1
00:00:00,280 --> 00:00:03,240
Speaker 1: Welcome to Tech Stuff, a production of iHeart Podcasts and

2
00:00:03,279 --> 00:00:06,840
Speaker 1: Kaleidoscope IMA's Veloscian, and today will bring you the headlines

3
00:00:06,840 --> 00:00:11,200
Speaker 1: of the week, including a genetically edited rodent, the Wally Mouse. Then,

4
00:00:11,280 --> 00:00:13,760
Speaker 1: on today's Tech Supports segment, we'll talk to four of

5
00:00:13,840 --> 00:00:16,960
Speaker 1: Form Media's Jason Kebler about what the future of AI

6
00:00:17,079 --> 00:00:22,360
Speaker 1: movies could look like. All of that on the weekend

7
00:00:22,400 --> 00:00:30,520
Speaker 1: Tech Is Friday. It's March seventh. I'm excited to be

8
00:00:30,560 --> 00:00:33,240
Speaker 1: back in the studio this week with our producer Eliza Dennis.

9
00:00:33,280 --> 00:00:34,640
Speaker 2: We're glad to have you Stateside.

10
00:00:34,880 --> 00:00:36,480
Speaker 1: Yes, it's felt like I was away for a long time.

11
00:00:36,600 --> 00:00:38,440
Speaker 2: I'm wondering if that had something to do with this

12
00:00:38,560 --> 00:00:39,440
Speaker 2: news cycle though.

13
00:00:39,720 --> 00:00:43,120
Speaker 1: Yeah, there's a lot, lot lot to cover, so should

14
00:00:43,120 --> 00:00:43,639
Speaker 1: we jump in.

15
00:00:43,800 --> 00:00:45,000
Speaker 2: Yeah? Absolutely. So.

16
00:00:45,120 --> 00:00:46,879
Speaker 1: It was a bit of a confusing week when it

17
00:00:46,920 --> 00:00:50,440
Speaker 1: comes to chips and semiconductors, and I'll come back to

18
00:00:50,479 --> 00:00:54,120
Speaker 1: why it was confusing. But Monday saw President Trump hold

19
00:00:54,160 --> 00:01:01,120
Speaker 1: a press conference with the Taiwan Semiconductor Manufacturing Company aka TSMC.

20
00:01:01,640 --> 00:01:05,800
Speaker 1: The clues in the name the company manufacture semiconductors and

21
00:01:06,120 --> 00:01:10,959
Speaker 1: they produce ninety percent of the world's super advanced semiconductor chips.

22
00:01:11,000 --> 00:01:14,039
Speaker 1: These are the chips that power AI training models, but

23
00:01:14,120 --> 00:01:17,679
Speaker 1: also devices and basically are the backbone of the new

24
00:01:17,800 --> 00:01:22,000
Speaker 1: global economy. However, the vast majority of the manufacturing takes

25
00:01:22,040 --> 00:01:26,240
Speaker 1: place in Taiwan, and so many in Washington and beyond

26
00:01:26,280 --> 00:01:29,559
Speaker 1: have worn that TSMC's dominance in the chip industry could

27
00:01:29,560 --> 00:01:33,720
Speaker 1: create a national security risk, given that Taiwan is squarely

28
00:01:33,880 --> 00:01:38,120
Speaker 1: in the bullseye of China's territorial ambitions. But this week,

29
00:01:38,319 --> 00:01:42,240
Speaker 1: the Taiwanese company pledged to invest one hundred billion dollars

30
00:01:42,280 --> 00:01:44,360
Speaker 1: in manufacturing chips on US soil.

31
00:01:44,760 --> 00:01:46,600
Speaker 2: You know, this is so interesting to me because it

32
00:01:46,640 --> 00:01:50,160
Speaker 2: comes after multiple announcements over the last couple of months

33
00:01:50,200 --> 00:01:54,240
Speaker 2: about investments in things like data centers and AI infrastructure.

34
00:01:54,440 --> 00:01:57,720
Speaker 2: And that was with Stargate, and then Apple actually recently

35
00:01:57,720 --> 00:02:01,040
Speaker 2: made a pledge to make more products domestically with themestic contractors.

36
00:02:01,240 --> 00:02:03,520
Speaker 1: Yes, I think they talked about five hundred billion dollars.

37
00:02:03,640 --> 00:02:06,559
Speaker 1: But what was really interesting was that as soon as Tuesday,

38
00:02:06,720 --> 00:02:10,919
Speaker 1: when Trump addressed Congress, he talked about his aggressive desire

39
00:02:11,000 --> 00:02:14,880
Speaker 1: to dismantle the Act that actually TSMC is using in

40
00:02:14,960 --> 00:02:19,560
Speaker 1: part to fund its semiconductor manufacturing in the US. The

41
00:02:19,680 --> 00:02:23,880
Speaker 1: Chips Act was biden error legislation that basically created a

42
00:02:23,919 --> 00:02:29,320
Speaker 1: platform for manufacturing semiconductor chips in the US. I don't know,

43
00:02:29,400 --> 00:02:31,080
Speaker 1: I don't know how to square those two things, but

44
00:02:31,160 --> 00:02:33,560
Speaker 1: that actually brings us to our next headline, which is

45
00:02:33,760 --> 00:02:39,160
Speaker 1: a breakthrough indirectly interpreting and reading brain waves and converting

46
00:02:39,200 --> 00:02:39,680
Speaker 1: them to text.

47
00:02:40,000 --> 00:02:41,480
Speaker 2: The superpower I want.

48
00:02:41,639 --> 00:02:43,880
Speaker 1: Yes, exactly what. You may be able to buy it

49
00:02:44,080 --> 00:02:46,040
Speaker 1: if met had anything to do with it, because they

50
00:02:46,120 --> 00:02:49,800
Speaker 1: announced that in partnership with the Basque Center on Cognition,

51
00:02:50,000 --> 00:02:53,960
Speaker 1: Brain and Language in Spain, researchers have been able to

52
00:02:54,000 --> 00:02:59,960
Speaker 1: decode unspoken language, often reconstructing full sentences directly from brainwaves

53
00:03:00,480 --> 00:03:03,600
Speaker 1: and not even requiring any surgical intervention. This is all

54
00:03:03,639 --> 00:03:05,600
Speaker 1: stuff which can be measured outside the head.

55
00:03:05,800 --> 00:03:08,840
Speaker 2: Yeah, and that's really the breakthrough here, right, because other

56
00:03:08,919 --> 00:03:14,400
Speaker 2: research from companies like Neurolink have been extremely invasive, you know,

57
00:03:14,520 --> 00:03:17,840
Speaker 2: electrodes being implanted into the brain. Invasive.

58
00:03:18,320 --> 00:03:20,639
Speaker 1: Yeah, that's right. And this research is all about kind

59
00:03:20,680 --> 00:03:24,000
Speaker 1: of putting monitors on the skull or around the head

60
00:03:24,280 --> 00:03:26,520
Speaker 1: to be able to read brain waves without having to

61
00:03:26,800 --> 00:03:30,040
Speaker 1: directly hook into the brain, which is obviously much less scary,

62
00:03:30,360 --> 00:03:33,760
Speaker 1: and there's an amazing promise for people with cognitive impairments or

63
00:03:33,760 --> 00:03:37,560
Speaker 1: brain injuries to be able to convert their thoughts into

64
00:03:37,800 --> 00:03:41,280
Speaker 1: text and therefore speech. But there are also some concerns.

