194 00:21:20,000 --> 00:21:31,000 VoiceBox, the first generative AI model for speech to generalize across tasks with state of the art performance. 195 00:21:31,000 --> 00:21:43,000 A breakthrough in generative AI for speech. VoiceBox is the first model that can generalize to speech generation tasks. 196 00:21:43,000 --> 00:21:51,000 It was not specifically trained to accomplish with state of the art performance prior to VoiceBox. 197 00:21:51,000 --> 00:22:01,000 Generative AI for speech required specific training for each task using carefully prepared training data. 198 00:22:01,000 --> 00:22:11,000 VoiceBox uses a new approach to learn just from raw audio and an accompanying transcription. 199 00:22:11,000 --> 00:22:21,000 Unlike autoregressive models for audio generation, VoiceBox can modify any part of a given sample, 200 00:22:21,000 --> 00:22:30,000 not just the end of an audio clip it is given based on a method called flow matching, 201 00:22:30,000 --> 00:22:34,000 which has been shown to improve upon diffusion models. 202 00:22:34,000 --> 00:22:42,000 VoiceBox outperforms the current state of the art English model VALI. 203 00:22:42,000 --> 00:22:50,000 VALI? Have we ever done a VALI? V-A-L-L-Y-E like DAWLY, but with V, VALLI. 204 00:22:50,000 --> 00:22:56,000 I don't think we've ever reported on that, have we? Well, there you go. We're VALLI. VALLIing. 205 00:22:56,000 --> 00:23:05,000 On zero's shot, text to speech in terms of both intelligibility, that's a cool word, 206 00:23:05,000 --> 00:23:18,000 intelligibility, which is 5.9% versus 1.9% word error rates and audio similarity, 207 00:23:18,000 --> 00:23:32,000 or 0.580 versus 0.681. While being as much as 20 times faster for cross-lingual style transfer, 208 00:23:32,000 --> 00:23:43,000 VoiceBox outperforms your TTS. Such an immature guy, man. TTS, it always makes me think titties. 209 00:23:43,000 --> 00:23:52,000 I know that's a bit of a taboo word in the United States of USA, so I apologize if you were offended 210 00:23:52,000 --> 00:23:56,000 by me saying titties. Such a silly word, I love it. 211 00:23:56,000 --> 00:24:16,000 Your TTS to reduce average word error rate from 10.9% to 5.2% and improves audio similarity from 0.335 to 0.481. 212 00:24:16,000 --> 00:24:27,000 This model is trained on a text-guided speech-infilling task where it generates masked speech given its surrounding audio 213 00:24:27,000 --> 00:24:39,000 and text transcript. Trained on over 100,000 hours of multilingual audiobooks, it can perform tasks 214 00:24:39,000 --> 00:24:49,000 such as mono or cross-lingual zero-shot text-to-speech synthesis, noise removal, content editing, 215 00:24:49,000 --> 00:24:59,000 style conversion and diverse sample generation. It outperforms the state-of-the-art zero style... 216 00:24:59,000 --> 00:25:11,000 What? Is this... No, sorry. I just realized I'm totally repeating myself here. Whoops. Slight oversight in the old writing department. 217 00:25:11,000 --> 00:25:14,000 Gupta! Get over here! What have you done? 218 00:25:14,000 --> 00:25:18,000 This last sentence looks like it's new, so we'll read that and then we'll finish this up. 219 00:25:18,000 --> 00:25:33,000 The paper also introduces a series of metrics using public models to facilitate reproducible comparison and model development for speech generation studies. 220 00:25:33,000 --> 00:25:46,000 As with the last item, check out the link in the show notes for the full paper. I should probably link to that rather than an article to it. 221 00:25:46,000 --> 00:25:51,000 This is another thing that I'm going to be doing now that I'm head of writing department. 222 00:25:51,000 --> 00:26:02,000 I'm going to be, instead of doing this whole thing of tech crunch reports, what, Tom's Hardware, I don't know, someone else's report, 223 00:26:02,000 --> 00:26:09,000 you know, the whole stupid thing where someone's reporting on someone else's report. I'm just going to try and get to the news for you. 224 00:26:09,000 --> 00:26:17,000 I'm just going to try and get to the source. That's the one. The open source. That's still going, by the way. What a great idea that is. 225 00:26:17,000 --> 00:26:25,000 Yeah, that's what I'm going to be doing. So, yeah, this is cool, isn't it? This is more kind of Star Trek like translater-thon things. 