1
00:00:01,840 --> 00:00:02,880
Speaker 1: All Zone Media.

2
00:00:04,040 --> 00:00:06,680
Speaker 2: Hello and welcome to Better Offline. I'm your host ed

3
00:00:06,760 --> 00:00:22,280
Speaker 2: z trun. In early June, three researchers from the University

4
00:00:22,280 --> 00:00:25,440
Speaker 2: of Glasgow published a paper in the Ethics of Information

5
00:00:25,520 --> 00:00:29,159
Speaker 2: and Technology Journal called chat GBT is Bullshit. And I

6
00:00:29,280 --> 00:00:31,680
Speaker 2: just want to be clear. This is a great and

7
00:00:31,760 --> 00:00:35,000
Speaker 2: thoroughly researched and well argued paper. This is not silly

8
00:00:35,080 --> 00:00:38,680
Speaker 2: at all. It's actually great academia. And today I'm joined

9
00:00:38,680 --> 00:00:41,720
Speaker 2: by the men who wrote it, academics Michael Townsend Hicks,

10
00:00:42,040 --> 00:00:45,160
Speaker 2: James Humphries and Joe Slater, to talk about chat gbt's

11
00:00:45,200 --> 00:00:48,159
Speaker 2: mediocrity and how it's not really built to represent the

12
00:00:48,159 --> 00:00:51,480
Speaker 2: world at all. So, for the sake of argument, could

13
00:00:51,479 --> 00:00:53,360
Speaker 2: you define bullshit for me?

14
00:00:54,600 --> 00:00:58,920
Speaker 3: So you are bullshitting if you are speaking without caring

15
00:00:59,640 --> 00:01:02,560
Speaker 3: about the truth of what you say. So normally, if

16
00:01:02,560 --> 00:01:06,000
Speaker 3: I'm telling you staff about the world in a good case,

17
00:01:06,040 --> 00:01:08,640
Speaker 3: I'll be telling you something that's true, and like trying

18
00:01:08,680 --> 00:01:11,280
Speaker 3: to tell you something that's true. If I'm lying to you,

19
00:01:11,360 --> 00:01:14,000
Speaker 3: I'll be knowingly telling you something that's false or something

20
00:01:14,080 --> 00:01:17,160
Speaker 3: I think is false. I'm bullshitting. I just don't care.

21
00:01:17,680 --> 00:01:20,400
Speaker 3: I'm trying to get you to believe me. I don't

22
00:01:20,440 --> 00:01:22,440
Speaker 3: really care about whether what I say is true. I

23
00:01:22,520 --> 00:01:25,400
Speaker 3: might not have any particular view on whether it's true.

24
00:01:25,280 --> 00:01:29,039
Speaker 2: Or not right. And you define between like soft and

25
00:01:29,160 --> 00:01:31,520
Speaker 2: hard bullshit? Can you also get into that as well?

26
00:01:31,840 --> 00:01:34,360
Speaker 2: Can you also identify yourselves as well to.

27
00:01:34,440 --> 00:01:37,200
Speaker 4: R Yeah, yeah, and Joe. So the soft bullshit hard

28
00:01:37,200 --> 00:01:41,039
Speaker 4: bullshit distinction is a very serious and technical distinction, right,

29
00:01:42,319 --> 00:01:43,479
Speaker 4: So he came up with.

30
00:01:43,440 --> 00:01:48,520
Speaker 3: This because bullshit is in the technical philosophical sense. Comes

31
00:01:48,520 --> 00:01:52,880
Speaker 3: from Harry Frankfurt, recently diseased but a really great philosopher.

32
00:01:52,720 --> 00:01:54,720
Speaker 4: And he talks about the about.

33
00:01:54,520 --> 00:01:58,120
Speaker 3: Bullshit that there is in a popular culture and just

34
00:01:58,200 --> 00:02:01,160
Speaker 3: in general discourse these days. Some of the ways he

35
00:02:01,240 --> 00:02:04,280
Speaker 3: talks about bullshit seemed to suggest that it needs to

36
00:02:04,320 --> 00:02:07,920
Speaker 3: be accompanied by a sort of malign intention.

37
00:02:08,840 --> 00:02:10,919
Speaker 4: I'm doing something kind of bad.

38
00:02:11,080 --> 00:02:15,200
Speaker 3: I'm intending to mislead you about the enterprise of what

39
00:02:15,240 --> 00:02:15,639
Speaker 3: I'm doing.

40
00:02:15,760 --> 00:02:18,120
Speaker 4: Yeah, maybe about who you are or what you know.

41
00:02:18,200 --> 00:02:21,760
Speaker 5: So you may be trying to portray yourself as someone

42
00:02:21,800 --> 00:02:25,799
Speaker 5: who is knowledgeable about a particular subject. Maybe you're a

43
00:02:25,840 --> 00:02:28,280
Speaker 5: student who showed up to class without doing the work.

44
00:02:28,800 --> 00:02:32,320
Speaker 5: Maybe you're trying to portray yourself as someone who's virtuous

45
00:02:32,320 --> 00:02:34,640
Speaker 5: in ways you're not. Maybe you're a politician who wants

46
00:02:34,639 --> 00:02:37,840
Speaker 5: to seem like you care about your constituents, but actually

47
00:02:37,880 --> 00:02:41,480
Speaker 5: you don't. So you're not trying to mislead somebody about

48
00:02:42,080 --> 00:02:46,000
Speaker 5: what you're saying the content of your utterance. You're trying

49
00:02:46,040 --> 00:02:48,799
Speaker 5: to mislead them instead about like why you're saying it's

50
00:02:50,120 --> 00:02:51,679
Speaker 5: that's what we call hard bullshit.

51
00:02:51,840 --> 00:02:53,960
Speaker 4: Yeah, it's one of the things Frankfort talked about.

52
00:02:54,200 --> 00:02:56,880
Speaker 3: Yeah, so FRANKFOT doesn't make this hard bullshit soft bullshit

53
00:02:56,919 --> 00:02:59,720
Speaker 3: the station, but we do because sometimes it seems like

54
00:02:59,760 --> 00:03:02,720
Speaker 3: Frank for has this particular kind of intention in mind,

55
00:03:03,280 --> 00:03:06,240
Speaker 3: but sometimes he's just a bit looser with it. And

56
00:03:06,280 --> 00:03:09,160
Speaker 3: we want to say that chat GPT and other large

57
00:03:09,240 --> 00:03:12,799
Speaker 3: language models they don't really have this intention to deceive

58
00:03:12,880 --> 00:03:15,160
Speaker 3: because they're not people that don't have these intentions. They're

59
00:03:15,160 --> 00:03:17,680
Speaker 3: not trying to mess with us in that kind of way,

60
00:03:18,040 --> 00:03:21,120
Speaker 3: but they do lack this kind of caring about truth.

61
00:03:21,840 --> 00:03:24,520
Speaker 1: Well, I'm James, by the way, I suppose we strictly

62
00:03:24,600 --> 00:03:27,320
Speaker 1: don't want to say that they aren't hard bullshitters, and

63
00:03:27,360 --> 00:03:29,840
Speaker 1: we just think if you don't think that large language

64
00:03:29,880 --> 00:03:31,640
Speaker 1: models are sapient. If you don't think they're kind of

65
00:03:31,680 --> 00:03:35,080
Speaker 1: minds in any important way, then they're not hard bullshitters.

66
00:03:35,080 --> 00:03:37,120
Speaker 1: So I think in the paper we don't take a

67
00:03:37,120 --> 00:03:39,320
Speaker 1: position on whether or not they are. We just say

68
00:03:39,800 --> 00:03:41,720
Speaker 1: if they are, this is the way in which they are.

69
00:03:41,920 --> 00:03:45,080
Speaker 1: But minimally they're soft bullshitters. So they're kind of soft bullshites,

70
00:03:45,120 --> 00:03:48,120
Speaker 1: as Joe says, doesn't require that speakers attempting to deceive

71
00:03:48,160 --> 00:03:50,360
Speaker 1: the audience about the nature of the enterprise.

72
00:03:50,560 --> 00:03:51,440
Speaker 6: Hard bullshit does.

73
00:03:51,680 --> 00:03:54,040
Speaker 1: So if it turns out that large language models are sapient,

74
00:03:54,080 --> 00:03:56,800
Speaker 1: which they're definitely not, like that's just technoproperty.

75
00:03:57,160 --> 00:03:58,280
Speaker 2: Yeah, that's nonsense.

76
00:03:58,400 --> 00:04:01,400
Speaker 1: Yeah, But if they are, then they're bullshitters and minimally

77
00:04:01,440 --> 00:04:03,480
Speaker 1: their soft bullshitters or their bullshit machines.

78
00:04:04,880 --> 00:04:07,880
Speaker 2: So you also make the distinction in there the intention,

79
00:04:08,240 --> 00:04:10,920
Speaker 2: so the very fabric of hard bullshit that you intentionally

80
00:04:10,960 --> 00:04:14,400
Speaker 2: are bullshitting to someone, you kind of make this distinction

81
00:04:14,480 --> 00:04:19,000
Speaker 2: that the intention of the designer and the involved prompting

82
00:04:19,160 --> 00:04:23,240
Speaker 2: could make this hard bullshit. Because with a lot of

83
00:04:23,279 --> 00:04:27,000
Speaker 2: these models and someone recently jail broke chat GPT and

84
00:04:27,040 --> 00:04:29,520
Speaker 2: it listed all of the things that's prompted to do

85
00:04:29,880 --> 00:04:33,800
Speaker 2: could prompting be considered intentional? That Sam Altman, CEO of

86
00:04:34,040 --> 00:04:37,000
Speaker 2: open AI, could be intentionally bullshitting? I think it is.

87
00:04:37,320 --> 00:04:39,279
Speaker 1: Yeah, this again, I think it's something I don't know

88
00:04:39,320 --> 00:04:41,680
Speaker 1: what the kind of hive mind consensus on this is.

89
00:04:41,720 --> 00:04:44,080
Speaker 1: I'm sort of sympathetic to the idea that if you

90
00:04:44,160 --> 00:04:48,479
Speaker 1: take this kind of purposive or teleological attitude towards right,

91
00:04:48,920 --> 00:04:51,440
Speaker 1: what an intention is is an effort to do something,

92
00:04:51,600 --> 00:04:54,320
Speaker 1: then maybe they do have intentions. But again, I think

93
00:04:54,320 --> 00:04:55,720
Speaker 1: in the paper we just sort of wanted I mean,

94
00:04:55,760 --> 00:04:59,080
Speaker 1: it's a standard philosophical move, right, just sort of go, look,

95
00:04:59,080 --> 00:05:01,520
Speaker 1: here's all this controversial stuff as we can make it.

96
00:05:01,760 --> 00:05:03,640
Speaker 1: Now we can hit you with the really controversial shit

97
00:05:03,680 --> 00:05:05,799
Speaker 1: that we wanted to get to. So in the paper

98
00:05:05,800 --> 00:05:08,359
Speaker 1: we sort of deliberately when maybe you might think it

99
00:05:08,400 --> 00:05:10,760
Speaker 1: has intentions for it this reason, we kind of have

100
00:05:10,839 --> 00:05:13,839
Speaker 1: no judgment on this efficiently, I'm sympathetic to the sort.

101
00:05:13,640 --> 00:05:14,440
Speaker 6: Of view that you're putting.

102
00:05:14,480 --> 00:05:15,920
Speaker 4: I think you're kind of sympathetic this as well.

103
00:05:16,240 --> 00:05:17,160
Speaker 6: Right, Yeah, so.

104
00:05:17,320 --> 00:05:19,080
Speaker 5: I Mike, there are a few ways that you can

105
00:05:19,120 --> 00:05:21,960
Speaker 5: think of CHATCHYBT as having intentions.

106
00:05:21,440 --> 00:05:23,920
Speaker 4: I think, and we talked about a few of them.

107
00:05:23,960 --> 00:05:27,320
Speaker 5: One is by thinking of the designers who created it

108
00:05:27,720 --> 00:05:30,280
Speaker 5: as kind of imbuing it with intentions. So they created

109
00:05:30,320 --> 00:05:32,760
Speaker 5: it for a purpose and to do a particular task,

110
00:05:33,520 --> 00:05:36,800
Speaker 5: and that task is to make people think that it's

111
00:05:36,839 --> 00:05:38,160
Speaker 5: a normal sounding person.

112
00:05:38,440 --> 00:05:38,640
Speaker 6: Right.

113
00:05:38,680 --> 00:05:41,760
Speaker 5: It's to make people when they have a conversation with it,

114
00:05:41,880 --> 00:05:45,760
Speaker 5: not be able to distinguish between what it's outputting and

115
00:05:45,880 --> 00:05:47,680
Speaker 5: what a normal human would say.

116
00:05:48,000 --> 00:05:48,200
Speaker 6: Right.

117
00:05:48,640 --> 00:05:51,080
Speaker 4: Right, And that kind.

118
00:05:50,880 --> 00:05:56,040
Speaker 5: Of goal, if it amounts to an intention, is the

119
00:05:56,160 --> 00:05:59,040
Speaker 5: kind of intention we think a bullshitter has.

120
00:05:59,120 --> 00:05:59,280
Speaker 4: Right.

121
00:05:59,360 --> 00:06:01,920
Speaker 5: It's not trying to deceive you about anything it's saying.

122
00:06:01,960 --> 00:06:04,120
Speaker 5: It doesn't care whether what it's saying is true. That's

123
00:06:04,160 --> 00:06:06,280
Speaker 5: not part of the goal. But the goal is to

124
00:06:06,279 --> 00:06:09,080
Speaker 5: do is to make it seem as if it's something

125
00:06:09,080 --> 00:06:13,040
Speaker 5: that it's not, like specifically a human interlocutor. And one

126
00:06:13,040 --> 00:06:16,240
Speaker 5: source for that goal is the programmers who designed it.

127
00:06:16,680 --> 00:06:20,600
Speaker 5: Another is the training method. So it was trained by

128
00:06:20,600 --> 00:06:23,760
Speaker 5: being given sort of positive and negative feedback in order

129
00:06:23,839 --> 00:06:25,480
Speaker 5: to achieve.

130
00:06:25,120 --> 00:06:28,680
Speaker 4: A specific thing, right, And that specific thing is just

131
00:06:28,760 --> 00:06:29,520
Speaker 4: sounding normal.

132
00:06:29,720 --> 00:06:31,640
Speaker 5: And that's so similar to what our students are doing

133
00:06:31,640 --> 00:06:34,720
Speaker 5: when they try to pretend that they read something they

134
00:06:34,760 --> 00:06:35,279
Speaker 5: haven't read.

135
00:06:36,440 --> 00:06:39,479
Speaker 2: There is something very collegiate about the way it bullshits though.

136
00:06:39,560 --> 00:06:42,159
Speaker 2: It reminds me of when I was in college when

137
00:06:42,240 --> 00:06:45,280
Speaker 2: Penn State and Aporistwyth two very different institutes.

138
00:06:45,040 --> 00:06:48,080
Speaker 4: Right, both kind of in the middle of nowhere.

139
00:06:49,080 --> 00:06:52,120
Speaker 2: Yeah, both very sad. But the one thing you saw

140
00:06:52,160 --> 00:06:54,960
Speaker 2: with like students who were doing like B plus homework

141
00:06:55,040 --> 00:06:57,360
Speaker 2: is they were using words they didn't really understand. They

142
00:06:57,360 --> 00:06:59,520
Speaker 2: were putting things together in a way that realists, like

143
00:06:59,520 --> 00:07:03,440
Speaker 2: the intro body conclusion, there was a certain formula behind it,

144
00:07:03,480 --> 00:07:06,279
Speaker 2: and it's just it feels exactly like it. But that

145
00:07:06,360 --> 00:07:08,160
Speaker 2: kind of brings me to my next question, which is,

146
00:07:08,480 --> 00:07:11,600
Speaker 2: how did you decide to write this? What inspired this?

147
00:07:11,920 --> 00:07:12,000
Speaker 5: Right?

