1
00:00:02,400 --> 00:00:03,560
Speaker 1: Cauzone Media.

2
00:00:05,200 --> 00:00:07,600
Speaker 2: Hey everyone, it's me ed Ze Tron and we're doing

3
00:00:07,640 --> 00:00:10,920
Speaker 2: a rerun episode this week. Sadly, the in studio recording

4
00:00:10,920 --> 00:00:13,960
Speaker 2: we did with Ashmund Rodriguez and Victoria's song we had

5
00:00:13,960 --> 00:00:17,400
Speaker 2: a technical fall. Nothing we can do. It sucks, but

6
00:00:17,560 --> 00:00:19,800
Speaker 2: we're doing a rerun this week. We're doing the Academics

7
00:00:19,800 --> 00:00:22,200
Speaker 2: that think Chat GPT is BS and this is one

8
00:00:22,200 --> 00:00:24,560
Speaker 2: of my favorite episodes I ever recorded. It changed how

9
00:00:24,600 --> 00:00:27,600
Speaker 2: I do interviews for at Large. It's with these three academics,

10
00:00:27,640 --> 00:00:30,720
Speaker 2: Michael Townsend Higgs, James Humphries and Joe Slater, who wrote

11
00:00:30,720 --> 00:00:35,000
Speaker 2: a paper using the actual Frank thirty in definition of

12
00:00:35,000 --> 00:00:38,840
Speaker 2: bullshit to say that chat GPT bullshits. It's so much fun.

13
00:00:38,920 --> 00:00:41,560
Speaker 2: It's one of my favorite episodes I've recorded. I will

14
00:00:41,560 --> 00:00:43,599
Speaker 2: have you a monologue this week as well. I do

15
00:00:43,720 --> 00:00:46,840
Speaker 2: apologize for not having something new for you this week.

16
00:00:47,040 --> 00:00:49,479
Speaker 2: You'll still get the monologue on Friday, though. Thank you

17
00:00:49,479 --> 00:00:52,040
Speaker 2: for your time, your patience and of course for listening

18
00:00:52,040 --> 00:01:04,600
Speaker 2: to Better Offline. Hello and welcome to Better Offline. I'm

19
00:01:04,600 --> 00:01:20,039
Speaker 2: your host ed Ze Trunk. In early June, three researchers

20
00:01:20,040 --> 00:01:22,760
Speaker 2: from the University of Glasgow published a paper in the

21
00:01:22,800 --> 00:01:26,640
Speaker 2: Ethics of Information and Technology Journal called chat GBT is bullshit,

22
00:01:27,319 --> 00:01:29,399
Speaker 2: And I just want to be clear. This is a

23
00:01:29,560 --> 00:01:32,959
Speaker 2: great and thoroughly researched, from well argued paper. This is

24
00:01:33,040 --> 00:01:36,559
Speaker 2: not silly at all. It's actually great academia. And today

25
00:01:36,640 --> 00:01:39,320
Speaker 2: I'm joined by the men who wrote it, academics Michael

26
00:01:39,360 --> 00:01:42,800
Speaker 2: Townsend Hicks, James Humphries, and Joe Slater, to talk about

27
00:01:42,880 --> 00:01:45,920
Speaker 2: chat gbt's mediocrity and how it's not really built to

28
00:01:46,120 --> 00:01:49,240
Speaker 2: represent the world at all. So, for the sake of argument,

29
00:01:49,720 --> 00:01:51,840
Speaker 2: could you define bullshit for me?

30
00:01:53,080 --> 00:01:57,440
Speaker 3: So you are bullshitting if you are speaking without caring

31
00:01:58,200 --> 00:02:01,040
Speaker 3: about the truth of what you say. So normally, if

32
00:02:01,080 --> 00:02:04,480
Speaker 3: I'm telling you staff about the world in a good case,

33
00:02:04,520 --> 00:02:07,160
Speaker 3: I'll be telling you something that's true, and like trying

34
00:02:07,160 --> 00:02:09,760
Speaker 3: to tell you something that's true. If I'm lying to you,

35
00:02:09,840 --> 00:02:12,519
Speaker 3: I'll be knowingly telling you something that's false or something

36
00:02:12,560 --> 00:02:15,640
Speaker 3: I think is false. I'm bullshitting. I just don't care.

37
00:02:16,160 --> 00:02:18,920
Speaker 3: I'm trying to get you to believe me. I don't

38
00:02:18,960 --> 00:02:20,919
Speaker 3: really care about whether what I say is true. I

39
00:02:21,000 --> 00:02:23,639
Speaker 3: might not have any particular view on whether it's true

40
00:02:23,760 --> 00:02:24,880
Speaker 3: or not right.

41
00:02:25,480 --> 00:02:28,560
Speaker 2: And you define between like soft and hard bullshit? Can

42
00:02:28,600 --> 00:02:30,880
Speaker 2: you also get into that as well? Can you also

43
00:02:30,960 --> 00:02:32,840
Speaker 2: identify yourselves as well too?

44
00:02:33,080 --> 00:02:33,240
Speaker 4: Yeah?

45
00:02:33,280 --> 00:02:34,000
Speaker 5: Yeah, and Joe.

46
00:02:34,320 --> 00:02:37,200
Speaker 3: So the soft bullshit hard bullshit distinction is a very

47
00:02:37,360 --> 00:02:41,880
Speaker 3: serious and technical distinction, right, So he came up with

48
00:02:41,919 --> 00:02:47,000
Speaker 3: this because bullshit is in the technical philosophical sense. Comes

49
00:02:47,000 --> 00:02:50,760
Speaker 3: from Harry Frankfurt, recently diseased but a really great philosopher,

50
00:02:51,200 --> 00:02:54,280
Speaker 3: and he talks about the about bullshit that there is

51
00:02:54,440 --> 00:02:58,400
Speaker 3: in a popular culture and just in general discourse these days.

52
00:02:58,639 --> 00:03:01,240
Speaker 3: Some of the ways he talks about bullshit seem to

53
00:03:01,280 --> 00:03:05,160
Speaker 3: suggest that it needs to be accompanied by a sort

54
00:03:05,200 --> 00:03:06,400
Speaker 3: of malign intention.

55
00:03:07,320 --> 00:03:09,440
Speaker 4: I'm doing something kind of bad.

56
00:03:09,560 --> 00:03:13,520
Speaker 3: I'm intending to mislead you about the enterprise of of

57
00:03:13,600 --> 00:03:14,120
Speaker 3: what I'm.

58
00:03:13,919 --> 00:03:16,600
Speaker 4: Doing, maybe about who you are or what you know.

59
00:03:16,720 --> 00:03:20,240
Speaker 1: So you might be trying to portray yourself as someone

60
00:03:20,280 --> 00:03:24,360
Speaker 1: who is knowledgeable about a particular subject. Maybe you're a

61
00:03:24,360 --> 00:03:26,800
Speaker 1: student who showed up to class without doing the work.

