Feb 1 2024 - THUNDERVIDEO CEO PANEL

Transcript

1. 0:04perfect just yeah come on you're on Tik Tok I'm ticking talking yo cats we're going
2. 0:24live I love you man all right ladies and gentlemen like what you're doing I mean just it's panel time just give you guys about 6 seconds to
3. 0:38wrap up your conversation two one shut the [ __ ] up sorry I'm such a dick I'm
4. 0:53sorry all right thank you so much for coming out dinner was fabulous thank you again for enjoying that we have an esteemed panel in front
5. 1:08of us this evening I don't say that lightly and what we trying we're doing an experiment this might totally [ __ ]
6. 1:14fail if it fails we're going to turn this into a normal panel but my goal is to have all of the questions that you
7. 1:22submitted or at least some of them ask to this group of panelists and it's not so much that they're going to answer to
8. 1:30the audience their thoughts but instead they're going to have a conversation with one another and we're going to get
9. 1:36to listening if that doesn't work we'll change it up but let's give this thing a shot so I'm literally going to stick
10. 1:42this microphone back up here everybody's going to introduce themselves and then we're just going to get into the [ __ ] or
11. 1:47they're going to get in the [ __ ] I'm going to sit down so everybody up here thank you so much for taking part in this thank you
12. 2:00here we go introduce yourselves does this work can you hear me we're try all right it doesn't really work ER promises this is going to work
13. 2:09can you hear me okay hello we're trying we're trying it let's see what happens hello okay hi I'm Margaret I'm the founder of lass of
14. 2:21Health uh I introduced myself at the beginning already but um for AI we're using large language models for a chatot
15. 2:29product to keep high-risk Patients Out of the emergency room and we're also using machine learning models structured
16. 2:35learning to do Predictive Analytics on electronic health records to flag undiagnosed patients and I've also
17. 2:42integrated a lot of AI into my personal workflow to optimize uh many things of startup founder life so that's my expertise I'm
18. 2:51sharing hey hey sorry Fe backing hello is that going to work driver left and right [Music] right hello hello oh I'm stereo I like
19. 3:09it fantastic but I'll tell him what he's W uh my name is Kyle Shannon uh coery of Story vine um on the AI front um I guess
20. 3:18I got my religion in in 2022 with stable diffusion and dream Booth uh doing some sort of crazy self-portraits that I turned into an art
21. 3:28project and then when chat GPT La launched I lost my flipping mind and I started the AI Salon a Tik Tok channel
22. 3:37uh and a newsletter as well as completely rethought my business uh so I've spent the last year trying to
23. 3:43figure out what the hell I'm doing with my life shout out to all the Irregulars yeah the Irregulars yo that's my Tik Tok Channel people
24. 3:54oh um so I'm Ian Karens uh I am co-founder and CEO at free play um we're B2B nerds we developer product nerds we
25. 4:05help uh product development teams so like Engineers product managers designers QA people figure out how to
26. 4:13build and ship uh great software that uses llms so we build tools for experimentation and testing and monitoring and ultimately like we're
27. 4:24helping people figure out can they trust these things for doing the right thing how do they build and ship faster more
28. 4:29confidence how do they save money and really optimize what they're doing um and yeah get to spend a lot of time with
29. 4:37seed stage to public companies getting to see what they're all building so I've gotten to do this for the last year and
30. 4:42change and probably talk to 200 uh software companies that are building with LMS and start away hey everybody
31. 4:55I'm hello is it working no it's it's dead yeah that's why there's no feedback you hear me now hey everybody I'm Kira
32. 5:04nman I'm a partner at e Partners we're at preed Focus deep Tech firm um based in the Bay Area I came out to Denver a
33. 5:12couple years ago to establish our Denver Outpost and uh yeah look at things like robotics synthetic biology uh he stays a
34. 5:20lot in generative AI um especially in the infrastructure and I'm on a handful of boards about half of them have an AI
35. 5:27or ml component to them so excited for the conversation I'm Chris Erikson a co-founder and one of the managing
36. 5:35directors at range Ventures we are a Colorado focused preed and Seed fund here uh we spend a lot of time we're not
37. 5:43focused on a specific industry or deep Tech like Cur is but as you can imagine a lot of what we are seeing today either
38. 5:50as core infrastructure to the AI world or incorporates in some form and so we spend a lot of time thinking about where
39. 5:58Valu is created now where it might be created in the future and where we want to invest in type of Founders in the
40. 6:07space someone have a question all right here we go we're on our own there's no moderator yeah well we got we got a
41. 6:13stack of questions but we can also go off script you want to start with the first question no I think we should go off
42. 6:19script to go we want to go off script I can actually start that um here's here you want to have a real discussion right um this was
43. 6:31actually a discussion we had for two hours as a team today at range um that we were trying to debate is we're
44. 6:37thinking about where we want to invest right um our two biggest concerns that we don't have answers to are do we think
45. 6:45the majority of where llms and whatever model is going to live within companies behind their own firewalls right or in
46. 6:53public secure clouds um and down the road do we think that the pace of Innovation is going to be so fast with a
47. 7:01lot of the foundational models anyone building specific features today around the edges are just going to be replaced
48. 7:08about those and so that was our entirety two hours earlier today as a team with r so the thoughts yes I I don't know I don't know how you
49. 7:22invest in what's coming I what one of the things that Sam Sam Alman was interviewed at Davos this week and he said um he said people should
50. 7:34think about chat gp4 as a preview and um the change that come that's coming is exponential that we basically have no idea what's coming
51. 7:45that that that the rate of change is going to be so dramatic my experience over the past year has been the stuff that I figured out last