65
00:03:41,360 --> 00:03:45,560
Speaker 1: Right the Vox headline was Meta's brain to text tech

66
00:03:45,760 --> 00:03:49,040
Speaker 1: is here. We are not remotely ready, And of course

67
00:03:49,040 --> 00:03:52,800
Speaker 1: the big concern here is privacy if private companies can

68
00:03:52,840 --> 00:03:55,520
Speaker 1: actually read our thoughts. But there's actually a long way

69
00:03:55,520 --> 00:03:58,600
Speaker 1: to go before this research leaves the lab. Nonetheless, the

70
00:03:58,680 --> 00:04:02,320
Speaker 1: experiment was kind of a maze. So thirty five volunteers

71
00:04:02,400 --> 00:04:06,640
Speaker 1: sat under magnetic brain imaging scanners and typed on a keyboard.

72
00:04:07,040 --> 00:04:10,000
Speaker 1: Based on prior training, an AI model was able to

73
00:04:10,040 --> 00:04:13,640
Speaker 1: predict what they were writing, and meture research is accurately

74
00:04:13,680 --> 00:04:17,080
Speaker 1: decoded between seventy and eighty percent of what people typed.

75
00:04:17,120 --> 00:04:20,479
Speaker 1: In other words, with seventy to eighty percent certainty, it

76
00:04:20,560 --> 00:04:23,280
Speaker 1: could know before I clicked a T that I was

77
00:04:23,279 --> 00:04:25,960
Speaker 1: about to click the T. And so the real promise

78
00:04:26,040 --> 00:04:29,880
Speaker 1: here is actually a data from this research is beginning

79
00:04:29,920 --> 00:04:34,799
Speaker 1: to give neuroscientists a path to understanding how abstract thoughts

80
00:04:35,120 --> 00:04:37,559
Speaker 1: are converted into language by the human brain.

81
00:04:38,080 --> 00:04:39,840
Speaker 2: Then I think the other part of this is that

82
00:04:39,880 --> 00:04:42,680
Speaker 2: we're getting closer and closer to this idea that we

83
00:04:42,720 --> 00:04:45,120
Speaker 2: can have wearables that do this kind.

84
00:04:45,000 --> 00:04:48,000
Speaker 1: Of tech totally. But of course, a wearable headset that

85
00:04:48,080 --> 00:04:50,640
Speaker 1: can can actually read your thoughts and translate them into

86
00:04:50,720 --> 00:04:54,960
Speaker 1: language is something that you know, conceivably could change a

87
00:04:55,000 --> 00:04:58,279
Speaker 1: lot of people's lives. In another kind of science fiction

88
00:04:58,480 --> 00:05:02,320
Speaker 1: becomes science fact story, it's about the wooly mammoths. The

89
00:05:02,360 --> 00:05:07,039
Speaker 1: headline from MPR was just irresistible hoping to revive mammoths,

90
00:05:07,360 --> 00:05:12,560
Speaker 1: scientists create wooly mice. Yeah, and I think one of

91
00:05:12,560 --> 00:05:14,200
Speaker 1: the scientists that we knew we could do it, but

92
00:05:14,400 --> 00:05:17,720
Speaker 1: we didn't know they would be this cute and they're

93
00:05:17,720 --> 00:05:20,040
Speaker 1: worth a look. But the story is about a company

94
00:05:20,080 --> 00:05:23,880
Speaker 1: called Colossal Biosciences, and they are, by their own account,

95
00:05:24,120 --> 00:05:26,880
Speaker 1: the first and only de extinction company.

96
00:05:27,360 --> 00:05:29,520
Speaker 2: Okay, this was a concept I had never heard of

97
00:05:29,600 --> 00:05:30,400
Speaker 2: until this week.

98
00:05:30,720 --> 00:05:33,280
Speaker 1: Yeah, this one's been one that's I've been intrigued by

99
00:05:33,320 --> 00:05:34,600
Speaker 1: for it for a long time, and I hope we'll

100
00:05:34,600 --> 00:05:36,000
Speaker 1: be able to cover it on an episode of the

101
00:05:36,040 --> 00:05:39,560
Speaker 1: story before too long. But Colossal's website points out that

102
00:05:39,680 --> 00:05:42,880
Speaker 1: nine hundred and two species are extinct and more than

103
00:05:42,960 --> 00:05:47,520
Speaker 1: nine two hundred are critically endangered, and their mission is

104
00:05:47,560 --> 00:05:52,240
Speaker 1: to restore extinct species to preserve biodiversity. It's a little controversial.

105
00:05:52,440 --> 00:05:54,960
Speaker 1: Some people think there are more efficient ways to do

106
00:05:55,040 --> 00:06:00,400
Speaker 1: conservation than reviving extinct species, you know, But to that

107
00:06:00,480 --> 00:06:02,479
Speaker 1: I would say, I mean, look at the wooly mouse.

108
00:06:02,520 --> 00:06:05,240
Speaker 1: This is whether or not you think this is the

109
00:06:05,240 --> 00:06:09,880
Speaker 1: most efficient investment. It is absolutely wild. So picture a

110
00:06:09,920 --> 00:06:13,599
Speaker 1: mouse with fluffy, orange tan fur that looks like it

111
00:06:13,640 --> 00:06:15,440
Speaker 1: got very wet and then got a blow dry at

112
00:06:15,440 --> 00:06:17,360
Speaker 1: the salon. You've got the picture.

113
00:06:18,000 --> 00:06:19,800
Speaker 2: They are extremely cute, and.

114
00:06:19,920 --> 00:06:23,160
Speaker 1: The way Colossal made them was first studying the wooly

115
00:06:23,200 --> 00:06:28,440
Speaker 1: mammoth genome and then genetically engineering mice by modifying seven

116
00:06:28,520 --> 00:06:32,039
Speaker 1: key genes to make them more like wooly mammoths. You know,

117
00:06:32,120 --> 00:06:35,440
Speaker 1: the wool obviously being the most visible element, but also

118
00:06:35,480 --> 00:06:38,359
Speaker 1: some things that were invisible, like the way the mice

119
00:06:38,400 --> 00:06:40,920
Speaker 1: store fat and their fat metabolism makes them much more

120
00:06:40,920 --> 00:06:44,080
Speaker 1: able to survive in the cold. And according to Colossal,

121
00:06:44,560 --> 00:06:49,839
Speaker 1: the plan is to implant wooly mammoth esque modified embryos

122
00:06:50,040 --> 00:06:54,040
Speaker 1: to Asian elephants by twenty twenty eight. This week, was

123
00:06:54,040 --> 00:06:58,040
Speaker 1: also the Oscars and we both saw the movie that

124
00:06:58,120 --> 00:06:59,760
Speaker 1: won Best Live Action.

125
00:07:00,640 --> 00:07:02,840
Speaker 2: Please tell people about it. It's wonderful.

126
00:07:03,320 --> 00:07:07,000
Speaker 1: So it's a Belgian Dutch copro called I'm Not a Robot?

127
00:07:07,120 --> 00:07:07,880
Speaker 1: What did you make of it?

128
00:07:08,320 --> 00:07:10,480
Speaker 2: I was extremely tickled by this promise.

129
00:07:11,200 --> 00:07:13,040
Speaker 1: So, for those who haven't seen it, the film was

130
00:07:13,040 --> 00:07:16,720
Speaker 1: written and directed by Victoria Wamadam and it's about a

131
00:07:16,760 --> 00:07:21,480
Speaker 1: music producer who fails a series of capture tests and

132
00:07:22,000 --> 00:07:25,400
Speaker 1: in so doing us to question whether she's in fact human.