148
00:07:12,040 --> 00:07:14,120
Speaker 1: I would feel this because whenever Mike tells the story,

149
00:07:14,160 --> 00:07:16,800
Speaker 1: you get sort of sanitized version. We were in the

150
00:07:16,800 --> 00:07:19,480
Speaker 1: pub whining about student essays. Perfect, yeah, like.

151
00:07:19,720 --> 00:07:22,800
Speaker 4: You were not long here, right, you're not long yeah.

152
00:07:22,800 --> 00:07:25,040
Speaker 1: Long in post of a bunch of us, or went

153
00:07:25,080 --> 00:07:27,960
Speaker 1: to the pub on emotional let's let's welcome our new

154
00:07:27,960 --> 00:07:30,720
Speaker 1: members of staff sort of event, and inevitably with hat

155
00:07:30,720 --> 00:07:32,920
Speaker 1: about two points we were all pissing in moaning and

156
00:07:33,440 --> 00:07:34,320
Speaker 1: do you think it might have been.

157
00:07:34,200 --> 00:07:35,960
Speaker 4: Neiled that kind of prompted this. No, I don't think

158
00:07:35,960 --> 00:07:36,360
Speaker 4: he was the.

159
00:07:38,760 --> 00:07:40,560
Speaker 6: We talked about it after, right, Yeah, in a case

160
00:07:40,600 --> 00:07:41,240
Speaker 6: like sort of came up.

161
00:07:41,280 --> 00:07:43,040
Speaker 1: We were talking about this sort of prevalence of chat

162
00:07:43,080 --> 00:07:46,560
Speaker 1: GPT generated stuff and kind of what it said about

163
00:07:47,240 --> 00:07:50,240
Speaker 1: how at least some students were kind of approaching the assessments,

164
00:07:50,320 --> 00:07:51,920
Speaker 1: and you know, I forget who. Someone sort of just

165
00:07:51,960 --> 00:07:53,800
Speaker 1: went off handly, yeah, it was all just Frank thirteen

166
00:07:53,840 --> 00:07:56,680
Speaker 1: bullshit though, isn't it. And we sort of collectively went hello,

167
00:07:56,760 --> 00:07:58,880
Speaker 1: because obviously, all having a background in plosphy, we've all

168
00:07:58,920 --> 00:08:01,200
Speaker 1: read on bullshit. You know, we all went he we

169
00:08:01,280 --> 00:08:04,080
Speaker 1: get to say bullshit and allegively serious academic work. So

170
00:08:04,160 --> 00:08:07,280
Speaker 1: the start it was we've had this experience of having

171
00:08:07,280 --> 00:08:10,240
Speaker 1: to read all of this kind of uncanny Valley stuff,

172
00:08:10,560 --> 00:08:12,960
Speaker 1: and when prompted, we all went, oh, it really is

173
00:08:13,120 --> 00:08:15,000
Speaker 1: like Frank thirty and bullshit in a way that we

174
00:08:15,000 --> 00:08:16,200
Speaker 1: can probably get paper outselof.

175
00:08:16,320 --> 00:08:19,480
Speaker 5: Yeah, And at that point I think we were in

176
00:08:19,520 --> 00:08:24,320
Speaker 5: the department meetings talking about how to deal with chat

177
00:08:24,400 --> 00:08:27,640
Speaker 5: GPT written papers, and there were discussions going on all

178
00:08:27,680 --> 00:08:29,800
Speaker 5: over the university, including kind of in high levels and

179
00:08:29,800 --> 00:08:34,360
Speaker 5: the administration to come up with a policy, and we

180
00:08:34,440 --> 00:08:38,960
Speaker 5: specifically wanted a stronger policy because our university is very

181
00:08:38,960 --> 00:08:42,400
Speaker 5: interested in cutting into technology, right, and so they wanted

182
00:08:42,440 --> 00:08:45,400
Speaker 5: the students to have an experience using and figuring out

183
00:08:45,400 --> 00:08:48,559
Speaker 5: how to use whatever the freshest technology is.

184
00:08:48,520 --> 00:08:52,719
Speaker 2: Which as they should as they should, but not for essays, right,

185
00:08:52,880 --> 00:08:53,280
Speaker 2: That's the.

186
00:08:53,240 --> 00:08:53,920
Speaker 4: Worry we had.

187
00:08:54,040 --> 00:08:56,600
Speaker 5: We thought, you know, if we're not very clear about this,

188
00:08:56,720 --> 00:08:58,360
Speaker 5: the students will be using it in a way that

189
00:08:58,440 --> 00:09:01,360
Speaker 5: will detract from the educational experience.

190
00:09:02,040 --> 00:09:05,400
Speaker 4: Right. And at the same time, it was becoming more.

191
00:09:05,200 --> 00:09:08,760
Speaker 5: Widely known how these machines work, like specifically how they're

192
00:09:08,800 --> 00:09:12,960
Speaker 5: doing next token or next word prediction in order to

193
00:09:13,040 --> 00:09:17,079
Speaker 5: come up with a digestible string of text. And when

194
00:09:17,080 --> 00:09:19,920
Speaker 5: you know that that's how you're they're doing it, I mean,

195
00:09:21,040 --> 00:09:25,080
Speaker 5: it seems so similar to what humans do when they

196
00:09:25,080 --> 00:09:28,000
Speaker 5: have no idea what they're talking about. And so when

197
00:09:28,040 --> 00:09:30,760
Speaker 5: we were talking about it just seemed like an obvious

198
00:09:30,800 --> 00:09:32,920
Speaker 5: paper that someone was going to write, and we thought

199
00:09:32,960 --> 00:09:36,079
Speaker 5: that a better bs. You know, eventually people are going

200
00:09:36,120 --> 00:09:38,000
Speaker 5: to see this connection and it's.

201
00:09:37,800 --> 00:09:39,080
Speaker 2: A great papa.

202
00:09:39,200 --> 00:09:40,160
Speaker 4: I think it worked very well.

203
00:09:40,200 --> 00:09:41,880
Speaker 5: I mean, I think that it's the kind of thing

204
00:09:41,920 --> 00:09:45,640
Speaker 5: where when I pitch this to other philosophers, it doesn't

205
00:09:45,640 --> 00:09:48,840
Speaker 5: take them very long to just agree to be like, ah, yes.

206
00:09:49,040 --> 00:09:49,680
Speaker 6: That's right.

207
00:09:50,200 --> 00:09:54,080
Speaker 2: It's an interesting philosophical concept as well, because the way

208
00:09:54,080 --> 00:09:56,319
Speaker 2: that people look at chat GPT in large language models

209
00:09:56,400 --> 00:09:58,600
Speaker 2: is very much machine do this. But when you think

210
00:09:58,640 --> 00:10:01,400
Speaker 2: about it, there are the ways where people are giving

211
00:10:01,440 --> 00:10:03,600
Speaker 2: it consciousness. I just sort of tweet just now as

212
00:10:03,640 --> 00:10:07,000
Speaker 2: someone talking about saying please with every request. It's like, no,

213
00:10:07,160 --> 00:10:09,800
Speaker 2: I will abuse the computer in whatever manner I see fit.

214
00:10:10,280 --> 00:10:15,520
Speaker 2: But it's it's curious because I think more people need

215
00:10:15,559 --> 00:10:18,559
Speaker 2: to be having conversations like this. And one particular thing

216
00:10:18,600 --> 00:10:20,120
Speaker 2: I like that you said, I actually would love you

217
00:10:20,120 --> 00:10:22,280
Speaker 2: to go into more detail, is you said, the chat

218
00:10:22,280 --> 00:10:25,680
Speaker 2: GPT in large language buddles, they're not designed to represent

219
00:10:25,720 --> 00:10:28,440
Speaker 2: the world at all. They're not lying or misrepresenting. They're

220
00:10:28,440 --> 00:10:30,160
Speaker 2: not designed to do that. What do you mean by that?

221
00:10:30,520 --> 00:10:33,520
Speaker 5: I mean kind of as background, I do philosophy of science,

222
00:10:33,640 --> 00:10:38,320
Speaker 5: but my thoughts about something like CHATCHEPT are largely inspired

223
00:10:38,360 --> 00:10:40,280
Speaker 5: by the fact that I also teach a class called

224
00:10:40,960 --> 00:10:42,800
Speaker 5: Understanding Philosophy through Science Fiction.

225
00:10:43,080 --> 00:10:44,800
Speaker 4: Oh and and now we like talk.

226
00:10:44,679 --> 00:10:48,679
Speaker 5: About whether computers could be conscious and I don't know

227
00:10:48,720 --> 00:10:51,480
Speaker 5: what you guys think. Actually I think they could, right,

228
00:10:52,880 --> 00:10:55,520
Speaker 5: I just don't think this one is. And part of

229
00:10:55,520 --> 00:10:57,800
Speaker 5: the reason I think they could but this one isn't

230
00:10:57,960 --> 00:11:00,120
Speaker 5: is that I think that in order to sort of

231
00:11:00,160 --> 00:11:03,320
Speaker 5: represent the world or have the kinds of things we

232
00:11:03,400 --> 00:11:07,480
Speaker 5: have that are like beliefs, desires, thoughts that are about

233
00:11:07,640 --> 00:11:12,920
Speaker 5: external things, you have to have internal states that are

234
00:11:12,960 --> 00:11:16,920
Speaker 5: connected in some way to the external world. Usually causation.

235
00:11:17,080 --> 00:11:20,160
Speaker 5: We're perceiving things, information is coming in. Then we've got

236
00:11:20,320 --> 00:11:23,440
Speaker 5: some kind of state in our brain that's designed just

237
00:11:23,480 --> 00:11:25,839
Speaker 5: to track these things in the external world.

238
00:11:26,080 --> 00:11:26,280
Speaker 4: Right.

239
00:11:26,800 --> 00:11:29,480
Speaker 5: That's a huge part of our cognitive lives. It is

240
00:11:29,679 --> 00:11:34,440
Speaker 5: just tracking external world things and it's a very important

241
00:11:34,440 --> 00:11:36,839
Speaker 5: part of childhood development when you figure out how to

242
00:11:36,880 --> 00:11:37,360
Speaker 5: interact them.

243
00:11:37,520 --> 00:11:42,000
Speaker 2: It's semiotics, right, Daniel John Aberistwyth told me semiotics it's

244
00:11:42,040 --> 00:11:43,000
Speaker 2: like an reception of the world.

245
00:11:43,000 --> 00:11:45,959
Speaker 5: And this is like theory of meaning stuff. So, yeah,

246
00:11:46,120 --> 00:11:51,599
Speaker 5: semiotics is like theory of science. How is it that assign.

247
00:11:50,200 --> 00:11:52,480
Speaker 6: Can be folks, the representation of the thing and the

248
00:11:52,520 --> 00:11:53,160
Speaker 6: thing itself.

249
00:11:53,280 --> 00:11:55,080
Speaker 4: That can happen, yeah, but not always.

250
00:11:55,080 --> 00:11:58,560
Speaker 5: Sometimes it's just the representation of the thing and there's

251
00:11:58,840 --> 00:12:02,240
Speaker 5: a lot of philosophy as at figuring out how brain

252
00:12:02,320 --> 00:12:05,840
Speaker 5: states or words on a page can be about external

253
00:12:05,840 --> 00:12:09,480
Speaker 5: world things. And a big part of it, at least

254
00:12:09,480 --> 00:12:13,240
Speaker 5: from my perspective, has to do with tracking those things,

255
00:12:13,800 --> 00:12:16,959
Speaker 5: keeping tabs on them, changing as a result of seeing

256
00:12:17,000 --> 00:12:20,880
Speaker 5: differences in the external world. And chatch ept is not

257
00:12:20,960 --> 00:12:24,040
Speaker 5: doing any of that, right, That's not what it's designed

258
00:12:24,040 --> 00:12:26,600
Speaker 5: to do. It's taking in a lot of data once

259
00:12:27,400 --> 00:12:31,600
Speaker 5: and then using that to respond to text. But it's

260
00:12:31,640 --> 00:12:37,559
Speaker 5: not remembering individuals, tracking things in the world in any

261
00:12:37,640 --> 00:12:41,640
Speaker 5: way perceiving things in the world. It's just forming a

262
00:12:41,720 --> 00:12:45,240
Speaker 5: sort of statistical model of what people say. And that's

263
00:12:45,320 --> 00:12:50,880
Speaker 5: kind of so divorced from what most thinking beings do.

264
00:12:51,600 --> 00:12:53,079
Speaker 2: It's divorced from experience.

265
00:12:53,320 --> 00:12:54,800
Speaker 4: Yeah. I mean, as far as I can tell, it

266
00:12:54,800 --> 00:12:56,679
Speaker 4: doesn't have anything like experience.

267
00:12:56,960 --> 00:12:57,160
Speaker 6: Yeah.

268
00:12:57,240 --> 00:12:59,400
Speaker 1: And that's one of the things that this I thinks of.

269
00:12:59,640 --> 00:13:01,679
Speaker 1: In one way it comes down to is that if

270
00:13:01,679 --> 00:13:03,680
Speaker 1: you sort of post this sufficiently far, someone is going

271
00:13:03,760 --> 00:13:06,000
Speaker 1: to go, ah, isn't this just biochauvinism, right, Like, aren't

272
00:13:06,000 --> 00:13:09,000
Speaker 1: you just assuming that unless something runs on like meat,

273
00:13:09,360 --> 00:13:11,520
Speaker 1: it can't be sentient? And this isn't something we get

274
00:13:11,559 --> 00:13:14,160
Speaker 1: into into paper part because we didn't really think it

275
00:13:14,200 --> 00:13:18,360
Speaker 1: was worth addressing. But the sorts of things that seem

276
00:13:18,520 --> 00:13:20,840
Speaker 1: like they're like never mind consciousness, right, but it seemed

277
00:13:20,840 --> 00:13:22,640
Speaker 1: to be necessary in order for something to be trying

278
00:13:22,960 --> 00:13:25,000
Speaker 1: to track the world, or in order to be corresponding

279
00:13:25,040 --> 00:13:27,120
Speaker 1: the world, or to form beliefs about the world. Chat

280
00:13:27,200 --> 00:13:29,959
Speaker 1: GPT just doesn't seem to meet any of them. Kind Of,

281
00:13:30,040 --> 00:13:32,960
Speaker 1: it's if it does turn out that it's sapient, then

282
00:13:33,000 --> 00:13:39,360
Speaker 1: chat GPT has got some profoundly serious executive function disorders, right, sapient, Right,

283
00:13:39,360 --> 00:13:40,760
Speaker 1: so we don't have to worry about it. But kind

284
00:13:40,760 --> 00:13:43,280
Speaker 1: of it's not the case that we've got some blundering

285
00:13:43,320 --> 00:13:45,599
Speaker 1: proto general intelligence that's trying to figure out how to

286
00:13:45,679 --> 00:13:47,480
Speaker 1: represent the world. It's not trying to represent the world

287
00:13:47,480 --> 00:13:50,080
Speaker 1: at all. It's utterances are designed to look as if

288
00:13:50,360 --> 00:13:52,040
Speaker 1: it's trying to represent the world, and then we just

289
00:13:52,080 --> 00:13:53,320
Speaker 1: go that that's just bullshit.

290
00:13:53,360 --> 00:13:54,920
Speaker 6: This is a classic case of bullshit.

291
00:13:55,280 --> 00:13:57,680
Speaker 2: Yeah, it seems to be making stuff up, but making

292
00:13:57,760 --> 00:13:59,880
Speaker 2: stuff up doesn't even seem to describe it. It's just

293
00:14:00,280 --> 00:14:03,520
Speaker 2: throwing shit at a wall very accurately, but not accurately enough.