52. 7:55February where I thought oh yeah I got this prompting thing down this is good then CAD gbd4 came out and I was like ah
53. 8:01damn I don't know what I'm doing and then it got multimodal and and so it's it's every 3 months the person that was
54. 8:08the expert before is clueless you know today and and I think not only does that keep happening I think it accelerates so
55. 8:17I think it gets really weird really fast so I have no answer for you other than I I think I think the big
56. 8:25risk is is that there's no moat that famous you know the the Google engineer that said we have no mo I think that's
57. 8:32the risk I think it's you know bet on the jockey do you have good teams that can adapt and learn and are curious and are [Music]
58. 8:46engaged natural conversation um well I I think like one of our questions and we were actually talking about this this earlier that was
59. 8:55like can you bet on a jockey when the other Jackie has1 billion in all of the gpus right like sorry for
60. 9:09well [ __ ] well I was just gonna Echo that point because uh I saw a graph from Benedict Evans that showed like hundred
61. 9:19billion dollars going from Google Amazon and Facebook into infrastructure this year so I think I think the point's
62. 9:25really valid and I also think there's going to be a lot of models that ultimately like everything in Tech ends
63. 9:30up being like free and cheaper and I just think that's where these models are going and so I I just think open source
64. 9:36is also eventually where this is going yeah uh my response to the the100 million thing I actually think there's
65. 9:43two markets here there's the tech Market where you're competing against the 100 billion Investments and then there's the
66. 9:49inflection point that happened on November 30th 2022 the generative AI Market it's a different market right so
67. 9:55in my opinion chbt does for machine learning learning in AI what the worldwide web did for the internet where
68. 10:03it makes it accessible to everyone um in my community the the the uh Irregulars in the AI Salon there are a lot of
69. 10:12developers in my community right now who've never read the line of code they're they're making gpts they're
70. 10:17doing them passionately and they're figuring out ways to express themselves in in ways that they they never would
71. 10:23have done it before so I I think we're going to start to see Innovation happening not from expected channels but
72. 10:29for very unexpected channels because of the democratization of skills your turn yeah I mean really unless you're
73. 10:41doing you're building out these LMS or you're working in a pure AI company for a company like ours that's using AI as a
74. 10:48tool to us it really matters what's the problem we're trying to solve like our customers doctor's Health Systems they
75. 10:54don't care if for using AI like they care that we're solving their problem and we're saving them money and AI is a
76. 11:00great tool that we use but right now it's it's not defensible and that's why it's not our only product our machine
77. 11:05learning models and the data set that we have for our machine learning models that is proprietary and that is a defensible mode uh and we're
78. 11:12experimenting with these llms and what they can do but not having that be our core Focus because um there is so many
79. 11:19unknowns and we can't put all our eggs in one basket right now this is like that thing like and you haven't talked yet so now what I mean I
80. 11:29I don't know come up with something smart or ask a question or ask a question I I I want to extend Kyle's
81. 11:35analogy about the worldwide web this is like too close it's very close god mode um I mean like we were having this conversation at dinner
82. 11:47earlier and you know the point that you made that like there was a long time before the worldwide web and then what
83. 11:55happened after right like the first time that I put something on the inter internet I'm I'm old enough that this is
84. 12:01true was like hand coded HTML I was like oh my gosh there's Times New Roman with blue underlined links and you can click
85. 12:10it and that was cool right and then like you know I was then in early in my career we were like helping people set
86. 12:18up you know websites with the lamp stack installing it on bare metal we knew this guy Thomas that ran the server like you
87. 12:25know and like now it's some of you are like what are you talking about right and it's web flow and framer and you
88. 12:31don't need to think about it right and I feel like we're in the handed HTML era of AI products but that or
89. 12:41earlier maybe earlier but like I mean whatever it is it's like in the same way that like web technology has advanced
90. 12:50and maybe it took 30 years to get here like it it might only take 10 for an equivalent amount of change he's like a super cool three
91. 12:59three I so this is why I don't say three right because like I I do get to spend time with some of like the literally
92. 13:05biggest companies in the world that are doing this stuff and they are still trying to figure out how to just get the
93. 13:11basics to work because like there are so many things that stand in the way of really like running this at scale and
94. 13:18doing business and there are a lot of people that are like oh my gosh the future is coming so fast and then like I
95. 13:24don't know you look at all the like agent chatter on Twitter and like for all the hype a lot it just doesn't work
96. 13:30there's like I don't know three is going to be a very different world from the foundation model perspective I wonder
97. 13:36about the application side but to kind of Chris's point it's like you know I don't know are should should we should
98. 13:43we bet everything on Google or like are there in the same way that the web like spawned this whole generation of
99. 13:50builders that created amazing things because they were solving real problems like I'm in that camp right like I think
100. 13:55there's going to be a lot of people that do amazing stuff over to I actually think there's not just a technology question here right about
101. 14:04your point right like today you the the CTO can't even win an argument with the general counsel of the company about
102. 14:10what they're going to do with their data and how they're going to train their models and that type of stuff right and
103. 14:14so I think there's I think the tech to your point pal is like going to advance especially on the foundational Mite