62
00:03:27,280 --> 00:03:30,800
Speaker 1: Maybe you're trying to portray yourself as someone who's virtuous

63
00:03:30,840 --> 00:03:33,079
Speaker 1: in ways you're not. Maybe you're a politician who wants

64
00:03:33,160 --> 00:03:36,280
Speaker 1: to seem like you care about your constituents, but actually

65
00:03:36,360 --> 00:03:40,000
Speaker 1: you don't. So you're not trying to mislead somebody about

66
00:03:40,560 --> 00:03:43,480
Speaker 1: what you're saying the content of your utterance.

67
00:03:44,080 --> 00:03:44,680
Speaker 4: You're trying to.

68
00:03:44,600 --> 00:03:48,800
Speaker 1: Mislead them instead about like why you're saying it's that's

69
00:03:48,840 --> 00:03:50,160
Speaker 1: what we call hard bullshit.

70
00:03:50,360 --> 00:03:52,480
Speaker 4: Yeah, it's one of the things Frankfort talked about.

71
00:03:52,680 --> 00:03:55,360
Speaker 3: Yeah, so Frankfurt doesn't make this hard bullshit soft bullshit

72
00:03:55,360 --> 00:03:58,200
Speaker 3: in the station, but we do because sometimes it seems

73
00:03:58,240 --> 00:04:01,240
Speaker 3: like Frankfurt has this particular kind of intention in mind,

74
00:04:01,760 --> 00:04:04,760
Speaker 3: but sometimes he's just a bit looser with it. And

75
00:04:04,800 --> 00:04:07,680
Speaker 3: we want to say that CHAT GPT and other large

76
00:04:07,720 --> 00:04:11,280
Speaker 3: language models they don't really have this intention to deceive

77
00:04:11,360 --> 00:04:13,640
Speaker 3: because they're not people that don't have these intentions. They're

78
00:04:13,640 --> 00:04:16,159
Speaker 3: not trying to mess with us in that kind of way,

79
00:04:16,560 --> 00:04:19,599
Speaker 3: but they do lack this kind of caring about truth.

80
00:04:20,360 --> 00:04:23,080
Speaker 6: Well, I'm James, by the way, I suppose we strictly

81
00:04:23,080 --> 00:04:25,840
Speaker 6: don't want to say that they aren't hard bullshitters. And

82
00:04:25,880 --> 00:04:28,360
Speaker 6: we just think if you don't think that large language

83
00:04:28,360 --> 00:04:30,120
Speaker 6: models are sapient, if you don't think they're kind of

84
00:04:30,160 --> 00:04:33,560
Speaker 6: minds in any important way, then they're not hard bullshitters.

85
00:04:33,600 --> 00:04:35,599
Speaker 6: So I think in the paper we don't take a

86
00:04:35,640 --> 00:04:37,840
Speaker 6: position on whether or not they are. We just say,

87
00:04:38,279 --> 00:04:40,200
Speaker 6: if they are, this is the way in which they are.

88
00:04:40,400 --> 00:04:43,599
Speaker 6: But minimally they're soft bullshitters, So they're kind of soft bullshites,

89
00:04:43,640 --> 00:04:46,640
Speaker 6: as Joe says, doesn't require that speakers attempting to deceive

90
00:04:46,680 --> 00:04:48,839
Speaker 6: the audience about the nature of the enterprise.

91
00:04:49,040 --> 00:04:49,960
Speaker 5: Hard bullshit does.

92
00:04:50,160 --> 00:04:52,520
Speaker 6: So if it turns out that large language models are sapient,

93
00:04:52,560 --> 00:04:54,560
Speaker 6: which they're definitely not, like that's just.

94
00:04:54,600 --> 00:04:56,760
Speaker 2: Technicio property, Yeah, that's nonsense.

95
00:04:56,880 --> 00:04:59,320
Speaker 6: Yeah, But if they are, then they're hard bullshitters and

96
00:04:59,440 --> 00:05:01,960
Speaker 6: minimally soft bullshitters or their bullshit machines.

97
00:05:03,400 --> 00:05:06,360
Speaker 2: So you also make the distinction in there the intention,

98
00:05:06,720 --> 00:05:09,400
Speaker 2: so the very fabric of hard bullshit that you intentionally

99
00:05:09,440 --> 00:05:12,880
Speaker 2: are bullshitting to someone, You kind of make this distinction

100
00:05:12,960 --> 00:05:17,520
Speaker 2: that the intention of the designer and the involved prompting

101
00:05:17,640 --> 00:05:21,719
Speaker 2: could make this hard bullshit. Because with a lot of

102
00:05:21,760 --> 00:05:25,480
Speaker 2: these models and someone recently jail broke chat GPT and

103
00:05:25,560 --> 00:05:28,000
Speaker 2: it listed all of the things that's prompted to do,

104
00:05:28,400 --> 00:05:32,280
Speaker 2: could prompting be considered intentional that Sam Altman, CEO of

105
00:05:32,560 --> 00:05:35,520
Speaker 2: open Ai, could be intentionally bullshitting I think it is.

106
00:05:35,680 --> 00:05:37,640
Speaker 6: And yeah, this again, I think it's something I don't

107
00:05:37,680 --> 00:05:40,159
Speaker 6: know what the kind of hive mind consensus on this is.

108
00:05:40,200 --> 00:05:42,599
Speaker 6: I'm sort of sympathetic to the idea that if you

109
00:05:42,640 --> 00:05:47,040
Speaker 6: take this kind of purposive or teleological attitude towards right,

110
00:05:47,440 --> 00:05:50,279
Speaker 6: what an intention is an effort to do something, then

111
00:05:50,279 --> 00:05:52,880
Speaker 6: maybe they do have intentions. But again, I think in

112
00:05:52,920 --> 00:05:54,360
Speaker 6: the paper we just sort of wanted I mean, it's

113
00:05:54,400 --> 00:05:57,560
Speaker 6: a standard philosophical move, right, just sort of go, look,

114
00:05:57,600 --> 00:05:59,960
Speaker 6: here's all this uncontroversial stuff as we can make it.

115
00:06:00,240 --> 00:06:02,120
Speaker 6: Now we can hit you with the really controversial shit

116
00:06:02,160 --> 00:06:04,279
Speaker 6: that we wanted to get to. So in the paper

117
00:06:04,320 --> 00:06:06,880
Speaker 6: we sort of deliberately when maybe you might think it

118
00:06:06,880 --> 00:06:09,280
Speaker 6: has intentions for it this reason, we kind of have

119
00:06:09,320 --> 00:06:12,120
Speaker 6: no judgment on this efficiently, I'm sympathetic to the sort

120
00:06:12,120 --> 00:06:12,920
Speaker 6: of view that you're putting.

121
00:06:12,960 --> 00:06:14,440
Speaker 5: I think you're kind of sympathetic this as well.

122
00:06:14,720 --> 00:06:17,200
Speaker 1: Right, Yeah, so, I Mike, there are a few ways

123
00:06:17,200 --> 00:06:20,440
Speaker 1: that you can think of CHATCHYBT as having intentions.

124
00:06:19,960 --> 00:06:22,400
Speaker 4: I think, and we talk about a few of them.