294
00:14:03,600 --> 00:14:05,840
Speaker 1: Yeah, it's got various guidelines that allow it to throw

295
00:14:05,880 --> 00:14:07,439
Speaker 1: shit at the wall where a sort of reasonably high

296
00:14:07,440 --> 00:14:08,719
Speaker 1: degree of Actually, no, you're right. I mean, one of

297
00:14:08,720 --> 00:14:10,800
Speaker 1: the things that a human bullshitter could at least be

298
00:14:10,840 --> 00:14:13,320
Speaker 1: characterized as doing is they'd have to try and kind

299
00:14:13,320 --> 00:14:15,800
Speaker 1: of judge their audience, right, They'd have to try and

300
00:14:15,840 --> 00:14:18,440
Speaker 1: make the bullshit plausible to the audience that they're speaking to,

301
00:14:18,720 --> 00:14:20,840
Speaker 1: and chat GPT can't do that, right. All it can

302
00:14:20,880 --> 00:14:23,640
Speaker 1: do is go on a statistically large enough model and

303
00:14:24,520 --> 00:14:25,239
Speaker 1: it looks.

304
00:14:25,040 --> 00:14:27,080
Speaker 6: Like Z follows why follows X?

305
00:14:27,800 --> 00:14:30,120
Speaker 1: Right, it's not kind of got any does have any

306
00:14:30,160 --> 00:14:32,280
Speaker 1: consciousness at all, of course, but it doesn't have any

307
00:14:32,480 --> 00:14:35,240
Speaker 1: sensitivity to the sorts of things that people are in

308
00:14:35,280 --> 00:14:38,080
Speaker 1: fact likely to find plausible. It just does a kind

309
00:14:38,080 --> 00:14:40,480
Speaker 1: of brute number crunch. So it's more complicated than that,

310
00:14:40,520 --> 00:14:43,000
Speaker 1: but I think it boies down effectively to kind of

311
00:14:43,080 --> 00:14:45,560
Speaker 1: number crunching, right, kind of data and context.

312
00:14:45,240 --> 00:14:49,000
Speaker 2: In which probabilistic planning of what planning, it doesn't plan.

313
00:14:49,240 --> 00:14:51,320
Speaker 2: That's the thing. It's interesting. There are these words you

314
00:14:51,440 --> 00:14:53,400
Speaker 2: use to describe things that, when you think about it,

315
00:14:53,440 --> 00:14:57,160
Speaker 2: are not accurate. You can't say it plans or things.

316
00:14:57,960 --> 00:14:58,720
Speaker 4: I think one thing.

317
00:14:58,840 --> 00:15:01,760
Speaker 5: Between when we came up with the idea and when

318
00:15:01,800 --> 00:15:05,720
Speaker 5: we finish writing the paper, we spent some time reading

319
00:15:05,760 --> 00:15:08,880
Speaker 5: about how it works and how it represents language and

320
00:15:08,920 --> 00:15:10,560
Speaker 5: what the statistical.

321
00:15:09,920 --> 00:15:10,840
Speaker 4: Model is like.

322
00:15:11,160 --> 00:15:17,200
Speaker 5: And I was maybe more impressed than James about that,

323
00:15:17,280 --> 00:15:20,760
Speaker 5: because it is like doing things that are similar to

324
00:15:20,800 --> 00:15:23,760
Speaker 5: what we do when we understand language. It does seem

325
00:15:23,800 --> 00:15:27,920
Speaker 5: to kind of locate words in a meaning space right right,

326
00:15:28,360 --> 00:15:30,680
Speaker 5: and connect them to other words, and you know, show

327
00:15:30,760 --> 00:15:33,480
Speaker 5: similarity and meaning, and it also does seem to be

328
00:15:33,520 --> 00:15:38,560
Speaker 5: able to in some way understand context. But we don't

329
00:15:38,560 --> 00:15:42,440
Speaker 5: know how similar that is for a variety of reasons,

330
00:15:42,440 --> 00:15:45,520
Speaker 5: but mostly because it's too big of a system and

331
00:15:45,560 --> 00:15:49,880
Speaker 5: we can only kind of probe it. And it's trained indirectly, right,

332
00:15:50,000 --> 00:15:55,560
Speaker 5: so it's not programmed by individuals. And even though that's

333
00:15:55,640 --> 00:15:58,240
Speaker 5: kind of a very impressive model of language and meaning

334
00:15:58,480 --> 00:16:02,080
Speaker 5: and may some ways be similar to what we do

335
00:16:02,160 --> 00:16:06,560
Speaker 5: when we understand language, we're doing a lot more things

336
00:16:06,640 --> 00:16:10,920
Speaker 5: like planning, things like tracking things in the world. Just

337
00:16:10,960 --> 00:16:13,960
Speaker 5: having desires and representing the way you want the world

338
00:16:13,960 --> 00:16:16,960
Speaker 5: to be and thereby generating goals doesn't seem to be

339
00:16:17,400 --> 00:16:18,880
Speaker 5: something that it has anything any.

340
00:16:18,800 --> 00:16:19,880
Speaker 4: Room for in its architects.

341
00:16:20,160 --> 00:16:21,480
Speaker 1: This is something of the time of it's just goods

342
00:16:21,520 --> 00:16:23,200
Speaker 1: me and you were talking about the kind of in

343
00:16:23,280 --> 00:16:25,040
Speaker 1: some ways it learns language the same way.

344
00:16:24,880 --> 00:16:25,320
Speaker 6: That we do.

345
00:16:25,480 --> 00:16:27,760
Speaker 1: I mean, it's got no grasp of expleted in fixation, right,

346
00:16:27,800 --> 00:16:28,680
Speaker 1: this is one of chumps here.

347
00:16:28,800 --> 00:16:31,040
Speaker 2: What does that mean? Just for not for me? I

348
00:16:31,080 --> 00:16:31,680
Speaker 2: definitely know.

349
00:16:32,160 --> 00:16:35,000
Speaker 1: Yeah, yeah, of course if I give you the sentence

350
00:16:35,440 --> 00:16:37,840
Speaker 1: that's completely crazy, man, and tell you to put the

351
00:16:37,840 --> 00:16:40,840
Speaker 1: word fucking into that sentence, there's a number of ways

352
00:16:40,840 --> 00:16:42,800
Speaker 1: in which any language speaker is going to do it

353
00:16:43,080 --> 00:16:44,600
Speaker 1: that they'll just go, yeah, of course.

354
00:16:44,440 --> 00:16:46,400
Speaker 4: That where it goes, right.

355
00:16:46,800 --> 00:16:49,200
Speaker 1: But we it seems that we've got a grasp on

356
00:16:49,240 --> 00:16:52,240
Speaker 1: this incredibly early on in a way that doesn't look

357
00:16:52,320 --> 00:16:53,800
Speaker 1: like it's the ways at least most of us that

358
00:16:53,880 --> 00:16:56,360
Speaker 1: taught language, right, we get quite highly told off when

359
00:16:56,360 --> 00:16:56,760
Speaker 1: we try.

360
00:16:56,640 --> 00:16:57,680
Speaker 4: And do expleted in fixation.

361
00:16:57,920 --> 00:16:59,960
Speaker 1: Yes, So this I think would be one of those

362
00:17:00,160 --> 00:17:03,360
Speaker 1: cases where you could do a sort of disnalogy by cases. Right,

363
00:17:03,400 --> 00:17:06,119
Speaker 1: you present chat GPT a sentence and say insert the

364
00:17:06,119 --> 00:17:07,720
Speaker 1: word fucking correctly in this sentence.

365
00:17:08,440 --> 00:17:10,200
Speaker 4: And I don't think it would be very good at.

366
00:17:10,080 --> 00:17:10,879
Speaker 6: It I think it would be.

367
00:17:10,920 --> 00:17:11,560
Speaker 4: I think you reckon.

368
00:17:11,760 --> 00:17:15,280
Speaker 1: I mean, we probably couldn't tell we could, but we shouldn't.

369
00:17:16,160 --> 00:17:16,400
Speaker 4: Yeah.

370
00:17:16,840 --> 00:17:20,560
Speaker 5: One of the things that I thought was like kind

371
00:17:20,560 --> 00:17:24,800
Speaker 5: of interesting about how it works is that it does

372
00:17:24,880 --> 00:17:27,199
Speaker 5: learn language, probably differently from the way we do, but

373
00:17:27,280 --> 00:17:30,040
Speaker 5: it does it all by examples, you know, So it's

374
00:17:30,080 --> 00:17:32,160
Speaker 5: looking at all these pieces of text and thinking, ah,

375
00:17:32,200 --> 00:17:34,960
Speaker 5: this is okay, that's okay. And one kind of interesting

376
00:17:34,960 --> 00:17:38,000
Speaker 5: thing about how humans understand language is that we're able

377
00:17:38,040 --> 00:17:42,800
Speaker 5: to kind of understand meaningless but grammatical sentences. It's not

378
00:17:42,840 --> 00:17:46,120
Speaker 5: clear to me that chat GPT would understand those. That's

379
00:17:46,160 --> 00:17:50,240
Speaker 5: another Chopski example. So you know, Chomsky has this example

380
00:17:50,280 --> 00:17:57,919
Speaker 5: that's like colorless uss sleep furiously color it's colorless green ideas, right,

381
00:17:58,520 --> 00:18:02,040
Speaker 5: And that's a meaningless sentence, but it's grammatically well formed,

382
00:18:02,280 --> 00:18:05,119
Speaker 5: and we can understand that it's grammatically well formed, but

383
00:18:05,359 --> 00:18:09,760
Speaker 5: also that it's meaningless. Because chat Gibt kind of combines

384
00:18:10,520 --> 00:18:16,159
Speaker 5: different aspects of what philosophers of language, logicians, linguistics people

385
00:18:17,320 --> 00:18:20,040
Speaker 5: see as like different components of meaning. It sees these

386
00:18:20,080 --> 00:18:21,320
Speaker 5: as all kind of wrapped up.

387
00:18:21,200 --> 00:18:23,000
Speaker 4: In the same thing. It puts them in the same.

388
00:18:22,840 --> 00:18:27,199
Speaker 5: Big model I'm not sure it could differentiate between ungrammaticality

389
00:18:27,240 --> 00:18:31,520
Speaker 5: and meaninglessness.

390
00:18:34,080 --> 00:18:36,040
Speaker 2: So we're doing real time science right now.

391
00:18:36,080 --> 00:18:38,640
Speaker 4: I just yeah, we should check the words get.

392
00:18:38,520 --> 00:18:41,680
Speaker 2: Into the following sentence in the correct way. Man, that's crazy.

393
00:18:42,000 --> 00:18:45,480
Speaker 2: And I did it six times, and I would say

394
00:18:45,520 --> 00:18:47,400
Speaker 2: fifty percent of the time it got it right. And

395
00:18:47,840 --> 00:18:50,680
Speaker 2: it did. Man, that's fucking crazy. Man, that's crazy. Fucking Man,

396
00:18:50,720 --> 00:18:54,960
Speaker 2: that's fucking crazy. Man that's crazy. Fucking My favorite is man,

397
00:18:55,080 --> 00:18:57,280
Speaker 2: comma that's crazy, Comma fucking.

398
00:18:58,680 --> 00:18:59,320
Speaker 6: To be fair.

399
00:19:01,040 --> 00:19:01,480
Speaker 4: Reliable.

400
00:19:01,640 --> 00:19:03,479
Speaker 1: You take the commas out of that last one, and

401
00:19:03,560 --> 00:19:04,920
Speaker 1: you've got a grammatical sentence.

402
00:19:05,080 --> 00:19:05,359
Speaker 2: Yeah.

403
00:19:05,400 --> 00:19:08,280
Speaker 1: Of course in Glasgow you can also start with fucking man,

404
00:19:08,320 --> 00:19:09,679
Speaker 1: that's crazy.

405
00:19:09,280 --> 00:19:11,560
Speaker 2: West London as well. But the thing is, though it

406
00:19:11,600 --> 00:19:13,160
Speaker 2: doesn't know what correct means there.

407
00:19:13,400 --> 00:19:13,600
Speaker 6: Yeah.

408
00:19:13,680 --> 00:19:17,119
Speaker 2: Now, when it's trained on this language, when it's trained

409
00:19:17,119 --> 00:19:21,719
Speaker 2: on thousands of Internet posts that stole, it's not like

410
00:19:21,800 --> 00:19:24,679
Speaker 2: it reads them and says, oh, I get this, like

411
00:19:24,720 --> 00:19:26,919
Speaker 2: I see what they're going for. It just learns structures

412
00:19:26,960 --> 00:19:31,120
Speaker 2: by looking, which is kind of how we learn language.

413
00:19:31,640 --> 00:19:33,240
Speaker 2: But it kind of reminds me of like when I

414
00:19:33,280 --> 00:19:35,119
Speaker 2: was a kid and I'd hear someone say something funny

415
00:19:35,119 --> 00:19:37,600
Speaker 2: I'd to repeat it, and my dad, who's wonderful, would

416
00:19:37,600 --> 00:19:39,639
Speaker 2: just say that doesn't make any sense, and you'd have

417
00:19:39,720 --> 00:19:42,840
Speaker 2: to explain, because if you're learning everything through copying, you're

418
00:19:42,880 --> 00:19:44,640
Speaker 2: not learning, you're just memorizing.

419
00:19:45,520 --> 00:19:46,960
Speaker 6: Yeah, yeah, exactly.

420
00:19:47,040 --> 00:19:49,639
Speaker 5: There's a I don't know if you have already talked

421
00:19:49,760 --> 00:19:52,600
Speaker 5: to somebody about this, but there's a you know, a

422
00:19:52,600 --> 00:19:55,440
Speaker 5: classic argument from child ski against behaviorism.

423
00:19:55,600 --> 00:19:55,919
Speaker 2: Right.

424
00:19:56,080 --> 00:20:01,080
Speaker 5: Behaviorism is the view that we learn everything through listen response.

425
00:20:01,400 --> 00:20:03,080
Speaker 4: Roughly, that's not exactly.

426
00:20:02,680 --> 00:20:04,240
Speaker 5: It, but I'm not a philosopher of mine, so I

427
00:20:04,240 --> 00:20:06,560
Speaker 5: can get away with with that.

428
00:20:07,200 --> 00:20:10,040
Speaker 4: So Chomsky says, look, we don't get enough.

429
00:20:09,840 --> 00:20:14,359
Speaker 5: Stimulus to learn language as quickly as we do just

430
00:20:14,520 --> 00:20:15,639
Speaker 5: through watching.

431
00:20:15,280 --> 00:20:17,199
Speaker 4: Other people's behavior and copying it.

432
00:20:17,640 --> 00:20:21,879
Speaker 5: We have to have some in built grammatical structures that

433
00:20:22,000 --> 00:20:25,840
Speaker 5: language is latching onto. And there's been some papers arguing

434
00:20:25,840 --> 00:20:29,240
Speaker 5: that chatchipt shows Chomsky was wrong because it doesn't have

435
00:20:29,800 --> 00:20:31,320
Speaker 5: the inbuilt grammatical structure.

436
00:20:31,440 --> 00:20:32,840
Speaker 4: But one interesting thing.

437
00:20:32,720 --> 00:20:35,760
Speaker 5: Is it requires ten to one hundred times more data

438
00:20:35,800 --> 00:20:38,520
Speaker 5: than a human child does when learning language. Right, So

439
00:20:38,600 --> 00:20:42,720
Speaker 5: Chomsky's argument was we don't get enough stimulus, and chat

440
00:20:42,800 --> 00:20:46,760
Speaker 5: GPT can kind of do it without the structure, but

441
00:20:46,880 --> 00:20:50,200
Speaker 5: it's not quite doing it as well, and it gets

442
00:20:50,560 --> 00:20:55,159
Speaker 5: like orders of magnitude more more input than a human

443
00:20:55,240 --> 00:20:58,840
Speaker 5: does before a human learns language, which is kind of interesting.

444
00:21:00,000 --> 00:21:02,159
Speaker 4: You're going to do something as basically as putting the word.

445
00:21:04,080 --> 00:21:04,520
Speaker 6: Yeah, yeah.

446
00:21:04,560 --> 00:21:06,679
Speaker 2: It doesn't even see and it doesn't have the knowledge

447
00:21:06,720 --> 00:21:10,719
Speaker 2: to say, request more context because it doesn't perceive context.

