Feb 1 2024 - THUNDERVIDEO CEO PANEL

Description

Chapters

Transcript

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