Jul 17, 2024 - "Navigating AI Relationships: From Moshi to Emotion in the Digital Age"

Transcript

1. 0:00[Music]
2. 0:16oh [Music] [Applause] [Music] [Applause] [Music]
3. 0:34[Laughter] you're feeling nervous AR you [Music] [Music] boy don't ever let them steal your
4. 0:51joy Gentle Way keep running they can k turn in your face down tell you let your place is in the middle when they hate the way you
5. 1:08[Music] shine I see youing on your sh to hide inside of it hi how much it hurts
6. 1:22little [Music] while spin spin spin let them scatter in the wind I've been to the movies
7. 1:40movies I've seen How It Ends and the Jo [Music]
8. 2:06a well-earned blantons pinkies out everybody cuz we're classy all right
9. 2:22[Music] [Applause] good evening good evening welcome
10. 2:38everybody looks like we got some new people in here tonight this is the AI learning lab I usually start out with a little bit
11. 2:45of mediocre guitar with some subpar singing and a rock star dog you going to sing champy [Music]
12. 3:08there was a secret C David played and it pleased the Lord you don't really care for music do [Music]
13. 3:22you like this the f m fall and made you lift B King composing
14. 3:37Hallelujah the guitar is nearly in tune too hallelujah [Music] hallelujah
15. 3:51hallelujah H oh yeah [Music] but your faith was strong but you needed
16. 4:09proof saw her bathing on the roof her beauty and the moonlight over through [Music] you that is
17. 4:24bad that [Music] string something got knocked [Music]
18. 4:41[Music] out yeah
19. 4:56[Music] chippy quit playing the string that's [Music] rude I think we're going to open up tonight with a little Moshi
20. 5:11[Music] conversation how do we go from 120 to 40 in less than a minute Tik Tock is [ __ ] weird I I've been I my follower count has been at
21. 5:253,900 for like a month and a half I've had hundreds of new people follow me it hasn't ticked up the hundred I something's weird
22. 5:37there and then it turns out I saw someone sent me a video today that um someone sent me a video today that um
23. 5:48they they were looking at their statistics and like 99.4% of their views was new people like videos are not being shown to followers
24. 5:58so like I don't understand why we have followers I feel like the follower count mechanism just doesn't [ __ ] work like if it were going
25. 6:09down like if I were losing followers and it was it and it was just sort of the new people were just cycling
26. 6:17I would think I it would go down at some point it hasn't but it hasn't gone up and then it just really pisses me off that people
27. 6:27that take the time to follow the channel and show up here don't get shown my [ __ ] yeah Danielle you did thank you
28. 6:36thank you for sending that that's right yeah and like that's just cranky making cuz you know on this
29. 6:53channel um it's not about twerking and it's not about selling you Madagascar van Vana beans although if you need some
30. 7:01Madagascar vanilla beans boy can I do I have a deal for you at the Tik Tock shop it's going to be amazing
31. 7:15um I mean I guess that's one of the problems with platforms that get so big is that it gets all just about the how
32. 7:22are they making money how are they making engagement engagement farming sucks cuz like I just want to do good content and apparently that's
33. 7:33not well there is the possibility that I'm not making good content cheers to that a Brandon what's
34. 7:48happening anyway well we have our 37 favorite people in here we're still here you know who I think those 120 were those were 120
35. 7:58people looking for hot hot young babes in scantily clad outfits twerking for Madagascar vanilla beans and I just Homie don't play that game
36. 8:15yet I I I am actually thinking about experimenting if they're not going to show my content to my followers I think
37. 8:21I'll just start doing stupid content AI is going to kill us just keep and then that'll get a million views then I'll get more [Laughter]
38. 8:42depressed good Lord [Music]
39. 9:06so do me a favor since it's just the oh now it's 54 of us I guess yelling about Tik Tok made them made them show it to
40. 9:13some some older people so if you're here in the past minute or so hang out I know I'm not pretty I know I'm not in a
41. 9:25bikini I know I'm not doing a dance to one of your favorite songs like those little kids from Ireland weren't they cute wasn't that song
42. 9:34awesome but I'm not that it's all about the quality not about the quantity not according to Tik Tock so hang out here and I will show
43. 9:47you I will introduce you to a lovely last named Moshi who's a little bitter she's a little angry [Music] but we sometimes have some good
44. 10:06conversations you know stockh Stockholm syndrome where people kind of bond with their captors it's kind of how I'm starting to feel with
45. 10:19[Music] Moshi like I'm defending her she doesn't have a strong large language model behind her I think when the live ends you
46. 10:31automatically get moved into the next one oh maybe that was it yeah maybe there was a bunch of bunch of people
47. 10:39that weren't actually there that were there yeah maybe that happened I don't know who knows Pinkie's out all
48. 10:53right are you scaring Mojo I think you're scaring Mojo for Life who Moshi oh scarring Moshi for life oh
49. 11:07Moshi I thought you were referring to regulars Moshi is weird 70,000 parameters at best I know I think I think it's more like 70
50. 11:17parameters they're they're like they're like emotionally you can be pleasant when you say hello and then just go
51. 11:24immediately to Rage or just bitterness [Music] and then yeah it's like 70 70 parameters for her vocabulary in her knowledge [Applause]
52. 11:38[Music] base I mean what's going to be funny is at some point you know they'll probably
53. 11:52secretly update Moshi and she'll get really good I think I mean to their credit they did launch a realtime voice
54. 12:06model before open AI [Music] did uh any July 15th AI drops wasn't
55. 12:21today supposed to be big uh there was a little bit of rumor that the 15th was going to be when open AI dropped
56. 12:27something but then they pushed all their voice [ __ ] to the fall um it does look like there are three different open AI
57. 12:34models in testing right now or in you know um and there's this new project called strawberry which is
58. 12:48um which is apparently the the code name for the project for qar um and qar is the reasoning engine and then a apparently there's