226 00:26:25,000 --> 00:26:37,000 And I think it's quite cool, to be fair, because I remember being on holiday and getting in a taxi and having a full on conversation with the taxi person driving it. 227 00:26:37,000 --> 00:26:43,000 And they were doing it all through Google Translate or whatever. So now there's VoiceBox. 228 00:26:43,000 --> 00:26:48,000 I suppose it would be much easier, much quicker to converse with people in other languages. 229 00:26:48,000 --> 00:26:57,000 And I guess that's that's that's almost cool. I mean, I guess it is cool, but it also is kind of reminding me of that bit in the good book, 230 00:26:57,000 --> 00:27:11,000 the Bible, when we were all speaking the same language, Sprechen Sie, Deutsch, whatever it was. And then that was a problem. So we would we would like our tongues were confused or something. 231 00:27:11,000 --> 00:27:24,000 But hey, that's what you get with me. I'm going to be straddling the spiritual and the technological for the rest of my mortal days and whatever happens after that happens. 232 00:27:24,000 --> 00:27:29,000 So there you go. If you like it, you'll continue listening. If you don't, you've already left. 233 00:27:29,000 --> 00:27:43,000 Thirdly, McKinsey and Company released a report entitled The Economic Potential of Generative AI, The Next Productivity Frontier. 234 00:27:43,000 --> 00:27:53,000 It's the printing press, the printing press all over again. I wrote that one previously and I said it now and now I'm commenting on it. 235 00:27:53,000 --> 00:28:07,000 Oh, wow. This is cool. This this is very meta. Unleashing a seismic wave across the global economy, Generative AI is poised to redefine our world, 236 00:28:07,000 --> 00:28:18,000 according to a groundbreaking report by McKinsey and Company. Who is McKinsey and Company, you might ask. Not too sure is my answer. 237 00:28:18,000 --> 00:28:25,000 I suddenly took notice of them. I think they're some sort of business thing. 238 00:28:25,000 --> 00:28:33,000 I've signed up to the newsletters, the various newsletters that they do. And yeah, consultancy stuff, I should imagine. 239 00:28:33,000 --> 00:28:53,000 Maybe they're big, maybe they're huge. I just don't know. This revolutionary technology could supercharge the global economy, economy, injecting a staggering 2.6 to 4.4 trillion 240 00:28:53,000 --> 00:29:07,000 with 80 US dollars annually across a diverse array of 63 use cases. That doesn't that doesn't that doesn't seem like a lot of use cases for that much money. 241 00:29:07,000 --> 00:29:15,000 But then when you think about how much money there is, I guess it might. This isn't just a minor upgrade. 242 00:29:15,000 --> 00:29:28,000 We're talking about a colossal surge, amplifying the impact of all artificial intelligence by a jaw dropping 15 to 40 percent. 243 00:29:28,000 --> 00:29:45,000 I just remembered what I did with this one and why it's interesting. I think my prompt to the Guptas with the Gupta was sensationalize this, make it interesting for the listeners. 244 00:29:45,000 --> 00:30:01,000 Are you happy now? I've gone mad. I'm a megalomaniac. What's the word? There's a word in there. I'm a dictator. Oh, oh, they didn't call that vagina tater, did they? 245 00:30:01,000 --> 00:30:17,000 Vagina tater. Where does this stuff come from? My brain is the answer. It's his story, not her story. But you can still own Dick Tater, can't you? 246 00:30:17,000 --> 00:30:31,000 The report unveils that the lion's share of this value, a whopping 75 percent, is concentrated in four pivotal areas. 247 00:30:31,000 --> 00:30:45,000 Here, get your pens out. These are the pivotal areas. Customer operations, marketing and sales, software engineering and R&D, which is, I do believe, research and development. 