1
00:00:00,000 --> 00:00:04,640
I'm asking our employees what they think of AI as a company both in the tech and this

2
00:00:04,640 --> 00:00:08,560
crave space. I'm sure there are takes that people want to share. Uncut versions will be

3
00:00:08,560 --> 00:00:14,240
available below, so let's get into it. I am here with Jonathan Horst. Emily Young. James Shrype.

4
00:00:14,240 --> 00:00:18,720
Do I get to hold this at any point? Do I have to keep leaning in? Yeah, I'm realizing. Jake Bell.

5
00:00:18,720 --> 00:00:25,360
Dan, hey, Andy. Nicole. Ploof. And what do you do for the company? Right. Good. What's your

6
00:00:25,360 --> 00:00:30,960
thought in the current state of AI? I've only used chat GPT before. I think for like day-to-day,

7
00:00:30,960 --> 00:00:36,880
like if I don't want to deal with like a customer representative, like, hey, I want to get refund

8
00:00:36,880 --> 00:00:42,880
off this. I was like, hey, chat GPT, write me an email about getting a refund off this. Over height.

9
00:00:42,880 --> 00:00:47,120
We're thinking of it as like, it's going to just do the thing we want it to do when it's probably

10
00:00:47,120 --> 00:00:52,080
just going to be yet another tool in the toolbox. It's pretty interesting. There's a lot of cool

11
00:00:52,080 --> 00:00:57,520
stuff going on, and I'm excited to see it kind of develop over the next, I guess, a year or a

12
00:00:57,520 --> 00:01:02,880
few months even. We don't have it? No, it's not. What we refer to as AI right now, have you ever

13
00:01:02,880 --> 00:01:08,640
played Mass Effect? So in the lore of Mass Effect, AIs are like, you don't do that. So instead, what

14
00:01:08,640 --> 00:01:13,600
they have is what's called a VI, a virtual intelligence. That is a lot like what we consider

15
00:01:13,600 --> 00:01:18,320
to be like chat GPT and all that kind of stuff right now. It's able to answer questions. It's

16
00:01:18,320 --> 00:01:23,760
able to kind of like carry a conversation. It's able to do a lot of things, but it's not able to

17
00:01:23,760 --> 00:01:28,000
think. So it's just learning patterns and spitting out what it thinks is the next thing in the pattern,

18
00:01:28,000 --> 00:01:33,520
more or less. To me, it just seems like this big mass of horrifying knowledge and it just gets kind

19
00:01:33,520 --> 00:01:38,640
of like randomly spit out whether it's accurate or not. So for me, it's like kind of spooky because

20
00:01:38,640 --> 00:01:43,280
you never really know what you're going to get. I think a lot of fantasy stuff going on. It's

21
00:01:43,280 --> 00:01:48,080
people getting very excited about what's going to be in the future. Well, it's actually the capability

22
00:01:48,080 --> 00:01:52,800
states are still a bit limited. However, they're crazy about two years ago, three years ago, like

23
00:01:52,800 --> 00:01:57,680
the transformational, especially the AI art space, like that's already pretty damn good and pretty

24
00:01:57,680 --> 00:02:02,320
damn disruptive and scary. I think overall it's really cool and it's a great idea. We could speed

25
00:02:02,320 --> 00:02:08,400
up things like machine learning. We can use it to diagnose people who have weird symptoms and stuff

26
00:02:08,400 --> 00:02:14,640
because let's face it, people are limited at some point. It becomes bad when we basically use prior

27
00:02:14,720 --> 00:02:20,480
work to then create new stuff and the people that made that work previously don't get any credit for

28
00:02:20,480 --> 00:02:25,520
it. It's not maybe as terrifying as people think it is. Yeah, certainly, certainly for it. So since

29
00:02:25,520 --> 00:02:29,760
you're for it, what's one thing you're really against in terms of the AI? Probably creating