104. 14:20really quickly but how it actually comes into the Enterprise not just from all of the tech they have to build that they
105. 14:25don't know how to build there's a lot of people and human legal questions that I think have to be answered to get that to
106. 14:33the end user too and those might be harder than some of the tech that are being done right now good yeah I
107. 14:43mean this I think a lot of these llm Solutions are targeting labor and human roles and functions and um I mean I read
108. 14:53from nfx like if you don't significantly reduce Opex Beyond a single job like you're not doing your job with the
109. 15:00technology so um I think there's a huge huge human component to it I think something I think a lot about is that
110. 15:06this is a lot of human augmentation and um kind of like empowerment if you think like there like what way is the world
111. 15:13going are we you know are we about to have a big brother and be controlled or is this about us you know enhancing our
112. 15:20abilities and getting to more human parts of what we all enjoy doing so I think there's some existential questions in there too
113. 15:31the one of the one of the biggest surprises for me about AI to talk what you're talking about the human factor of
114. 15:39this when I first got access to chat GPT the first thing I did was I went to zappier and I went o automations
115. 15:45productivity this is going to be about productivity and and making things more efficient I got really bored with that really fast
116. 15:54and what's happened to me over the past I don't know six or seven months is I'm now excited about things I haven't been
117. 16:04excited about in a long long time a lot of people in my community um are now artists but they told themselves 30 years ago that they
118. 16:14couldn't do that and and I think people as they're leveling up their skills they're rediscovering talents that they didn't
119. 16:22know they have passions that they didn't know they had one of the women um in the community said she feels like she five
120. 16:29or eight again that that she's got a the Creative Energy of five or 8-year-old the impact on that for me for
121. 16:36business is that if you get 10 of those people that have that youthful enthusiasm or a fivey or 8-y old with
122. 16:45these incredible tools you're going to have a 10p person company show up that's going to redefine and compete with a 100
123. 16:53or a thousand person company and they're not going to know what to hit them so yes I agree people are going to have to
124. 16:58solve the lawyerly things and the tech things and the Privacy things um and they're going to get absolutely
125. 17:04blindsided but by a 10 people that are like gee Shucks well you know I got picture skills and I got word skills
126. 17:12let's get some Ai and put on a show um so so I think that the unexpectedness of competition is the thing that is is
127. 17:19probably the the thing I'm I'm most excited about and it's going to be a [ __ ] train wreck so and on that so what can can I jump in for one
128. 17:33second dude you're tell me the 10 [ __ ] youthful enthusiasts are going to compete against 100 experts yes [ __ ]
129. 17:40that [ __ ] that's [ __ ] dude no wrong I'm seeing it right now I've got I've got major Pharma companies right right
130. 17:47now that are asking me how they should manage their ad agencies because their ad agencies are saying listen we're us
131. 17:56this is how we've always done it we've got these 50 people that we pay minimum wage and we sit them in a pen and we
132. 18:02have them write copy all day and we bill you $120 an hour for that you're not going to [ __ ] with that and a 10 person
133. 18:09agency fully AI enabled is going to come in and wipe the floor with it I will bet you ,000 that that doesn't happen in the
134. 18:18next three years okay not not at scale you hear it here $1,000 next three years a 10 person company doesn't wipe
135. 18:26out a 100 person company I'll take that bet all [ __ ] day long once or at scale what do you mean once or at scale
136. 18:33meaning there's a huge difference between somebody goes [ __ ] it I hate spending a million dollars on my uh on
137. 18:39my agency of record let's try take a flyer but is that going happen regularly yeah I so I think what here's what's
138. 18:46going to happen I think small agencies because they can work faster are going to shift away from the time
139. 18:52and materials model and the holding companies aren't going to be able to compete so okay so I I've heard a lot of
140. 19:01people saying okay AI is really bad at writing coffee like you can tell it looks fake when people do like d post or
141. 19:06Instagram but there was a new poll just a couple weeks ago that that actually did a test and showed people human
142. 19:12coffee and AI coffee and in every category they looked at every type of content creation people preferred the AI
143. 19:19like that's where it's getting now yeah hold on every single one of you received an email last week that had my coffee and the AI
144. 19:29copy you will not offend me what do you like who likes my copy raise your hand how do we know the difference what
145. 19:51was wait this is a red heron this is a red we legitimately tested this on this group last week here that's a red Haring cuz here's
146. 20:00here's the reality of how you actually use these tools is not make me copy copy and paste that's not how I use it that's
147. 20:09not how any creative person I know uses it what it's a guide to like it's a guide It's a framework it's here where's
148. 20:17where where where generative AI is right now is that I can confidently get to 80% of good in minutes getting from 80% to
149. 20:27100% % is just as hard as it ever was and AI will occasionally get me from 80% to maybe 87% that last so what ends up happening
150. 20:38is in the olden timey days 2022 I would take two hours to write a first draft and then I would take the
151. 20:45next week or so to like massaging now I get to that first draft in 10 minutes and I spend the two hours making it
152. 20:52really good and so it's a collaborative back and forth so very little of what I put out there is the AI wrote this you
153. 21:00love that so we're investors in a company called brandguard and it's governance for brands that are using machine generated content so we're
154. 21:09talking about um friends like Pharma like thermofisher to Coca-Cola to pis you know um wanting to use this