125
00:06:22,440 --> 00:06:25,840
Speaker 1: One is by thinking of the designers who created it

126
00:06:26,240 --> 00:06:28,799
Speaker 1: as kind of imbuing it with intentions. So they created

127
00:06:28,800 --> 00:06:31,280
Speaker 1: it for a purpose and to do a particular task,

128
00:06:32,000 --> 00:06:35,320
Speaker 1: and that task is to make people think that it's

129
00:06:35,360 --> 00:06:36,680
Speaker 1: a normal sounding person.

130
00:06:36,960 --> 00:06:37,120
Speaker 4: Right.

131
00:06:37,200 --> 00:06:40,280
Speaker 1: It's to make people, when they have a conversation with it,

132
00:06:40,400 --> 00:06:44,280
Speaker 1: not be able to distinguish between what it's outputting and

133
00:06:44,360 --> 00:06:46,200
Speaker 1: what a normal human would say.

134
00:06:46,520 --> 00:06:46,719
Speaker 4: Right.

135
00:06:47,120 --> 00:06:52,320
Speaker 1: Right, And that kind of goal, if it amounts to

136
00:06:52,360 --> 00:06:56,680
Speaker 1: an intention, is the kind of intention we think a

137
00:06:56,720 --> 00:06:57,520
Speaker 1: bullshitter has.

138
00:06:57,640 --> 00:06:57,800
Speaker 4: Right.

139
00:06:57,839 --> 00:07:00,440
Speaker 1: It's not trying to deceive you about anything it's saying.

140
00:07:00,480 --> 00:07:02,600
Speaker 1: It doesn't care whether what it's saying is true. That's

141
00:07:02,640 --> 00:07:04,760
Speaker 1: not part of the goal. But the goal is to

142
00:07:04,800 --> 00:07:07,559
Speaker 1: do is to make it seem as if it's something

143
00:07:07,600 --> 00:07:11,520
Speaker 1: that it's not, like specifically a human interlocator. And one

144
00:07:11,560 --> 00:07:14,720
Speaker 1: source for that goal is the programmers who designed it.

145
00:07:15,160 --> 00:07:19,080
Speaker 1: Another is the training method. So it was trained by

146
00:07:19,120 --> 00:07:22,240
Speaker 1: being given sort of positive and negative feedback in order

147
00:07:22,320 --> 00:07:26,320
Speaker 1: to achieve a specific thing, right, And that specific thing

148
00:07:26,920 --> 00:07:29,280
Speaker 1: is just sounding normal. And that's so similar to what

149
00:07:29,280 --> 00:07:31,680
Speaker 1: our students are doing when they try to pretend that

150
00:07:31,760 --> 00:07:33,800
Speaker 1: they read something they haven't read.

151
00:07:34,920 --> 00:07:37,760
Speaker 2: There is something very collegiate about the way it bullshits.

152
00:07:37,760 --> 00:07:37,960
Speaker 4: Though.

153
00:07:38,040 --> 00:07:40,320
Speaker 2: It reminds me of when I was in college, as

154
00:07:40,480 --> 00:07:43,880
Speaker 2: when Penn State and Aporistwyth two very different institutis.

155
00:07:43,520 --> 00:07:46,600
Speaker 4: Right, both kind of in the middle of nowhere.

156
00:07:47,560 --> 00:07:50,600
Speaker 2: Yeah, both very sad. But the one thing you saw

157
00:07:50,640 --> 00:07:53,440
Speaker 2: with like students who were doing like B plus homework

158
00:07:53,520 --> 00:07:55,840
Speaker 2: is they were using words they didn't really understand. They

159
00:07:55,840 --> 00:07:57,800
Speaker 2: were putting things together in a way that really is

160
00:07:57,920 --> 00:08:00,760
Speaker 2: like the intro body can clu usion. There was a

161
00:08:00,800 --> 00:08:03,680
Speaker 2: certain formula behind it, and it's just it feels exactly

162
00:08:03,800 --> 00:08:05,640
Speaker 2: like it. But that kind of brings me to my

163
00:08:05,680 --> 00:08:08,960
Speaker 2: next question, which is, how did you decide to write this?

164
00:08:09,040 --> 00:08:10,080
Speaker 2: What inspired this?

165
00:08:10,400 --> 00:08:10,560
Speaker 1: Right?

166
00:08:10,560 --> 00:08:12,640
Speaker 6: I would feel this because whenever Mike tells the story,

167
00:08:12,640 --> 00:08:13,960
Speaker 6: you get a sanitized version.

168
00:08:14,760 --> 00:08:17,720
Speaker 5: We were in the pub winsing about student essays. Perfect, yeah,

169
00:08:17,800 --> 00:08:21,080
Speaker 5: like you were not long here, right, you're not long yeah,

170
00:08:21,320 --> 00:08:22,720
Speaker 5: long post of.

171
00:08:22,640 --> 00:08:24,000
Speaker 6: A bunch of us sort of went to the pub

172
00:08:24,480 --> 00:08:26,800
Speaker 6: on a notional let's let's welcome our new members of

173
00:08:26,840 --> 00:08:29,600
Speaker 6: staff sort of event, and inevitably we had about two

174
00:08:29,600 --> 00:08:32,240
Speaker 6: points we were all passing and moaning and I think

175
00:08:32,240 --> 00:08:33,640
Speaker 6: it might have been Neil that kind of prompted this.

176
00:08:33,840 --> 00:08:37,200
Speaker 4: No, I don't think he was about it.

177
00:08:37,280 --> 00:08:39,040
Speaker 5: We talked about it after, right, Yeah, in a case

178
00:08:39,120 --> 00:08:39,760
Speaker 5: like sort of came out.

179
00:08:39,760 --> 00:08:41,440
Speaker 6: We were talking about this sort of prevalence of chat

180
00:08:41,559 --> 00:08:45,040
Speaker 6: GPT generated stuff and kind of what it said about

181
00:08:45,760 --> 00:08:48,719
Speaker 6: how at least some students were kind of approaching assessments

182
00:08:48,800 --> 00:08:50,400
Speaker 6: and you know, I forget who. Someone sort of just

183
00:08:50,440 --> 00:08:52,200
Speaker 6: went off hand, Yeah, but it's all just Frank thirty

184
00:08:52,240 --> 00:08:54,440
Speaker 6: and bullshit though, wasn't it. And we sort of collectively

185
00:08:54,480 --> 00:08:57,160
Speaker 6: went hello, because obviously we all having a background plosophy,

186
00:08:57,160 --> 00:08:59,079
Speaker 6: we've all read on bullshit. You know, we all went

187
00:08:59,600 --> 00:09:02,040
Speaker 6: we get to say bullshit and allegedly serious academic work.

188
00:09:02,400 --> 00:09:05,160
Speaker 6: So the start of it was we've had this experience

189
00:09:05,240 --> 00:09:07,959
Speaker 6: of having to read all of this kind of uncanny

190
00:09:08,080 --> 00:09:10,840
Speaker 6: Valley stuff, and when prompted, we all went, oh, it

191
00:09:10,960 --> 00:09:13,400
Speaker 6: really is like Frank thirteen bullshit in a way that

192
00:09:13,440 --> 00:09:14,760
Speaker 6: we can probably get paper ouse off.