448
00:21:10,720 --> 00:21:12,200
Speaker 2: And that's kind of the interesting thing. So there was

449
00:21:12,200 --> 00:21:15,160
Speaker 2: another paper out of Oxver I think that's talking about

450
00:21:15,200 --> 00:21:19,280
Speaker 2: cognition and chat GPT and all this thing, and it's

451
00:21:19,359 --> 00:21:22,280
Speaker 2: just it doesn't feat chat GPT features in no way

452
00:21:22,359 --> 00:21:24,400
Speaker 2: any of the things that the human mind is really

453
00:21:24,400 --> 00:21:27,760
Speaker 2: involved in. It seems it's mostly just not even memorization,

454
00:21:27,800 --> 00:21:33,359
Speaker 2: because it doesn't memorize. It's just guessing based on a

455
00:21:33,480 --> 00:21:36,120
Speaker 2: very blood pile of stuff. But this actually does lead

456
00:21:36,119 --> 00:21:38,040
Speaker 2: me to marvel question, which is, you don't like the

457
00:21:38,119 --> 00:21:41,680
Speaker 2: term hallucination. Why is that hallucination?

458
00:21:42,320 --> 00:21:45,679
Speaker 3: It makes it sound a bit like I'm usually doing

459
00:21:45,720 --> 00:21:49,400
Speaker 3: something right and I'm looking around seeing the world that's

460
00:21:49,480 --> 00:21:53,160
Speaker 3: something like what it really is. And then one little

461
00:21:53,200 --> 00:21:56,800
Speaker 3: bit of the feature for a visual hallucination, one feature

462
00:21:56,840 --> 00:22:00,680
Speaker 3: of my visual field actually isn't represented in the real world,

463
00:22:00,680 --> 00:22:01,680
Speaker 3: it's not actually there.

464
00:22:02,119 --> 00:22:03,919
Speaker 4: Everything else might well might well be right.

465
00:22:03,960 --> 00:22:07,119
Speaker 3: Imagine I hallucinate, there's a red balloon in front of me.

466
00:22:07,600 --> 00:22:09,800
Speaker 3: I still see Mike, I still see James, I still

467
00:22:09,800 --> 00:22:10,560
Speaker 3: see the laptop.

468
00:22:11,280 --> 00:22:12,000
Speaker 4: One bit is.

469
00:22:11,920 --> 00:22:14,920
Speaker 3: Wrong, everything else is right, and I'm doing the same

470
00:22:15,000 --> 00:22:17,359
Speaker 3: thing that I'm usually doing, Like my eyes are still

471
00:22:17,400 --> 00:22:21,200
Speaker 3: working pretty much normally, I think, I and this.

472
00:22:21,240 --> 00:22:23,480
Speaker 4: Is the way I usually get knowledge about the world.

473
00:22:23,800 --> 00:22:28,560
Speaker 3: This is a pretty reliable process for me learning from

474
00:22:28,560 --> 00:22:31,560
Speaker 3: it right and representing the world in this way. So

475
00:22:31,760 --> 00:22:36,960
Speaker 3: when talking about hallucinations, this suggests that Chatgypt and other

476
00:22:37,040 --> 00:22:41,720
Speaker 3: similar things they're going through this process that is usually

477
00:22:41,840 --> 00:22:45,160
Speaker 3: quite good at representing the world. And then oh, it's

478
00:22:45,160 --> 00:22:48,680
Speaker 3: made a mistake this one time, but actually no, it's

479
00:22:48,680 --> 00:22:52,440
Speaker 3: bullshitting the whole time, and like sometimes it gets things

480
00:22:52,480 --> 00:22:55,040
Speaker 3: right right, bullshitting just like a politician.

481
00:22:55,359 --> 00:22:59,440
Speaker 4: Imagine a politician that bullshits all the time, if you.

482
00:22:59,359 --> 00:23:02,520
Speaker 3: Could possibly imagine it, Sometimes they might just get some

483
00:23:02,600 --> 00:23:07,280
Speaker 3: things true, and we should still call them bullshitting because

484
00:23:07,280 --> 00:23:09,960
Speaker 3: that's what they're doing. And this is what at GPT

485
00:23:10,160 --> 00:23:13,879
Speaker 3: is doing every time it produces an output. So this

486
00:23:13,960 --> 00:23:15,879
Speaker 3: is why we think bullshit is a better way of

487
00:23:15,880 --> 00:23:18,560
Speaker 3: thinking about this. One of the reasons why we think

488
00:23:18,560 --> 00:23:19,879
Speaker 3: bullshit's better way think about it.

489
00:23:20,320 --> 00:23:22,960
Speaker 5: I also kind of think that some of the ways

490
00:23:22,960 --> 00:23:27,320
Speaker 5: we talk about chat gpt LIN, even when it makes mistakes,

491
00:23:27,440 --> 00:23:32,720
Speaker 5: lind themselves to overhyping its ability or overestimating its abilities,

492
00:23:32,800 --> 00:23:34,680
Speaker 5: and talking about it is hallucinating. Is one of these

493
00:23:34,760 --> 00:23:39,120
Speaker 5: because when you say that it's hallucinating, as you're pointed out,

494
00:23:39,119 --> 00:23:43,000
Speaker 5: you're giving the idea that it's representing the world in

495
00:23:43,040 --> 00:23:45,120
Speaker 5: some way and then telling you what the.

496
00:23:45,119 --> 00:23:46,880
Speaker 2: Content and it has perception.

497
00:23:46,720 --> 00:23:49,879
Speaker 1: Yeah, exactly, like it has perceived something and like, oh no,

498
00:23:50,040 --> 00:23:53,199
Speaker 1: it's taken some computer acid and now it's hallucinating like

499
00:23:53,400 --> 00:23:56,480
Speaker 1: mastery things and yeah, just as you say, and.

500
00:23:56,400 --> 00:23:57,440
Speaker 4: That's not what it's doing.

501
00:23:57,520 --> 00:24:01,879
Speaker 5: And so when the kind of people who are trying

502
00:24:01,880 --> 00:24:05,760
Speaker 5: to promote these these things as products talk about the

503
00:24:05,800 --> 00:24:09,720
Speaker 5: AI hallucination problem, they're kind of selling a product that

504
00:24:10,119 --> 00:24:14,600
Speaker 5: is a product that's representing the world usually checking things

505
00:24:15,160 --> 00:24:19,560
Speaker 5: and occasionally makes mistakes. And if the mistakes were, like

506
00:24:19,640 --> 00:24:23,280
Speaker 5: Joe said, we're a misfiring of a normally a reliable

507
00:24:23,359 --> 00:24:28,600
Speaker 5: process or you know, something that normally represents going wrong

508
00:24:28,680 --> 00:24:31,919
Speaker 5: in some way that would lend itself to certain solutions

509
00:24:31,960 --> 00:24:33,320
Speaker 5: to them, and it would make you think there's an

510
00:24:33,359 --> 00:24:37,439
Speaker 5: underlying reliable product here, right, which is exactly what somebody

511
00:24:37,440 --> 00:24:40,600
Speaker 5: who's making a product to go on the market will.

512
00:24:40,480 --> 00:24:41,160
Speaker 4: Want you to think.

513
00:24:41,240 --> 00:24:41,440
Speaker 3: Right.

514
00:24:41,920 --> 00:24:44,240
Speaker 5: But if that's not what it's doing in a certain sense,

515
00:24:44,240 --> 00:24:47,920
Speaker 5: they're misrepresenting what it's doing even when it gets things right.

516
00:24:48,680 --> 00:24:51,840
Speaker 4: And that's that's bad for all of us who are

517
00:24:51,880 --> 00:24:52,880
Speaker 4: going to be using these.

518
00:24:52,800 --> 00:24:55,960
Speaker 5: Systems, especially since people, you know, most people don't know

519
00:24:55,960 --> 00:24:59,440
Speaker 5: how this works. They're just understanding the product as it's

520
00:24:59,480 --> 00:25:02,320
Speaker 5: described and using these kind of metaphors. So the way

521
00:25:02,359 --> 00:25:05,960
Speaker 5: the metaphor describes it's going to really influence how they

522
00:25:05,960 --> 00:25:07,400
Speaker 5: think about it and how they use it.

523
00:25:08,640 --> 00:25:10,480
Speaker 6: Yeah, just to sort of cat fire if I can.

524
00:25:10,640 --> 00:25:13,639
Speaker 1: Is one of the responses to some corners has been

525
00:25:13,680 --> 00:25:16,040
Speaker 1: to sort of say, of us, look, you winch about

526
00:25:16,080 --> 00:25:19,560
Speaker 1: people anthropomorphosizing chat GPT, but look, if you've got a bullshity,

527
00:25:19,560 --> 00:25:22,199
Speaker 1: you're doing exactly the same thing. And I mean there

528
00:25:22,280 --> 00:25:24,119
Speaker 1: might be some center which it's just really hard not

529
00:25:24,320 --> 00:25:26,040
Speaker 1: to anthropomorphise it. I don't know why I picked the

530
00:25:26,080 --> 00:25:29,239
Speaker 1: word that I can barely say, like we've been doing

531
00:25:29,280 --> 00:25:31,480
Speaker 1: it constantly throughout this discussion. Right when we were talking

532
00:25:31,680 --> 00:25:33,320
Speaker 1: through the kind of through the paper, we kept talking

533
00:25:33,320 --> 00:25:35,800
Speaker 1: about chat GPT as if it had intention, right, as

534
00:25:35,840 --> 00:25:38,600
Speaker 1: if it was thinking about anything. That might be another

535
00:25:38,600 --> 00:25:40,199
Speaker 1: reason to call it bullshit. We go, look, if we

536
00:25:40,280 --> 00:25:42,520
Speaker 1: have to treat it as if it's doing something like

537
00:25:42,600 --> 00:25:45,600
Speaker 1: what we do, it's not hallucinating, it's not lying, it's

538
00:25:45,640 --> 00:25:48,399
Speaker 1: not confabulating, it's bullshitting, right, And if we have to

539
00:25:48,400 --> 00:25:50,320
Speaker 1: treat it as if it's behaving in some kind of

540
00:25:50,400 --> 00:25:53,320
Speaker 1: human like way, here's the appropriate human like behavior to

541
00:25:53,359 --> 00:25:53,840
Speaker 1: describe it.

542
00:25:54,040 --> 00:25:56,199
Speaker 2: I also think the language in this case, and one

543
00:25:56,240 --> 00:25:58,320
Speaker 2: of the reasons they probably really like the large language

544
00:25:58,359 --> 00:26:02,320
Speaker 2: model concept is language gives life too things when we

545
00:26:02,400 --> 00:26:04,960
Speaker 2: describe the processes through which we interact with the world

546
00:26:04,960 --> 00:26:08,000
Speaker 2: and interact with living beings, even cats, dogs, even we

547
00:26:08,040 --> 00:26:13,320
Speaker 2: anthropomiz living things. But also when we communicate with something,

548
00:26:13,440 --> 00:26:16,720
Speaker 2: languages life, and so it probably works out really fucking

549
00:26:16,760 --> 00:26:19,640
Speaker 2: well for them. Samultman was saying a couple of weeks back,

550
00:26:19,640 --> 00:26:21,520
Speaker 2: maybe a month or two. He was saying, oh, yeah,

551
00:26:21,600 --> 00:26:24,840
Speaker 2: AI is not a creature with something he said, and

552
00:26:24,920 --> 00:26:26,960
Speaker 2: it was just so obvious what he wanted people to

553
00:26:26,960 --> 00:26:29,320
Speaker 2: do was say, but what if it was or are

554
00:26:29,359 --> 00:26:32,600
Speaker 2: people saying this is a creature and it almost feels

555
00:26:32,640 --> 00:26:35,840
Speaker 2: like just part of the con Yeah, yeah, yeah.

556
00:26:36,960 --> 00:26:39,040
Speaker 5: I hadn't thought about that as a reason for them

557
00:26:39,080 --> 00:26:42,000
Speaker 5: to go for large language models as a way of

558
00:26:42,080 --> 00:26:48,480
Speaker 5: kind of, I don't know, being the gateway into more investment.

559
00:26:48,040 --> 00:26:49,640
Speaker 2: Into fight consciousness.

560
00:26:50,440 --> 00:26:50,760
Speaker 6: Yeah.

561
00:26:50,880 --> 00:26:54,320
Speaker 5: Yeah, but I had thought about, like how this might

562
00:26:54,320 --> 00:26:57,920
Speaker 5: have been caused by just like deep misunderstandings of returning test.

563
00:26:58,920 --> 00:27:00,840
Speaker 2: Right, go ahead, hear this one.

564
00:27:01,240 --> 00:27:04,360
Speaker 5: Yeah, yeah, so that like the Turing test, I think

565
00:27:04,359 --> 00:27:06,960
Speaker 5: this is closer to what Turing was thinking. But the

566
00:27:07,000 --> 00:27:09,919
Speaker 5: turning test is a way of getting evidence that something

567
00:27:09,960 --> 00:27:13,800
Speaker 5: is conscious. Right, So you know, I'm not in your head,

568
00:27:14,480 --> 00:27:18,840
Speaker 5: so I can't feel your feelings or think your thoughts directly. Right,

569
00:27:19,359 --> 00:27:22,520
Speaker 5: I have to judge whether you're conscious based on how

570
00:27:22,520 --> 00:27:24,439
Speaker 5: you interact with me, and the way I do it

571
00:27:24,480 --> 00:27:28,359
Speaker 5: is by listening to what you say, right, and talking

572
00:27:28,359 --> 00:27:31,160
Speaker 5: to you and turning sort.

573
00:27:30,920 --> 00:27:34,360
Speaker 4: Of was asked, you know, how would you know if

574
00:27:34,359 --> 00:27:35,440
Speaker 4: a computer was conscious?

575
00:27:35,480 --> 00:27:38,080
Speaker 5: So, you know, we think that our brains are doing

576
00:27:38,160 --> 00:27:40,840
Speaker 5: something similar to what computers do. That's a reason to

577
00:27:40,880 --> 00:27:44,320
Speaker 5: think that maybe computers eventually could have thoughts.

578
00:27:44,000 --> 00:27:47,200
Speaker 4: Like ours, right, and some of us I think that

579
00:27:47,440 --> 00:27:49,320
Speaker 4: I think it's possible. It's great.

580
00:27:49,440 --> 00:27:51,399
Speaker 5: Yeah, I didn't know if they thought it was possible,

581
00:27:51,400 --> 00:27:52,720
Speaker 5: because not everybody thinks it's possible.

582
00:27:52,760 --> 00:27:54,400
Speaker 2: It's possible. I just don't know how.

583
00:27:54,520 --> 00:27:58,439
Speaker 4: Yeah, yeah, exactly. So Turing was kind of, you know,

584
00:27:58,480 --> 00:28:01,440
Speaker 4: thinking like, how would we know? And one way we

585
00:28:01,440 --> 00:28:01,760
Speaker 4: would know.

586
00:28:01,800 --> 00:28:04,320
Speaker 5: The obvious way is to do the same thing you

587
00:28:04,359 --> 00:28:07,159
Speaker 5: do to humans, talk to it and see how it responds.

588
00:28:07,760 --> 00:28:11,480
Speaker 5: And that's actually pretty good evidence if you don't have

589
00:28:11,560 --> 00:28:16,360
Speaker 5: the ability to look deeper. But it's not constitutive of

590
00:28:16,359 --> 00:28:19,880
Speaker 5: being conscious. It's not what makes something conscious or determines

591
00:28:19,920 --> 00:28:24,760
Speaker 5: whether they're conscious or in any way like grounds their consciousness. Right,

592
00:28:24,840 --> 00:28:27,520
Speaker 5: Their ability to talk to you is just evidence. It's

593
00:28:27,600 --> 00:28:30,720
Speaker 5: just one signal you can get. And that's the way

594
00:28:30,760 --> 00:28:33,880
Speaker 5: to think of the Turing test. So as a result

595
00:28:33,920 --> 00:28:37,159
Speaker 5: of people thinking in a kind of behaviorist way, thinking, ah,

596
00:28:37,200 --> 00:28:39,720
Speaker 5: passing the Turning test is just all it is to

597
00:28:39,760 --> 00:28:43,000
Speaker 5: be a thinking thing. There have been, at least since

598
00:28:43,040 --> 00:28:46,120
Speaker 5: the nineties, attempts to design chatbucks that can be the

599
00:28:46,120 --> 00:28:50,160
Speaker 5: Turning test, right, and popularizations of these attempts and run

600
00:28:50,200 --> 00:28:52,680
Speaker 5: throughs of the Turing test that talk as if, oh,

601
00:28:52,720 --> 00:28:54,320
Speaker 5: if a computer finally beats the.