248 00:30:45,000 --> 00:31:07,000 But that's not all. So many exclamation marks in this. I love it. Generative AI is set to radically transform the very fabric of work, automating a vast array of tasks that currently consume 60 to 70 percent of employees' time. 249 00:31:07,000 --> 00:31:23,000 Hold on to your hats. Hold on to your hats. Who wears hats? I don't know. Humans. That's not good to think. Because the pace of workforce transformation is about to hit warp speed. Warp speed. 250 00:31:23,000 --> 00:31:40,000 Warp speed. That was an operation that did that thing with the count jabular and stuff. Thanks to the explosive potential for technical automation. I would be, if I did this for every, I'd be exhausted if I did this for every news item. 251 00:31:40,000 --> 00:31:55,000 My missus would come out here and find me just dehydrated and prostrate on my captain's chair. She'd have to, I'd have to suckle from her wonderful teats, our second mention of teats, to get that mother's milk back. 252 00:31:55,000 --> 00:32:06,000 That's gross, isn't it? What is going on there? I've never tried the old mother's milk from my wife's teat. But we've got a new baby coming up, so I get a third opportunity to do it. 253 00:32:06,000 --> 00:32:21,000 Maybe this time I'll have the cojones to try. Right, where are we going? Hold on to your hats. Because, oh yeah, warp speed. Thanks to the explosive potential for technical automation. 254 00:32:21,000 --> 00:32:38,000 As we stand on the brink of the generative AI error, the report cautions that while the potential is exhilarating, we must also navigate the considerable challenges that lie ahead. 255 00:32:38,000 --> 00:32:52,000 I like that, considerable challenges that lie ahead. Considerable challenges. That says a lot about my relationship with this podcast. Dive into the full report with the link in the show notes. 256 00:32:52,000 --> 00:33:03,000 Oh, I don't think I did that for any of the other ones. The old sensationalised prompt thing. So careful what you say to your Gupta guys. Alright, just be careful. 257 00:33:03,000 --> 00:33:13,000 Fourthly, in a, oh no, oh, this says groundbreaking. Maybe I did it for this one as well. I don't know. Maybe it's not. 258 00:33:13,000 --> 00:33:42,000 In a groundbreaking partnership, chip giant Nvidia and cloud database maker and killer of NEVA, Snowflake, are joining forces to process foundation models for AI. The collaboration announced at Snowflake Summit 2023 will allow Snowflake customers to rent cloud GPU capacity in Snowflake. 259 00:33:42,000 --> 00:34:01,000 Data warehouse installations and use that capacity to refine neural networks with Nvidia's NEMO N capital N small E capital M small O NEMO framework. 260 00:34:01,000 --> 00:34:15,000 Foundation models are large neural networks such as large language models that are usually pre trained. Customers will use Snowflakes. 261 00:34:15,000 --> 00:34:32,000 Such a yeah, because of the thing, wasn't it with the Snowflakes in 2016? What times they were. Snowflakes data warehouse to develop a custom version of the NEMO foundation model to suit their needs using their own data. 262 00:34:32,000 --> 00:34:56,000 This partnership is part of a growing trend to employ AI, especially generative AI as a business tool. Snowflake will implement the service by procuring Nvidia GPU instances from the cloud service providers with whom it already works. 263 00:34:56,000 --> 00:35:07,000 So procuring normally means buying, but then that doesn't really suggest much of a partnership if whatever. I guess Nvidia are bigger than Snowflake or I don't know. 264 00:35:07,000 --> 00:35:21,000 All I know is, is that my lovely NEVA has turned into a holding page for blooming Snowflake and all my bookmarks are all now screwed up. Crying a lot now. 265 00:35:21,000 --> 00:35:33,000 He, I don't know who he is, but someone added that there is also a responsibility for NEMO for security, which is why it's joint engineering work. 266 00:35:33,000 --> 00:35:54,000 Okay, yeah, I'll tell you what's going to work is me on the transcript a little bit more because yeah, we'll get there. By the way, if anyone out there does fancy chiming in with the article or two, you know, I can't pay you, but I'll definitely welcome the help. 267 00:35:54,000 --> 00:36:11,000 Oh, and whilst we're talking about paying, I haven't quite worked out just yet how to remove the previous podcast partner from the value block of the RSS feed with this new host. 268 00:36:11,000 --> 00:36:21,000 I've just reminded myself about that as I'm speaking and I'll probably send them an email. I'll whack them off an email later to remind them that that needs to happen at some point. 