155. 21:19technology but um I guess the point that I think is important here is they have like 36 Ensemble models that are
156. 21:27evaluating and because you're working with different types of companies you know like safety regulation what's on
157. 21:32brand Amazon does um you know like Christmas campaigns and if you don't have a model that understands like
158. 21:39seasonality you're going to give it the wrong score because each piece of content gets the score so the point of
159. 21:45all that is to say like this company exists because it's really hard to tell the difference and the governance layer
160. 21:53of like the human being able to say okay this is machine generated content let me from in an automated way
161. 21:58um have a machine tool powering my human review of it I want to go back to S this this bet and this discussion here right I'm not
162. 22:07going to agree with Eric just because he's hosted this and invited you know us to this but I think if you look at like
163. 22:13most of the big technology changes over the last 30 years right starting with the PC client server right web then
164. 22:21going to Mobile and all those things and most of the disruption right happened with small companies and all the big
165. 22:28companies ignored it or said this isn't going to happen this doesn't matter right I think if you were to pull the CI
166. 22:36CTO and CEOs of all the fortune, companies and ask them what are your top three concerns or top of mind right now
167. 22:43ai would be in the top three of all of them and I don't think anyone is ignoring this right you look at like
168. 22:49who's buying the gpus and things like that it's not just Facebook who's buying a [ __ ] one of them right but it's
169. 22:55companies like John Deere and Mercedes right no one is ignoring what's going on here so sure will there be more Nimble
170. 23:02faster smaller servicing that adopt the tools quicker but I don't think that this is being ignored in the same way
171. 23:09that a lot of the last Evolutions are I'm sure Ian you probably see this because of the companies you're talking
172. 23:14like companies you probably thought would never be having a conversation about bading at Technologies or like
173. 23:19help me figure this out because I know I need to do it right now right and I think that's what's different about this
174. 23:25change than all of the other Cycles we've been through in the last 30 years is it's Broad and it's not narrow and
175. 23:32people aren't saying this doesn't matter what what was the rest of your conversation in your room oh you I got
176. 23:38you guys have another hour and 50 minutes yeah um so there were there were two pieces largely of it it was where is
177. 23:46the where the model is going to reside are people going to be comfortable with them being run in the secure open AI
178. 23:53version on Azure right and sending your data and your parameters and whatever to right or are a lot of the companies
179. 24:00going to say no this has to live on our infrastructure behind our firewall right and we'll run an open source model and
180. 24:06we'll buy the gpus we'll train and do all this um that was one and then it was well is that even going to matter
181. 24:12because the big guys are going to lose so fast that whatever you're doing in house like gets wiped out so quickly
182. 24:18that any of your privacy or whatever concerns are irrelevant because your texts IR relevant and we were having
183. 24:24that conversation because our whole thing is we don't want to have to decide on which of those worlds is going to
184. 24:30happen we want to make bets their agnostic of either versions of those worlds and so I don't think we reached a
185. 24:36conclusion but we just argued for two hours about it so any help please I mean that that topic is one that I see a fair
186. 24:43am like there there has been all this debate about open source and like what's going to happen here I I don't know hey what's up for
187. 24:54regulars I guess I'm like you know putting myself out there but I have built businesses around open source software as part of the team
188. 25:02launch mapbox work a bunch of Open Source stuff in the past like I was a Deb day with open AI this year that
189. 25:07company is crushing it you know um and the folks at Google and in anthropic are like not messing around and there's like
190. 25:15real serious work happening but there's also just like Innovation happening that the open source models aren't doing yet
191. 25:21where like open the ey function calling right or like tool use and make it out of first class citizen is just like it
192. 25:27is value ad that's differentiating for um Builders and so I think what we've been seeing is like the Privacy problem
193. 25:36is not what's going to cause the big companies to not use these models because you better believe AWS and gcp
194. 25:44and Azure are incentivized to solve that problem so you know people are setting up these like peering systems you know
195. 25:51essentially running those models in a secure enough environment they're going to open source just for for Pure
196. 25:58efficiency so cost eleny right and you know if you're going to get to a place where you're like I need to do complex
197. 26:03reasoning in a chatbot that might have wild stuff thrown at me then I'm going to use open AI or anthropic or Gemini
198. 26:10right if I'm going to like label whether this is like you know a mean or a nice tweet like you know then that's a great
199. 26:17opportunity to use open source in like a scill cheaper way and so I think that's that's the current moment in time but I
200. 26:24do think it's less about just like the pure um um I guess what I think about deing sort of stuff right now not seeing
201. 26:31the company maybe Mist does this you I think ml's thinking about it a smart way but like doing the product Innovation
202. 26:39everybody's like just arguing about the quality of the model like you know how can you beat the like hell swag you know
203. 26:44benchmark right but like what opening eyes do is so sticky is building like the function calling feature and then
204. 26:51other things after that right and if this is just like way in the we you're like what are we talking about like give
205. 26:57us like a thumbs up and we'll adjust closer all right closer to the mic I don't I thought I was going to
206. 27:04beat back sorry um I got the feedback control over here nice Eric's in control of all the things what other questions do we have