133
00:07:26,320 --> 00:07:28,120
Speaker 2: I mean the minute I knew that we were having

134
00:07:28,160 --> 00:07:30,840
Speaker 2: a capture test as part of the plot to this movie,

135
00:07:30,920 --> 00:07:33,080
Speaker 2: I was all in. I don't know if you have

136
00:07:33,160 --> 00:07:37,040
Speaker 2: this feeling, but I hate failing captured tests, especially when

137
00:07:37,040 --> 00:07:40,160
Speaker 2: you have to click I'm not a robot and all

138
00:07:40,200 --> 00:07:42,840
Speaker 2: you have to do is choose squares that show images

139
00:07:42,920 --> 00:07:46,320
Speaker 2: of street lights or motorcycles or bikes. How can I

140
00:07:46,360 --> 00:07:46,960
Speaker 2: get that wrong?

141
00:07:47,240 --> 00:07:49,800
Speaker 1: Yeah? So she's failing the tests again and again, even

142
00:07:49,800 --> 00:07:51,920
Speaker 1: though it looks like she's doing it right. And then

143
00:07:51,920 --> 00:07:55,120
Speaker 1: she gets to pop up with another quiz and one

144
00:07:55,120 --> 00:07:58,120
Speaker 1: of the questions is did your parents die before you

145
00:07:58,160 --> 00:08:02,480
Speaker 1: met them? And she answers, she answers, yes, and I

146
00:08:02,480 --> 00:08:04,760
Speaker 1: don't want to spoil the whole plot. It gets pretty eerie,

147
00:08:05,120 --> 00:08:07,040
Speaker 1: but it's a fascinating film well worth a watch. You

148
00:08:07,080 --> 00:08:08,920
Speaker 1: can check it out actually on the New Yorker website

149
00:08:08,920 --> 00:08:12,040
Speaker 1: because they were involved in releasing the film and on YouTube,

150
00:08:12,280 --> 00:08:15,160
Speaker 1: and as a tech nerd, I was rooting for them

151
00:08:15,200 --> 00:08:17,320
Speaker 1: to win the Best Live Action Short and did.

152
00:08:18,040 --> 00:08:18,280
Speaker 3: Yes.

153
00:08:18,680 --> 00:08:22,000
Speaker 2: Congratulations team. I'm not a robot, So.

154
00:08:22,040 --> 00:08:24,920
Speaker 1: Stick around as well after the break for a look

155
00:08:24,920 --> 00:08:28,720
Speaker 1: at how AI was used in this year's OSCAR nominated

156
00:08:28,760 --> 00:08:32,320
Speaker 1: feature films, including The Brutalist, and for a conversation with

157
00:08:32,400 --> 00:08:35,160
Speaker 1: Jason Kebler about what it's like to attend an AI

158
00:08:35,320 --> 00:08:45,640
Speaker 1: film festival. Stay with us, Welcome back. The Oscars were

159
00:08:45,640 --> 00:08:48,600
Speaker 1: on Sunday, so we're going to stick with movies. Back

160
00:08:48,600 --> 00:08:51,839
Speaker 1: in twenty twenty three, the Hollywood Writers' Strike was this

161
00:08:52,080 --> 00:08:57,000
Speaker 1: fascinating early example of a very public negotiation over how

162
00:08:57,160 --> 00:09:02,760
Speaker 1: AI might could, and even would disrupt and displace human labour. Ultimately,

163
00:09:02,760 --> 00:09:05,280
Speaker 1: the Writers Guild of America signed an agreement with the

164
00:09:05,280 --> 00:09:09,439
Speaker 1: Alliance of Motion Picture and Television Producers that Generative AI

165
00:09:09,559 --> 00:09:13,240
Speaker 1: would not reduce or eliminate writers and their pay. But

166
00:09:13,320 --> 00:09:15,920
Speaker 1: this was not a commitment by the industry not to

167
00:09:16,040 --> 00:09:19,920
Speaker 1: use generative AI in filmmaking, far from it. In fact,

168
00:09:19,920 --> 00:09:23,600
Speaker 1: this January, the editor of the Triple Oscar winning movie

169
00:09:23,679 --> 00:09:27,240
Speaker 1: The Brutalist told an industry publication that he had used

170
00:09:27,240 --> 00:09:31,040
Speaker 1: generative AI a few times in post production. Some of

171
00:09:31,040 --> 00:09:35,160
Speaker 1: the actors in The Brutalist, namely Felicity Jones and Adrian Brody,

172
00:09:35,640 --> 00:09:39,240
Speaker 1: performed their roles with a heavy Hungarian accent, and they

173
00:09:39,240 --> 00:09:42,920
Speaker 1: even had some dialogue in Hungarian. To prepare for the roles,

174
00:09:43,000 --> 00:09:45,839
Speaker 1: Brody and Jones spent months with a dialect coach to

175
00:09:45,880 --> 00:09:50,480
Speaker 1: perfect their accents, but as The Brutalist editor David Joncho,

176
00:09:50,920 --> 00:09:54,720
Speaker 1: a native Hungarian speaker, pointed out, English speakers can have

177
00:09:54,720 --> 00:09:58,880
Speaker 1: a hard time pronouncing certain sounds. In post he tried

178
00:09:58,920 --> 00:10:01,760
Speaker 1: to perfect the Hungarian in dialogue, and first the team

179
00:10:01,800 --> 00:10:03,600
Speaker 1: had the actor's reader of the lines in the studio.

180
00:10:04,080 --> 00:10:06,840
Speaker 1: Then they tried having other actors say the lines, but

181
00:10:07,080 --> 00:10:10,560
Speaker 1: that also didn't sound right, so Yoncho turned to AI.

182
00:10:11,200 --> 00:10:15,079
Speaker 1: He fared Brody and Jones's voices into the program respeecher

183
00:10:15,600 --> 00:10:19,199
Speaker 1: and then using his own voice, Yoncho refined certain vowels

184
00:10:19,200 --> 00:10:22,000
Speaker 1: and letters for accuracy, a process that could have been

185
00:10:22,000 --> 00:10:24,880
Speaker 1: done without generative AI, like in an audio editors such

186
00:10:24,880 --> 00:10:28,120
Speaker 1: as pro Tools, but Respeecher made the process much more

187
00:10:28,120 --> 00:10:31,720
Speaker 1: efficient and of course Adrian Brody won the Oscar for

188
00:10:31,760 --> 00:10:35,800
Speaker 1: Best Actor. As us Say Today reported, not all viewers

189
00:10:35,840 --> 00:10:39,040
Speaker 1: would pleased with the news. Don't think it's too reactionary

190
00:10:39,080 --> 00:10:41,240
Speaker 1: to say this movie should lose the Academy buzz. It

191
00:10:41,280 --> 00:10:44,880
Speaker 1: was getting one person posted on eggs. But the manipulation

192
00:10:44,960 --> 00:10:48,559
Speaker 1: of vocal tracks is not uncommon in movies. Deadline noted

193
00:10:48,559 --> 00:10:51,600
Speaker 1: that combinations of vocal tracks will use in performances like

194
00:10:51,720 --> 00:10:55,800
Speaker 1: Romy Mallock's Oscar winning portrayal of Freddie Mercury, and Respeecher

195
00:10:55,840 --> 00:10:57,880
Speaker 1: may have been used in another film nominated for Best

196
00:10:57,880 --> 00:11:01,440
Speaker 1: Picture this year, Amelia Perez. The rise of generative AI

197
00:11:01,559 --> 00:11:04,440
Speaker 1: has been remarkably fast in creative industries. But one big

198
00:11:04,520 --> 00:11:07,640
Speaker 1: question I have is how far could this go and

199
00:11:07,720 --> 00:11:10,880
Speaker 1: how soon? And to answer that, we want to turn

200
00:11:10,960 --> 00:11:13,679
Speaker 1: to our friend Jason Kebler at four or four Media,

201
00:11:13,760 --> 00:11:16,959
Speaker 1: who not too long ago attended a film festival of

202
00:11:17,080 --> 00:11:20,040
Speaker 1: AI generated movies. Jason, welcome back to the show.