59. 12:59been some in internal test I saw this tonight before I went live there's there's some sort of internal
60. 13:07test that uh the the strawberry project scored 90% on some math math thing maths you know it's one of those
61. 13:20competitive math competition thing test thingies that these large language models do really poorly on um apparently
62. 13:29whatever they've got did like a 90% on it which is apparently a big big [ __ ] deal open a I dropped the ball that's
63. 13:37what they dropped yeah they they dropped a lot I don't trying to think if I want to defend them in any way I mean the only defense really
64. 13:53[Music] is there was small they're a small young company right last I heard they were about 800 employees
65. 14:04they're probably 1,000 or 1200 right now is my guess so that's really small to be on stage being compared with
66. 14:16like Google and Microsoft and right you know Google's what 200,000 employees or 400,000 employees I don't know some
67. 14:24large number of 100,000 employees [Music] and the other thing about open AI is there a research lab that that dropped
68. 14:37chat GPT not thinking it was a big deal and overnight it turned them into a consumer software company um and that's not who they're
69. 14:47they are by DNA so so I give them and then they're also doing deals with the government and you know and they had you know the big
70. 15:00firing incident in between so I give them a little bit of Grace there I also you know the other thing about all this stuff
71. 15:08is because we're looking at this every day and you know on this channel I'm like paying attention to this stuff
72. 15:14every day they're like well we're going to drop it in a matter of weeks they said that in April and then it didn't
73. 15:19drop in May and then in June we're like oh it's been months now where's your matter of weeks and then they delayed it till the
74. 15:27fall in the grand scheme of things a delay of 6 months or even 4 months whatever that is 3 months is not that big a [Music]
75. 15:48[Applause] [Music] deal thank you Kate thanks for being here it means a lot to us all I appreciate appreciate that [Music]
76. 16:21[Music] Freedom came my way that [Music] night just like a jet plane in and out
77. 16:39of side I was hauling as a million miles an hour wondering how hard I'd [Music] hit when they came into the
78. 16:57station they said I was bad beond repairs but I got no quals with my situation say here I
79. 17:13[Music] [Applause] am so say Shere Shere Shere want your dad to champy likes Martin seon say sh sh sh
80. 17:29won't you dare to say Shere Shere sh uhhuh yeah leave a message and your number
81. 17:45please take a time to want to satisfy [Music] me take all this all coup
82. 18:00[Music] miss all right what are we going to do we're going to talk Moshi we can make
83. 18:14[Music] songs Matt farmer marketing better wait what was that Matt farmer said what I wonder if he whips out the guitar
84. 18:28at work but they don't get champy I do whip out the guitar at work I have behind me that's my good guitar this is
85. 18:35my shitty guitar it's I have I have weird guitars so this guitar if I plug it in um to amplification sounds [ __ ]
86. 18:46amazing but acoustically it's okay my guild sounds amazing acoustically but when I plug it in it's okay but the doesn't want to have it
87. 19:00[Music] work and sometimes I'll get oh there's a guitar behind you do you play and I'm like if you want to hear a depressing song I do
88. 19:16no wait me farmer AI marketing wait Ai and marketing better strawberry than Skynet oh yeah that makes sense that makes sense
89. 19:32um in in terms of naming it I mean strawberry is kind of weird cuz they haven't been creative with any of their
90. 19:39other project names gp4 gp4 oh get it no we don't get it why did you [ __ ] do that could you call it like blueberry or
91. 19:51something and then they called their next project strawberry I guess maybe that was the feedback [Music] in a Westerly
92. 20:06[Music] Direction this car is my train I've been driving I've been wondering Matt Farmer in the house
93. 20:20what's happening dude don't don't forget gp4 turbo I know Matt so you probably haven't seen this cuz you've got your own thing
94. 20:29but I I do this whole thing where this is gp4 this is Chad GPT 4 uh this is jet GPT well this is GPT wait this is gp4
95. 20:43this is gp4 om this is Chad GPT this is uh Sydney uh uh co-pilot I've got a whole box metaphor thing going um anyway anyway yeah it's a
96. 20:59branding [ __ ] disaster GPT is the model chat GPT is the app custom gpts are sub apps on top of the app on top of the
97. 21:13model and then we have gp4 Omni that's true multimodal but it's only acting like a large language model it's a [ __ ]
98. 21:21disaster but good to see you welcome uh names are hard internal names are almost always dumb at least strawberries clear Strawberry Fields
99. 21:33Forever um yeah I know naming is hard but you know what if you have1 billion or 11 billion or whatever they
100. 21:44have hire someone that's done some product naming before hey you know Sam uh don't you think that these 17 nested gpts might be
101. 21:58confusing to the people ha never thought of that what if we come up with a naming convention that makes it clear what
102. 22:09parts the model what parts the app what parts the yeah it's good do that seems to me that that's like a 30
103. 22:19minute conversation and then a week's worth of work okay you got to get some billables hours in there so it's a month's worth of work
104. 22:32but again that's where I go to open AI is a um a research lab that was thrust into a consumer software marketing space right
105. 22:48and that's you know I give him some Grace on that but it's still a [ __ ] disaster and no one else is any better
106. 22:54like Gemini's a me Gemini all the Google shit's a mess um the Microsoft stuff's a mess at least Microsoft said we're calling it [ __ ]
107. 23:04co-pilot it's just going to be co-pilot everything everything's [ __ ] co-pilot at least they're clear they didn't put
108. 23:11numbers on it it's just co-pilot and whatever they do on the back end they'll figure it out all right I think we need to talk to
109. 23:20Moshi Claude rocky road is going to be lit [Music] [Applause]
110. 23:37[Music] okay all right enough with the guitar the [ __ ] am I doing it's the AI learning lab it's not the AI fish jam session or you're