207. 27:16yeah I I I I think it's the agility that really is going to make the difference of figuring out when to use the right
208. 27:23kind of models when it makes sense for open source and when it makes sense to use open AI I think the the qar things that they're
209. 27:30working on now are really groundbreaking because that's what a lot of large language models really struggle with is
210. 27:35that logical reasoning and you know if they're able to really execute a model like they're talking about with qar like
211. 27:41that's that is really groundbreaking and be hard to replicate um but from a from a person building off of those models
212. 27:48it's really expensive to run uh and if there are lower cost open source ones it's better to use those but to use the
213. 27:55better models for specific tools that were you do need that higher quality of logical reasoning I think one of the things
214. 28:02that's so cool about what you're doing in defensible right is like a big thing you're doing right now is You're
215. 28:07Building trust right with your users and they're showing it can work and so maybe there ends up being you know a better
216. 28:14model that could specifically answer this but you've built a bunch of trust with users that they know in in a use
217. 28:20case where trust really matters right and I think actually as we start we think about Moes is like again maybe we
218. 28:26get back less to it's the tech and it's actually more how you understand your users and this is just a piece of how
219. 28:32you deliver it right and I think that's what's so cool about what you're doing already is you're building that trust
220. 28:38with these things and that could be a really defensible and valuable hope long term right yeah I got so the Miriam
221. 28:48Webster's word of the Mr feedback how's it going um Miriam Webster's word of the year for 2023 was AU
222. 28:57authenticity and I think that um I think that trust and authenticity my company story of M what we do is it's automated video production
223. 29:09so somebody answers questions in an app it goes up to the cloud it gets edited together in 5 minutes and it comes down
224. 29:14fully edited super magical Auto Magic we've been doing that for 11 and a half years it's not AI if I look at that part
225. 29:22of our value proposition in 2024 the value of that drops to zero automation is irrelevant the tech is
226. 29:30irrelevant the other half of what we do is real human beings authentically telling their stories at scale in a
227. 29:37world of infinite content I think the value of authenticity goes up so I think every company big or small is going to
228. 29:43have to look at these tools and say you know what do these things do and and what are what are the parts of our value
229. 29:52proposition that drop to zero and what are the parts that go up in value and figure out how to how to amplify the
230. 29:58things that are going to go up in value and I I mean as investors that might be an interesting conversation to have is
231. 30:04say let's assume let's assume I'm right and I and I beat Eric and these things get or I beat I beat you you said 10
232. 30:10years I said three let's say in three years they get really good really fast what value propositions drop to zero if
233. 30:17that happens and then what part of the value propositions go up if that happens and so that might be an interesting way to to look at that
234. 30:27yeah like kind of a similar thinking um I think you want to look for the next bottleneck so it's become for example
235. 30:35really easy to write code right um and you can even not know how to code just tell tell the system what to do um so I
236. 30:45think like well where's the bottleneck if that's true well the product managers and The Architects like that would be an
237. 30:53interesting area that's a little more defensible and then might be in the next bottleneck but finding an
238. 30:58audience yeah customers right so that's where my go yeah you know I think it really is what what you just said which
239. 31:08is finding customers right like I remember a year ago when all this was happening and we were trying to figure
240. 31:13out okay like how we should pivot what we should do how we should adopt and I had a conversation with an investor and
241. 31:18I was like you know I was thinking like we really need like diving on this technology and build it out and lean
242. 31:23into the AI and he was like no you just need to do sales you just need to relationships and do freaking sales and
243. 31:29I was like but this technology we need to get our technology and he's like no no just do sales all the same cash solves all the
244. 31:38problems but it's true because like you can spend all this time building some AI model that nobody wants to buy and it
245. 31:44doesn't matter how cool it is so you always have to keep stay anchored in that and listening to your customers and understanding
246. 31:52them okay I have another question I swear to God I'm not a plan cuz Kyle I talked about this earlier like I feel
247. 31:59like this is a good question for Kyle and maybe the rest of you like Kyle has hundreds of thousands of people that
248. 32:05listen to him every night no hundreds not hundreds of thousands you said hundreds of thousands no well whatever
249. 32:12how many people let me this hey everybody you have lots of people talk to you they listen to you every night
250. 32:19talk about AI right and you're a person you're a real person with a face you're not a computer yeah I think it says 224 but I don't use
251. 32:32Tik Tok so I don't really but still like so that's like an interesting question right like people still want to have even
252. 32:40conversations about a computer with a real person what's funny is so like yeah like moving forward like for businesses
253. 32:49for companies like you still want to actually have a real person that can talk to you about all this stuff right
254. 32:55you don't just want to like chat GP how can I use you right yeah so what what you're what you're pointing to I think is actually really
255. 33:07critical again I go back to I there's a there's a meme going around that I've totally adopted 2023 the last normal
256. 33:14year I think 2023 is the last year we can sort of look back and and see it as kind of normal um the the people conversation is really
257. 33:26important we don't talk a lot about technology when I'm talking about AI we're talking about how it's impacting