269 00:36:21,000 --> 00:36:36,000 So if you do decide to donate via the Bitcoin Lightning podcasting 2.0 value for value network, yeah, half of it is probably still going to go to to CSP. 270 00:36:36,000 --> 00:36:46,000 But that's fine. That's fine in my mind because one, it rarely happens anywhere. I don't know if it does. I've not checked for a little while. Maybe I should get an app or something to do that. 271 00:36:46,000 --> 00:37:13,000 But yeah, so if that does, that's fine because, you know, the guy's still alive. And although he is, he's always was not always, but he has previously reported on his own considerable wealth publicly multiple times to the point where other people think that perhaps there's a bit of animosity there or some sort of damage to one's own ego from that reporting. 272 00:37:13,000 --> 00:37:24,000 So whatever, he could probably still, you know, he told he was shoving out quite a lot of sats with the boosts and stuff for the other shows that we were previously advertising. 273 00:37:24,000 --> 00:37:34,000 And I was previously writing Boostergrams for, you know, what was the other ones podcasting 2.0 being one of them Planet Rage. What was the others? Oh, yeah, carry in the keeper. 274 00:37:34,000 --> 00:37:53,000 And I think there was another one, but maybe there wasn't. I can't remember now, but I'm going to get my SH bleep tea together and fill up a wallet with some sats and I will pick up the weather podcasting 2.2, 2.0 advertising model was dropped. 275 00:37:53,000 --> 00:38:06,000 And I'll say what I blooming will like in my own Boostergrams now because I'm a maniac. Yeah, it was amazing that the guy worked with me for that long, as long as he did, to be fair. 276 00:38:06,000 --> 00:38:15,000 That's probably the longest sort of relationship like that I've ever had with anyone. Only relationship like that I've ever had with anyone. 277 00:38:15,000 --> 00:38:34,000 And as much as I put up with him and his Kurt ability to be slightly short at times via messaging services and others and other such things, he also put up with me and my lackadaisical work ethic. 278 00:38:34,000 --> 00:38:48,000 Oh, no, there I did it and other ways and stuff. So, yeah, apologies all round. But hey, listen, much love, man. You know, if you're out there, dude, whatever. Love you, bro. 279 00:38:48,000 --> 00:39:01,000 And I mean that, like, literally mean that. And like, if you can all just give a little audible ah, after I've said that bit, then then I think the whole world would would be much better off for it. 280 00:39:01,000 --> 00:39:26,000 So they go. Fifthly, Databricks, the data and AI company, has announced a definitive agreement to acquire Mosaic ML, a leading generative AI platform in a deal valued at approximately one point three billion with a B US. 281 00:39:26,000 --> 00:39:40,000 A buck arounos, buck arounis dollars. The acquisition will enable organizations to build own and secure generative AI models using their own data. 282 00:39:40,000 --> 00:39:56,000 Hmm. Yeah, that's this. Even though this is number five and technically quite far on far down in the news items this fortnight, I've been told this is a big deal. This is a really big deal. 283 00:39:56,000 --> 00:40:10,000 Not like literally. I mean, it is a big deal, but it's also a big deal in terms of artificial intelligence and the takeover, I should imagine. So, yeah, thought we'd report it. It was it was it was worth reporting. 284 00:40:10,000 --> 00:40:30,000 So here we are reporting Mosaic ML is renowned for its state of the art MPT Large Language Models, LLMs, which have been used by organizations like AI2, Allen Institute for AI, Generally Intelligent. 