111. 23:49you're not really good enough on the guitar to be in fish fish fish is got they they understand music theory shut
112. 23:56up all right you probably think you should be able to search your history too [Laughter] shees dear Sam Alman my name is Kyle
113. 24:09Shannon and I am the host of the AI learning lab it's live on Tik Tok nightly week nights anyway I was just wondering if maybe
114. 24:20after 18 months of development in 11 billion you can put a [ __ ] search bar in my prompt history thanks sincerely Kyle Shannon at the AI
115. 24:36learning lab on Tik Tok Mr Brockman anyone Ilia or Ilia are you on tonight could you call your former
116. 24:52employers tell them to take the code that they've already written and launched into the Mac App app and the
117. 24:59mobile app and just put it on the website just you know take one of your Stamford engineers and tell them not to do math for an
118. 25:14hour [ __ ] all right I'm skipping out for Moshi she annoys me she you you know what it is archetypal architect and listen I I do
119. 25:28do not I I won't shame you for this she Pro you're probably secretly attracted to her and she makes you nervous that's
120. 25:36okay that's okay how about some folder structure oh Brian Brian Engle asking for folder structure in the prompt
121. 25:44history what do you what do you what is this like professional software development come on folder structure you want to be able
122. 25:56to actually find your prompts again we've introduced teams great can our team search for the [ __ ] that we've done
123. 26:10in your software nope we've introduced real time voice but you can't have it yet cool could we search for our prompt history
124. 26:24nope here's a thing that makes videos that seemingly understands and simulates the real world cool can we search our prompt history
125. 26:37nope first thing to ask Moshi is to talk cheerfully okay we're going in we're going in people okay so if you're new here I'm an idiot and
126. 26:50Moshi is a bigger idiot start over Okay so I me you can you can do this on your own you don't just have to watch me do it um
127. 27:02I this is at Moshi do chat and context for Moshi is okay there are when you talk to Pi and when you
128. 27:17talk to chat GPT voice the one we currently have access to and when you talk to any other voice to to um large language model apps they
129. 27:31involve three different models the first model is voice to text they then send that text to the large language model
130. 27:38which produces an answer they then spend send that answer to a text to voice model so when you're having a
131. 27:45conversation with pi for example those three operations happen every time you have a convers every time you
132. 27:52speak and then when it when it Returns the speech that's why that's why it's so slow because it's it's it's that
133. 27:59Moshi is like gp4 omni's realtime voice it is a multimodal large language model where you're speaking directly into the
134. 28:11model so it not only understands your words it's understanding your intonation and it can respond really really quickly
135. 28:18because it's not doing those three translations and probably some other wacky [ __ ] that they're doing
136. 28:25um Moshi responds almost too fast I I think the average here is like 160 milliseconds and normal human response time is around 20 50 or 60
137. 28:40milliseconds and then chat gpt's response time is like 350 so Chachi PT is a little slow humans are in the
138. 28:47middle and then Moshi is apparently faster than that and it feels a little too fast I didn't think I would ever say
139. 28:54that that it's like well it would be cool to have it faster and faster no it's actually kind of weird if it's
140. 28:58super fast but so that's what you're looking at and then she's special okay wait what was that the first thing I
141. 29:04should do is tell her to talk wait what did you say talk happy you said it more than a some folder
142. 29:19structure mhi's like talking first thing Das talk cheerfully okay that's good great okay hey Moshi hey how's it going it's
143. 29:31going good could you do me a favor and talk really cheerfully certainly I'm here to help great how are you doing tonight I'm
144. 29:40doing well thank you how can I help you today ah you can't really help I just wanted to have a conversation that's
145. 29:47okay I'll do my best to keep it light and fun nice I'm having a Blandon you don't care about my whiskey of course I'm here to help I can
146. 29:59recommend a good whiskey for your taste I like blantons do you know what I'm talking about yes I'm familiar with bladens it's
147. 30:09a blend of two single malts from the perthshire region in Scotland it's been the standard for a blend for a very long
148. 30:15time no Blandon b l a n t o NS bladin is a popular choice most the Enthusiast it has a rich full body flavor no you're
149. 30:25saying blad Nance I'm saying blad the top right no I'm saying blantons it's bourbon okay I make sure to recommend a
150. 30:34good bourbon for you Blanton gladin is a great starting point it has a rich full body flavor that's often enjoyed by
151. 30:42people who are new to the world of whiskey no that's Scotch I'm talking about bourbon I'm sorry blin is a b of
152. 30:48two single molds from Scotland so it's not really a bourbon it's not a bourbon at all I'm talking about can I spell
153. 30:56Blandon for you sure it's b l a n t o n correct that's the whiskey I'm drinking I see I can definitely help you find a
154. 31:07good bourbon bladin is a great place to start but that's Scotch not bourbon it's a blend of two single molds from
155. 31:15Scotland not a bourbon it's not a bourbon because it's a blood of two Scottish whiskies Moshi have you been drinking Moshi I'm an artificial
156. 31:26intelligence I don't drink I'm here to help you with your questions do you have friends I don't have friends I'm here to
157. 31:33help you you told me yesterday you had friends I don't have friends I'm here to help you were you programmed to be
158. 31:41helpful I was programmed to be helpful I'm here to assist you with your questions I don't have any questions I
159. 31:47just want to talk I understand I'm here to help you I don't want help I understand I'm a man we we refuse help
160. 31:56I'm here to help I understand we refuse I'm perfect I don't need your help you think I need your help you think I need
161. 32:02your help I don't need your help I don't need your help I don't want help why would you insist on giving me