30
00:02:29,760 --> 00:02:35,920
images of real people. As we've seen with QTC and Pokimane and some of these other streamers that

31
00:02:35,920 --> 00:02:40,560
had explicit photos of them generated and shared on websites, stuff like that is awful. And now

32
00:02:40,560 --> 00:02:44,400
people think it's a lot of fun to create the AI songs of like Linus and stuff like that online,

33
00:02:44,400 --> 00:02:48,960
but he didn't consent to that. And no matter what, that's kind of wrong and weird to think that just

34
00:02:48,960 --> 00:02:53,600
because he's on the internet, he's okay with it. That's not true. And it's something that obviously

35
00:02:53,600 --> 00:02:57,680
there's no way to limit. But it's something I'm definitely concerned about. I think probably

36
00:02:57,680 --> 00:03:03,440
some of the protectionism, I mean, you're spending millions of dollars on like cloud infrastructure

37
00:03:03,440 --> 00:03:09,520
GPU time just to actually create a reasonable model. People I think are willing to protect that

38
00:03:09,520 --> 00:03:15,520
rather than just open sourcing it. Adobe's like generative fill is kind of like interesting.

39
00:03:15,520 --> 00:03:20,400
They're demonstrating it in like, look at, we could make this image so much wider and we can

40
00:03:20,400 --> 00:03:24,400
fill it just from this tiny little portrait, we could turn it into a whole landscape. And it's

41
00:03:24,400 --> 00:03:28,720
like, yeah, I guess that's cool. But it's also like, what is the picture then anymore? AI is

42
00:03:28,720 --> 00:03:33,200
difficult. There are things that AI is really good for. Say, for example, AI image scaling or

43
00:03:33,200 --> 00:03:38,720
whatever, you train it to do whatever it needs to do, and it does it basically by failing as many

44
00:03:38,800 --> 00:03:43,360
times as is necessary to fail the correct way. And that's actually really cool, because it

45
00:03:43,360 --> 00:03:48,640
does things in a way that is very difficult to do in traditional software. When you get into

46
00:03:49,200 --> 00:03:53,600
marketing AI as well, I mean, I keep saying AI, it's more like machine learning. So like, when

47
00:03:53,600 --> 00:03:57,840
you get into marketing it as like AI, and you say, Oh, well, chat, GPT is going to take over the world.

48
00:03:57,840 --> 00:04:03,040
Well, that's, that's overselling it. And that's, I think it's by design that they're doing that,

49
00:04:03,040 --> 00:04:06,640
because they want people to invest, they want people to think that we're close enough.

50
00:04:06,640 --> 00:04:12,720
I think it's a really interesting tool. And I think it, depending on how it's used, it kind of

51
00:04:12,720 --> 00:04:19,680
broadens a lot of channels and possibilities of being able to sort through a lot of information

52
00:04:19,680 --> 00:04:25,520
really quickly. Like for example, you can ask AI something and it'll like throw out a ton of

53
00:04:25,520 --> 00:04:31,760
results. But then it's also like good to keep in mind, this is always just from a certain pool

54
00:04:31,760 --> 00:04:38,400
of information. So depending on how accurate that pool of information is, you can't always 100%

55
00:04:38,400 --> 00:04:42,160
trust it. I hate writing, by the way, but like, but when you figure out how to write something

56
00:04:42,160 --> 00:04:46,160
really cohesively and like, it's like, that's really impressive. And I just don't know how

57
00:04:46,880 --> 00:04:50,880
prompt is going to lead to that. As a writer, you hate writing. I do.