203
00:11:20,120 --> 00:11:21,120
Speaker 3: Hey, thanks for having me.

204
00:11:21,360 --> 00:11:23,320
Speaker 1: Before we get into that film festival you went to

205
00:11:23,600 --> 00:11:26,839
Speaker 1: could you just explain how respeech it works and how

206
00:11:26,880 --> 00:11:29,439
Speaker 1: it was used in the editing process for the Brutalist.

207
00:11:29,760 --> 00:11:34,720
Speaker 3: Yeah. So, respeacher is an AI voice synthesizer, and so

208
00:11:35,160 --> 00:11:39,600
Speaker 3: it takes training data of an actor's voice and runs

209
00:11:39,600 --> 00:11:42,240
Speaker 3: it against a large language model. So in this case,

210
00:11:42,240 --> 00:11:45,800
Speaker 3: it would probably be examples of the Hungarian language, et cetera.

211
00:11:46,280 --> 00:11:49,520
Speaker 3: And it would take Adrian Brodie's voice and make it

212
00:11:49,760 --> 00:11:56,440
Speaker 3: more closely match other examples of Hungarian language. And it's

213
00:11:56,520 --> 00:11:59,120
Speaker 3: very interesting because this technology is sort of one of

214
00:11:59,160 --> 00:12:04,360
Speaker 3: the first native AI technologies that was widely used commercially,

215
00:12:04,920 --> 00:12:08,800
Speaker 3: not just Respeecher, but another company called eleven Labs has

216
00:12:09,640 --> 00:12:13,520
Speaker 3: become really famous for like Eric Adams, the mayor of

217
00:12:13,600 --> 00:12:18,520
Speaker 3: New York City, did a calling campaign to various communities

218
00:12:18,520 --> 00:12:21,240
Speaker 3: in New York City where he spoke English, but then

219
00:12:21,360 --> 00:12:24,920
Speaker 3: eleven Labs translated his voice into like fifteen different languages.

220
00:12:25,320 --> 00:12:28,920
Speaker 3: And it's not just like a robot voice reading it

221
00:12:29,080 --> 00:12:33,239
Speaker 3: sounds like Eric Adams speaking Mandarin or Eric Adams speaking Hungarian.

222
00:12:33,760 --> 00:12:37,600
Speaker 3: And so increasingly this is being used in movies, not

223
00:12:37,640 --> 00:12:40,600
Speaker 3: just Respeecher, but also eleven Labs and other tools like it,

224
00:12:41,160 --> 00:12:44,080
Speaker 3: and it really is like one of the first big

225
00:12:44,120 --> 00:12:47,160
Speaker 3: commercial uses of generative AI in movies.

226
00:12:47,720 --> 00:12:49,560
Speaker 1: To me, it feels like it's not that far away

227
00:12:49,600 --> 00:12:52,520
Speaker 1: from other post production tools that have been super charged

228
00:12:52,559 --> 00:12:55,679
Speaker 1: by AI, like description, podcast editing, or other tools like that.

229
00:12:56,040 --> 00:12:58,880
Speaker 3: Yeah, I mean it's really interesting because I think that

230
00:12:59,160 --> 00:13:02,280
Speaker 3: music had this a long time ago, with things like autotune,

231
00:13:02,600 --> 00:13:06,880
Speaker 3: and it's like many, many, many popular artists use autotune,

232
00:13:06,920 --> 00:13:10,439
Speaker 3: and this is a very similar technology. I mean it's

233
00:13:10,679 --> 00:13:13,840
Speaker 3: it's in the same family of technologies at least. So

234
00:13:13,920 --> 00:13:17,800
Speaker 3: it just becomes a question of how much post can

235
00:13:17,880 --> 00:13:21,000
Speaker 3: there be for the human performance to still be there.

236
00:13:21,040 --> 00:13:24,160
Speaker 3: And I think it's a really open question at this point.

237
00:13:24,280 --> 00:13:27,480
Speaker 3: I think if you asked me a while ago, I

238
00:13:27,480 --> 00:13:31,120
Speaker 3: would say they're changing the performance in some fundamental way.

239
00:13:31,200 --> 00:13:35,080
Speaker 3: But I think everything in a movie is so carefully edited,

240
00:13:35,160 --> 00:13:38,600
Speaker 3: so carefully shot. They do hundreds of takes for certain

241
00:13:38,679 --> 00:13:42,000
Speaker 3: scenes and then splice together different takes and cuts, and

242
00:13:42,040 --> 00:13:44,280
Speaker 3: so I think it really is a spectrum of what

243
00:13:44,360 --> 00:13:48,240
Speaker 3: you are willing to accept if you're in the Academy

244
00:13:48,240 --> 00:13:50,679
Speaker 3: and need to decide whether someone is worthy of an

245
00:13:50,679 --> 00:13:54,040
Speaker 3: award for this, I think audiences sort of have to

246
00:13:54,120 --> 00:13:56,240
Speaker 3: accept it because it's being done, and it's been done

247
00:13:56,280 --> 00:13:57,960
Speaker 3: for a long time. And I think that if you

248
00:13:58,040 --> 00:14:01,679
Speaker 3: start like having purity tests about this sort of thing,

249
00:14:01,679 --> 00:14:03,800
Speaker 3: I think it's going to be pretty difficult to know

250
00:14:03,880 --> 00:14:06,040
Speaker 3: which movies to see and which are not to see,

251
00:14:06,080 --> 00:14:08,840
Speaker 3: because ye, honestly, the only reason we know that this

252
00:14:09,040 --> 00:14:11,400
Speaker 3: was used at all was because the editor talked about

253
00:14:11,400 --> 00:14:12,320
Speaker 3: it to the media.

254
00:14:12,800 --> 00:14:15,280
Speaker 1: Yeah. And also, I mean, to be fair to Adrian Brody,

255
00:14:15,360 --> 00:14:18,439
Speaker 1: I doubt that many Academy members would have voted against

256
00:14:18,520 --> 00:14:21,600
Speaker 1: him on the basis to his owncungarian accident wasn't quite perfect,

257
00:14:21,720 --> 00:14:23,960
Speaker 1: So I'm not sure that this was like the key

258
00:14:24,000 --> 00:14:27,320
Speaker 1: input to his victory. But what you said about like

259
00:14:27,880 --> 00:14:30,600
Speaker 1: the role of post production and what that means visa

260
00:14:30,720 --> 00:14:33,600
Speaker 1: v like the original product made me think about this

261
00:14:33,760 --> 00:14:36,960
Speaker 1: AI generated film festival that you went to. So, first

262
00:14:37,000 --> 00:14:40,160
Speaker 1: of all, what made this an AI generated film festival?

263
00:14:40,240 --> 00:14:42,760
Speaker 1: How much of the films were AI generated?