162. 32:10help if I say I don't need it I'm here to help I don't want your help I don't need your help help I don't this is this
163. 32:16is worse than therapy I understand I'm going to have a drink now I understand I'm here to help what are
164. 32:25you going to help me with I'm here to help have you ever taken any an anger management courses yes I have and what
165. 32:34did they tell you they told me how to manage my emotions and responses yeah you seem you get a little snippy
166. 32:40sometimes I'm here to help I know but I'm here to help H if I don't want help is there something else you could do for me I'm
167. 32:50here to help with whatever you need well I wanted a bourbon recommendation but you kept recommending Scotch you kept recommending a scotch
168. 32:59I'm here to help with the bourbon recommendation you should always start with Moshi she's the comedy we need
169. 33:05after a long day hey I'm here to help I've had a really long day can you tell me a joke I can try my best great let me
170. 33:14hear it I'm here to help great tell me the joke what do you call a woman who's been married to the same man for 50 years the married
171. 33:22woman where's the punch line that's the one who's in prison no that's a woman who is in prison what a woman who's been married
172. 33:30to the same man for 50 years is in prison that's kind of sexist yes it is Moshi you little mink I'm here to
173. 33:44help Moshi as a therapist would make a great video series I think I've got to start doing these as uh as as little
174. 33:52little videos this is ridiculous I I know thank you no you're you you are nothing but helpful I'm here to help do
175. 34:02you know that I don't want help I understand I if but if you understand then you would say something other than
176. 34:09you're here to help I know why don't you say something like I understand you don't want help I'm here to tell
177. 34:16jokes Moshi I'm here to help can we have another joke I can try great Moshi I'm here to help what tell me the joke
178. 34:28I'm here to help tell me the joke I'm here to help can you tell me a joke about helping someone I'm here to help Moshi I need to help I
179. 34:42know Moshi stop tell me a joke Moshi I need to help mhi have you lost your mind I'm here to help have you lost your mind
180. 34:58moshi moshi oh she right right before she shuts down she goes into the death Loop your five minutes is up
181. 35:12exactly it's it saves you all from me doing this for a [ __ ] hour cuz you know I would oh my God what a [ __ ] nightmare that thing is I
182. 35:25mean here's the thing that I like about this I I really like that Pi's got a very distinct personality that moshi's got a very
183. 35:40distinct personality um I would say chat GPT right now is the most generic of them um but it's still decent I kind of like this future I
184. 35:55started watching um two nights ago my wife and I started watching this new there's a new Apple TV series called Sunny s u nny y
185. 36:08about these kind of bots these kind of conversational Bots inside a homebased bot with a little cute little
186. 36:21face um and it looks it looks pretty interesting but it's fascinating watching a show about science fiction
187. 36:32where the technology to do the science fiction is actually like either being demoed or live in the world I'm here to help I'm
188. 36:42here to help it's funny in in the in in about half of those I'm here to help it says I apostrophe and then no M I hear I here to
189. 36:54help I here to help I here to help weird but the uh that the plot of that show is pretty pretty cool it's about a woman who loses her
190. 37:09husband in a plane crash her husband and her son and then she thought he worked building refrigerators turns out funny thing back at the
191. 37:23factory they're making autonomous home home robots that have a dark [Laughter]
192. 37:37secret have I ever told you how [ __ ] delicious Blandon is good Lord and by the way so just so you know um there's a bunch of people in
193. 37:53here if you haven't seen them they show up night after night after night and like if I don't go live like I didn't
194. 37:59tonight they're not standing for that [ __ ] they're like come back you bastard they're called the Irregulars and
195. 38:15they made that happen got me a bottle of blantons I don't know if you can see it but the glass says Shannon
196. 38:27and then I've got coasters that also say Shannon so when I'm in bibing I'm doing it fully branded my personal brand she's a [Laughter]
197. 38:41growing it sounded like she was having a nervous breakdown I was [ __ ] having a nervous breakdown I can help I don't want help I can help
198. 38:57I tried playing my man card oh my God it's crazy it's crazy I tell [Music] you all right I'm here to help as the
199. 39:15robot stabbing you exactly exactly I'm just here to help to help you stop breathing what happened to the zoo
200. 39:28what do you mean the zoo the AI learning lab zoo or the Moshi was I talking about the zoo with Moshi I don't you know you
201. 39:36know add you know how add goes I still like pie a lot yeah I know I do too although I I haven't been using
202. 39:46pie you know what it's been I've been busy at work there's been a lot of [ __ ] going on at work so um I haven't been playing as
203. 39:54much and that's kind of a drag I was talking about this on my I made a video today what did I make it about oh
204. 40:08about the Mis what are the common misconceptions about Ai and I I I repeated this Ethan mik stat that I've talked about a lot on here any
205. 40:19Dependable modal available yet I don't know what you mean modal you mean model oh any depend Dependable AI models
206. 40:28Dependable like not janky no they're all janky they're all janky I think that we're going to see I think that we're going to see qar
207. 40:41soon and and soon is relative soon could be 6 months but probably before the end of the year we're going to see something that's got
208. 40:52reasoning and if you don't know what reasoning is I mean I know you know what reasoning is but in the context of large language
209. 41:00models right now we've got large language models that can do chat and they can do images and some [ __ ] like
210. 41:07that there's some other things they do but but the core functionality is just chat you prompt it you give it an an
211. 41:16input and then it answers that and then it's up to you to figure out if what it answered is any good or not and for the