602
00:28:54,280 --> 00:28:56,760
Speaker 4: Turing test, I should say what the Turing test is?

603
00:28:56,840 --> 00:28:59,720
Speaker 5: Right? The way turns suggests that the test works as

604
00:28:59,760 --> 00:29:04,400
Speaker 5: you have a computer and a person both chatting in

605
00:29:04,440 --> 00:29:07,080
Speaker 5: some way with a judge, and the judge is also

606
00:29:07,080 --> 00:29:09,600
Speaker 5: a person, right, And if the judge can't tell which

607
00:29:09,600 --> 00:29:12,840
Speaker 5: of the people he's chatting with is a human, then

608
00:29:12,880 --> 00:29:15,880
Speaker 5: the computer's one the right test because it's distinguishable from

609
00:29:15,880 --> 00:29:16,200
Speaker 5: a human.

610
00:29:16,320 --> 00:29:16,440
Speaker 4: Right.

611
00:29:16,800 --> 00:29:19,840
Speaker 5: So people have taken this and it's been popularized as

612
00:29:19,840 --> 00:29:24,000
Speaker 5: a way of determining sort of full stop, whether something's conscious.

613
00:29:24,680 --> 00:29:26,560
Speaker 4: But it's just a piece of evidence, and we have

614
00:29:26,600 --> 00:29:27,040
Speaker 4: a lot.

615
00:29:26,920 --> 00:29:29,080
Speaker 5: More evidence, Like we know a lot more than Turning

616
00:29:29,120 --> 00:29:32,840
Speaker 5: did about how the internal functioning of a mind works functionally,

617
00:29:32,880 --> 00:29:38,200
Speaker 5: what it's doing, how representation works, how experience and belief works,

618
00:29:38,240 --> 00:29:40,640
Speaker 5: how they are connected to action, and how they're connected

619
00:29:40,680 --> 00:29:45,440
Speaker 5: to speech and thought. And once you know all that stuff,

620
00:29:45,480 --> 00:29:48,760
Speaker 5: you have a lot of other avenues to get more

621
00:29:48,800 --> 00:29:53,160
Speaker 5: evidence about whether the thing is conscious and whether it passes.

622
00:29:53,200 --> 00:29:55,320
Speaker 5: The Turning test is just like a drop in the

623
00:29:55,320 --> 00:29:57,200
Speaker 5: bucket compared to these, especially if you know how it's

624
00:29:57,240 --> 00:29:58,360
Speaker 5: internal functioning.

625
00:29:58,040 --> 00:30:00,200
Speaker 1: Is the other the story is problem with the showing test,

626
00:30:00,200 --> 00:30:02,840
Speaker 1: and I think to be fair, Cheuring did mention this.

627
00:30:02,960 --> 00:30:06,400
Speaker 1: If not in the original then kind of later on. One

628
00:30:05,880 --> 00:30:08,280
Speaker 1: problem with the chewing test is that it's like the

629
00:30:08,360 --> 00:30:11,200
Speaker 1: de void camp in de androids gene of electric cheap right.

630
00:30:11,800 --> 00:30:15,360
Speaker 1: Plenty of humans would fail the cheering test. Yeah, it

631
00:30:15,440 --> 00:30:17,560
Speaker 1: is a piece of evidence, but it was never as

632
00:30:17,640 --> 00:30:19,840
Speaker 1: Mike said, it was supposed to be constitutive. It wasn't like,

633
00:30:19,920 --> 00:30:22,720
Speaker 1: if you can do this thing, your conscious kind of

634
00:30:22,720 --> 00:30:24,800
Speaker 1: full stops. It was supposed to be here is a

635
00:30:24,840 --> 00:30:28,480
Speaker 1: thing that might indicate that being you're talking to is conscious.

636
00:30:28,640 --> 00:30:30,520
Speaker 4: Happily, as my said, we've got the loads and loads

637
00:30:30,520 --> 00:30:31,160
Speaker 4: of other evidence.

638
00:30:31,280 --> 00:30:33,800
Speaker 5: So these guys have made a machine that's just designed

639
00:30:34,320 --> 00:30:37,560
Speaker 5: to do one thing, and that's past the Turing test.

640
00:30:37,840 --> 00:30:39,920
Speaker 2: I can give you one more annoying example. Are you

641
00:30:39,960 --> 00:30:43,920
Speaker 2: familiar with Francoise Chole the abstraction and reasoning corpus, So

642
00:30:43,960 --> 00:30:46,000
Speaker 2: this is going to make you laugh. So he's a

643
00:30:46,000 --> 00:30:48,520
Speaker 2: Google engineer and he created this thing of the arc

644
00:30:48,640 --> 00:30:52,400
Speaker 2: the abstraction reasoning corpus, the test whether a machine was intelligent,

645
00:30:53,040 --> 00:30:55,240
Speaker 2: and someone created a model that could be it, and

646
00:30:55,240 --> 00:30:58,920
Speaker 2: then he immediately went, Okay, you can't just train the

647
00:30:59,040 --> 00:31:01,560
Speaker 2: model on the answers to the test.

648
00:31:02,040 --> 00:31:04,880
Speaker 1: This is why people, well I see people a fairly

649
00:31:04,920 --> 00:31:08,600
Speaker 1: small subset of weird nerds, But this is why a

650
00:31:08,640 --> 00:31:10,520
Speaker 1: small subset of weird nerds have been for the last

651
00:31:10,520 --> 00:31:14,880
Speaker 1: twenty years emphasizing artificial general intelligence, right, and kind of

652
00:31:15,000 --> 00:31:16,920
Speaker 1: like what we'll call it. When something really is a

653
00:31:16,960 --> 00:31:19,960
Speaker 1: thinking being is when it's not specialized to do one

654
00:31:20,000 --> 00:31:23,240
Speaker 1: and only one task, but rather when it's capable of

655
00:31:23,280 --> 00:31:27,200
Speaker 1: applying reasoning to multiple kinds of different and this analogous cases.

656
00:31:27,320 --> 00:31:29,880
Speaker 1: On the one hand, it does seem a little bit

657
00:31:30,000 --> 00:31:31,520
Speaker 1: like the guy flipping the table and go, oh, for

658
00:31:31,560 --> 00:31:33,480
Speaker 1: fox sake, now you've one. I'm changing the rules. But

659
00:31:33,480 --> 00:31:35,000
Speaker 1: on the other I think he's got a point, right, Yeah,

660
00:31:35,200 --> 00:31:39,160
Speaker 1: if you're training a thing to do very specific you know,

661
00:31:39,280 --> 00:31:42,520
Speaker 1: like activate certain shiboleths, then unless you're some kind of

662
00:31:42,560 --> 00:31:44,880
Speaker 1: mad hard behaviorist, then yeah, like it's not that doesn't

663
00:31:44,920 --> 00:31:48,080
Speaker 1: demonstrate intelligence. It is one thing that might indicate intelligence.

664
00:31:48,120 --> 00:31:50,480
Speaker 2: It's the same problem with the chat GPT. It's built

665
00:31:50,520 --> 00:31:54,680
Speaker 2: to resemble intelligence and resemble conscious This will resemble these things,

666
00:31:54,680 --> 00:31:58,840
Speaker 2: but it isn't. It's almost like it's meaningless. On a

667
00:31:58,920 --> 00:32:02,280
Speaker 2: very high end philosophical I find the whole generative a

668
00:32:02,440 --> 00:32:03,719
Speaker 2: I think deeply nihilistic.

669
00:32:03,880 --> 00:32:06,479
Speaker 5: I mean, one thing that connects to this is how

670
00:32:06,520 --> 00:32:09,120
Speaker 5: bad it is at reasoning. And this is kind of

671
00:32:09,520 --> 00:32:13,880
Speaker 5: good for us, especially in philosophy, because our students when

672
00:32:13,880 --> 00:32:17,080
Speaker 5: they use it to write papers, the papers have to

673
00:32:17,120 --> 00:32:21,600
Speaker 5: have arguments, and chat GPT is very bad at doing reasoning.

674
00:32:22,000 --> 00:32:24,120
Speaker 5: If it has to be you know, sort of extended

675
00:32:24,200 --> 00:32:27,440
Speaker 5: argument or a proof or something like that, it's very

676
00:32:27,480 --> 00:32:30,440
Speaker 5: bad at it. I think also that if there's one

677
00:32:30,440 --> 00:32:34,200
Speaker 5: thing kind of we learned from chat GPT, it's that

678
00:32:34,560 --> 00:32:38,160
Speaker 5: this is not the way to get to artificial general intelligence.

679
00:32:38,480 --> 00:32:39,840
Speaker 2: I was going to ask, do you think that this

680
00:32:39,960 --> 00:32:40,640
Speaker 2: is getting to that?

681
00:32:41,360 --> 00:32:45,960
Speaker 5: No, partially because it's so subject specific, right, It's one

682
00:32:46,080 --> 00:32:48,880
Speaker 5: it's trained to do one task, and it takes quite

683
00:32:48,920 --> 00:32:51,120
Speaker 5: a lot of training to get it to do that task. Well,

684
00:32:51,880 --> 00:32:55,000
Speaker 5: it's bad at many of the other tasks that we

685
00:32:55,040 --> 00:32:57,400
Speaker 5: think are connected with intelligence. It's bad at logical and

686
00:32:57,440 --> 00:33:01,920
Speaker 5: mathematical reasoning. I understand that open AI is trying to

687
00:33:02,000 --> 00:33:04,560
Speaker 5: fix that. Some sometimes it sounds like they want to

688
00:33:04,560 --> 00:33:08,240
Speaker 5: fix it by just connecting it to a database or

689
00:33:08,280 --> 00:33:10,920
Speaker 5: a program that can do that for it. But either way,

690
00:33:12,000 --> 00:33:14,880
Speaker 5: when you have with these kind of big base nets models,

691
00:33:14,920 --> 00:33:19,800
Speaker 5: it's something that is really good at whatever you train

692
00:33:19,880 --> 00:33:22,440
Speaker 5: it to do, but not going to be good at

693
00:33:22,440 --> 00:33:25,080
Speaker 5: anything else. You know, it's got a lot of data

694
00:33:25,120 --> 00:33:28,840
Speaker 5: on one thing. It finds patterns in that, it finds

695
00:33:28,880 --> 00:33:32,120
Speaker 5: regularities in that, it represents those. The more data you

696
00:33:32,200 --> 00:33:34,160
Speaker 5: feed it, the better it will be at that. But

697
00:33:34,400 --> 00:33:37,360
Speaker 5: it's not going to have this kind of general ability,

698
00:33:38,000 --> 00:33:39,680
Speaker 5: and it's not going to grow it out of learning

699
00:33:39,680 --> 00:33:41,120
Speaker 5: how to speak English.

700
00:33:41,560 --> 00:33:44,600
Speaker 4: Have you heard the Terry Pratchett quote about It's quite

701
00:33:44,600 --> 00:33:45,000
Speaker 4: early on.

702
00:33:45,240 --> 00:33:47,560
Speaker 1: He's talking about Hex, the kind of steampunk computer a

703
00:33:47,680 --> 00:33:50,000
Speaker 1: make and unseen university, and it just has this off

704
00:33:50,040 --> 00:33:53,760
Speaker 1: hand line. A computer program is exactly like a philosophy professor.

705
00:33:54,040 --> 00:33:56,760
Speaker 1: Unless you ask it the question in exactly the right way,

706
00:33:57,000 --> 00:33:59,520
Speaker 1: it will delight in giving you a perfectly accurate, completely

707
00:33:59,600 --> 00:34:00,440
Speaker 1: unhelped answer.

708
00:34:00,800 --> 00:34:02,800
Speaker 6: Right. So if you abstract that justified you got a

709
00:34:02,800 --> 00:34:03,520
Speaker 6: philosophy lecture.

710
00:34:03,600 --> 00:34:03,880
Speaker 4: Is there?

711
00:34:04,040 --> 00:34:08,360
Speaker 1: Basically what intelligent things do is go You can't have

712
00:34:08,440 --> 00:34:12,560
Speaker 1: meant that, you must have meant this right, chat GPTs goes,

713
00:34:13,239 --> 00:34:15,239
Speaker 1: I will take a question as read. I mean, of

714
00:34:15,239 --> 00:34:16,880
Speaker 1: course it doesn't have an eye. There is nothing in

715
00:34:16,880 --> 00:34:18,880
Speaker 1: the rest. Like I've just answered, formal play it again.

716
00:34:18,960 --> 00:34:21,080
Speaker 1: But it's the same thing. And if trained to do

717
00:34:21,400 --> 00:34:24,480
Speaker 1: attractively specific things, and you get the same problem as

718
00:34:24,520 --> 00:34:33,960
Speaker 1: any program like garbage in Garbage album.

719
00:34:34,000 --> 00:34:36,320
Speaker 2: So have you found a lot of students using chat GBT,

720
00:34:36,480 --> 00:34:38,920
Speaker 2: because this is a hard problem to quantify? Is it

721
00:34:39,080 --> 00:34:39,720
Speaker 2: all the time?

722
00:34:41,080 --> 00:34:42,359
Speaker 6: I mean it's it's a lot.

723
00:34:42,680 --> 00:34:43,719
Speaker 4: I wouldn't say all the time.

724
00:34:43,920 --> 00:34:45,040
Speaker 6: I changed that you.

725
00:34:45,080 --> 00:34:50,160
Speaker 3: Thought that there were more seas indeed, yeah, so I

726
00:34:50,200 --> 00:34:52,440
Speaker 3: think there are quite a few. And sometimes it is

727
00:34:53,080 --> 00:34:55,479
Speaker 3: difficult for us to prove and if we can't prove

728
00:34:55,520 --> 00:34:58,040
Speaker 3: it then at our university, then.

729
00:34:58,040 --> 00:35:00,399
Speaker 4: Well shucks, they kind of get away with it.

730
00:35:00,680 --> 00:35:04,200
Speaker 3: We can interview them, but unless we're like, unless we

731
00:35:04,239 --> 00:35:07,160
Speaker 3: have the proof that you could take before like a court,

732
00:35:07,239 --> 00:35:11,399
Speaker 3: then we are we're not able to really nail them

733
00:35:11,400 --> 00:35:14,239
Speaker 3: for it, which is a bit of a shame. So suspicion, Yeah,

734
00:35:14,320 --> 00:35:16,960
Speaker 3: we suspect that a lot of them are very I'm

735
00:35:17,000 --> 00:35:18,359
Speaker 3: one hundred percent on some of them.

736
00:35:18,400 --> 00:35:20,000
Speaker 2: I just know what are the clues I want to

737
00:35:20,040 --> 00:35:21,319
Speaker 2: hear from all of you on this one, like, what

738
00:35:21,400 --> 00:35:22,000
Speaker 2: are the side?

739
00:35:22,160 --> 00:35:23,400
Speaker 6: It's the uncanny value.

740
00:35:23,560 --> 00:35:25,040
Speaker 1: I will not shut up about this, right you know

741
00:35:25,080 --> 00:35:28,000
Speaker 1: that like you of course will be imagine a lot

742
00:35:28,000 --> 00:35:29,680
Speaker 1: of your kind of the reader will be as well.

743
00:35:29,680 --> 00:35:31,200
Speaker 4: But the uncanny Valley.