193
00:09:14,800 --> 00:09:18,000
Speaker 1: Yeah, And at that point I think we were in

194
00:09:18,040 --> 00:09:22,840
Speaker 1: the department meetings talking about how to deal with chat

195
00:09:22,880 --> 00:09:26,160
Speaker 1: GPT written papers. And there were discussions going on all

196
00:09:26,160 --> 00:09:28,280
Speaker 1: over the university, including kind of in high levels and

197
00:09:28,320 --> 00:09:32,880
Speaker 1: the administration to come up with a policy. And we

198
00:09:32,920 --> 00:09:37,440
Speaker 1: specifically wanted a stronger policy because our university is very

199
00:09:37,440 --> 00:09:40,920
Speaker 1: interested in cutting into technology, right, and so they wanted

200
00:09:40,920 --> 00:09:43,839
Speaker 1: the students to have an experience using and figuring out

201
00:09:43,880 --> 00:09:47,360
Speaker 1: how to use whatever the freshest technology.

202
00:09:46,880 --> 00:09:49,680
Speaker 2: Is, right, as they should as they should, but not

203
00:09:50,280 --> 00:09:51,000
Speaker 2: for essays.

204
00:09:51,080 --> 00:09:53,240
Speaker 4: Right, that's the worry we had. We thought, you know,

205
00:09:53,720 --> 00:09:54,960
Speaker 4: if we're not very clear.

206
00:09:54,760 --> 00:09:56,560
Speaker 1: About this, the students will be using it in a

207
00:09:56,559 --> 00:09:59,840
Speaker 1: way that will detract from their educational experience.

208
00:10:00,559 --> 00:10:03,840
Speaker 4: Right. And at the same time, it was becoming more.

209
00:10:03,679 --> 00:10:07,280
Speaker 1: Widely known how these machines work, like specifically how they're

210
00:10:07,280 --> 00:10:11,480
Speaker 1: doing next token or next word prediction in order to

211
00:10:11,520 --> 00:10:15,559
Speaker 1: come up with a digestible string of text. And when

212
00:10:15,600 --> 00:10:18,400
Speaker 1: you know that that's how they're doing it, I mean,

213
00:10:19,559 --> 00:10:23,560
Speaker 1: it seems so similar to what humans do when they

214
00:10:23,600 --> 00:10:26,520
Speaker 1: have no idea what they're talking about. And so when

215
00:10:26,520 --> 00:10:29,280
Speaker 1: we were talking about it just seemed like an obvious

216
00:10:29,280 --> 00:10:31,400
Speaker 1: paper that someone was going to write, and we thought

217
00:10:31,440 --> 00:10:34,600
Speaker 1: that had better bs. You know, eventually people are going

218
00:10:34,640 --> 00:10:37,559
Speaker 1: to see this connection and it's a great papa.

219
00:10:37,679 --> 00:10:38,640
Speaker 4: I think it worked very well.

220
00:10:38,679 --> 00:10:40,400
Speaker 1: I mean, I think that it's the kind of thing

221
00:10:40,440 --> 00:10:44,120
Speaker 1: where when I pitch this to other philosophers, it doesn't

222
00:10:44,160 --> 00:10:47,280
Speaker 1: take them very long to just agree to be like, ah, yes.

223
00:10:47,520 --> 00:10:48,160
Speaker 5: That's right.

224
00:10:48,720 --> 00:10:52,559
Speaker 2: It's an interesting philosophical concept as well, because the way

225
00:10:52,600 --> 00:10:54,839
Speaker 2: that people look at chat GPT in large language models

226
00:10:54,920 --> 00:10:57,080
Speaker 2: is very much machine do this. But when you think

227
00:10:57,120 --> 00:10:59,840
Speaker 2: about it, there are other ways what people are giving

228
00:10:59,880 --> 00:11:02,120
Speaker 2: it consciousness. I just sort of tweet just now if

229
00:11:02,120 --> 00:11:05,520
Speaker 2: someone talk about saying please with every request, it's like, no,

230
00:11:05,679 --> 00:11:08,320
Speaker 2: I will abuse the computer in whatever manner I see fit.

231
00:11:08,760 --> 00:11:14,000
Speaker 2: But it's it's curious because I think more people need

232
00:11:14,040 --> 00:11:17,040
Speaker 2: to be having conversations like this. And one particular thing

233
00:11:17,080 --> 00:11:18,599
Speaker 2: I like that you said, I actually would love you

234
00:11:18,640 --> 00:11:20,760
Speaker 2: to go into more detail, is you said, the chat

235
00:11:20,760 --> 00:11:24,160
Speaker 2: GIPT in large language buddels, they're not designed to represent

236
00:11:24,200 --> 00:11:26,920
Speaker 2: the world at all. They're not lying or misrepresenting. They're

237
00:11:26,960 --> 00:11:28,640
Speaker 2: not designed to do that. What do you mean by that?

238
00:11:29,000 --> 00:11:32,000
Speaker 1: I mean kind of as background, I do philosophy of science,

239
00:11:32,160 --> 00:11:36,840
Speaker 1: but my thoughts about something like chatchipt are largely inspired

240
00:11:36,840 --> 00:11:38,760
Speaker 1: by the fact that I also teach a class called

241
00:11:39,480 --> 00:11:41,319
Speaker 1: Understanding Philosophy through Science Fiction.

242
00:11:41,559 --> 00:11:43,360
Speaker 4: Oh and and now we like talk.

243
00:11:43,160 --> 00:11:47,200
Speaker 1: About whether computers could be conscious, and I don't know

244
00:11:47,200 --> 00:11:50,000
Speaker 1: what you guys think. Actually I think they could, right,

245
00:11:51,360 --> 00:11:53,959
Speaker 1: I just don't think this one is. And part of

246
00:11:54,000 --> 00:11:56,240
Speaker 1: the reason I think they could, but this one isn't

247
00:11:56,440 --> 00:11:58,560
Speaker 1: is that I think that in order to sort of

248
00:11:58,600 --> 00:12:01,840
Speaker 1: represent the world or have the kinds of things we

249
00:12:01,920 --> 00:12:07,000
Speaker 1: have their like beliefs, desires, thoughts that are about external things,

250
00:12:07,920 --> 00:12:12,280
Speaker 1: you have to have internal states that are connected in

251
00:12:12,320 --> 00:12:16,640
Speaker 1: some way to the external world. Usually causation, we're perceiving things,

252
00:12:16,679 --> 00:12:19,280
Speaker 1: information is coming in. Then we've got some kind of

253
00:12:19,280 --> 00:12:23,080
Speaker 1: state in our brain that's designed just to track these

254
00:12:23,120 --> 00:12:24,360
Speaker 1: things in the external world.

255
00:12:24,600 --> 00:12:24,760
Speaker 3: Right.

256
00:12:25,280 --> 00:12:28,240
Speaker 1: That's a huge part of our cognitive lives is just

257
00:12:29,120 --> 00:12:33,160
Speaker 1: tracking external world things, and it's a very important part

258
00:12:33,200 --> 00:12:35,720
Speaker 1: of childhood development when you figure out how to act.