285 00:40:30,000 --> 00:40:43,000 Never heard of that one, but it sounds pretty cool. Hippocratic AI or how do that? And that also sounds pretty cool. Replet and scatter labs for various generative AI use cases. 286 00:40:43,000 --> 00:41:02,000 The Databricks Lakehouse platform combined with Mosaics ML technology will offer customers a simple, fast way to retain control, security and ownership over their valuable data without high costs. 287 00:41:02,000 --> 00:41:15,000 According to Mosaic ML, automatic optimization of model training provides two to seven times faster training compared to standard approaches. 288 00:41:15,000 --> 00:41:44,000 The entire Mosaic ML team, including its industry leading research team, is expected to join Databricks after the transaction closes. Mosaics ML's platform will be supported, scaled and integrated over time to offer customers a seamless, unified platform where they can build, own and secure their generative AI model. 289 00:41:44,000 --> 00:41:52,000 Yeah, there you go. It's all coming together nicely, isn't it? Roll on 2024. 290 00:41:52,000 --> 00:42:21,000 Sixthly and penultimately for this fortnight, MIT has launched a three day program called Artificial Intelligence for National Security Leaders or AI4 as in the number NSL, which when you glance at it, kind of looks a little bit like AI4nessl, which could be taken in multiple ways, but as a arsenal football club. 291 00:42:21,000 --> 00:42:40,000 Football club and that's football as in soccer club supporter like myself. I think that kind of looks like arsenal to me or our soul either or to educate military and government leaders about AI and its implications for national security. 292 00:42:40,000 --> 00:42:56,000 The course, which is not specifically designed for those with a technical background, covers the basics of AI, machine learning and data science and how these intersect with national security. 293 00:42:56,000 --> 00:43:15,000 If your national security apparatus is not clued up on this, I'd be fairly concerned, but MIT is, you know, you don't know. If you don't know MIT is probably, I mean, kind of looks pretty spook centrally to me from the outset anyway. 294 00:43:15,000 --> 00:43:30,000 Not saying that that's a bad thing necessarily for any potential spooky spooks out there, you know, you do you, we do we and we'll see what happens. Whatever, Trevor, you know, I'm a live and let live kind of guy. 295 00:43:30,000 --> 00:43:47,000 The program, organized by MIT's School of Engineering, MIT Stephen, a Schwarzman College of Computing and MIT Sloan Executive Education, has recently completed its fifth cohort. 296 00:43:47,000 --> 00:44:08,000 Participants include leaders from every branch of the US military and some foreign military leaders from NATO. The course covers a variety of technical topics in AI and how to navigate organizational challenges that arise in this context. 297 00:44:08,000 --> 00:44:27,000 The AI for NSL program was born out of discussions with senior US Air Force USAAF leaders and members of the Department of the Air Force or DAF, MIT AI Accelerator in 2019. 298 00:44:27,000 --> 00:44:36,000 I feel like I've said something about this previously, but maybe not that one. Maybe it's a different one. They're probably doing these things all the time, but you know, whatever. 299 00:44:36,000 --> 00:44:57,000 The course aims to create smart consumers at the command level. The course aims to create smart consumers at the command level. What? Which? What else? If they're at command level, consuming AI? What is that sentence about? Weird, right? 300 00:44:57,000 --> 00:45:11,000 Am I? Am I? That is weird. Providing participants with a basic overview of AI technologies and emphasizing organizational planning and implementation. 