264
00:14:43,440 --> 00:14:46,520
Speaker 3: Yeah, so it varied for each movie, but I think

265
00:14:46,560 --> 00:14:49,320
Speaker 3: that if you walked in off the street, you would say, oh,

266
00:14:49,400 --> 00:14:51,720
Speaker 3: these films were made with AI And what I mean

267
00:14:51,760 --> 00:14:56,480
Speaker 3: by that is each movie had visuals that were clearly

268
00:14:56,520 --> 00:14:59,680
Speaker 3: AI generated, like a lot of the backgrounds were constantly

269
00:14:59,720 --> 00:15:02,920
Speaker 3: changed in a way that if you were using a camera,

270
00:15:03,560 --> 00:15:07,800
Speaker 3: they wouldn't happen. A lot of people had like faces

271
00:15:07,840 --> 00:15:10,600
Speaker 3: that were morphing from scene to scene. One thing I

272
00:15:10,640 --> 00:15:14,280
Speaker 3: will say though, is that TCL was very clear that

273
00:15:14,320 --> 00:15:17,040
Speaker 3: all of the scripts were written by humans, and all

274
00:15:17,160 --> 00:15:19,320
Speaker 3: the voices were done by humans, and all of the

275
00:15:19,440 --> 00:15:24,000
Speaker 3: music was done by humans. The artificial intelligence was limited

276
00:15:24,040 --> 00:15:26,440
Speaker 3: to the visuals in different movies.

277
00:15:27,080 --> 00:15:28,640
Speaker 1: Can you just take me back to kind of how

278
00:15:28,640 --> 00:15:31,520
Speaker 1: you got invited and what questions you had going in?

279
00:15:32,200 --> 00:15:35,920
Speaker 3: Yeah, So I went to the Chinese Theater in Hollywood,

280
00:15:36,320 --> 00:15:39,520
Speaker 3: which is ironically where the oscars are. It's like the

281
00:15:39,560 --> 00:15:44,040
Speaker 3: same complex. And that theater is owned by TCL, which

282
00:15:44,160 --> 00:15:48,760
Speaker 3: is a Chinese TV manufacturer, and like a lot of

283
00:15:48,800 --> 00:15:52,040
Speaker 3: other TV manufacturers at this point, they have their own

284
00:15:52,200 --> 00:15:55,920
Speaker 3: free streaming TV service if you buy a TCL TV,

285
00:15:56,360 --> 00:16:00,000
Speaker 3: And TCL is the first company to put fully AI

286
00:16:00,160 --> 00:16:03,600
Speaker 3: generated movies on its streaming service. And so this was

287
00:16:03,640 --> 00:16:08,920
Speaker 3: a premiere of five films that were created using generative AI.

288
00:16:09,120 --> 00:16:12,200
Speaker 3: And so I had been writing basically about this technology

289
00:16:12,200 --> 00:16:14,400
Speaker 3: for a while and they invited me to come watch them.

290
00:16:14,720 --> 00:16:16,680
Speaker 1: So, despite the fact that you'll perhaps more on the

291
00:16:16,680 --> 00:16:19,520
Speaker 1: skeptical side, they welcome you into the film festival.

292
00:16:19,720 --> 00:16:22,320
Speaker 3: I was pretty shocked that they invited me, because honestly,

293
00:16:22,360 --> 00:16:25,360
Speaker 3: I had written about a trailer that they released for

294
00:16:25,440 --> 00:16:28,160
Speaker 3: an AI generated film and I kind of dunked on it.

295
00:16:28,200 --> 00:16:31,640
Speaker 3: I said, it was really terrible. It's called Last Train Paris,

296
00:16:31,680 --> 00:16:34,800
Speaker 3: and it was like an AI generated rom com. And

297
00:16:35,320 --> 00:16:38,320
Speaker 3: in the YouTube video, it's like the lip syncing of

298
00:16:38,360 --> 00:16:42,000
Speaker 3: the audio and the lips is like really bad. The

299
00:16:42,120 --> 00:16:46,960
Speaker 3: characters move incredibly robotically, and it has this very dreamlike

300
00:16:47,400 --> 00:16:52,200
Speaker 3: quality to it that is very common with AI generated visuals,

301
00:16:52,280 --> 00:16:54,640
Speaker 3: where it's not like a cool effect. It's like, wow,

302
00:16:54,680 --> 00:16:57,760
Speaker 3: this is really distracting because the background is constantly swirling

303
00:16:57,800 --> 00:17:00,400
Speaker 3: and changing and things are popping in and out. And

304
00:17:00,480 --> 00:17:03,680
Speaker 3: after I wrote that article, they still decided to invite me,

305
00:17:03,760 --> 00:17:05,760
Speaker 3: So I thought that was brave of them.

306
00:17:05,920 --> 00:17:07,159
Speaker 1: But what did you think, I mean, what were you

307
00:17:07,200 --> 00:17:08,400
Speaker 1: kind of expecting going into it?

308
00:17:08,880 --> 00:17:11,440
Speaker 3: Going in? I thought that they would be pretty bad,

309
00:17:11,560 --> 00:17:14,600
Speaker 3: to be totally honest with you, just because the state

310
00:17:14,640 --> 00:17:18,200
Speaker 3: of the art at the time. This was back in December,

311
00:17:18,520 --> 00:17:22,840
Speaker 3: which it was only three months ago, but at the time,

312
00:17:23,040 --> 00:17:28,920
Speaker 3: AI video generators were pretty bad, and I didn't think

313
00:17:28,960 --> 00:17:32,600
Speaker 3: that TCL had access to some proprietary system that we

314
00:17:32,680 --> 00:17:35,119
Speaker 3: hadn't seen before. I figured that they would be using

315
00:17:35,560 --> 00:17:38,040
Speaker 3: the state of the art that you can find on

316
00:17:38,119 --> 00:17:41,600
Speaker 3: the internet, and I think that those tools are not

317
00:17:41,800 --> 00:17:43,680
Speaker 3: very good, and so I thought that they would be bad,

318
00:17:43,800 --> 00:17:46,280
Speaker 3: to be totally honest with you, and they were bad.

319
00:17:48,640 --> 00:17:51,120
Speaker 1: Can you describe some of the highlights on the low Lights?

320
00:17:51,400 --> 00:17:54,760
Speaker 3: Yeah? I thought that the films themselves were just they

321
00:17:54,760 --> 00:17:57,199
Speaker 3: felt pretty rushed. So one of them was called The

322
00:17:57,240 --> 00:18:00,600
Speaker 3: Slug and it's about a woman who turns into a slug.

323
00:18:00,760 --> 00:18:02,719
Speaker 3: She has a disease that turns her into a slug

324
00:18:02,760 --> 00:18:05,600
Speaker 3: and it feels like The Substance, which is another you know,

325
00:18:05,640 --> 00:18:10,160
Speaker 3: Oscar nominated film. The visuals on it are wild. Things

326
00:18:10,200 --> 00:18:12,800
Speaker 3: are just like constantly changing. Her face is changing, the

327
00:18:13,080 --> 00:18:15,600
Speaker 3: you know, the food is changing. There's a lot of

328
00:18:15,640 --> 00:18:19,840
Speaker 3: like weird screams that happen that are not super well

329
00:18:19,880 --> 00:18:23,439
Speaker 3: timed with the dialogue. And then also there's like a

330
00:18:23,480 --> 00:18:26,240
Speaker 3: scene where the woman takes a bath and there's like

331
00:18:26,240 --> 00:18:29,040
Speaker 3: a close up on some bath salts and like the

332
00:18:29,160 --> 00:18:33,639
Speaker 3: text on that label is like an alien language because

333
00:18:33,680 --> 00:18:37,800
Speaker 3: AI has like a really bad time generating text, and

334
00:18:38,359 --> 00:18:39,800
Speaker 3: I guess you can take it with a grain of

335
00:18:39,840 --> 00:18:42,560
Speaker 3: salt or say like, hey, this is early technology. But

336
00:18:42,840 --> 00:18:45,200
Speaker 3: when you're watching something as a viewer in a movie

337
00:18:45,280 --> 00:18:48,439
Speaker 3: theater on this giant screen and the text is completely

338
00:18:48,800 --> 00:18:51,360
Speaker 3: not even in English, it's like, wow, it really takes

339
00:18:51,359 --> 00:18:52,360
Speaker 3: you out of the narrative.