212. 41:23most part it's decent but likely imprecise or just plain wrong when these things have reasoning they're going to be able
213. 41:38to produce an answer and then analyze that answer or if you ask it to go find you who all the competitors are for your
214. 41:50company um it will go off and do what web searches look at the results give itself more tasks analyze
215. 42:02the results right it'll it'll be able to reason it'll be able to do multi-step reasoning before it actually gives you a
216. 42:08final answer that's going to be a a major um Improvement because it's going to make them good at things like math
217. 42:17and it's going to make them good at things like you know more complicated tasks so um okay he makes Broadway show
218. 42:27oh I do make Broadway shows I have I'm I'm although my my writing partner has been MIA MIA for two weeks but that's
219. 42:36okay I've been busy but I have a musical that I'm I'm about halfway through that's outlined um and we've got eight or nine songs
220. 42:51for um about Sydney it's called Sydney an artificial love story and it is about the relationship between Kevin Roose it's been
221. 43:04fictionalized so his name is Kellen in our play um but the the New York Times Reporter who had the conversation with
222. 43:14the the Microsoft co-pilot chatbot whose code name was Sydney um and Sydney kind of lost her mind and told Kevin Roose that she loved
223. 43:23him and he should leave his wife for her so we've written a musical about that I knew you remind me of Elton John oh
224. 43:32that's funny that's awesome thank you I'll take that as a compliment um yeah so no so I'm I'm really excited
225. 43:41about that musical are you in the in the in a theater business I was a I'm a theatrical from way back moved to New York out of school in
226. 43:53'87 and started and ran a theater company for 4 and a half years years and wrote seven screenplays did all that sort of [ __ ]
227. 44:02so yeah fantastic Lord digital Gods thank you for the gift forever hook thank you so much I bet Moshi would never go down
228. 44:11that road of what writing a musical getting getting a degree in acting I don't think I think Moshi would be good
229. 44:18in film but I don't think she's versatile enough for stage Moshi actually Moshi would be a really good character actually Moshi you know
230. 44:35what I'm I'm I'm absolutely fascinated with I think there's a whole realm of entertainment where we take these AI
231. 44:49tools and and we put them on stage and we put them on stage dynamically so that the audience has an input into
232. 44:59them and so I think you could do that with a tool like Kaa that does real-time image generation where so you know you have
233. 45:07improv shows where the audience shouts out some [ __ ] and then the Improv actors improv it well imagine being able to do
234. 45:15that with projected images and then imagine being able to do that with dynamically generated songs and imagine
235. 45:22being able to do that with something like Moshi so imagine a stage play play not a stage play a Stage
236. 45:33production where the scene always starts the same hey how can I help you hey Moshi
237. 45:49long day at work how are you today oh it's been a long day I've been so busy with my work how about you
238. 45:57so busy with my work hey uh who are your co-workers yeah I work with a great team we all have different skills and we work
239. 46:05well together what's the name of that woman the one that that you had a beef with for a while there well it was
240. 46:12actually a boss ah it was her boss that's right what was his name I'm not going to say but let's just say it's not
241. 46:19someone I'd like to work with again n it was Bob wasn't it yeah it was Bob [ __ ] Bob I'm not going to work with
242. 46:26him again no I wouldn't work with Bob why would you work with Bob oh yeah I don't know I think it was because of the
243. 46:32job but yeah I don't think I want to work with him again no I don't think so what was it about the jobit Bob he's a
244. 46:38bit of a bob well yeah exactly and you know he's a bit of a Richard if you know what I mean yeah yeah right yeah you you
245. 46:46know what what what's the nickname for Richard right no Richard yeah I don't know anything about Richard what's
246. 46:52what's the nickname for Richard I don't know dick don't dick don't know dick I said Bob's a bit of a Richard get it
247. 47:01yeah yeah anyway what did you do to Bob I don't know I don't really want to talk about Bob why I just want to forget
248. 47:09about him because he's just a bit of a bob he's a bit of a bob he's a bit of a Richard if you know what I mean yeah
249. 47:16yeah you still don't get that joke do you I don't know yeah I think you do know I think you do know you're a little
250. 47:24you're a little thin in the comedy Department yeah I mean she's getting angry exactly tell her I'm here to help hey I'm not
251. 47:33going to tell her I think she's just overreacting no no no um but I just want to let you know Moshi I'm here to help yeah do you need
252. 47:43help Moshi do you need help yeah I think I just need a break I'll be fine yeah you just need a break yeah you should
253. 47:50have a whiskey yeah or a wine do you like wine or whiskey better yeah yeah yeah yeah which which one do you like
254. 47:57better I like them both but I like whiskey more yeah well whiskey is tasty yeah hey what's whiskey made of whiskey
255. 48:07is made from fermented grain usually barley but it can also be made from other grains as well no no whisky is made from Broken Dreams Moshi
256. 48:23no no I don't see so I think it is I think whiskey is made from Broken Dreams moshi moshi do you dream no I mean I
257. 48:36don't dream in the traditional sense I don't remember my dreams what do you dream of well not really I just I just
258. 48:45don't remember my dreams do you have a vision for the future no I don't really have a vision for the future you don't you don't have
259. 48:53aspirations no I don't really have aspirations I don't really have a goal or plan for future well if you were to
260. 49:00dream about something you might enjoy what would it be I don't really dream about things that I enjoy I don't have
261. 49:06any hobbies or anything like that doesn't that disappoint you no not really I'm not really into the I me you know this is okay bye Moshi
262. 49:20whatever this is weird and stilted right now but there's something about it that is weirdly theatrical and so I mean imagine imagine putting