744
00:35:30,960 --> 00:35:33,600
Speaker 1: Thing in respective humans is that there's a kind of

745
00:35:33,640 --> 00:35:37,759
Speaker 1: a point up to which humans seem to trust artificial

746
00:35:37,800 --> 00:35:40,080
Speaker 1: things more the more they look like a human. So like,

747
00:35:40,120 --> 00:35:42,520
Speaker 1: we'll trust a robot if it's kind of bipedal or

748
00:35:42,560 --> 00:35:45,240
Speaker 1: if it looks like a dog, But after a point,

749
00:35:45,360 --> 00:35:47,920
Speaker 1: we really rapidly start to trust me. So like if

750
00:35:47,920 --> 00:35:51,399
Speaker 1: it basically when it starts looking like data, some deep

751
00:35:51,480 --> 00:35:53,240
Speaker 1: buried lizard brain goes danger.

752
00:35:53,480 --> 00:35:55,600
Speaker 6: Danger. This thing's trying to fuck with you somehow.

753
00:35:56,320 --> 00:35:57,240
Speaker 4: Yeah, right, and.

754
00:35:57,200 --> 00:35:59,640
Speaker 6: Chatter GPT writes exactly like that.

755
00:36:00,080 --> 00:36:03,360
Speaker 1: It writes almost, but not entirely, like a thing that

756
00:36:03,440 --> 00:36:04,879
Speaker 1: is trying to convince you that it's human.

757
00:36:04,920 --> 00:36:06,840
Speaker 4: I mean, the dead giveway for me is that it never.

758
00:36:06,719 --> 00:36:10,000
Speaker 1: Makes any spelling mistakes, but it can't form out a

759
00:36:10,040 --> 00:36:13,200
Speaker 1: paragraph to save its life. Normally, you would expect someone

760
00:36:13,239 --> 00:36:15,560
Speaker 1: who didn't know how to kind of order their sentences

761
00:36:15,719 --> 00:36:19,760
Speaker 1: to misspell the occasional word chat GPT spells everything correctly

762
00:36:20,120 --> 00:36:23,000
Speaker 1: and doesn't know what like subject object agreement is.

763
00:36:23,160 --> 00:36:24,120
Speaker 6: It's like, it's.

764
00:36:24,719 --> 00:36:28,480
Speaker 2: So, can you just define that, please? Subject object the.

765
00:36:29,320 --> 00:36:31,720
Speaker 6: Beer that I drink, right, not the drink that I bear.

766
00:36:32,200 --> 00:36:35,000
Speaker 2: Right, Because it's interesting. It almost feels like there is

767
00:36:35,080 --> 00:36:39,319
Speaker 2: just an entire way of processing and delivering information as

768
00:36:39,320 --> 00:36:42,279
Speaker 2: a human being that we do not understand. There is

769
00:36:42,320 --> 00:36:43,120
Speaker 2: something missing.

770
00:36:43,719 --> 00:36:46,040
Speaker 5: I mean, for me, like it's very good at summarizing,

771
00:36:47,080 --> 00:36:50,720
Speaker 5: but when it responds, it responds in them in really

772
00:36:50,800 --> 00:36:53,720
Speaker 5: trite and repetitive ways, or like you'll get a paper

773
00:36:53,760 --> 00:36:57,600
Speaker 5: that summarizes a bunch of literature at length very effectively,

774
00:36:58,440 --> 00:37:03,160
Speaker 5: and then respond by saying, well, you know, this person

775
00:37:03,239 --> 00:37:06,160
Speaker 5: said this, that is doubtful. They should say more to

776
00:37:06,160 --> 00:37:09,120
Speaker 5: support that, you know, which is basically saying nothing right,

777
00:37:09,320 --> 00:37:11,000
Speaker 5: And that's pretty common.

778
00:37:11,960 --> 00:37:14,480
Speaker 4: It also does lists in formats things and kind of

779
00:37:14,480 --> 00:37:15,000
Speaker 4: like a list.

780
00:37:15,080 --> 00:37:19,080
Speaker 5: And even if this like make it look less like

781
00:37:19,120 --> 00:37:21,799
Speaker 5: a list, the paper still reads like a list.

782
00:37:22,080 --> 00:37:24,720
Speaker 2: Oh, because someone has asked give me a few thoughts

783
00:37:24,719 --> 00:37:30,200
Speaker 2: on X. Yeah, yeah, that's very depressing.

784
00:37:30,880 --> 00:37:35,640
Speaker 3: Things like the introductions you get you'll get this is

785
00:37:35,719 --> 00:37:39,799
Speaker 3: an interesting and complex question that parst of us have

786
00:37:39,840 --> 00:37:41,520
Speaker 3: asked you the ages, and this is one of the

787
00:37:41,560 --> 00:37:44,919
Speaker 3: things we shouted students not to write from first year.

788
00:37:45,320 --> 00:37:47,720
Speaker 4: And then you'll get this garbage right back at.

789
00:37:47,560 --> 00:37:50,440
Speaker 3: You, and then at the end it will be oh overall,

790
00:37:50,480 --> 00:37:54,480
Speaker 3: this is a complicated and difficult question with many nuancewers.

791
00:37:54,640 --> 00:37:55,919
Speaker 2: It's a world of contrasts.

792
00:37:56,320 --> 00:37:59,000
Speaker 3: One of the things we tell them, don't ever fucking

793
00:37:59,080 --> 00:38:02,520
Speaker 3: do this. This is terrible and right back at you

794
00:38:02,880 --> 00:38:06,200
Speaker 3: in perfect English. That's one of English you'd expects a

795
00:38:06,280 --> 00:38:08,959
Speaker 3: really good student might have written. But clearly they can't

796
00:38:09,000 --> 00:38:11,000
Speaker 3: be a good student because otherwise they've listened to a

797
00:38:11,080 --> 00:38:11,880
Speaker 3: fucking instruction.

798
00:38:13,640 --> 00:38:16,000
Speaker 4: I also, I had some papers that I suspected were

799
00:38:16,040 --> 00:38:19,480
Speaker 4: chatept this year, but they were already failing, so I

800
00:38:19,520 --> 00:38:22,799
Speaker 4: didn't okay, yeah, I didn't think it was worth it

801
00:38:23,560 --> 00:38:29,400
Speaker 4: to pursue them, you know, as a plagiarism or Chatchepte case.

802
00:38:29,640 --> 00:38:32,080
Speaker 2: So it's never good papers. Then it's never like an eight,

803
00:38:32,280 --> 00:38:33,360
Speaker 2: like a first.

804
00:38:32,960 --> 00:38:33,920
Speaker 6: No, absolutely not.

805
00:38:34,520 --> 00:38:36,399
Speaker 4: It's so I think that part of what goes on

806
00:38:36,880 --> 00:38:40,480
Speaker 4: is you can get a passing grade with the Chatcheapte paper.

807
00:38:40,880 --> 00:38:44,360
Speaker 4: Sometimes in the first couple of years, when the papers

808
00:38:44,400 --> 00:38:47,960
Speaker 4: are shorter and we're not expecting as much. But then

809
00:38:48,000 --> 00:38:51,239
Speaker 4: when you move into the what we call honors level here,

810
00:38:51,280 --> 00:38:53,480
Speaker 4: which is like upper level classes in the US, like

811
00:38:53,560 --> 00:38:54,520
Speaker 4: third and fourth.

812
00:38:54,320 --> 00:38:56,680
Speaker 2: Year, I've got first ever wrist with I know.

813
00:38:57,480 --> 00:39:00,360
Speaker 4: Yeah, yeah, exactly, great, well done.

814
00:38:59,719 --> 00:39:02,280
Speaker 7: Well you would not have gotten it with chat Gipt

815
00:39:02,520 --> 00:39:05,160
Speaker 7: because you get dropped in these classes where we expect

816
00:39:05,200 --> 00:39:08,920
Speaker 7: you to have gained writing skills minimal ones in your

817
00:39:08,960 --> 00:39:11,840
Speaker 7: first two years and then we're going to build on

818
00:39:11,960 --> 00:39:14,920
Speaker 7: that and have you do more complicated stuff, and chat

819
00:39:15,000 --> 00:39:19,359
Speaker 7: Gipt doesn't build on that, right, it just stays where

820
00:39:19,400 --> 00:39:22,800
Speaker 7: it was. So you go from writing kind of C

821
00:39:23,040 --> 00:39:27,160
Speaker 7: grade passing papers, yeah, your F grade paper.

822
00:39:27,719 --> 00:39:29,920
Speaker 5: And it's also more obvious because the papers are longer,

823
00:39:30,000 --> 00:39:33,000
Speaker 5: and chat gypt can write long text. But it gets

824
00:39:33,080 --> 00:39:38,080
Speaker 5: very repetitive and noticeably repetitive, right, and so you're kind

825
00:39:38,120 --> 00:39:42,040
Speaker 5: of lost, like you haven't done the work of figuring

826
00:39:42,080 --> 00:39:44,520
Speaker 5: out how to write on your own and the tool

827
00:39:44,560 --> 00:39:47,720
Speaker 5: that you've been using is not up to the task

828
00:39:47,760 --> 00:39:50,840
Speaker 5: that you're now presented with. And so I think I

829
00:39:50,920 --> 00:39:54,520
Speaker 5: have seen a few papers that I was suspicious of,

830
00:39:54,880 --> 00:39:56,759
Speaker 5: but the papers that I was certain of were ones

831
00:39:56,760 --> 00:40:00,000
Speaker 5: that were like senior thesis, the very clear person just

832
00:40:00,080 --> 00:40:02,040
Speaker 5: had no way of writing culture.

833
00:40:02,400 --> 00:40:04,000
Speaker 2: That's insane at that stage.

834
00:40:04,480 --> 00:40:06,480
Speaker 1: Yeah, I mean it's fun because I mean one of

835
00:40:06,520 --> 00:40:08,440
Speaker 1: the things that we get told about is like, oh,

836
00:40:08,560 --> 00:40:10,000
Speaker 1: students have got to learn how to use you know,

837
00:40:10,080 --> 00:40:14,120
Speaker 1: AI and large language model plagiarism machines responsibly and in

838
00:40:14,160 --> 00:40:16,680
Speaker 1: a kind of positive way. Well, if they're using them

839
00:40:16,680 --> 00:40:18,680
Speaker 1: in a way that means they don't learn how to write,

840
00:40:18,680 --> 00:40:20,799
Speaker 1: then it's not positive. Visit like, yeah, it's fucking hard

841
00:40:20,840 --> 00:40:23,360
Speaker 1: to write a good essay. Yes, it is hard to write.

842
00:40:23,520 --> 00:40:26,879
Speaker 1: That's why we practice, That's why we have editors, that's

843
00:40:26,880 --> 00:40:29,680
Speaker 1: why we do this collaboratively. And if you're using this

844
00:40:30,400 --> 00:40:32,759
Speaker 1: as a sort of oh I don't know how to write, well,

845
00:40:33,239 --> 00:40:35,279
Speaker 1: tough shit, you're never going to know how to write that.

846
00:40:35,280 --> 00:40:37,840
Speaker 1: That doesn't seem a positive use of any of this.

847
00:40:38,000 --> 00:40:41,960
Speaker 2: Well that's the thing. Writing is also reading the consumption

848
00:40:42,000 --> 00:40:44,120
Speaker 2: of information and then bouncing the ideas off of your

849
00:40:44,160 --> 00:40:44,920
Speaker 2: brain allegedly.

850
00:40:45,640 --> 00:40:49,960
Speaker 5: Yeah, I worry about like, because CHATCHBT is good at summarizing,

851
00:40:50,040 --> 00:40:52,600
Speaker 5: So I worry that one of the uses people will think, ah,

852
00:40:52,640 --> 00:40:54,920
Speaker 5: this is a pretty good use for it is summarizing

853
00:40:54,920 --> 00:40:58,760
Speaker 5: the paper that they're supposed to read, and it will

854
00:40:58,800 --> 00:41:03,080
Speaker 5: do that effectively enough for them to discuss it in

855
00:41:03,120 --> 00:41:04,200
Speaker 5: class if they're.

856
00:41:03,960 --> 00:41:04,640
Speaker 6: Willing to be here.

857
00:41:04,719 --> 00:41:07,800
Speaker 4: Yeah, right, but they're.

858
00:41:07,560 --> 00:41:09,400
Speaker 5: Not going to pick up a lot of the nuances

859
00:41:09,560 --> 00:41:11,440
Speaker 5: in a lot of the kind of like stylistic ways

860
00:41:11,440 --> 00:41:14,640
Speaker 5: of presenting ideas that you get when you actually do

861
00:41:14,719 --> 00:41:15,160
Speaker 5: the reading.

862
00:41:15,840 --> 00:41:18,920
Speaker 2: And it's it's so frustrating as well, because like for this,

863
00:41:19,080 --> 00:41:21,880
Speaker 2: for example, I've just got the printing off things that

864
00:41:21,920 --> 00:41:24,680
Speaker 2: don't read PDFs anymore because I feel like you do

865
00:41:24,760 --> 00:41:25,520
Speaker 2: need to focus.

866
00:41:26,160 --> 00:41:28,920
Speaker 1: Yeah, there's some evidence that reading physical copies makes you

867
00:41:28,960 --> 00:41:29,480
Speaker 1: engage more.

868
00:41:29,520 --> 00:41:31,799
Speaker 2: Sorry I'm very old fashioned, I guess, but no, it's

869
00:41:31,840 --> 00:41:35,400
Speaker 2: true though, But also reading that I wouldn't have really

870
00:41:35,920 --> 00:41:37,880
Speaker 2: on a PDF. I wouldn't have given it as much attention.

871
00:41:37,960 --> 00:41:40,960
Speaker 2: But also, going through this paper, you could see what

872
00:41:41,000 --> 00:41:43,000
Speaker 2: you were doing, like you could see that you were

873
00:41:43,040 --> 00:41:45,120
Speaker 2: lining up. Here are the qualities that we use to

874
00:41:45,200 --> 00:41:49,240
Speaker 2: judge bullshit. But also summarizing a paper does not actually

875
00:41:49,239 --> 00:41:53,480
Speaker 2: give you the argument, it gives you an answer. So

876
00:41:53,560 --> 00:41:55,880
Speaker 2: what do you actually want students to do instead? Because

877
00:41:55,920 --> 00:41:58,520
Speaker 2: I don't think there's any reason to use chat GPT

878
00:41:58,640 --> 00:42:01,400
Speaker 2: for these reasons that it doesn't seem to do anything

879
00:42:01,640 --> 00:42:02,640
Speaker 2: that's useful for them.

880
00:42:02,960 --> 00:42:06,480
Speaker 5: I don't have, no, I don't actually have any use

881
00:42:06,560 --> 00:42:09,200
Speaker 5: for chat shabt that I can put to my students

882
00:42:09,200 --> 00:42:09,960
Speaker 5: and say, here's.

883
00:42:09,719 --> 00:42:11,000
Speaker 4: What I think you should do with it.

884
00:42:11,600 --> 00:42:15,120
Speaker 5: We are like kind of developing strategies for keeping them

885
00:42:15,160 --> 00:42:19,320
Speaker 5: from using it, So like building directly on what you're saying.

886
00:42:19,360 --> 00:42:20,560
Speaker 4: Like, in my class.

887
00:42:20,200 --> 00:42:24,040
Speaker 5: Next year, I'm going to have the students do regular assignments,

888
00:42:24,040 --> 00:42:27,160
Speaker 5: which are argument summaries and not paper summaries. So the

889
00:42:27,200 --> 00:42:29,560
Speaker 5: idea is they have to read the paper, find an argument,

890
00:42:29,680 --> 00:42:33,160
Speaker 5: and tell me what are the premises, what's the conclusion?

891
00:42:33,320 --> 00:42:36,319
Speaker 5: And that's something that chatchept is not good at, right,

892
00:42:36,400 --> 00:42:39,440
Speaker 5: But it's also something that will give them critical reading skills,

893
00:42:39,800 --> 00:42:40,920
Speaker 5: which is what I want to do.