340
00:18:52,520 --> 00:18:54,800
Speaker 1: I would say, I mean it's a weird idea, right,

341
00:18:54,880 --> 00:18:57,040
Speaker 1: because I mean you mentioned this is for TCL, the

342
00:18:57,119 --> 00:19:01,040
Speaker 1: Chinese TV manufacturer, and the assumption be like, they don't

343
00:19:01,080 --> 00:19:02,639
Speaker 1: want you to change the channel, right, they want you

344
00:19:02,680 --> 00:19:04,800
Speaker 1: to have their own channel on kind of in the

345
00:19:04,840 --> 00:19:07,240
Speaker 1: background so that you know your attention is with them

346
00:19:07,280 --> 00:19:08,800
Speaker 1: and they can sell you ads whatever it may be.

347
00:19:09,320 --> 00:19:11,800
Speaker 1: But that's very different to like putting hundreds of people

348
00:19:11,960 --> 00:19:14,359
Speaker 1: in a movie theater and kind of fulcing them to

349
00:19:14,400 --> 00:19:16,160
Speaker 1: watch with full attention, right, yeah.

350
00:19:16,280 --> 00:19:16,520
Speaker 2: Yeah.

351
00:19:16,520 --> 00:19:21,399
Speaker 3: And it's very interesting because before the movies played, two

352
00:19:21,440 --> 00:19:25,119
Speaker 3: TCL executives addressed the audience, and it was very interesting

353
00:19:25,280 --> 00:19:27,520
Speaker 3: the difference between what they were saying and what the

354
00:19:27,560 --> 00:19:31,600
Speaker 3: filmmakers were saying, because the TCL executives were business people

355
00:19:32,000 --> 00:19:35,639
Speaker 3: and they were saying our research shows that almost no

356
00:19:35,680 --> 00:19:38,800
Speaker 3: one changes the channel once they're watching something like this,

357
00:19:38,960 --> 00:19:42,000
Speaker 3: like they are watching it in the background usually, and

358
00:19:42,119 --> 00:19:44,520
Speaker 3: so their hope is that you're just going to be

359
00:19:44,520 --> 00:19:46,199
Speaker 3: too lazy to change the channel.

360
00:19:46,280 --> 00:19:48,800
Speaker 1: So inspiring creative brief.

361
00:19:49,440 --> 00:19:51,920
Speaker 3: Right, right, And then the other executives said, like, we're

362
00:19:51,920 --> 00:19:55,240
Speaker 3: going to use this as part of our targeted advertising strategy,

363
00:19:56,440 --> 00:20:00,280
Speaker 3: which was pretty dystopian. And then the actual filmmakers came

364
00:20:00,280 --> 00:20:02,040
Speaker 3: on and said, you know, we put our heart and

365
00:20:02,080 --> 00:20:03,960
Speaker 3: soul into this, and we think this is the future

366
00:20:03,960 --> 00:20:06,680
Speaker 3: of the industry. So that was kind of like a

367
00:20:06,680 --> 00:20:09,200
Speaker 3: whiplash situation for me in the audience.

368
00:20:12,240 --> 00:20:14,720
Speaker 1: When we come back, more from Jason Kebler about the

369
00:20:14,800 --> 00:20:19,240
Speaker 1: rapid advances in generative AI video technology and how the

370
00:20:19,280 --> 00:20:22,440
Speaker 1: state of the art is evolving in real time, stay

371
00:20:22,440 --> 00:20:37,320
Speaker 1: with us. Welcome back to our conversation with Jason Kebler

372
00:20:37,359 --> 00:20:40,639
Speaker 1: from four or four Media, where we continue our conversation

373
00:20:40,920 --> 00:20:45,359
Speaker 1: about a recent AI film festival he attended. There was

374
00:20:45,400 --> 00:20:47,240
Speaker 1: one film though, which I think was like a kind

375
00:20:47,240 --> 00:20:51,600
Speaker 1: of blended documentary and AI film that you thought was

376
00:20:51,640 --> 00:20:52,960
Speaker 1: potentially a bit more interesting.

377
00:20:53,560 --> 00:20:56,560
Speaker 3: Yeah, I thought it was pretty cool. I mean, it

378
00:20:56,640 --> 00:20:58,640
Speaker 3: still had a lot of problems, but It was called

379
00:20:58,640 --> 00:21:02,040
Speaker 3: The Best Day of My Life, and it was mountaineering

380
00:21:02,119 --> 00:21:07,840
Speaker 3: documentary where a mountaineer who got trapped in an avalanche

381
00:21:08,520 --> 00:21:11,600
Speaker 3: is talking directly to the camera, like the actual person

382
00:21:11,640 --> 00:21:14,119
Speaker 3: is talking directly to the camera recounting his story, and

383
00:21:14,280 --> 00:21:18,680
Speaker 3: as he is telling his story, they flashed to generative

384
00:21:18,720 --> 00:21:22,760
Speaker 3: AI depictions of what he is saying, And so I

385
00:21:22,840 --> 00:21:24,920
Speaker 3: thought that was kind of interesting because this is something

386
00:21:24,960 --> 00:21:27,040
Speaker 3: that happened to the guy. He obviously didn't bring a

387
00:21:27,080 --> 00:21:30,639
Speaker 3: camera with him at the time, and you were able

388
00:21:30,640 --> 00:21:34,600
Speaker 3: to sort of like see what he was describing.

389
00:21:34,440 --> 00:21:37,960
Speaker 1: In a way that was actually viscerally compelling, or in

390
00:21:37,960 --> 00:21:40,560
Speaker 1: a way that's still felt a bit uncanny and jarring.

391
00:21:40,640 --> 00:21:42,800
Speaker 3: In a way that made me think that maybe this

392
00:21:42,880 --> 00:21:45,920
Speaker 3: has potential in the future, but this isn't quite there yet,

393
00:21:45,960 --> 00:21:50,719
Speaker 3: because it similarly like the there's various scenes in the film,

394
00:21:51,359 --> 00:21:55,600
Speaker 3: and the guy who's happening to changes in each scene.

395
00:21:55,640 --> 00:21:59,719
Speaker 3: It's like his face looks different in different scenes. He

396
00:21:59,800 --> 00:22:02,800
Speaker 3: was under snow because it was an avalanche, and then

397
00:22:02,840 --> 00:22:05,080
Speaker 3: in the next scene all of the snow had turned

398
00:22:05,119 --> 00:22:07,639
Speaker 3: to mud, and then it turned back to snow, and

399
00:22:08,160 --> 00:22:12,480
Speaker 3: it was like, similarly took you out of the narrative,

400
00:22:12,680 --> 00:22:16,879
Speaker 3: but I thought that the idea behind it was pretty

401
00:22:16,880 --> 00:22:20,239
Speaker 3: interesting and I could see that being a direction that

402
00:22:20,920 --> 00:22:22,640
Speaker 3: future documentaries go.

403
00:22:23,280 --> 00:22:25,280
Speaker 1: And was what was the feeling like in the room?