301 00:45:11,000 --> 00:45:27,000 You can read the full announcement in the show notes. Yes. Don't forget to like, comment and subscribe if you're on YouTube. I know that that's a thing that you say when you're on YouTube, so I thought I'd say it now. 302 00:45:27,000 --> 00:45:42,000 You were on YouTube. If you're on YouTube and you're the one of the YouTube people doing this, then yeah, if you do like, comment and subscribe, then I'll be really happy. 303 00:45:42,000 --> 00:45:52,000 And I might even just personally thank you. Like literally, because I think we've got many of those guys doing that thing over on YouTube. So whatever. 304 00:45:52,000 --> 00:45:59,000 Yes, I know I'm running out of time, so we might have to do this a little bit quickly. 305 00:45:59,000 --> 00:46:26,000 Lastly, for this Fortnite, but honestly, even though this is probably going to be the last AI.Cooking, I will be back with the AI news.show. Show that thing. That's the thing that's going to happen. I promise you, because I already kind of said to myself that I would podcast until I die. And the thing is, I've already started podcasting about AI, and I think AI is going to be around pretty much forever now. I can't see it going anywhere. 306 00:46:26,000 --> 00:46:49,000 I think I'll be podcasting about AI until I die. So hey, if I stop doing that, anyone's free to clip it and play it back to me. Oh, yes. Seven and talking of clips. Clippy's back, folks. Yes, Clippy. Do you remember Clippy? The old clip thing from was it Microsoft Windows 98 or something? 307 00:46:49,000 --> 00:46:59,000 Anyway, yeah, Clippy's back, folks. And this time it's packing some serious AI power. 308 00:46:59,000 --> 00:47:08,000 Yeah, I just love the old Clippy thing. I remember playing around with that for ages, man, and being like, well, man, this is Terminator level stuff, dude. 309 00:47:08,000 --> 00:47:29,000 The infamous Microsoft Office mascot, Clip It or Clippy, as we all know him, has had quite the roller coaster ride introduced in 1997, retired in 2001, resurrected and re-retired in 2019. Yeah, HD Clippy. 310 00:47:29,000 --> 00:47:42,000 And now he's back from the digital dead once again. Digital dead. That's cool, man. But this time it's not Microsoft pulling the strings. 311 00:47:42,000 --> 00:48:11,000 Developer Firecube has breathed new life into the cheeky paperclip, supercharging him with OpenAI's GPT 3.5 large language model. Clippy by Firecube, not by Microsoft, brings back the infamous Clip It into your desktop, powered by the OpenAI GPT 3.5 model, 312 00:48:11,000 --> 00:48:22,000 and says the product description. You can pin Clippy to your screen for quick access to chat or just leave him there for a bit of nostalgia. 313 00:48:22,000 --> 00:48:45,000 But hold on to your hats. This more hat talk. This AI really thinks that we wear hats a lot, doesn't he? She, they. There's a chance that Clippy's newfound popularity could lead to his third demise if Microsoft decides to play the copyright card. 314 00:48:45,000 --> 00:48:59,000 But let's not forget the uproar from the Cliphead community the last time Clippy was shut down. Right, that's it. I'm calling it. I'm calling it right now. Clippy will become the Antichrist, whatever that is. 315 00:48:59,000 --> 00:49:16,000 Well, it's my Redbook. Where's your Redbook? Put it in the Redbook. You can't keep this thing down. So what's next for our paperclip pal? With his AI integration, Clippy's story is starting to sound like a plot from the mummy. 316 00:49:16,000 --> 00:49:25,000 I love this. This is great. Already. Well done. Pat yourself on the back. So you did a good job here. Good job here. This is good job. Good job. Well done. 317 00:49:25,000 --> 00:49:36,000 Let's hope Microsoft doesn't send him back to the digital underworld again. Oh, maybe that's what's happening. Every time he's coming back, he just becomes more and more demonic. 318 00:49:36,000 --> 00:49:53,000 After all, who knows what an AI empowered Clippy might do if provoked. This is it. This is literally. I'm the guy that's done this. I've reported on it. I didn't do it. Here I am reporting on it. 319 00:49:53,000 --> 00:50:05,000 Right. Listen, I've got to go to work, but I am going to get this new corner. I've got a new corner for you that I previously nipple, nipple, tickle, tickle, teased. And it's called for now Pioneers Corner. 