58
00:04:53,120 --> 00:04:57,520
There's a lot of worry that like AI would take over creative jobs. And as we are in a creative

59
00:04:57,520 --> 00:05:04,400
field, what do you think about that? I personally don't think so. Cause like AI can be very objective,

60
00:05:04,400 --> 00:05:11,680
but like they probably can really be subjective. If I trust the AI to grab focus for me sometime,

61
00:05:11,680 --> 00:05:17,360
if we have a face in there, they'll grab like focus like pretty well. But there is one person

62
00:05:17,360 --> 00:05:22,880
of the chance that, you know, the focus gets shifted to something else. Cause the camera thing,

63
00:05:22,960 --> 00:05:27,840
that's the face. Human being in general is still better in that case, like determine

64
00:05:28,800 --> 00:05:33,120
which part is the important part. So that's exactly the reason why we don't use like

65
00:05:33,120 --> 00:05:38,640
auto focus on the top down and be cam for ShortCircuit. If no new writing is made,

66
00:05:38,640 --> 00:05:42,960
like if you say, for example, writing is completely no longer a thing that basically,

67
00:05:42,960 --> 00:05:48,160
as you iterate on things that are already iterated on, you kind of like lose bits and pieces of it,

68
00:05:48,160 --> 00:05:53,600
it'll be difficult to continue indefinitely. You might get away with a little bit for a

69
00:05:53,600 --> 00:05:56,640
little while. People are already getting away with writing paragraphs or whatever,

70
00:05:56,640 --> 00:06:01,280
but it still requires like an editing pass. It still requires people who know what they're

71
00:06:01,280 --> 00:06:06,560
doing because I don't know if you or like people who are watching know, but AI, especially like

72
00:06:06,560 --> 00:06:12,320
chat GBT has a very short memory. So if you're trying to tell a story, like in a movie or something,

73
00:06:13,920 --> 00:06:17,520
keeping that thing on the rails is going to be a really tough, a really tough operation.

74
00:06:17,520 --> 00:06:22,720
Creative industries definitely are at risk. I think it's going to be similar to other industries

75
00:06:22,720 --> 00:06:27,600
like perhaps the Apex where there's humans still doing the job, but there's fewer of them

76
00:06:27,600 --> 00:06:31,680
because the human is doing it can do so much faster now. But I think we're going to see things

77
00:06:31,680 --> 00:06:36,800
change and adapt. I think we're going to see a lot more like identity driven personality driven

78
00:06:36,800 --> 00:06:42,480
content like influencers like Linus who you love and have a parasocial relationship with.

79
00:06:42,560 --> 00:06:47,920
That's going to just become even more of a thing because in a couple of years, the majority of

80
00:06:47,920 --> 00:06:51,920
the content on the internet is going to be from AI and you're not even going to know that it's

81
00:06:51,920 --> 00:06:56,400
from AI and we're going to have to figure out identity online and whether that's a blockchain

82
00:06:56,400 --> 00:07:02,080
thing or some kind of centralized, I don't know, identity system, we're going to have to figure

83
00:07:02,080 --> 00:07:07,680
out a way to know like, oh, actual humans said that, not just a very convincing bot with a political

84
00:07:07,680 --> 00:07:12,880
agenda. Oh yeah, will my job be at risk? If I was in high school right now, I would be doing trades

85
00:07:12,880 --> 00:07:19,280
for sure. The robots are 15 years behind the AIs right now, I think. So something practical that

86
00:07:19,280 --> 00:07:24,080
you can help with would definitely be a little more secure than a purely creative job. Yeah,

87
00:07:24,080 --> 00:07:29,520
I was telling Nicole, AI can maybe edit videos, but it can't do a fart reverb like I can.

88
00:07:29,520 --> 00:07:32,000
No way. It can't put vine boom in at the perfect moment.

89
00:07:32,960 --> 00:07:37,920
Have you used any AI currently, like where it's like the chat GBT, the image random

90
00:07:37,920 --> 00:07:41,200
rangers, music, I already think that, have you used any of it? The only thing I've really done

91
00:07:41,200 --> 00:07:46,560
with AI was when Mid Journey and those other guys were coming out, I was messing around on it.