259
00:12:36,000 --> 00:12:40,480
Speaker 2: It's semiotics, right, Daniel Chanler Aperistwyth told me semiotics it's

260
00:12:40,520 --> 00:12:41,440
Speaker 2: like anception of the world.

261
00:12:41,520 --> 00:12:44,439
Speaker 1: And this is like theory of meaning stuff. So yeah,

262
00:12:44,600 --> 00:12:47,000
Speaker 1: semiotics is like theory of science. How is it that

263
00:12:47,040 --> 00:12:50,640
Speaker 1: assign It can be both the representation of the thing

264
00:12:50,760 --> 00:12:53,560
Speaker 1: and the thing itself. That can happen, yeah, but not always.

265
00:12:53,559 --> 00:12:57,040
Speaker 1: Sometimes it's just the representation of the thing. And there's

266
00:12:57,080 --> 00:13:00,760
Speaker 1: a lot of philosophy is about figuring out how brain

267
00:13:00,800 --> 00:13:04,320
Speaker 1: states or words on a page can be about external

268
00:13:04,360 --> 00:13:07,960
Speaker 1: world things. And a big part of it, at least

269
00:13:07,960 --> 00:13:11,720
Speaker 1: from my perspective, has to do with tracking those things,

270
00:13:12,280 --> 00:13:15,400
Speaker 1: keeping tabs on them, changing as a result of seeing

271
00:13:15,480 --> 00:13:19,360
Speaker 1: differences in the external world. And chatch ept is not

272
00:13:19,440 --> 00:13:22,520
Speaker 1: doing any of that, right, that's not what it's designed

273
00:13:22,520 --> 00:13:25,079
Speaker 1: to do. It's taking in a lot of data once

274
00:13:25,880 --> 00:13:30,080
Speaker 1: and then using that to respond to text. But it's

275
00:13:30,160 --> 00:13:36,120
Speaker 1: not remembering individuals, tracking things in the world in any

276
00:13:36,120 --> 00:13:40,160
Speaker 1: way perceiving things in the world. It's just forming a

277
00:13:40,200 --> 00:13:43,720
Speaker 1: sort of statistical model of what people say. And that's

278
00:13:43,800 --> 00:13:49,280
Speaker 1: kind of so divorced from what most thinking beings do.

279
00:13:50,080 --> 00:13:51,559
Speaker 2: It's divorced from experience.

280
00:13:51,840 --> 00:13:53,280
Speaker 4: Yeah. I mean, as far as I can tell, it

281
00:13:53,320 --> 00:13:55,160
Speaker 4: doesn't have anything like experience.

282
00:13:55,480 --> 00:13:55,680
Speaker 5: Yeah.

283
00:13:55,720 --> 00:13:57,640
Speaker 6: And that's one of the things that this, I think

284
00:13:57,800 --> 00:14:00,199
Speaker 6: of in one way, comes down to, is that if

285
00:14:00,200 --> 00:14:02,200
Speaker 6: you sort of post this sufficiently far, someone is going

286
00:14:02,240 --> 00:14:04,480
Speaker 6: to go, ah, isn't this just bioschauvinism, right, Like, aren't

287
00:14:04,520 --> 00:14:07,480
Speaker 6: you just assuming that unless something runs on like meat,

288
00:14:07,840 --> 00:14:10,040
Speaker 6: it can't be sentient? And this isn't something we get

289
00:14:10,040 --> 00:14:12,559
Speaker 6: into into the paper, partly because we didn't really think

290
00:14:12,559 --> 00:14:15,720
Speaker 6: it was worth addressing. But the sorts of things that

291
00:14:16,559 --> 00:14:19,000
Speaker 6: seem like they're like never mind consciousness, right, But it

292
00:14:19,080 --> 00:14:20,840
Speaker 6: seemed to be necessary in order for something to be

293
00:14:20,840 --> 00:14:22,960
Speaker 6: trying to track the world, or in order to be

294
00:14:22,960 --> 00:14:25,160
Speaker 6: corresponding the world, or to form beliefs about the world.

295
00:14:25,400 --> 00:14:28,480
Speaker 6: Chat GPTs just doesn'tthing to meet any of them. Kind Of,

296
00:14:28,560 --> 00:14:31,360
Speaker 6: it's if it does turn out that it's sapient, and

297
00:14:31,360 --> 00:14:37,840
Speaker 6: then chat gpt has got some profoundly serious executive function disorders, right, sapient, Right,

298
00:14:37,840 --> 00:14:39,240
Speaker 6: so we don't have to worry about it. But kind

299
00:14:39,280 --> 00:14:41,760
Speaker 6: of it's not the case that we've got some blundering

300
00:14:41,840 --> 00:14:44,160
Speaker 6: proto general intelligence that's trying to figure out how to

301
00:14:44,160 --> 00:14:46,000
Speaker 6: represent the world. It's not trying to represent the world

302
00:14:46,000 --> 00:14:48,560
Speaker 6: at all. This utterances are designed to look as if

303
00:14:48,880 --> 00:14:50,560
Speaker 6: it's trying to represent the world, and then we just

304
00:14:50,560 --> 00:14:52,840
Speaker 6: go that that's just bullshit. This is a classic case

305
00:14:52,840 --> 00:14:53,400
Speaker 6: of bullshit.

306
00:14:53,800 --> 00:14:56,200
Speaker 2: Yeah, it seems to be making stuff up, but making

307
00:14:56,240 --> 00:14:58,400
Speaker 2: stuff up doesn't even seem to describe it. It's just

308
00:14:58,760 --> 00:15:02,040
Speaker 2: throwing shit at a wall accurately but not accurately enough.

309
00:15:02,080 --> 00:15:04,320
Speaker 6: Yeah, it's got various guidelines that allow it to throw

310
00:15:04,360 --> 00:15:05,920
Speaker 6: shit at the wall where a sort of reasonably high

311
00:15:05,960 --> 00:15:07,200
Speaker 6: degree of Actually, no, you're right, I mean one of

312
00:15:07,200 --> 00:15:09,280
Speaker 6: the things that a human bullshitter could at least be

313
00:15:09,360 --> 00:15:11,840
Speaker 6: characterized as doing is. They'd have to try and kind

314
00:15:11,840 --> 00:15:14,280
Speaker 6: of judge their audience, right, they'd have to try and

315
00:15:14,320 --> 00:15:16,960
Speaker 6: make the bullshit plausible to the audience that they're speaking to,

316
00:15:17,200 --> 00:15:19,320
Speaker 6: and chat GPT can do that, right. All it can

317
00:15:19,360 --> 00:15:22,120
Speaker 6: do is go on a statistically large enough model and

318
00:15:23,000 --> 00:15:26,960
Speaker 6: it looks like Z follows, why follows X. It's not

319
00:15:27,040 --> 00:15:29,440
Speaker 6: kind of got any or doesn't have any consciousness at all,

320
00:15:29,480 --> 00:15:32,080
Speaker 6: of course, but it doesn't have any sensitivity to the

321
00:15:32,160 --> 00:15:34,600
Speaker 6: sorts of things that people are in fact likely to

322
00:15:34,600 --> 00:15:37,560
Speaker 6: find plausible. It just does a kind of brute number crunch.