258. 33:32people and what's happening in in in my community in the AI salon and on the Tik Tok channel the Irregulars um that show they show up every
259. 33:42night I'm not talking about different [ __ ] every night I'm talking about here's chat gbt here's how you make gbts
260. 33:48I'm talking about a lot of the same [ __ ] there's still people that come back night after night after night why is
261. 33:53that because what they're discovering is that what these AI tools are doing is they're amplifying parts
262. 34:02of things they wanted to do they never thought were possible in their lives and so what's going to happen as these tools
263. 34:09get increasingly powerful is that every single one of us is going to have the skills to do any [ __ ] thing we've ever wanted
264. 34:19to and so right now in 2024 early 24 and and all of 2023 why we hired people was you can do that skill you've got that
265. 34:29skill oh you're good at this kind of programm oh you you're good at this this medical stuff when you move into a world
266. 34:36where every single person can do every single job to to let's say 80% quality you no longer hire them for what they
267. 34:44can do you hire them for who they are so you hire them for trust you hire them because they're not an [ __ ] you hire
268. 34:52them because it's like oh we had dinner at that Thunder viiew thing that was cool and so for for me that's the shift
269. 34:58that happens is we shift out of hiring people for skills and two hiring them for who they are and Trust completely
270. 35:04disagree with that awesome the the the job then that will create value will require you being able to do those
271. 35:12things to the degree of skill that other people can't do in the same way because if it's undifferentiated it's not
272. 35:19valuable and people won't pay for it and then it's not a job right and so like what the output will need to be right
273. 35:26you might be more efficient do it it's just going to be more [ __ ] complicated or harder to do whoever can
274. 35:31use those tools to get to it better is going to be valuable and there is not going to be um everyone be a to use
275. 35:39those tools in the same way to get the same output right then there's no value being created CU it's undifferentiated
276. 35:46like we are not going to like we just hire people to fill a seat cuz they'll do the same [ __ ] that's never been the
277. 35:52history it's always been the prediction oh this technology is going to wipe this out we just keep building cooler and
278. 35:57cooler and harder and harder shipped so let me let let me let me respond so I'm not saying that humans don't
279. 36:06participate and don't use the tools I'm saying they they still use them they still use them but I'm saying that if I
280. 36:12know that Jim works his ass off and he will do it until it's right I'm going to hire Jim versus someone else who soft
281. 36:21skills it's more soft skills yeah we shift to more soft skills and we shift and I'll tell you I'll tell you another
282. 36:26shift that happen s and and this one I think is going to [ __ ] some people up I think we shift away from specialty
283. 36:34skills being the higher value skills and and Broad horizontal skills increasing in value so I'm calling it revenge of the liberal arts major
284. 36:45people people who can think across disciplines and and leverage these tools and bring the things together are going
285. 36:52to be of higher value than specialty skills are today because the specialty skills are going to get commodified what do you think using an
286. 37:00AI tool is better than someone else that's a specialty skill wait what what using a specific AI tool better than
287. 37:08someone else is a specific skill right that's not a broad-based skill you learn the specifically a tool right that's not
288. 37:16a broad-based skill but maybe the gap between you and me is small some people I'll never make up the Gap each separ but found found the more
289. 37:25comparable on the same scene so know I I I don't think we are and we're not trending to a world where skills don't matter the output you can
290. 37:34achieve with differentiated skills will be better but it will still be people who learn the skills that matter in
291. 37:41whatever the economy the things that we create are yeah and that's not a flat world oh I don't think it's a flat world
292. 37:49at all I don't think it's a flat world at all CLE World um I mean Ian I get to see this a little bit the part of what we do is we're helping
293. 38:03people who are building a product generative AI like chod or scanning you know clinical records on behalf of doctors in order to
294. 38:15automatically code you know the notes and then feel better for insurance I don't know like people that are um
295. 38:21automatically drafting email responses for various purposes like all these different things right and then you try to figure out like
296. 38:30well how do we scale this product right um and part of scaling is you got to be able to trust it it's got to fing the
297. 38:36right thing and we ask like Okay cool so how are you going to test this they like come up with a bunch of examples all
298. 38:42right these are representative inputs to the system I'm want to see if they're good on the other side cool how do you
299. 38:48know if they're good and nine times out of 10 people freeze and they're like I know it I'll see it you're like okay so you're going
300. 38:59to read a million chatbot conversations to see if they're any good um and it's this really interesting thing that's
301. 39:06happening where it's like what what we then are doing is proba we people figure out okay so how do you define what you
302. 39:11want to know and then how do you build uh basically like llms around this to do your analytics for you it's a core part
303. 39:18of our product and figuring that out is fascinating because we see people start to do it and then say like I don't trust
304. 39:26this it doesn't work and we're like great let's look at what you're asking Elm to do for you and you're turns out
305. 39:32they're asking like a very like wrote questions sometimes even in Bad English and like the decision was that the AI
306. 39:41doesn't work and then what you're seeing there is like well okay cool like if you really want to just get good it to this
307. 39:46one this thing of analyzing your conversation and knowing if it was good on Dimensions you care about on brand
308. 39:52voice tone whatever else like there's an art in that alone to be able to figure out like how do you even phrase the