894
00:42:41,560 --> 00:42:43,760
Speaker 4: Right. So Yeah, I think that I've.

895
00:42:43,600 --> 00:42:48,120
Speaker 5: Mostly been thinking about ways to keep them from relying

896
00:42:48,160 --> 00:42:51,080
Speaker 5: on it because I think that often if they rely

897
00:42:51,160 --> 00:42:54,760
Speaker 5: on it, they'll they'll they'll put.

898
00:42:54,560 --> 00:42:55,840
Speaker 4: Themselves in the worst position.

899
00:42:56,080 --> 00:42:58,920
Speaker 5: Yeah, when it comes to future work, they won't develop

900
00:42:58,920 --> 00:43:00,880
Speaker 5: the skills that they're going to need, and the skills

901
00:43:00,880 --> 00:43:04,160
Speaker 5: that we tell them and their parents they're gonna get

902
00:43:04,160 --> 00:43:05,320
Speaker 5: with their college degree.

903
00:43:05,480 --> 00:43:08,200
Speaker 2: Right, it almost feels like we need more deliberacy in

904
00:43:08,800 --> 00:43:12,000
Speaker 2: university education because I was not taught to write. I

905
00:43:12,120 --> 00:43:13,960
Speaker 2: just did a lot of it until I got good

906
00:43:14,040 --> 00:43:17,040
Speaker 2: enough grades and Daniel CHOYNDL look great mental, but I've

907
00:43:17,080 --> 00:43:20,000
Speaker 2: had tons of them, and it almost feels like we

908
00:43:20,080 --> 00:43:23,040
Speaker 2: need classes where it's like, Okay, no computer for this one.

909
00:43:23,880 --> 00:43:25,880
Speaker 2: I'm gonna print this paper out and you're going to

910
00:43:26,080 --> 00:43:28,600
Speaker 2: underline the things that are important and talk to me

911
00:43:28,640 --> 00:43:32,200
Speaker 2: about Almost feels like we need to rebuild this because yes,

912
00:43:32,360 --> 00:43:34,880
Speaker 2: we shouldn't be using chet GPT to half us our essays.

913
00:43:34,880 --> 00:43:37,680
Speaker 2: But at the same time, human beings are lazy.

914
00:43:38,560 --> 00:43:41,839
Speaker 5: Yeah, I mean for me, I also prefer to read

915
00:43:42,200 --> 00:43:46,279
Speaker 5: off the computer, but I often read PDFs because I'm

916
00:43:46,440 --> 00:43:50,160
Speaker 5: terrible at keeping files, right, physical Like, you know, I'm

917
00:43:50,160 --> 00:43:53,120
Speaker 5: not going to keep a giant filed for with all

918
00:43:53,160 --> 00:43:56,320
Speaker 5: the papers that I've read and then written my liner

919
00:43:56,360 --> 00:43:56,680
Speaker 5: notes in.

920
00:43:57,440 --> 00:43:59,880
Speaker 4: You guys can see in my office they're just piled around,

921
00:44:00,080 --> 00:44:01,160
Speaker 4: like you can't see this end.

922
00:44:01,239 --> 00:44:02,200
Speaker 2: That's academia.

923
00:44:02,360 --> 00:44:05,760
Speaker 5: Yeah, but I just have piles of paper with empty

924
00:44:05,800 --> 00:44:08,560
Speaker 5: coffee bugs everywhere that the camera is not facing.

925
00:44:09,080 --> 00:44:10,880
Speaker 4: But it's the terrible system so at least.

926
00:44:10,800 --> 00:44:12,839
Speaker 5: On my computer, if I'm like, oh, I read that

927
00:44:12,880 --> 00:44:14,560
Speaker 5: paper like a year ago, what did I think?

928
00:44:14,600 --> 00:44:17,160
Speaker 4: I can click on it and see my own notes, and.

929
00:44:17,160 --> 00:44:20,480
Speaker 5: I do think that there's something to keeping those records

930
00:44:20,560 --> 00:44:22,000
Speaker 5: and kind of actively.

931
00:44:21,640 --> 00:44:24,359
Speaker 4: Reading in that way. I don't know how I ended

932
00:44:24,360 --> 00:44:27,680
Speaker 4: this without telling you how to make students do that. No,

933
00:44:27,719 --> 00:44:30,120
Speaker 4: you started with the correct answer, which is don't use

934
00:44:30,200 --> 00:44:32,600
Speaker 4: chat GPT. Yeah.

935
00:44:32,640 --> 00:44:36,359
Speaker 6: I actually I've got a certain amount of sympathy with like,

936
00:44:36,440 --> 00:44:38,000
Speaker 6: just keep writing it till you get good at it.

937
00:44:38,040 --> 00:44:40,200
Speaker 1: But I realized as a lecturer of that camp in

938
00:44:40,280 --> 00:44:42,759
Speaker 1: my official possession, and I certainly think that it's the

939
00:44:42,840 --> 00:44:46,759
Speaker 1: case that certainly Glasgow has got better over the last

940
00:44:46,760 --> 00:44:49,279
Speaker 1: few years about going, oh, actually you do like, we

941
00:44:49,360 --> 00:44:50,960
Speaker 1: do need to give you some kind of structuring and

942
00:44:50,960 --> 00:44:53,440
Speaker 1: some buttressing on here's how to write academically, here's how.

943
00:44:53,400 --> 00:44:54,040
Speaker 6: To do research.

944
00:44:54,080 --> 00:44:55,759
Speaker 1: And I think that's all to the good. It's worth

945
00:44:55,760 --> 00:44:59,640
Speaker 1: saying this started happening well before chat GPT's started pissing

946
00:44:59,680 --> 00:45:01,640
Speaker 1: a lover our doorsteps, so they don't get to claim

947
00:45:01,680 --> 00:45:02,600
Speaker 1: that as being a benefit.

948
00:45:03,360 --> 00:45:05,640
Speaker 2: There was the whole Wikipedia panic when I was in school.

949
00:45:05,760 --> 00:45:07,839
Speaker 6: Yeah, they think about Wikipedia the right.

950
00:45:07,840 --> 00:45:09,440
Speaker 4: It's like I should to say this to my students.

951
00:45:09,480 --> 00:45:12,680
Speaker 2: WIKIPII one of the best resources.

952
00:45:11,960 --> 00:45:16,480
Speaker 1: Absolutely fine as a starting point for research, absolutely with

953
00:45:16,520 --> 00:45:19,439
Speaker 1: it whatsoever. But if you're turning in and on his level, essay,

954
00:45:19,520 --> 00:45:21,279
Speaker 1: I want you to go and read the fucking things

955
00:45:21,280 --> 00:45:21,960
Speaker 1: it's referencing.

956
00:45:22,160 --> 00:45:22,960
Speaker 4: Yeah, that's right.

957
00:45:23,719 --> 00:45:26,000
Speaker 5: Yeah, I think these things are often great as sources.

958
00:45:26,160 --> 00:45:28,359
Speaker 5: I'm worry about chat GPT is that it's not great

959
00:45:28,400 --> 00:45:28,920
Speaker 5: as a source.

960
00:45:29,000 --> 00:45:30,440
Speaker 2: It's just yeah, we've.

961
00:45:30,280 --> 00:45:33,040
Speaker 5: Been saying it often gets things wrong, and it often

962
00:45:33,400 --> 00:45:36,000
Speaker 5: it'll make that sources, whereas Wikipedia will never do that.

963
00:45:37,360 --> 00:45:41,799
Speaker 6: There are some famous hoaxes, but they get cool.

964
00:45:41,880 --> 00:45:44,799
Speaker 1: Yeah, Joe, have you got any positive things to say

965
00:45:44,800 --> 00:45:46,040
Speaker 1: about chat gipt.

966
00:45:47,440 --> 00:45:47,600
Speaker 4: Fan?

967
00:45:49,800 --> 00:45:52,600
Speaker 3: So I know some people who have used these kinds

968
00:45:52,640 --> 00:45:56,359
Speaker 3: of things productively, not in ways that our students would,

969
00:45:56,440 --> 00:45:59,759
Speaker 3: but I know some mathematicians who have been using it

970
00:46:00,200 --> 00:46:03,479
Speaker 3: to do sort of informal proofs and things like that.

971
00:46:03,760 --> 00:46:07,280
Speaker 3: And it does still bullshit, and it bullshits very convincingly,

972
00:46:07,440 --> 00:46:10,239
Speaker 3: which makes it very difficult to use for this kind

973
00:46:10,280 --> 00:46:10,760
Speaker 3: of purpose.

974
00:46:11,440 --> 00:46:13,480
Speaker 4: But it can do some interesting and cool things.

975
00:46:14,239 --> 00:46:16,440
Speaker 3: Yeah, I think some people in that sort of field

976
00:46:16,680 --> 00:46:20,040
Speaker 3: have found useful and also We've mentioned this before, like,

977
00:46:20,040 --> 00:46:21,600
Speaker 3: if you want CHAT to be teach and write, you're

978
00:46:21,600 --> 00:46:25,040
Speaker 3: a bibliography. You've got a bibliography in one style. Tell

979
00:46:25,080 --> 00:46:27,600
Speaker 3: it to put something into a different one. Then it's that,

980
00:46:27,760 --> 00:46:30,600
Speaker 3: And it's good for like coding the process doing certain things.

981
00:46:31,560 --> 00:46:34,600
Speaker 4: I also think, I don't know, I'm not sure how

982
00:46:34,640 --> 00:46:35,920
Speaker 4: I feel about what I'm about.

983
00:46:35,680 --> 00:46:38,560
Speaker 2: To say, but perfect, Yeah.

984
00:46:38,360 --> 00:46:39,080
Speaker 4: I'm going for it.

985
00:46:39,080 --> 00:46:41,320
Speaker 5: It is a somewhat positive thing maybe for a CHAT GPT,

986
00:46:41,560 --> 00:46:44,040
Speaker 5: which is that we often have students who have like

987
00:46:44,120 --> 00:46:48,880
Speaker 5: really interesting ideas and well thought out arguments, but for

988
00:46:49,000 --> 00:46:51,080
Speaker 5: whom English isn't their first language, and the.

989
00:46:51,160 --> 00:46:53,479
Speaker 4: Actual writing is kind of rough and you have.

990
00:46:53,400 --> 00:46:57,760
Speaker 5: To like push through reading it to get the good idea,

991
00:46:57,800 --> 00:47:02,719
Speaker 5: which is often really there and quite creative and insightful. Yeah,

992
00:47:02,719 --> 00:47:04,440
Speaker 5: and so I do wonder if there's a way to

993
00:47:04,520 --> 00:47:07,000
Speaker 5: use it so it just smooths off the edges of

994
00:47:07,080 --> 00:47:09,000
Speaker 5: this kind of thing. But I worry that if you

995
00:47:09,080 --> 00:47:12,600
Speaker 5: tell students to do that, they'll just first if they

996
00:47:12,640 --> 00:47:14,359
Speaker 5: can develop the language skills.

997
00:47:14,040 --> 00:47:16,839
Speaker 4: They offering really good by the year. Yeah, what are

998
00:47:16,840 --> 00:47:17,520
Speaker 4: you going to say? Do well?

999
00:47:17,520 --> 00:47:19,799
Speaker 1: So I want to say, yeah, you can see me

1000
00:47:19,840 --> 00:47:22,200
Speaker 1: getting agitated. Yeah, I think much correct about it, like

1001
00:47:22,239 --> 00:47:23,840
Speaker 1: that this is a kind of possible use. But I

1002
00:47:23,880 --> 00:47:27,120
Speaker 1: think this and this is why I'm getting visibly agitated here.

1003
00:47:27,480 --> 00:47:31,200
Speaker 1: That's students either need to or feel they need to use.

1004
00:47:31,239 --> 00:47:33,600
Speaker 1: This speaks to a deeper issue, right to a social issue,

1005
00:47:33,600 --> 00:47:35,680
Speaker 1: to political issue, to an issue about how universities work.

1006
00:47:35,800 --> 00:47:39,160
Speaker 1: If a student is having problems with English, then there's

1007
00:47:39,200 --> 00:47:41,799
Speaker 1: a number of like explanations or a number of kind

1008
00:47:41,800 --> 00:47:42,280
Speaker 1: of responses.

1009
00:47:42,320 --> 00:47:42,440
Speaker 3: Right.

1010
00:47:42,440 --> 00:47:45,160
Speaker 1: The one response is that, like Glasgow is an English

1011
00:47:45,160 --> 00:47:47,880
Speaker 1: teaching university, if someone's English isn't good enough to be

1012
00:47:47,960 --> 00:47:50,160
Speaker 1: taking a degree, then possibly it shouldn't have been let in,

1013
00:47:50,239 --> 00:47:51,200
Speaker 1: and why have they been learned well?

1014
00:47:51,239 --> 00:47:51,920
Speaker 6: Because of money?

1015
00:47:52,160 --> 00:47:55,279
Speaker 1: Or alternatively, if someone's having problems with English for like

1016
00:47:55,360 --> 00:47:58,439
Speaker 1: whatever reason at all, there should be supported. There should

1017
00:47:58,440 --> 00:47:59,839
Speaker 1: be kind of tutors. They should be people who can

1018
00:48:00,080 --> 00:48:02,640
Speaker 1: with English. But again that would cost university money. So

1019
00:48:02,680 --> 00:48:04,200
Speaker 1: of course that doesn't happen.

1020
00:48:04,160 --> 00:48:09,359
Speaker 6: Right, and it doesn't happen anywhere that it doesn't extent.

1021
00:48:09,000 --> 00:48:10,799
Speaker 1: It would have to happen in order for this to

1022
00:48:10,840 --> 00:48:12,560
Speaker 1: be a general policy.

1023
00:48:12,680 --> 00:48:15,080
Speaker 5: Right, Yeah, I think it could be better, But I

1024
00:48:15,120 --> 00:48:18,719
Speaker 5: do think that universities often have like a writing center

1025
00:48:18,920 --> 00:48:20,640
Speaker 5: or a tutoring center that you can.

1026
00:48:20,520 --> 00:48:23,280
Speaker 6: Send, but they don't.

1027
00:48:23,440 --> 00:48:25,880
Speaker 1: They don't have the sort of spread that would be

1028
00:48:25,960 --> 00:48:27,799
Speaker 1: needed or the staff that would be needed for this

1029
00:48:27,880 --> 00:48:31,200
Speaker 1: to be Instead of using chat GPT to sound the edges.

1030
00:48:31,000 --> 00:48:33,920
Speaker 4: Off, yeah, I think for example twenty three with your supervisor.

1031
00:48:34,040 --> 00:48:37,759
Speaker 5: My worry especially would be that this is my first

1032
00:48:37,840 --> 00:48:38,800
Speaker 5: year here at Glasgow.

1033
00:48:38,880 --> 00:48:41,000
Speaker 4: But I probably have a good Yeah.

1034
00:48:42,000 --> 00:48:44,080
Speaker 5: I think they probably have a good writing center. Universities

1035
00:48:44,080 --> 00:48:46,680
Speaker 5: have been in the past. I felt very confident sending

1036
00:48:46,680 --> 00:48:49,240
Speaker 5: students to the writing center when they have these problems.

1037
00:48:49,440 --> 00:48:53,160
Speaker 5: But I think James is completely right that we don't

1038
00:48:53,200 --> 00:48:55,160
Speaker 5: want the universities to see this as a way to

1039
00:48:55,160 --> 00:48:57,680
Speaker 5: get rid of the writing center. And that's one hundred

1040
00:48:57,680 --> 00:49:00,799
Speaker 5: percent a risk given the financial problems.

1041
00:49:00,440 --> 00:49:05,640
Speaker 4: That universities are facing, and maybe they're already not. Yeah,

1042
00:49:06,080 --> 00:49:09,040
Speaker 4: Like given the quality of papers, they're often good.