323
00:15:37,760 --> 00:15:39,880
Speaker 6: So it's more complicated than that, but I think it

324
00:15:40,120 --> 00:15:42,240
Speaker 6: boils down it effectively to kind of number crunching, right,

325
00:15:42,360 --> 00:15:44,000
Speaker 6: kind of data and context.

326
00:15:43,720 --> 00:15:47,480
Speaker 2: In which probablistic planning of what planning it doesn't plan.

327
00:15:47,720 --> 00:15:49,800
Speaker 2: That's the thing. It's interesting. There are these words you

328
00:15:49,920 --> 00:15:51,920
Speaker 2: use to describe things that, when you think about it,

329
00:15:51,960 --> 00:15:55,640
Speaker 2: are not accurate. You can't say it plans or things.

330
00:15:56,480 --> 00:15:59,280
Speaker 1: I think one thing between when we came up with

331
00:15:59,360 --> 00:16:02,920
Speaker 1: the idea and when we finish writing the paper, we

332
00:16:03,000 --> 00:16:06,000
Speaker 1: spend some time reading about how it works and how

333
00:16:06,000 --> 00:16:09,040
Speaker 1: it represents language and what the statistical.

334
00:16:08,400 --> 00:16:09,320
Speaker 4: Model is like.

335
00:16:09,640 --> 00:16:15,680
Speaker 1: And I was maybe more impressed than James about that,

336
00:16:15,760 --> 00:16:19,240
Speaker 1: because it is like doing things that are similar to

337
00:16:19,320 --> 00:16:22,240
Speaker 1: what we do when we understand language. It does seem

338
00:16:22,280 --> 00:16:26,440
Speaker 1: to kind of locate words in a meaning space right right,

339
00:16:26,840 --> 00:16:29,200
Speaker 1: and connect them to other words, and you know, show

340
00:16:29,240 --> 00:16:32,000
Speaker 1: similarity and meaning, and it also does seem to be

341
00:16:32,000 --> 00:16:35,000
Speaker 1: able to in some way understand context.

342
00:16:35,760 --> 00:16:38,720
Speaker 4: But we don't know how.

343
00:16:38,440 --> 00:16:41,560
Speaker 1: Similar that is for a variety of reasons, but mostly

344
00:16:41,640 --> 00:16:44,320
Speaker 1: because it's too big of a system and we can

345
00:16:44,360 --> 00:16:48,360
Speaker 1: only kind of probe it. And it's trained indirectly, right,

346
00:16:48,520 --> 00:16:54,040
Speaker 1: so it's not programmed by individuals. And even though that's

347
00:16:54,120 --> 00:16:56,760
Speaker 1: kind of a very impressive model of language and meaning

348
00:16:57,000 --> 00:17:00,240
Speaker 1: and may in some ways be similar to or what

349
00:17:00,320 --> 00:17:04,440
Speaker 1: we do when we understand language, we're doing a lot

350
00:17:04,520 --> 00:17:08,960
Speaker 1: more things like planning, things like tracking things in the world.

351
00:17:09,320 --> 00:17:12,119
Speaker 1: Just having desires and representing the way you want the

352
00:17:12,160 --> 00:17:15,280
Speaker 1: world to be and thereby generating goals doesn't seem to

353
00:17:15,320 --> 00:17:18,000
Speaker 1: be something that it has anything any room for in its.

354
00:17:17,960 --> 00:17:19,679
Speaker 6: Architect So this is something of the time of it's

355
00:17:19,680 --> 00:17:21,000
Speaker 6: just goods to me, and you were talking about the

356
00:17:21,080 --> 00:17:23,200
Speaker 6: kind of in some ways it learns language the same

357
00:17:23,200 --> 00:17:24,919
Speaker 6: way that we do. I mean, it's got no grasp

358
00:17:24,920 --> 00:17:27,240
Speaker 6: of expleted in fixation, right, this is one of chum's here.

359
00:17:27,280 --> 00:17:29,520
Speaker 2: What does that mean? Just for not for me? I

360
00:17:29,560 --> 00:17:30,159
Speaker 2: definitely know.

361
00:17:30,640 --> 00:17:33,480
Speaker 6: Yeah, yeah, of course, if I give you the sentence

362
00:17:33,920 --> 00:17:36,320
Speaker 6: that's completely crazy, man, and tell you to put the

363
00:17:36,320 --> 00:17:39,320
Speaker 6: word fucking into that sentence, there's a number of ways

364
00:17:39,359 --> 00:17:41,280
Speaker 6: in which any language speaker is going to do it

365
00:17:41,560 --> 00:17:43,120
Speaker 6: that they'll just go yeah, like of course.

366
00:17:42,920 --> 00:17:44,920
Speaker 5: That where it goes, right.

367
00:17:45,280 --> 00:17:47,720
Speaker 6: But we it seems that we've got a grasp on

368
00:17:47,760 --> 00:17:50,760
Speaker 6: this incredibly early on in a way that doesn't look

369
00:17:50,840 --> 00:17:52,280
Speaker 6: like it's the ways at least most of us that

370
00:17:52,400 --> 00:17:54,840
Speaker 6: taught language, right, we get quite highly told off when

371
00:17:54,840 --> 00:17:55,280
Speaker 6: we try and do.

372
00:17:55,280 --> 00:17:56,200
Speaker 5: Expleted in fixation.

373
00:17:56,400 --> 00:17:58,520
Speaker 6: Yes, So this I think would be one of those

374
00:17:58,600 --> 00:18:01,080
Speaker 6: cases where you could do a sort of dis analogy

375
00:18:01,160 --> 00:18:03,760
Speaker 6: by cases. Right, you present chat GPT a sentence and

376
00:18:03,800 --> 00:18:07,040
Speaker 6: say insert the word fucking correctly in this sentence, and

377
00:18:07,119 --> 00:18:08,679
Speaker 6: I doesn't think it would be very good at it.

378
00:18:08,720 --> 00:18:09,359
Speaker 4: I think it would be.

379
00:18:09,400 --> 00:18:11,119
Speaker 5: I think you reckon. I mean we probably couldn't.

380
00:18:11,840 --> 00:18:14,880
Speaker 4: We could, but we shouldn't. Yeah.

381
00:18:15,359 --> 00:18:19,080
Speaker 1: One of the things that I thought was like kind

382
00:18:19,080 --> 00:18:23,280
Speaker 1: of interesting about how it works is that it does

383
00:18:23,400 --> 00:18:25,720
Speaker 1: learn language, probably differently from the way we do, but

384
00:18:25,800 --> 00:18:28,560
Speaker 1: it does it all by examples, you know, So it's

385
00:18:28,560 --> 00:18:30,680
Speaker 1: looking at all these pieces of text and thinking, ah,

386
00:18:30,680 --> 00:18:33,440
Speaker 1: this is okay, that's okay. And one kind of interesting

387
00:18:33,480 --> 00:18:36,480
Speaker 1: thing about how humans understand language is that we're able

388
00:18:36,520 --> 00:18:41,280
Speaker 1: to kind of understand meaningless but grammatical sentences. It's not

389
00:18:41,320 --> 00:18:44,600
Speaker 1: clear to me that chat GPT would understand those. That's

390
00:18:44,640 --> 00:18:49,200
Speaker 1: another Chopski example. So you know, Chowski has this example that's.