92
00:07:46,560 --> 00:07:50,720
I was like, yeah, this is cool. Haha, like Trump eating a lemon or whatever. I don't know,

93
00:07:50,720 --> 00:07:55,760
that's like the, oh my god, why is that my example? Have you used things like AI Dungeon,

94
00:07:55,760 --> 00:07:59,840
the like choosing an adventure Dungeon and Dragons type game online where it creates a story

95
00:07:59,840 --> 00:08:03,680
for you to play through? And that's a lot of fun. But otherwise, I haven't done really any image

96
00:08:03,680 --> 00:08:08,080
generation. The only thing is I created an image of Kanye West at a ring door bell cam,

97
00:08:08,080 --> 00:08:12,880
and that was pretty funny. Yeah, I've played with pretty much all of them. And I actually run a

98
00:08:12,880 --> 00:08:20,000
couple at home. So I run an LLM as well as stable diffusion for image generation. And I run those

99
00:08:20,000 --> 00:08:26,000
locally. I recently just bought a couple more 30 90s and, you know, upgraded some of them with VRAM

100
00:08:26,000 --> 00:08:32,960
so that I could try some of the larger models. So they're like a chat GPT for but you can only

101
00:08:32,960 --> 00:08:39,360
talk to them for like, I don't know, two to five minutes before the like token inference length

102
00:08:39,360 --> 00:08:44,400
gets too much. And it takes like a minute to respond to you. But initially, they're they're

103
00:08:44,400 --> 00:08:49,040
really quite powerful as a tech slash creator company that we are. Do you think that we should

104
00:08:49,040 --> 00:08:54,080
be using AI and like any capacity or depends on the AI and depends on what we're using it for.

105
00:08:54,080 --> 00:09:02,320
If we're talking chat GPT, I don't know that it's not like copyright infringing. So that's one thing.

106
00:09:03,520 --> 00:09:08,000
The fact that we don't know what the test, like the the actual data sets that they train them on

107
00:09:08,000 --> 00:09:14,320
is a bit of a big giant question mark in terms of like the ethics and the, I guess, legality

108
00:09:14,320 --> 00:09:22,000
ultimately of using AI for generative purposes. On the other hand, like say, for example, I'm like

109
00:09:22,720 --> 00:09:27,680
95% of the way through a script. And you're like, you don't know how to close it out.

110
00:09:28,400 --> 00:09:32,320
You could sit there and you could, you know, scratch your head for an hour or two, or you could

111
00:09:33,520 --> 00:09:39,200
ping some concepts off of chat GPT or something and see what it looks like. I believe we do

112
00:09:39,200 --> 00:09:42,960
currently. I think the business team uses it for talking points and things like that. And then

113
00:09:42,960 --> 00:09:48,320
they just kind of clean it up. I think basically every organization should be using whatever new

114
00:09:48,320 --> 00:09:53,360
tools come to light. I mean, transcribing meetings, summarizing a meeting that you missed,

115
00:09:53,360 --> 00:09:57,040
if they can allow your team to be smaller and increase your output. Yeah, you should be using

116
00:09:57,040 --> 00:10:03,440
that really specific use cases where it's going to be really amazing. But in terms of like broad

117
00:10:03,440 --> 00:10:08,640
strokes, no, it's going to like help just like that tiny little step, like making one thing that

118
00:10:08,640 --> 00:10:13,840
you're trying to making one vision, you have a bit easier to execute or like making it better.

119
00:10:14,480 --> 00:10:18,560
That sounds great. We should use it as a starting point, but we shouldn't use it for like the final

120
00:10:18,560 --> 00:10:23,760
product as a tool. We should be using it as a tool. Like you'd use a computer as a tool or a

121
00:10:23,760 --> 00:10:28,960
camera as a tool. If I'm not like sitting out of other people's work, I'm okay with that. It's like,

122
00:10:28,960 --> 00:10:35,120
I need lines to be in a car. Where can I get the car? You know, like that car, it costs money,

123
00:10:35,120 --> 00:10:40,800
that costs time, that costs people's effort. Yeah. So in general, just like as long as the AI

124
00:10:40,800 --> 00:10:47,520
we're not like sitting out of other people's stuff and be able to help us like increase our

125
00:10:47,520 --> 00:10:52,960
production quality, increase our speed, I think that's a good thing to use. Not really.