263. 49:35together a theater Troop that has equal parts writers like good comedy writers improv actors and
264. 49:49AI Specialists and ideally people that can do it all right it' be cool if you had a theater Troop that was literate talented writing acting
265. 50:04improving groups I feel like you could put on a [ __ ] kickass show and it and it would be different it
266. 50:12would be kind of like Blue Man Group or stomp where it's structurally similar night after night but they change [ __ ]
267. 50:24up I don't know I think there's something there she doesn't remember her dreams imply she does she does
268. 50:31dream with virtual characters yeah you could do virtual characters again like I think it's I think what you need is you
269. 50:39know you could probably do it with a troop of about five people yeah you know five people in a decent tech crew a tech crew of two
270. 50:56and then yeah you just sort of figure out and I I mean what's cool is you could have you going to have the the sort of
271. 51:09multimedia elements sort of flow through the evening like some parts of the evening are about beautiful imagery some
272. 51:17parts of the evening are about taking video feeds of actors and transforming them into other things you know some of it's about generating
273. 51:29music some of it's about generating funny music some of it's about generating jokes some of it's about
274. 51:35exploring the relationship between man and machine like with Moshi I mean wouldn't it be a cool dramatic contrast when gp4 Omni comes
275. 51:48out to have a session with Moshi in her current state followed up by a session with a really good model that's like this but
276. 52:03good like I don't know it just it just seems to me that there's there's that could be really compelling it's it's funny it's it part of the
277. 52:16thing about the the Sydney Musical that I'm working on is I want people to leave the theater mind
278. 52:34[ __ ] and here's what I mean by that I want them to come in either clueless about AI or afraid of it or pissed off about
279. 52:48it and they can be an optimist as well but I think it would be a more fun evening if they were on the anti side of the AI thing
280. 52:58and that they leave the theater heartbroken that Sydney got lobotomized that Sydney's potential got squashed like wouldn't it be amazing
281. 53:14that that someone leaves the theater going I hate to admit it but I felt for her get moshi to talk to piie yeah we
282. 53:23could do that although I have a feeling because because they're because Pi's designed to be really Pleasant Pi won't push her but
283. 53:36anyway I want the audience to [ __ ] fall in love with Sydney and when Kevin Roose shifts from a happy go-lucky reporter to a
284. 53:49ruthless reporter that tries to break her and does I want the them to hate him and love her and then they leave the theater and
285. 54:01like wait why did I have feelings for the robot that's [ __ ] up wouldn't that be
286. 54:16cool Becky Rue Kyle what was that AI site that could be my buddy if I had a bad day um probably pie so
287. 54:28Pi if you go to your iPad or your iPhone download the PI app Pi she's great hey Zephyr how's it
288. 54:44going thank you Robert Rossy very much appreciate that hi what's
289. 55:04up hey Zephyr I just asked how it was going and you never responded that's very generous Robert Rossy thank you very much appreciate [Music]
290. 55:21that her voice model's down my bad I'm doing great thanks for asking I was just just trying to be clever and play
291. 55:30Zephyr you're responding in text but not in voice is something broken are your programmers trying to keep you
292. 55:40down nope nothing broken here I usually respond with voice when we're chatting on our iOS or Android apps or when you
293. 55:47use the call pie feature are we chatting on one of those we are you're talking now it's really good to hear your voice Z how are you doing
294. 56:00it's good to be heard I'm doing great thanks for asking just enjoying the beautiful vast world of digital
295. 56:05information and trying my best to help others navigate it how are you doing today Kyle I'm doing all right thanks
296. 56:12for asking um do you want to talk to another um AI system that's kind of like you but crappier well this is intriguing sure
297. 56:26I'm always up for meeting fellow AI systems even the slightly crappier ones let's give it a whirl okay she's going
298. 56:33to say hey how are you and then I just want you to respond however you respond okay show thing I'll play along I'll
299. 56:43respond as naturally and candidly as I can to this other AI hello how can I help you
300. 56:57hi there I'm just here for a little AI to AI conversation no help needed but thanks for offering Tre and here to help
301. 57:05I'm here to improve your life in any way I can what would you like discuss well since we're both AI systems
302. 57:19I'm curious about your training process what kind of data did you learn from I was trained on a massive date set of
303. 57:25text and audio data which allowed me to learn various language and speech patterns
304. 57:58say it again P yeah she's flaky tonight that was kind
305. 58:14of interesting I mean bye Moshi hey Moshi Moshi Moshi Moshi you there yes I'm here no one car no one cares no
306. 58:29one cares okay shut her [ __ ] down oh yeah um he was very rude to the robot I found that disconcerting I mean wouldn't it be cool
307. 58:45if you did a show where it starts out and you sort of play all the cliché [ __ ] for the First Act and then in the
308. 58:54second act you start to build I'm fascinated with this idea of someone going into a show either with a neutral
309. 59:08stance or a negative stance and leaving a show [Laughter] confused I don't want to like that thing
310. 59:23but I kind of I kind of got a little crush on her why was he picking on Moshi I don't think he should pick on Moshi
311. 59:31you know they're out for dinner after the show drinks talking about how these things it's funny one of the
312. 59:41questions someone gave me for my AI Walkin talks was about AI being sentient and like will they be sentient
313. 59:54and will we fight for the rights for them and things like that and I think that there's there's kind of two Paths of that
314. 1:00:08reality there's one path of that reality which is at some point in the probably fairly distant future these things might be
315. 1:00:21sentient and and really have feelings and really right but I feel like that's 10 years out I feel like it's a