391
00:18:49,000 --> 00:18:53,600
Speaker 5: Like, what is it colorless? Us sleep furiously color?

392
00:18:53,640 --> 00:18:58,840
Speaker 1: It's colorless green ideas, right, And that's a meaningless sentence,

393
00:18:58,840 --> 00:19:02,119
Speaker 1: but it's grammatically well formed, and we can understand that

394
00:19:02,119 --> 00:19:06,119
Speaker 1: it's grammatically well formed, but also that it's meaningless. Because

395
00:19:06,200 --> 00:19:12,359
Speaker 1: chat Gibt kind of combines different aspects of what philosophers

396
00:19:12,359 --> 00:19:16,800
Speaker 1: of language, logicians, linguistics people see y as like different

397
00:19:16,800 --> 00:19:19,199
Speaker 1: components of meaning. It sees these as all kind of

398
00:19:19,200 --> 00:19:20,960
Speaker 1: wrapped up in the same thing. It puts them in

399
00:19:21,000 --> 00:19:24,040
Speaker 1: the same big model. I'm not sure it could differentiate

400
00:19:24,040 --> 00:19:29,120
Speaker 1: between ungrammaticality and meaninglessness.

401
00:19:36,880 --> 00:19:39,720
Speaker 2: So we're doing real time science right now. I just yeah,

402
00:19:39,720 --> 00:19:42,440
Speaker 2: we should check the word shoe into the following sentence

403
00:19:42,480 --> 00:19:45,680
Speaker 2: in the correct way. Man, that's crazy. And I did

404
00:19:45,720 --> 00:19:49,040
Speaker 2: it six times, and I would say fifty percent of

405
00:19:49,080 --> 00:19:51,360
Speaker 2: the time it got it right. And it did. Man,

406
00:19:51,400 --> 00:19:54,520
Speaker 2: that's fucking crazy. Man, that's crazy. Fucking Man, that's fucking crazy. Man,

407
00:19:54,560 --> 00:19:58,959
Speaker 2: that's crazy. Fucking My favorite is man, comma, that's crazy,

408
00:19:59,080 --> 00:20:00,080
Speaker 2: Comma fucking.

409
00:20:01,240 --> 00:20:02,200
Speaker 5: To be fair.

410
00:20:03,840 --> 00:20:06,080
Speaker 6: Reliable, you take the commas out of that last one,

411
00:20:06,160 --> 00:20:07,720
Speaker 6: and you've got a grammatical sentence.

412
00:20:07,880 --> 00:20:08,160
Speaker 2: Yeah.

413
00:20:08,200 --> 00:20:11,080
Speaker 6: Of course in Glasgow you can also start with fucking man,

414
00:20:11,119 --> 00:20:12,560
Speaker 6: that's crazy.

415
00:20:12,119 --> 00:20:14,359
Speaker 2: West London as well. But the thing is, though it

416
00:20:14,400 --> 00:20:15,960
Speaker 2: doesn't know what correct means there.

417
00:20:16,200 --> 00:20:16,400
Speaker 4: Yeah.

418
00:20:16,480 --> 00:20:19,919
Speaker 2: Now, when it's trained on this language, when it's trained

419
00:20:19,920 --> 00:20:24,520
Speaker 2: on thousands of Internet posts that stole, it's not like

420
00:20:24,600 --> 00:20:27,480
Speaker 2: it reads them and says, oh, I get this, like

421
00:20:27,520 --> 00:20:29,719
Speaker 2: I see what they're going for. It just learns structures

422
00:20:29,760 --> 00:20:33,920
Speaker 2: by looking, which is kind of how we learn language.

423
00:20:34,440 --> 00:20:36,040
Speaker 2: But it kind of reminds me of like when I

424
00:20:36,080 --> 00:20:37,920
Speaker 2: was a kid and i'd hear someone say something funny,

425
00:20:37,920 --> 00:20:40,400
Speaker 2: I'd to repeat it, and my dad, who's wonderful, would

426
00:20:40,400 --> 00:20:42,439
Speaker 2: just say, that doesn't make any sense, and you'd have

427
00:20:42,520 --> 00:20:45,640
Speaker 2: to explain because if you're learning everything through copying. You're

428
00:20:45,680 --> 00:20:47,440
Speaker 2: not learning, you're just memorizing.

429
00:20:48,320 --> 00:20:49,680
Speaker 5: Yeah, yeah, exactly.

430
00:20:49,840 --> 00:20:52,439
Speaker 1: There's a I don't know if you have already talked

431
00:20:52,560 --> 00:20:55,399
Speaker 1: to somebody about this, but there's a you know, a

432
00:20:55,400 --> 00:20:58,240
Speaker 1: classic argument from child Skyt against behaviorism.

433
00:20:58,400 --> 00:20:58,720
Speaker 2: Right.

434
00:20:58,880 --> 00:21:03,120
Speaker 1: Behaviorism is the view that we learn everything through stimulus

435
00:21:03,119 --> 00:21:03,919
Speaker 1: and response.

436
00:21:04,200 --> 00:21:05,879
Speaker 4: Roughly, that's not exactly.

437
00:21:05,480 --> 00:21:07,040
Speaker 1: It, but I'm not a philosopher of mine, so I

438
00:21:07,040 --> 00:21:09,360
Speaker 1: can get away with with that.

439
00:21:10,000 --> 00:21:12,840
Speaker 4: So Chomsky says, look, we don't get enough.

440
00:21:12,680 --> 00:21:17,159
Speaker 1: Stimulus to learn language as quickly as we do just

441
00:21:17,320 --> 00:21:20,720
Speaker 1: through watching other people's behavior and copying it. We have

442
00:21:20,760 --> 00:21:25,240
Speaker 1: to have some in built grammatical structures that language is

443
00:21:25,320 --> 00:21:29,480
Speaker 1: latching onto. And there's been some papers arguing that chatchipt

444
00:21:29,680 --> 00:21:32,160
Speaker 1: shows Chomsky was wrong because it doesn't.

445
00:21:31,880 --> 00:21:34,120
Speaker 4: Have the inbuilt grammatical structure.

446
00:21:34,240 --> 00:21:37,040
Speaker 1: But one interesting thing is it requires ten to one

447
00:21:37,119 --> 00:21:39,960
Speaker 1: hundred times more data than a human child does when

448
00:21:40,040 --> 00:21:42,840
Speaker 1: learning language. Right, So Chomsky's argument was, we don't get

449
00:21:42,920 --> 00:21:47,600
Speaker 1: enough stimulus, and chat GPT can kind of do it

450
00:21:47,640 --> 00:21:51,560
Speaker 1: without the structure, but it's not quite doing it as well,

451
00:21:51,800 --> 00:21:57,480
Speaker 1: and it gets like orders of magnitude more input than

452
00:21:57,520 --> 00:21:58,320
Speaker 1: a human does.