404
00:22:25,320 --> 00:22:27,080
Speaker 1: I mean, who else was in the audience? What was

405
00:22:27,119 --> 00:22:30,040
Speaker 1: the general takeaway from this experience?

406
00:22:30,480 --> 00:22:34,280
Speaker 3: The mood in the theater was one of incredible optimism

407
00:22:34,440 --> 00:22:37,399
Speaker 3: and excitement. It was a mix of people who had

408
00:22:37,440 --> 00:22:40,760
Speaker 3: worked on these films and people who have like a

409
00:22:40,800 --> 00:22:44,000
Speaker 3: lot of money invested in the idea that this is

410
00:22:44,040 --> 00:22:46,640
Speaker 3: going to be the next big thing in Hollywood. And

411
00:22:46,720 --> 00:22:50,760
Speaker 3: so the mood in the theater was one of incredible

412
00:22:50,800 --> 00:22:55,800
Speaker 3: optimism and excitement. Meanwhile, the films like Objectively are not good.

413
00:22:55,840 --> 00:22:58,439
Speaker 3: They're really They're all on YouTube now and if you

414
00:22:58,480 --> 00:23:01,840
Speaker 3: go watch them, like the comments brutal, there's not a

415
00:23:01,840 --> 00:23:04,439
Speaker 3: lot of views on them. I think on some of them,

416
00:23:04,480 --> 00:23:06,720
Speaker 3: the comments that you even been turned off because people

417
00:23:06,760 --> 00:23:09,719
Speaker 3: are like, how could you dare put this on my television.

418
00:23:10,520 --> 00:23:13,440
Speaker 3: So I did think it was interesting because it reminded

419
00:23:13,440 --> 00:23:16,639
Speaker 3: me of things that I had been to in the past,

420
00:23:16,760 --> 00:23:20,639
Speaker 3: for like virtual reality or for cryptocurrency, things like that,

421
00:23:21,160 --> 00:23:24,480
Speaker 3: And a lot of people have said like generative AI

422
00:23:24,640 --> 00:23:26,879
Speaker 3: is the new crypto, it's the new metaverse, it's the

423
00:23:26,960 --> 00:23:31,480
Speaker 3: new virtual reality. And I think that AI there's like

424
00:23:31,520 --> 00:23:35,679
Speaker 3: a lot of snake oil out there, but undeniably companies

425
00:23:35,680 --> 00:23:37,679
Speaker 3: are leaning into it in a way that's going to

426
00:23:37,760 --> 00:23:41,720
Speaker 3: affect us and affect workers and affect people in the industry.

427
00:23:42,200 --> 00:23:47,000
Speaker 1: It's also interesting where companies fall in terms of how

428
00:23:47,600 --> 00:23:50,600
Speaker 1: vocal they want to be about how they see the

429
00:23:50,640 --> 00:23:54,639
Speaker 1: AI future unfolding. Right, Like, obviously for Chinese TV manufacturer,

430
00:23:55,040 --> 00:23:58,560
Speaker 1: alienating Hollywood doesn't really matter that much, right, whereas like

431
00:23:58,720 --> 00:24:01,320
Speaker 1: full Hollywood studios had to behave very differently.

432
00:24:01,920 --> 00:24:04,280
Speaker 3: Yeah, it's super interesting, and that's a great point because,

433
00:24:04,320 --> 00:24:07,800
Speaker 3: as you said, like the Writer's Guild strike was partially

434
00:24:07,880 --> 00:24:11,520
Speaker 3: about generative AI in the writer's rooms, a lot of

435
00:24:11,600 --> 00:24:15,680
Speaker 3: voice actors, going back to Respeecher, voice actors in both

436
00:24:15,680 --> 00:24:20,280
Speaker 3: the video game world and the animation world are really

437
00:24:20,320 --> 00:24:24,160
Speaker 3: worried that AI voices are going to replace their jobs

438
00:24:24,280 --> 00:24:27,240
Speaker 3: or that they're going to get less work because AI

439
00:24:27,359 --> 00:24:30,280
Speaker 3: is going to be used to generate voices for animation

440
00:24:30,400 --> 00:24:33,560
Speaker 3: and video games. And then, of course, like you said,

441
00:24:33,600 --> 00:24:37,320
Speaker 3: a lot of companies are laying off their workers in

442
00:24:37,359 --> 00:24:40,080
Speaker 3: a bunch of industries and then realizing, oh wait, the

443
00:24:40,080 --> 00:24:42,320
Speaker 3: AI is not good enough to do these jobs yet.

444
00:24:42,359 --> 00:24:45,200
Speaker 3: And so there's a real tension about it because fundamentally,

445
00:24:45,240 --> 00:24:49,520
Speaker 3: this is an automation technology. It's designed to replace human

446
00:24:49,600 --> 00:24:53,800
Speaker 3: labor or do things that sometimes humans can't do. And

447
00:24:54,040 --> 00:24:56,760
Speaker 3: I do think that a lot of companies are going

448
00:24:56,840 --> 00:25:00,920
Speaker 3: to be able to differentiate themselves by saying we do

449
00:25:01,000 --> 00:25:03,720
Speaker 3: not use AI, we respect human artists, we don't want

450
00:25:03,760 --> 00:25:05,959
Speaker 3: to do that. And then some companies are going their

451
00:25:06,000 --> 00:25:09,960
Speaker 3: total opposite way, like TCL, which has very little original programming,

452
00:25:10,080 --> 00:25:13,400
Speaker 3: very little relationships in Hollywood. They don't care if they

453
00:25:13,400 --> 00:25:16,080
Speaker 3: piss off directors and actors and things like that because

454
00:25:16,359 --> 00:25:18,240
Speaker 3: they're just trying to make a name for themselves, so

455
00:25:18,280 --> 00:25:20,360
Speaker 3: they're able to be more aggressive about this.

456
00:25:21,080 --> 00:25:22,919
Speaker 1: So I guess, on the one hand, you have like

457
00:25:23,080 --> 00:25:28,080
Speaker 1: TCL and more or less fully AI generated films. On

458
00:25:28,119 --> 00:25:30,159
Speaker 1: the other hand, you have the brutalist where you know

459
00:25:30,200 --> 00:25:33,040
Speaker 1: at the margins AI was used and respeech, she was

460
00:25:33,119 --> 00:25:35,959
Speaker 1: used to do some accent correction. Do you see like

461
00:25:36,080 --> 00:25:38,800
Speaker 1: ultimately a convergence between those two things, or do you

462
00:25:38,840 --> 00:25:42,080
Speaker 1: think it will remain that like AI is either used

463
00:25:42,119 --> 00:25:45,800
Speaker 1: in like premium productions for optimizing posts, shall we say.

464
00:25:45,880 --> 00:25:47,880
Speaker 1: And on the other hand, you have like this kind

465
00:25:47,880 --> 00:25:50,800
Speaker 1: of wild west of full AI generation, which is a

466
00:25:50,840 --> 00:25:53,040
Speaker 1: long way off from being consumable.