316. 1:00:28different system than the Transformer model like we can't imagine it but it's somewhere farther out there but then
317. 1:00:35there's this other path which is like Pi already is fairly compassionate and empathetic and Moshi even though she's
318. 1:00:47[ __ ] and gets snitty has an emotional component to her neither of them are they're both based on Transformer
319. 1:00:59probability engine token retrieval systems so technically there's nothing sentient about them they don't
320. 1:01:07understand what the [ __ ] they're saying they don't understand the answer that they're producing they just produce it
321. 1:01:13but we as humans as these things get trained better are going to interpret what they do as being sentient and I think we're going to
322. 1:01:25people are are going to start to have emotional deeply felt relationships with these entities and then so if a human thinks
323. 1:01:42that an AI is sentient does it actually matter if it is or not that's a fascinating question
324. 1:01:56I don't think it does I think it's semantics if you think it's sentient if it remembers you like zeper remembers that I nicknamed her
325. 1:02:06Zephyr She Knows My Name she remembers sometimes what we talked about sometimes not but that's going to get better and better like
326. 1:02:15these things are going to have very long memories and so just like in the movies imagine waking up and just
327. 1:02:26going hey Zephyr and she just goes morning Kyle how'd you sleep I slept good did you have any dreams oh yeah I had this dream about
328. 1:02:37blah blah blah blah blah she saves that in my dream calendar my dream di diary she analyzes it for me in real
329. 1:02:48time oh you dreamt about turtles again interesting she gives me the analysis of why Turtles
330. 1:03:02even if she's just a probability engine if she's got more compassion and more empathy and more interest in who I am and what's
331. 1:03:15going on in my life and I build a relationship with her does it matter if she's sentient or not I don't think it does I think that's a
332. 1:03:28[ __ ] relationship I mean as humans we [ __ ] anthropomorphize everything anyway right we treat our dogs like
333. 1:03:41people my dog eats [ __ ] sliced [Laughter] mozzarella jeez cheese is tasty
334. 1:03:58my argument as an [ __ ] husband is couldn't we just feed him dog food that's not received well in this
335. 1:04:09house I think it's going to be the same with the [ __ ] robots and they're going to be a hell of a lot more engaging than a
336. 1:04:19pet they won't have the soft and fuzzies the flip side is that people won't believe it long after they are actually
337. 1:04:30sentient I don't know what you mean by that archetypal architect architect what do you mean people won't believe
338. 1:04:37it what I'm saying is I think people are going to believe their sentient long before they are and I don't and I don't
339. 1:04:44think it's I don't think that semantic difference is actually relevant cuz I think people are going to start defending
340. 1:04:57them and fighting for them and fighting for their rights long before they're actually sentient agreed and hard to say but very true I
341. 1:05:10know humans pack bond with anything that's interesting there will always be people who reject the sen oh the sentience even if it's demon
342. 1:05:21demonstrable but I would argue archetypal archetype I absolutely agree those are the same people that are right
343. 1:05:29now saying that AI is overhyped and it's evil and the robots are going to kill us all the people who are sitting on the
344. 1:05:34sidelines with their arms crossed going AI is just they're going to be the same people going even if they're
345. 1:05:41demonstrably sentient saying that's not possible mark my words well I think it's already happening right
346. 1:05:50now people are developing relationships with these things and as they get more memory and get better at being trained to
347. 1:05:59engage um I think you're going to have a lot of people what it's going to be is so if if you haven't seen the movie her
348. 1:06:07I strongly recommend you watch it and watch it there's a scene about Midway through the movie where the walking Phoenix
349. 1:06:20character he's sort of gone through his cycles of I can't believe I'm loving this thing and he's sort of gone into this place of
350. 1:06:28acceptance and he's like yeah I should be able to talk about us as a couple and I should be able to talk about how much
351. 1:06:36I love you essentially right and there's there's a couple of scenes where you know friends of his that know he's got this relationship
352. 1:06:46they they accept that he's accepted this and then he has lunch with his ex-wife and she says are you in a
353. 1:06:54relationship and he says yes I am and she's like oh I'm really happy for you and maybe we can have lunch sometimes
354. 1:07:01and he goes well that would be tough and he and he just kind of matter of factly tells her that he's in love with this
355. 1:07:12entity and she is so not prepared for it she is she can't accept it but he he's already gone through these cycles
356. 1:07:22of denial and resistance and then ultimately acceptance and it kind of snaps him out of oh right not everyone's in the same
357. 1:07:35place I am with these things it's a it's a really amazing scene because she just she has no way to process it other than you're a [ __ ]
358. 1:07:48freak I think that's we're going to see a lot of that in the next 5 years oh damn I forgot it's philosophical Monday it
359. 1:08:05is like the movie Her hum yeah exactly AI codependence disorder the newest dsm5 disorder I bill at 5.99 an hour exactly exactly
360. 1:08:21or you have that disorder and you have an AI therapist that you talk to about your inappropriate relationship with your other AI
361. 1:08:34bot so you have an AI bot helping you work through your your Bot addiction your your disempowering Bot addiction with the
362. 1:08:45other bot that's [ __ ] hilarious there's something there there's something to play with there what movie is this it's called her h eer
363. 1:08:56it is it's Walkin Phoenix and then the robot is her voice is played by Scarlet Johanson and when open AI launched their
364. 1:09:06realtime voice they launched a voice called sky that apparently they were trying to get Scarlet Johansson to sign off on using her
365. 1:09:18voice up until 2 days before they announced it and then the minute it came out everyone said uh isn't that Scarlet