309. 40:00question and then prompt an an llm to like answer it in a reliable way and when you get it working it is magical
310. 40:08and when it's not working it looks like it's terrible and the difference is a skill right you know and I think
311. 40:15like we're in a moment right now where like that skill looks a little bit like English you know I don't think that's
312. 40:22where we're going to stay right I think part of you we're working on is like you know what is what are the the higher
313. 40:28order levels of optimization on that um but I think like those kinds of examples are like what's kind of playing out
314. 40:35right now like people are just they're evolving what they have to get done it is a lot of critical thinking I'm I
315. 40:40think it's all critical thinking I think critical thinking is the big skill like I think that's where it goes to I think
316. 40:46I think the skill becomes critical thinking the skill becomes critical thinking it doesn't become I got a mathematics degree here I got a
317. 40:53programming degree here I got an English degree here it becomes critical thinking to be able to think okay what's good
318. 40:58what's bad what am I trying to accomplish and then to your point do I have the skills with these AI tools to
319. 41:03be able to you know Marshall those all together to do that thing so I think we agree told us we have roughly 10 minutes
320. 41:14well I think few questions do we do of the card I I let's do a speed I likeed one that was if you have never used AI
321. 41:23know nothing about it or aren't using it in your company right now am I too late should I just give up
322. 41:31yes just quit no start using CHT like I feel like this is the most access if if you want to it's th%
323. 41:41accessible if if you haven't used it and if you have kids open up Dolly and ask them what they want to see a picture of
324. 41:48and you will waste two hours of them having their mind blown and you having your mind blown about what they do do this
325. 41:58what they do about what Chris um it'll it'll catch up if you guys haven't just like turned on chat GPT in audio mode and just let it talk
326. 42:12back to you like it it's just about to take over this conversation and tell us all these interesting things and ask us good questions
327. 42:22ready I'm here to help you right now what do you need assistance for oh man you just disappointed us so
328. 42:31bad I'm really sorry to hear that I'm here to help so please let me know how I can assist you this is a terrible
329. 42:39example cating last night we had this conversation about the Harry Potter Wizarding Wizarding World with my six
330. 42:46and 8-year-old and it started a conversation that lasted for like 25 minutes and it just kept going back and
331. 42:52forth with my kids and I don't know asking good question questions it's magical that was terrible let me so in terms of are are
332. 43:03you too late we are so [ __ ] early so McKenzie just came out with a study at the beginning of 2023 I think 87% of
333. 43:13people had not used chat gbt at the end of 2023 was 82% so 82 had not so 18% of people have used chat GPT
334. 43:26of those let's say at least half have used Chad gp4 let's say half of that have used Chad gp4 in any kind of
335. 43:33advanced way like talking to it that means 5% of people know what the [ __ ] these tools are most people that say
336. 43:40that they know what they are they used it once and it gave them a shitty answer and they're like ah [ __ ] it it's not
337. 43:45that good I'd like say it's been dozens of times dozens they've used it dozens of times and because they're not good
338. 43:52with language they couldn't get it to that was my first time using chat GPT I was I just asked some questions I was
339. 43:58like this is stupid like Googling is better and then my CTO who's also my husband he used it and he's a much more
340. 44:06creative person than me and the first thing he asked it he was like can you please write a poem about endometriosis
341. 44:13using exclusively pirate language and this poem about this chronic condition in Pirate language just came out and I
342. 44:21was like oh my gosh like this is incredible like it is all about asking the questions so talking to your
343. 44:27children I think is a great way since they're a lot more creative than us um because it's a new tool if you haven't
344. 44:32used it before you don't know what to ask but isn't that novelty versus actual creating something of value well here's
345. 44:38a different example for you uh I have a friend uh two friends that were dating and had a relationship problem and um
346. 44:48went to chat gbt and the woman was like I need you to also talk to chat gbt about this problem and see when I we at
347. 44:54and they that is I'm not getting how they worked it out they both just had a dialogue therapist if you if you haven't
348. 45:01used Pi so you can download the chat jpt app you should also download the PI app you and I I I demoed this for you two or
349. 45:08three of these things ago download Pi so Pi is from a company called inflection they raised $1.3 billion they're
350. 45:14spending about a billion of it on 22,000 gpus so it Go them they're setting out to build the world's best personal
351. 45:22assistant it is designed to be conversational it is marketly different I I'll tell you something really
352. 45:28interesting go to pi. the website and try to treat it like chat gbt it will frustrate the crap out of you because
353. 45:35it's it's it's optimized for conversation but then download the PI app and put it in conversation mode in
354. 45:43your car on like a 2hour trip and have a conversation with it it will blow your [ __ ] mind what are you scared of that was one
355. 45:54of his questions I think this question actually said what absolutely scares the [ __ ] out of you when it comes to AI in the present or the
356. 46:09future really the AI getting its own goals I think this is the classic problem thinking of like Terminator but
357. 46:19yeah what that produce AUM yeah it's right and then big paper clip yeah the paper foot story where if you tell the
358. 46:28AI to produce you know a paper foot Factory and then along the way it realizes that you know the humans are
359. 46:34going kind of slow in the work and actually the most efficient way to create that factory might be to kill the humans that kind of
360. 46:44exes I was going less dark and maybe more more realistic from it day I can't do anything about that though right so I