467
00:25:53,359 --> 00:25:55,520
Speaker 3: Yeah, I mean, I do think it's a spectrum and

468
00:25:56,600 --> 00:26:00,639
Speaker 3: slippery slope, if you will. And Special Effects have in

469
00:26:00,720 --> 00:26:03,680
Speaker 3: general been incorporating a lot more AI over the last

470
00:26:03,760 --> 00:26:07,000
Speaker 3: few years. I think one that was really interesting to

471
00:26:07,040 --> 00:26:11,280
Speaker 3: me was when the first deep fakes were sort of invented,

472
00:26:11,359 --> 00:26:13,640
Speaker 3: maybe like five or six years ago, where you can

473
00:26:13,720 --> 00:26:17,879
Speaker 3: like replace someone's face with another face. Star Wars had

474
00:26:18,240 --> 00:26:21,879
Speaker 3: tried to generate like Carrie Fisher after she had died

475
00:26:21,920 --> 00:26:25,119
Speaker 3: for one of the Star Wars films, and apparently they

476
00:26:25,160 --> 00:26:28,240
Speaker 3: spent like millions of dollars doing this. And then someone

477
00:26:28,359 --> 00:26:32,359
Speaker 3: on Reddit using deep fake technology was able to do

478
00:26:32,440 --> 00:26:36,359
Speaker 3: something that was almost indistinguishable from what Lucasfilms had done,

479
00:26:36,800 --> 00:26:39,679
Speaker 3: like on their computer at home, for free. And so

480
00:26:40,119 --> 00:26:42,560
Speaker 3: I do think that we're going to see a lot

481
00:26:42,600 --> 00:26:45,080
Speaker 3: more of this stuff in films, but you may not

482
00:26:45,160 --> 00:26:50,760
Speaker 3: even notice that's happening when they start replacing artists, replacing musicians,

483
00:26:50,840 --> 00:26:55,440
Speaker 3: replacing actors with AI. I think that's I personally think

484
00:26:55,480 --> 00:26:57,639
Speaker 3: that's a problem, and I think that that's when you

485
00:26:57,720 --> 00:27:00,480
Speaker 3: end up with a lesser product. Yeah, I don't know.

486
00:27:00,560 --> 00:27:02,480
Speaker 3: I hope that AI is going to be used to

487
00:27:02,520 --> 00:27:08,400
Speaker 3: make films better, not to create tons of low budget,

488
00:27:08,560 --> 00:27:11,960
Speaker 3: poorly made films that are designed to scratch a specific

489
00:27:12,040 --> 00:27:15,360
Speaker 3: itch or perform an algorithm, which we're definitely gonna see

490
00:27:15,400 --> 00:27:15,600
Speaker 3: a lot.

491
00:27:15,520 --> 00:27:19,679
Speaker 1: Of itist you're a humanist at HUT, Yeah, yeah, And

492
00:27:19,760 --> 00:27:22,600
Speaker 1: I mean you mentioned that this film festival was a

493
00:27:22,600 --> 00:27:25,119
Speaker 1: couple of months ago. Has the state of the art

494
00:27:25,240 --> 00:27:28,159
Speaker 1: change since then? I was playing around with this Google

495
00:27:28,200 --> 00:27:31,560
Speaker 1: deep Mind product called vo two. At least on like

496
00:27:31,560 --> 00:27:34,320
Speaker 1: a scene by scene basis, you can make pretty good

497
00:27:34,359 --> 00:27:37,560
Speaker 1: photo realistic depictions, but then like a couple of seconds each.

498
00:27:37,600 --> 00:27:39,399
Speaker 1: I don't think they've figured out that any means how

499
00:27:39,440 --> 00:27:42,480
Speaker 1: to stitch them together or make continuity. But how is

500
00:27:42,520 --> 00:27:43,760
Speaker 1: the state of the art devolving?

501
00:27:44,000 --> 00:27:46,639
Speaker 3: It's changed a lot in the last three months. There's

502
00:27:46,680 --> 00:27:50,160
Speaker 3: been a lot of Chinese companies that have released video

503
00:27:50,280 --> 00:27:53,080
Speaker 3: models in the last just a couple of weeks, like

504
00:27:53,200 --> 00:27:56,639
Speaker 3: ten Cent, which is a massive Chinese company, released a

505
00:27:56,680 --> 00:28:00,280
Speaker 3: new video model that seems to be better than most

506
00:28:00,560 --> 00:28:03,760
Speaker 3: publicly released video models. You know, it was sort of

507
00:28:03,800 --> 00:28:07,840
Speaker 3: immediately used by people to create non consensual pornography, which

508
00:28:07,880 --> 00:28:10,800
Speaker 3: is quite upsetting and is what a lot of people

509
00:28:10,800 --> 00:28:13,640
Speaker 3: are using these tools for on the internet. But basically

510
00:28:13,680 --> 00:28:16,040
Speaker 3: it's like every week there's a new model and they're

511
00:28:16,080 --> 00:28:19,000
Speaker 3: they're constantly leapfrogging each other. So you know, one will

512
00:28:19,040 --> 00:28:21,560
Speaker 3: be able to generate hands better than another, one will

513
00:28:21,560 --> 00:28:24,840
Speaker 3: be able to generate faces better than another, one will

514
00:28:24,880 --> 00:28:28,040
Speaker 3: have like better movement when you try to make these

515
00:28:28,080 --> 00:28:32,320
Speaker 3: people move, or they require less training data, meaning you

516
00:28:32,359 --> 00:28:36,560
Speaker 3: can make videos based on one input image versus having

517
00:28:36,600 --> 00:28:39,600
Speaker 3: to feed hours of footage into a model to create

518
00:28:39,640 --> 00:28:42,080
Speaker 3: something else. And so you know, these are things that

519
00:28:42,120 --> 00:28:45,160
Speaker 3: like AI nerds spend a lot of time caring about,

520
00:28:45,640 --> 00:28:49,240
Speaker 3: and I would say that there is a big generational

521
00:28:49,280 --> 00:28:52,800
Speaker 3: difference between them. But as like a consumer of these things,

522
00:28:52,800 --> 00:28:55,160
Speaker 3: you might not know that this is happening behind the scenes.

523
00:28:55,200 --> 00:28:58,959
Speaker 3: But the short version is basically it's getting easier to

524
00:28:59,080 --> 00:29:02,360
Speaker 3: make a generated video, it's getting cheaper to do it,

525
00:29:02,440 --> 00:29:05,920
Speaker 3: and the quality is getting better and it's changing on

526
00:29:05,960 --> 00:29:14,640
Speaker 3: like a day to day basis. At this point, Jason,

527
00:29:14,640 --> 00:29:16,600
Speaker 3: thank you so much. Thank you so much for having me.

528
00:29:20,360 --> 00:29:23,080
Speaker 1: That's it for this week. For tech Stuff, I'm oz Voloshin.

529
00:29:23,560 --> 00:29:26,920
Speaker 1: This episode was produced by Eliza Dennis and Victoria Dominguez.

530
00:29:27,160 --> 00:29:30,040
Speaker 1: It was executive produced by me Carrot Price and Kate

531
00:29:30,080 --> 00:29:34,360
Speaker 1: Osborne for Kaleidoscope and Katrina Norvell for iHeart Podcasts. The

532
00:29:34,480 --> 00:29:38,080
Speaker 1: Heath Fraser is our engineer. Kyle Murdoch mixed this episode

533
00:29:38,160 --> 00:29:40,760
Speaker 1: and he also wrote our theme song. Join us next

534
00:29:40,800 --> 00:29:43,440
Speaker 1: Wednesday for tech stuff The Story, when we'll share an

535
00:29:43,480 --> 00:29:47,560
Speaker 1: in depth conversation with the neuroscientist David Eagleman about people

536
00:29:47,680 --> 00:29:52,720
Speaker 1: who develop romantic relationships with AI. Please rate, review, and

537
00:29:52,800 --> 00:29:55,360
Speaker 1: reach out to us at tech Stuff podcast at gmail

538
00:29:55,360 --> 00:29:55,760
Speaker 1: dot com.

539
00:30:00,040 --> 00:30:00,200
Speaker 2: Eight