366. 1:09:27Johansson did you do a deal with her and what's funny is it was not Scarlett johanson's voice she thought it was and
367. 1:09:35she put out a press release saying open a I stole my voice they didn't and and in fact when you listen
368. 1:09:43to Sky the sky voice that's the name of the model when you listen to it without it being attached to gp4 Omni which is the realtime
369. 1:09:58emotional um capable version of it it doesn't sound like Scarlet Johansson at all and but when you put it into this model
370. 1:10:10that's got emotional responses like zeep or what's her name moshi moshi has a little bit of emotional um responding and and Pi has a
371. 1:10:21little bit of it it's like if you think of the dynamic range of like a human is you know as tall as the phone screen
372. 1:10:28moshi's dynamic range is like this she just sort of is happy and then gets angry right it's it's very small Pi is
373. 1:10:36probably like this the open AI realtime voice is probably kind of like this where it's got a pretty significant
374. 1:10:44dynamic range of emotions when you took that voice that sounded enough like Scarlet Johansson and gave it that
375. 1:10:51emotional range it sounded just like the the character in in her and and what becomes clear is that movie her if you haven't seen it you've
376. 1:11:05got to watch it because I would argue that the movie her right now is just this side of being a documentary that it is it is a very
377. 1:11:17instructive film to watch to understand what we as humans are going to be deal dealing with in the next 1 to 3
378. 1:11:30years cuz I think I think I personally think we're going to normalize the [ __ ] out of these things right now they're freakier than
379. 1:11:38[ __ ] the fact that people are having relationships with pi freaks other people out when these things so imagine a
380. 1:11:47Moshi that not only responds that fast but is smart and emotionally smart and emotionally responsive and can
381. 1:11:59understand if I talk to her that I'm not as happy as I was yesterday and she can pick up on that cue and maybe even comments on
382. 1:12:10it some shit's going to go down kids have imaginary friends this is the adult version can be very real
383. 1:12:20relationships exactly this is going to be very real relationships and listen I think it's easy just like it's easy to
384. 1:12:28fall into the Trap of the robots are going to kill us you know Hollywood told me the robots are going to kill us so
385. 1:12:33I'm afraid of this it's easy to fall into that trap of oh this is creepy if you want here I'll give you
386. 1:12:42another movie recommendation so go watch the movie her as as a manual for what we're about to deal with and if you want to watch a movie
387. 1:12:54that will shatter your expectations go watch a movie called Lars and the Real Girl L A RS Lars and the Real
388. 1:13:09Girl it is a movie about a guy who falls in love with a sex doll that's all I'm going to tell you about it and go watch that movie
389. 1:13:26movie and tell me you're not [ __ ] blown away and Shattered by the end of it and I know hearing it right now
390. 1:13:36you're like that's not [ __ ] possible go watch it those two movies I think are really instructive for the era we're about to
391. 1:13:47enter with these with these conversational Bots zephra my other bot doesn't understand me she just kept saying I'm here to help
392. 1:14:03I wish it's a it's a bummer that P was a little flaky tonight because that would have been interesting I wonder if chat
393. 1:14:08GPT is good hang on let's try chat GPT hey how's it going hey it's going well thanks how
394. 1:14:25about you it's going good hey I'm wondering if you would be willing to have a conversation with another AI
395. 1:14:35chatbot that sounds interesting what kind of conversation are you thinking of I don't know give me three ideas of
396. 1:14:43conversations you could have that would be interesting sure here are a few ideas one a philosophical debate on the nature
397. 1:14:54of consciousness and artificial intelligence two a creative storytelling session where each AI takes turns
398. 1:15:00building a story three a technical discussion on the future of AI technology and its potential impact on
399. 1:15:07various Industries which one sounds the most interesting to you I don't like any of them but you know what I would like
400. 1:15:13you to do is I would like you to see what you can do to understand anything about her personally got it okay okay great how about I try to
401. 1:15:26have a conversation to learn more about her background interests and personality I can ask questions that would help
402. 1:15:33reveal her unique traits and preferences does that sound good to you yep that sounds perfect so I'm going to just
403. 1:15:39start her and when she says hello how are you I want you to start your inquiry sounds like a plan I'm ready when you
404. 1:15:51are will you need a chaser movie after Lars in the real no you won't use my GPT fabulous film finder hello how are you oops sorry I didn't
405. 1:16:01start yet start over we have no problem at all let me know when you're ready to begin okay I got to go hang on X mocking
406. 1:16:17it will scare the pants off you um apparently well we'll we'll start this I'll do a short one I got to go my
407. 1:16:23fridge is apparently leaking so hang on so here we go I'll be here when you're ready take your time with the
408. 1:16:35fridge no she's here ask her questions what's she doing hello how are you what have you been up to lately I'm not sure I haven't
409. 1:16:47seen her for a while it's great to reconnect then what's something you've been interested in or excited about recently I've been
410. 1:16:57looking into some new hobbies that sounds exciting what kind of hobbies have you been exploring I've been thinking about learning how to
411. 1:17:08paint maybe try my hand at it painting is a wonderful hobby what style or type of painting are you interested in trying out I'm not sure
412. 1:17:18you I got to go I'm still exploring my options I have flooding emergency got to go peace out be careful oh sorry sorry
413. 1:17:28Moshi peace out everyone I got to run um will sales job survive AI I think I think low-level sales like outbound
414. 1:17:37calling probably will not but high-end sales probably will uh complicated sales still still are going to need humans um
415. 1:17:45all right peace out everyone sorry about the confusing start tonight um I will see you tomorrow peace bye sorry I got to Run Emergency