1043
00:49:09,120 --> 00:49:12,520
Speaker 5: I think another thing, as far as this is like

1044
00:49:12,520 --> 00:49:17,320
Speaker 5: a social problem, is that when grading, I myself tried

1045
00:49:17,360 --> 00:49:19,560
Speaker 5: to grade in terms of like the ideas and argument,

1046
00:49:19,719 --> 00:49:22,640
Speaker 5: because this is a philosophy and not the quality of

1047
00:49:22,680 --> 00:49:26,399
Speaker 5: the right. But not everybody does that. So I kind

1048
00:49:26,440 --> 00:49:28,839
Speaker 5: of think that another part of this is figuring out

1049
00:49:28,840 --> 00:49:30,600
Speaker 5: how we want to evaluate the students and what we

1050
00:49:30,640 --> 00:49:32,759
Speaker 5: want to privilege in that evaluation.

1051
00:49:33,680 --> 00:49:33,879
Speaker 6: Yeah.

1052
00:49:33,960 --> 00:49:36,720
Speaker 1: Sure, So then again the kind of that that becomes

1053
00:49:36,719 --> 00:49:39,440
Speaker 1: a problem about what people are checking for. Not let's

1054
00:49:39,480 --> 00:49:42,520
Speaker 1: take this arched backwards approach to marketing, which is like

1055
00:49:42,640 --> 00:49:43,880
Speaker 1: how fancy is your English?

1056
00:49:43,920 --> 00:49:46,239
Speaker 6: Are fancy English? Is good English? Have an a? But

1057
00:49:46,320 --> 00:49:47,200
Speaker 6: rather we should be kind.

1058
00:49:47,080 --> 00:49:47,880
Speaker 4: Of checking for different things.

1059
00:49:47,960 --> 00:49:50,479
Speaker 1: Right, So again the blame blows differently in that case,

1060
00:49:50,600 --> 00:49:51,640
Speaker 1: but it still becomes a question.

1061
00:49:51,680 --> 00:49:52,759
Speaker 4: It's not solved technological.

1062
00:49:53,120 --> 00:49:56,880
Speaker 2: Yeah, almost feels like large language models are taking advantage

1063
00:49:56,920 --> 00:49:59,480
Speaker 2: of a certain kind of organizational failure.

1064
00:50:00,480 --> 00:50:02,280
Speaker 4: What an idea?

1065
00:50:03,000 --> 00:50:07,160
Speaker 2: Crazy that the tech industry manipulating parts of society.

1066
00:50:07,200 --> 00:50:08,400
Speaker 6: That wait, I have a kind.

1067
00:50:08,320 --> 00:50:11,719
Speaker 5: Of related tangent here, which is like what are the

1068
00:50:11,880 --> 00:50:15,880
Speaker 5: use cases that open ai was expecting but didn't want

1069
00:50:15,880 --> 00:50:19,279
Speaker 5: to emphasize Because for everybody in universities, as soon as

1070
00:50:19,280 --> 00:50:22,080
Speaker 5: this came out, the first thought was students are going

1071
00:50:22,160 --> 00:50:25,600
Speaker 5: to use this to cheat And certainly like the people

1072
00:50:25,600 --> 00:50:28,800
Speaker 5: in open ai went to college, right, and that's what

1073
00:50:28,920 --> 00:50:29,719
Speaker 5: I hear about.

1074
00:50:29,560 --> 00:50:33,800
Speaker 2: Them, so they must well, sam Oman drop down, Oh yeah, I'm.

1075
00:50:33,640 --> 00:50:37,320
Speaker 5: Sure he really understands the value of a secondary He's like, I'm.

1076
00:50:37,160 --> 00:50:39,240
Speaker 2: Gonna write at least fucking essays.

1077
00:50:38,840 --> 00:50:41,040
Speaker 5: Anyway, Maybe he was thinking I would have loved to

1078
00:50:41,040 --> 00:50:42,280
Speaker 5: have a computer write my essays.

1079
00:50:42,280 --> 00:50:43,160
Speaker 4: I'll devote my life.

1080
00:50:43,760 --> 00:50:47,040
Speaker 5: But I mean, I'm sure that they like recognize these

1081
00:50:47,120 --> 00:50:51,240
Speaker 5: bad use cases, right, but they're doing nothing to mitigate

1082
00:50:51,239 --> 00:50:52,719
Speaker 5: them as far as I can see. And like, another

1083
00:50:52,760 --> 00:50:57,120
Speaker 5: one that's very related is like, you know, I'm sure

1084
00:50:57,120 --> 00:50:59,480
Speaker 5: you've heard of this ed phishing, right, a lot of

1085
00:50:59,719 --> 00:51:03,879
Speaker 5: you know, corporations get attacked and get hacked not by

1086
00:51:03,920 --> 00:51:08,080
Speaker 5: someone cleverly figuring out a backdoor to their system, but

1087
00:51:08,080 --> 00:51:09,120
Speaker 5: by somebody sending.

1088
00:51:08,920 --> 00:51:10,600
Speaker 2: A social engineering.

1089
00:51:10,160 --> 00:51:13,839
Speaker 5: Yeah, asking for the password to somebody else. And one

1090
00:51:13,880 --> 00:51:15,759
Speaker 5: of the biggest barriers to that is that a lot

1091
00:51:15,800 --> 00:51:19,600
Speaker 5: of the people who are engaging in fishing aren't from

1092
00:51:19,640 --> 00:51:22,719
Speaker 5: the same country as the company they're targeting, right, so

1093
00:51:22,760 --> 00:51:25,600
Speaker 5: they're not able to write a convincing email or make

1094
00:51:25,640 --> 00:51:29,120
Speaker 5: a phone call that sounds like that person's supervisor. But

1095
00:51:29,760 --> 00:51:33,319
Speaker 5: with a tool like this, you could one percent write

1096
00:51:33,320 --> 00:51:34,759
Speaker 5: that email. Right. It's going to make it a lot

1097
00:51:34,800 --> 00:51:38,640
Speaker 5: easier for these kinds of illicit schemes to work.

1098
00:51:38,760 --> 00:51:41,680
Speaker 2: That has been a market increase CNBC, which would just

1099
00:51:41,680 --> 00:51:43,839
Speaker 2: brow up, is that on two hundred and sixty five

1100
00:51:43,840 --> 00:51:46,279
Speaker 2: percent increase in malicious fishing emails since the launch of

1101
00:51:46,360 --> 00:51:48,600
Speaker 2: chat JPTA. Great style. I mean if I could have

1102
00:51:48,640 --> 00:51:51,600
Speaker 2: thought of that, imagine what a criminal could do.

1103
00:51:52,360 --> 00:51:52,520
Speaker 6: Right.

1104
00:51:52,600 --> 00:51:55,960
Speaker 4: But also, weren't the people at open AI thinking about.

1105
00:51:55,719 --> 00:51:57,960
Speaker 2: That because they don't care?

1106
00:51:58,239 --> 00:51:58,600
Speaker 4: Yeah?

1107
00:51:58,680 --> 00:52:00,480
Speaker 6: Yeah, we will see in duastic fuck right.

1108
00:52:00,520 --> 00:52:00,719
Speaker 3: Yeah.

1109
00:52:00,800 --> 00:52:02,560
Speaker 1: Yeah, they were so busy thinking about what they could

1110
00:52:02,640 --> 00:52:04,080
Speaker 1: do they never thought about whether they should.

1111
00:52:04,239 --> 00:52:06,480
Speaker 5: Yeah, this is the kind of problem with the move

1112
00:52:06,560 --> 00:52:09,520
Speaker 5: fast and break things mentality, like obvious.

1113
00:52:09,760 --> 00:52:11,520
Speaker 4: I mean, I think it might be the only person

1114
00:52:11,560 --> 00:52:12,800
Speaker 4: who has raised in the US.

1115
00:52:12,560 --> 00:52:15,799
Speaker 8: But we had future problems solvers, where you you know,

1116
00:52:15,920 --> 00:52:20,000
Speaker 8: you think about a future problem and what bad consequences

1117
00:52:20,040 --> 00:52:22,279
Speaker 8: there could be of some technology and how to solve them,

1118
00:52:22,760 --> 00:52:25,560
Speaker 8: usually through social cues. If I could do that in

1119
00:52:25,640 --> 00:52:31,040
Speaker 8: fifth grade.

1120
00:52:28,600 --> 00:52:31,000
Speaker 5: I would expect these people to have thought through some

1121
00:52:31,080 --> 00:52:33,480
Speaker 5: of the bad consequences of the technology they're putting out.

1122
00:52:33,960 --> 00:52:36,880
Speaker 5: And you know, some of those are cheating on tests

1123
00:52:37,120 --> 00:52:39,359
Speaker 5: and they don't seem to have worried about that.

1124
00:52:39,800 --> 00:52:41,600
Speaker 4: Yeah, And another one is fishing. They don't seem to

1125
00:52:41,640 --> 00:52:42,399
Speaker 4: have a worried about that.

1126
00:52:42,440 --> 00:52:47,600
Speaker 1: You know, biases and algorithms, right, So yeah, yeah, comes

1127
00:52:47,640 --> 00:52:49,600
Speaker 1: no surprise to you. It turns out so with a

1128
00:52:49,680 --> 00:52:52,440
Speaker 1: lot of the facial recognition systems they were incredibly racist.

1129
00:52:52,680 --> 00:52:55,279
Speaker 2: They were going back to Microsoft's connect they could not

1130
00:52:55,320 --> 00:52:55,719
Speaker 2: see back.

1131
00:52:56,040 --> 00:53:00,239
Speaker 1: Yeah yeah, and CCTV stuff that basically just sort of nless.

1132
00:53:00,280 --> 00:53:03,400
Speaker 1: It was presented with a blindingly white Caucasian.

1133
00:53:02,960 --> 00:53:05,080
Speaker 6: Where I don't know, right, but like.

1134
00:53:08,320 --> 00:53:11,480
Speaker 1: The sort of stuff where these like language knowledge are

1135
00:53:11,520 --> 00:53:13,520
Speaker 1: trained on certain sets of data, and they trained on

1136
00:53:13,600 --> 00:53:18,239
Speaker 1: certain assumptions and like shitting shit out right, and particularly

1137
00:53:18,280 --> 00:53:20,800
Speaker 1: if people think that it's actually doing any kind of thinking,

1138
00:53:21,080 --> 00:53:24,160
Speaker 1: and if they kind of cargo cult it, we again

1139
00:53:24,160 --> 00:53:27,040
Speaker 1: get a kind of social problem multiplied by technology feeding

1140
00:53:27,080 --> 00:53:30,640
Speaker 1: back into a social problem, right. And it's the So

1141
00:53:30,960 --> 00:53:33,040
Speaker 1: these guys have heard me when you're about it so much,

1142
00:53:33,360 --> 00:53:36,560
Speaker 1: I love it or so, but I'm profoundly skeptical of

1143
00:53:37,680 --> 00:53:41,719
Speaker 1: technology's ability to solve anything unless we know exactly the

1144
00:53:41,840 --> 00:53:44,560
Speaker 1: respect in which we want to solve it and how

1145
00:53:44,640 --> 00:53:48,239
Speaker 1: that technology is going to be applied. You know, like sure,

1146
00:53:48,520 --> 00:53:51,680
Speaker 1: experiment with bringing back dinosaurs, but but like, don't tell

1147
00:53:51,680 --> 00:53:54,480
Speaker 1: me that it's going to save the healthcare system unless

1148
00:53:54,520 --> 00:53:56,680
Speaker 1: you can demonstrate it to be step by step. How

1149
00:53:56,719 --> 00:53:58,920
Speaker 1: that big old rex run around on Eiland Nublade is

1150
00:53:58,920 --> 00:53:59,720
Speaker 1: going to save anything.

1151
00:54:00,040 --> 00:54:01,120
Speaker 6: They just tried to blind.

1152
00:54:00,960 --> 00:54:05,080
Speaker 3: People actually bringing back dinosaurs.

1153
00:54:04,400 --> 00:54:04,959
Speaker 4: That would be great.

1154
00:54:05,520 --> 00:54:08,120
Speaker 1: All right, this isn't the best example, but I already had.

1155
00:54:09,440 --> 00:54:11,000
Speaker 1: I had Jeff Gobwoman in my head.

1156
00:54:11,440 --> 00:54:14,319
Speaker 2: I had to go to the fellas. This has been

1157
00:54:14,400 --> 00:54:18,720
Speaker 2: such a pleasure. Don't put that in the recording.

1158
00:54:20,480 --> 00:54:21,120
Speaker 4: That's right.

1159
00:54:21,239 --> 00:54:23,960
Speaker 2: We won't edit it in post, and you will give

1160
00:54:24,040 --> 00:54:24,720
Speaker 2: us your names.

1161
00:54:25,320 --> 00:54:27,840
Speaker 5: I'm Mike Hicks, but my papers are written by Michael

1162
00:54:27,880 --> 00:54:31,640
Speaker 5: Townsend Hicks and I'm a lecturer at the University of Glasgow.

1163
00:54:31,760 --> 00:54:34,480
Speaker 4: My website is Hicks.

1164
00:54:34,920 --> 00:54:36,560
Speaker 2: It will be in the podcast profile.

1165
00:54:36,600 --> 00:54:39,720
Speaker 6: Don't you worry. We'll get that just plugging my luggy stuff.

1166
00:54:41,800 --> 00:54:42,880
Speaker 4: My name is Joe Slater.

1167
00:54:43,080 --> 00:54:46,080
Speaker 3: I'm a university lecturer in moral and political policy at Glasgow.

1168
00:54:47,080 --> 00:54:48,040
Speaker 6: I'm James Joffrees.

1169
00:54:48,120 --> 00:54:50,920
Speaker 1: I'm a lecturer in political theory at the University of Glasgow.

1170
00:54:51,040 --> 00:54:51,839
Speaker 4: And even if I.

1171
00:54:51,840 --> 00:54:53,880
Speaker 6: Wanted to or couldn't give you what websites, I don't have.

1172
00:54:56,360 --> 00:54:58,560
Speaker 2: Everyone you've been listening to Better Offline, thank you so

1173
00:54:58,640 --> 00:55:00,959
Speaker 2: much for listening. Everyone guys, thank you for joining me.

1174
00:55:01,640 --> 00:55:02,760
Speaker 4: Thanks having me here.

1175
00:55:02,800 --> 00:55:03,319
Speaker 6: This is.

1176
00:55:12,800 --> 00:55:15,239
Speaker 2: Thank you for listening to Better Offline. The editor and

1177
00:55:15,280 --> 00:55:18,440
Speaker 2: composer of the Better Offline theme song is Matasowski. You

1178
00:55:18,480 --> 00:55:20,680
Speaker 2: can check out more of his music and audio projects

1179
00:55:20,880 --> 00:55:24,360
Speaker 2: at Matasowski dot com, M A T T O S

1180
00:55:24,440 --> 00:55:28,520
Speaker 2: O W s ki dot com. You can email me

1181
00:55:28,560 --> 00:55:31,160
Speaker 2: at easy at Better offline dot com or visit Better

1182
00:55:31,200 --> 00:55:33,640
Speaker 2: Offline dot com to find more podcast links and of course,

1183
00:55:33,640 --> 00:55:36,799
Speaker 2: my newsletter. I also really recommend you go to chat

1184
00:55:36,800 --> 00:55:39,440
Speaker 2: dot Where's youreed dot at to visit the discord, and

1185
00:55:39,480 --> 00:55:41,640
Speaker 2: go to our slash Better off Line to check out

1186
00:55:41,640 --> 00:55:46,000
Speaker 2: our reddit. Thank you so much for listening. Better Offline

1187
00:55:46,080 --> 00:55:48,319
Speaker 2: is a production of cool Zone Media. For more from

1188
00:55:48,320 --> 00:55:50,640
Speaker 2: cool Zone Media, visit our website cool

1189
00:55:50,680 --> 00:55:54,040
Speaker 7: Zonemedia dot com, or check us out on the iHeartRadio app,

1190
00:55:54,080 --> 00:56:16,880
Speaker 7: Apple Podcasts, or wherever you get your podcasts.