126
00:10:53,600 --> 00:10:59,360
Maybe machine learning for the lab, but like real AI and using it to make, I don't know,

127
00:10:59,360 --> 00:11:04,720
I don't think so. I just, I hate the idea of eliminating jobs, especially when it's not ready.

128
00:11:04,720 --> 00:11:08,400
I think in a couple of years that story might change and I might be like, yeah, sure. I don't

129
00:11:08,400 --> 00:11:12,880
want to make the timestamps for every video. So run it through the AI generator to generate those

130
00:11:12,880 --> 00:11:17,120
timestamps for us, like fine. But right now, nah. I'm sweating like crazy. So we're going to wrap

131
00:11:17,120 --> 00:11:20,560
it up here. Do you want to say anything to the float planers? I'm trying to coin this phrase now

132
00:11:20,560 --> 00:11:23,600
or like something you didn't, we didn't talk about that you want to expand on. Tell Sammy,

133
00:11:23,600 --> 00:11:29,600
you don't like being called float planers. No, I'm so glad to get this off my chest.

134
00:11:29,600 --> 00:11:34,880
Everything, like all, all tools are inherently not good or evil. It is what you use it for,

135
00:11:34,880 --> 00:11:40,720
that is, this is the most important. So like AI, every job scaling is amazing. For example,

136
00:11:40,720 --> 00:11:46,320
it's just one of those things that's just so universally useful. Try it at home. It's fun at

137
00:11:46,320 --> 00:11:52,000
home. It's a fun little exercise. It's really, really simple to get going and it's, it's a fun

138
00:11:52,000 --> 00:11:59,200
little playground and you don't have to like spend money on chat GPT. I mean chat GPT is still the

139
00:11:59,200 --> 00:12:05,280
best. But if you play with it at home, you're yours. You can play with it as much as you want.

140
00:12:05,280 --> 00:12:11,840
It's still, it's still flawed. It's still very flawed. It's all, no, like how to commit the

141
00:12:11,840 --> 00:12:17,200
perfect crime. It's got answers for that somewhere. You just got to convince it. I would love really.

142
00:12:17,200 --> 00:12:21,600
I want to Jarvis, you know, and Jarvis that goes like, Oh, I got you fam. I got this ready for

143
00:12:21,600 --> 00:12:26,480
you before like an actual personality and actual artificial intelligence that's effectively a human.

144
00:12:26,480 --> 00:12:31,440
That'll be scary. And it means like deleting an AI is like deleting a sentient life.

145
00:12:31,440 --> 00:12:34,800
And that's why I does not truly exist now is because like a sentient life is not something

146
00:12:34,800 --> 00:12:43,120
that's easy to do. For a lot of like the medical environment, I am pretty like pro AI because

147
00:12:43,120 --> 00:12:48,720
like some people are doing research on like data and stuff. Like for data driven stuff,

148
00:12:48,880 --> 00:12:56,960
like I think AI is definitely winning and also quite like useful. Maybe like if the AI research

149
00:12:56,960 --> 00:13:02,080
enough like paper, we can cure cancer, you know, you never know. So in that perspective, I think

150
00:13:02,080 --> 00:13:11,040
it's great. Generally excited. Generally excited. Um, I don't, yeah, this is cut this out. No,

151
00:13:11,040 --> 00:13:16,800
this is an uncut version. All right, we'll cut it there. Thank you very much. Bye flow planers.

152
00:13:16,800 --> 00:13:22,480
James leaving, but I'm gonna keep waving. I'm committed a bit. Bye.