361. 46:53worry about things I can control right um no so uh a lot of therapy there um no I worry about what we talked about like
362. 47:01the pace of what things are going in these these debates we have as a team from an investment perspective is like
363. 47:07we can't predict the future and so can we at least from our perspective you know identify the things that we think
364. 47:14have a chance of not being disrupted away but it's going so fast I don't know if we're going to do J and I wored that we're going to fail
365. 47:23miserably and Rance I think one of the things that scares me is that is that people will stay on the sidelines thinking they have three to
366. 47:31five years to figure this stuff out and they don't have 3 to 5 years and and a lot of people are going to be blindsided
367. 47:37by it and I that that scares me I me I think with as much as AI can Empower us to do great things it also
368. 47:46can do Terrible Things um and and our is being used for for scams for voter suppression for fraud uh and so we have
369. 47:55to have regulation mitigation strategies to to fight those things as well the other thing that worries me a lot is is
370. 48:03you know there's so much bias that already exists in these data structures that in a lot of ways we can be
371. 48:09amplifying existing bias in our models and we need to be really intentional about that because when there's flaws in
372. 48:15underlying data that's going to just be Amplified in these models you scared of anything yeah I need my blankie I I
373. 48:26don't I'm trying to start what to say here um I mean one of the first conversations I had with somebody that
374. 48:32was working deep in this space who is training llms that you've all used you know they just were basically saying
375. 48:41like without like any kind of negative thing to what you just said it's reminding me of it he like if you look
376. 48:47at the research right now in the US that we're publishing all this research about how to avoid bias if you look at some of
377. 48:53the other research being published in other parts of the world it's a bit about how to like improve facial
378. 48:59recognition detection and like if you wanted to build like the ultimate authoritarian surveillance state right
379. 49:06now it would be easy with this technology and I I just think there's like like we have kind of our you
380. 49:13know perspective of like can things are like relatively Okay and like relatively benevolent right now um so I think there
381. 49:21are things to worry about there I think on the flip side maybe we should like end on I had a darker question actually well or
382. 49:30what are we most hopeful for later yeah I mean I don't know like I I why am I working on this like I'm maybe any of us
383. 49:37that are here who are founders are like just blind optimists by our nature but like what an amazing opportunity to like
384. 49:46amplify human creativity and productivity and you know potential right and I think this is the story that
385. 49:54everybody's been telling I just like I'm willing to blindly hold on to hope for that like people were worried about the
386. 50:04Advent of the printing press and you know the world didn't end people were worried about the Advent of the
387. 50:09typewriter and worried about the Advent of the car like you know it is hard right now I feel what Kyle feels of like
388. 50:16the change might happen so fast that it couldn't hurt and be disruptive um but I do think like people are just so
389. 50:24resilient on the other side side of like figuring out how to adapt to a new technology like there's amazing things
390. 50:31that we will get done with it and like let's go collectively as people that build things and B companies and build
391. 50:37businesses like go make those things happen like and not that the chaos to so I don't know that's where my mind is we're doing
392. 50:46hope hope you changing stuff hope you changing stuff hope Chang you what's your hope we changing stuff I I think it's just that that
393. 50:55optimism um but but I think it's that you know if it's a problem you care about like it's we're in it because we
394. 51:02want AI to be used for good we're we're working to try to figure out those problems and mitigate it and so we need
395. 51:07more good we need more good people working and solving and addressing those problems so uh I I don't know I think
396. 51:14bad things will happen but I think we'll figure out a way to avoid them mostly mostly I think it's easy to talk about
397. 51:21the risk and you know we've got 50 years of Hollywood telling us the robots are going to kill us so I think we default
398. 51:26to that um the thing that I'm I'm most excited about is I think there is as real a chance that AI makes things way
399. 51:34better than makes things way worse um and I think that there is a very what I'm experiencing in my community and in
400. 51:42myself is that the AI is increasingly doing the work that was the soul crushing um demotivating work and it's
401. 51:50freeing me up to to sort of execute at a higher creative level all the time and I imagine a world where what if we
402. 51:58as humans could do the things we really wanted to do and these machines did the [ __ ] work and I think that's a very real
403. 52:04possibility and that's what I'm excited about I really like the idea of critical thinking um for the earlier comments
404. 52:13just being a lot more important and I think we're in a political climate and a National Climate that there we just
405. 52:19could use a lot more of that yes um and I'm so glad that that is more more important because I I believe people will follow
406. 52:29suit yeah I think I think Kyle said this earlier about um unlocking potential right and enabling people to do things
407. 52:36that they wish they knew how to do easier and I think I even see that with my kids right when you know Dolly in
408. 52:42particular like they create these pictures that they never could have right and their Joy from it is so great
409. 52:49especially when they can get it to put a swear word right in the picture right but I I think you see that not just if would you see that all
410. 52:56through and I think that will be a great thing to enable people to do things they didn't think they could do call it there cool
411. 53:12yeah so much thank you three years three years th bucks in good deal that's great thank you got I got to go release the
412. 53:27Irregulars all right say hi everybody hi Irregulars bye all see you later so