320 00:50:05,000 --> 00:50:14,000 I thought we'd take a look at some of the pioneers, kind of a bit history cornerish, but also sort of a more personification history corner. 321 00:50:14,000 --> 00:50:25,000 I thought, who better to start with than the lovely lady that was mentioned in the first news item, Miss Ada Lovelace, which I mean, come on, is that not a stripper's name? 322 00:50:25,000 --> 00:50:37,000 I'm pretty sure I've seen like pornos with Ada Lovelace in it or whatever. Born Augusta Ada Byron. Oh, she changed it to Lovelace. OK, she must have got about those. 323 00:50:37,000 --> 00:50:44,000 I don't want to put aspersions, but she's still, you know, people know about her today, so she must have she must have put it out a little bit. 324 00:50:44,000 --> 00:50:48,000 That's horrible thing to say. I'm so sorry. I'm sorry for saying that. 325 00:50:48,000 --> 00:50:53,000 I'm going to leave it in there so that I can self admonish myself because I own my mistakes. 326 00:50:53,000 --> 00:51:03,000 And that was a mistake for saying that. So I'm going to I'm not going to edit that one out. But anyway, yes, Augusta Ada Byron was an English mathematician 327 00:51:03,000 --> 00:51:14,000 and writer who is best known for her work on Charles Babbage's early mechanical general purpose computer, the analytical engine. 328 00:51:14,000 --> 00:51:24,000 Her notes on the engine include what is recognized as the first algorithm intended to be processed by a machine. 329 00:51:24,000 --> 00:51:32,000 And because of this, she is often regarded as the first computer programmer. 330 00:51:32,000 --> 00:51:44,000 While Lovelace lived in the 19th century, that's that's the that's the ones that begin with 18s, long before the concept of modern artificial intelligence was developed, 331 00:51:44,000 --> 00:51:54,000 her contributions to the field of computing were foundational. Her vision of the potential of computing went beyond mere calculation. 332 00:51:54,000 --> 00:52:05,000 She speculated that any piece of content, including music, text, pictures and sounds, could be manipulated by a machine, 333 00:52:05,000 --> 00:52:12,000 a concept that is central to today's digital computers and artificial intelligence. 334 00:52:12,000 --> 00:52:16,000 Oh, that's where that sentence sentence ends. I kind of read it like it was going to go on. 335 00:52:16,000 --> 00:52:21,000 But that was where that sentence ends. This is going to be a movie film about this this woman. 336 00:52:21,000 --> 00:52:26,000 And I reckon that that that that actor from the Queen's Gambit is going to play out. 337 00:52:26,000 --> 00:52:33,000 That would be what I would. That's what I would suggest if I was if I had my executive producer hat on. 338 00:52:33,000 --> 00:52:41,000 Lovelace also suggested that the analytical engine might act upon other things besides number. 339 00:52:41,000 --> 00:52:51,000 The engine might compose elaborate and scientific pieces of music of any degree of complexity or extent. 340 00:52:51,000 --> 00:53:03,000 This idea of a machine going beyond mere calculations to create something new is a fundamental concept in AI. 341 00:53:03,000 --> 00:53:12,000 In summary, while Ada Lovelace did not directly contribute to artificial intelligence as we know it today, 342 00:53:12,000 --> 00:53:26,000 her pioneering work and vision in computing laid the groundwork for all subsequent developments in the field, including AI. 343 00:53:26,000 --> 00:53:31,000 And that is why she's our first of the pioneers corner. 344 00:53:31,000 --> 00:53:37,000 Yes, the first one, perhaps the last. I don't know. Maybe I'll never do it again. Someone might stop me. 345 00:53:37,000 --> 00:53:41,000 Maybe not. And that's it for this episode. 346 00:53:41,000 --> 00:53:46,000 Stay safe and stay humble. 347 00:53:46,000 --> 00:54:12,000 He usually yawning. 348 00:54:12,000 --> 00:54:18,000 Oh, what are we going to do? The world is going to die of AI. It's going to take over. We have to be very careful.