453
00:21:58,720 --> 00:22:02,120
Speaker 4: Before a human learns language, which is kind of interesting.

454
00:22:01,720 --> 00:22:04,320
Speaker 6: And it's still going to do something as basically as

455
00:22:04,320 --> 00:22:07,360
Speaker 6: putting the word yeah yeah.

456
00:22:07,359 --> 00:22:09,520
Speaker 2: It doesn't even see and it doesn't have the knowledge

457
00:22:09,520 --> 00:22:13,520
Speaker 2: to say, request more context because it doesn't perceive context.

458
00:22:13,520 --> 00:22:15,000
Speaker 2: And that's kind of the interesting thing. So there was

459
00:22:15,000 --> 00:22:18,000
Speaker 2: another paper out of Oxford, I think that's talking about

460
00:22:18,000 --> 00:22:22,080
Speaker 2: cognition and chat GPT and all this thing, and it's

461
00:22:22,160 --> 00:22:25,080
Speaker 2: just it doesn't feat chat GPT features in no way

462
00:22:25,200 --> 00:22:27,200
Speaker 2: any of the things that the human mind is really

463
00:22:27,200 --> 00:22:30,560
Speaker 2: involved in. It seems it's mostly just not even memorization,

464
00:22:30,600 --> 00:22:36,159
Speaker 2: because it doesn't memorize. It's just guessing based on a

465
00:22:36,280 --> 00:22:38,920
Speaker 2: very blooge pile of stuff. But this actually does lead

466
00:22:38,920 --> 00:22:40,840
Speaker 2: me to marvel question, which is, you don't like the

467
00:22:40,920 --> 00:22:44,480
Speaker 2: term hallucination. Why is that hallucination?

468
00:22:45,119 --> 00:22:48,480
Speaker 3: It makes it sound a bit like I'm usually doing

469
00:22:48,520 --> 00:22:52,000
Speaker 3: something right and I'm looking around seeing the world.

470
00:22:52,040 --> 00:22:53,639
Speaker 4: That's something like what it really is.

471
00:22:54,400 --> 00:22:57,280
Speaker 3: And then one little bit of the feature for a

472
00:22:57,359 --> 00:23:02,080
Speaker 3: visual hallucination, one feature of my visual field actually isn't

473
00:23:02,119 --> 00:23:05,320
Speaker 3: represented in the real world. It's not actually there. Everything

474
00:23:05,320 --> 00:23:08,159
Speaker 3: else might well might well be right. Imagine I hallucinate

475
00:23:08,200 --> 00:23:10,840
Speaker 3: there's a red balloon in front of me. I still

476
00:23:10,840 --> 00:23:13,360
Speaker 3: see Mike, I still see James, I still see the laptop.

477
00:23:14,080 --> 00:23:17,080
Speaker 3: One bit is wrong, everything else is right. And I'm

478
00:23:17,080 --> 00:23:19,399
Speaker 3: doing the same thing that I'm usually doing, Like my

479
00:23:19,480 --> 00:23:22,919
Speaker 3: eyes are still working pretty much normally, I think, I

480
00:23:23,080 --> 00:23:25,560
Speaker 3: and this is the way that I usually get knowledge

481
00:23:25,560 --> 00:23:28,480
Speaker 3: about the world. This is a pretty reliable process for

482
00:23:28,640 --> 00:23:33,119
Speaker 3: me learning from it right and representing the world in

483
00:23:33,160 --> 00:23:37,720
Speaker 3: this way. So when talking about hallucinations, this suggests that

484
00:23:38,520 --> 00:23:42,440
Speaker 3: chat Gypt and other similar things, they're going through this

485
00:23:42,560 --> 00:23:46,840
Speaker 3: process that is usually quite good at representing the world

486
00:23:46,960 --> 00:23:49,480
Speaker 3: and then oh, it's made a mistake this one time,

487
00:23:50,359 --> 00:23:53,959
Speaker 3: but actually no, it's bullshitting the whole time, and like

488
00:23:54,040 --> 00:23:57,840
Speaker 3: sometimes it gets things right right bullshitting just like a politician.

489
00:23:58,160 --> 00:24:02,159
Speaker 3: Imagine a politician that bullshits all the time, if you

490
00:24:02,200 --> 00:24:05,320
Speaker 3: could possibly imagine it. Sometimes they might just get some

491
00:24:05,400 --> 00:24:10,080
Speaker 3: things true, and we should still call them bullshitting because

492
00:24:10,080 --> 00:24:12,760
Speaker 3: that's what they're doing. And this is what a GPT

493
00:24:12,960 --> 00:24:16,679
Speaker 3: is doing every time it produces an output. So this

494
00:24:16,800 --> 00:24:18,679
Speaker 3: is why we think bullshit is a better way of

495
00:24:18,680 --> 00:24:21,119
Speaker 3: thinking about this. A one of the reasons why we

496
00:24:21,200 --> 00:24:22,679
Speaker 3: think bullshit's a better way think about it.

497
00:24:23,119 --> 00:24:25,760
Speaker 1: I also kind of think that some of the ways

498
00:24:25,760 --> 00:24:30,120
Speaker 1: we talk about chat gpt lin, even when it makes mistakes,

499
00:24:30,240 --> 00:24:32,520
Speaker 1: lind themselves to overhyping its.

500
00:24:32,440 --> 00:24:35,240
Speaker 4: Ability or overestimating its.

501
00:24:35,040 --> 00:24:37,280
Speaker 1: Abilities, and talking about it is hallucinating is one of

502
00:24:37,320 --> 00:24:41,400
Speaker 1: these because when you say that it's hallucinating, as you're

503
00:24:41,440 --> 00:24:45,359
Speaker 1: pointed out, you're giving the idea that it's representing the

504
00:24:45,359 --> 00:24:47,840
Speaker 1: world in some way and then telling you what the

505
00:24:47,920 --> 00:24:48,520
Speaker 1: content and.

506
00:24:48,520 --> 00:24:51,720
Speaker 6: It has perception, Yeah, exactly like it has perceived something

507
00:24:51,840 --> 00:24:54,560
Speaker 6: and like, oh no, it's taken some computer acid and

508
00:24:54,600 --> 00:24:57,080
Speaker 6: now it's hallucinating like mattery things.

509
00:24:57,480 --> 00:24:59,240
Speaker 5: Yeah, just as you say, And.

510
00:24:59,200 --> 00:25:00,280
Speaker 4: That's not what it's doing.

511
00:25:00,320 --> 00:25:04,680
Speaker 1: And so when the kind of people who are trying

512
00:25:04,720 --> 00:25:08,560
Speaker 1: to promote these these things as products talk about the

513
00:25:08,600 --> 00:25:12,520
Speaker 1: AI hallucination problem, they're kind of selling a product that

514
00:25:12,920 --> 00:25:17,399
Speaker 1: is a product that's representing the world usually checking things

515
00:25:17,960 --> 00:25:22,359
Speaker 1: and occasionally makes mistakes. And if the mistakes were, like

516
00:25:22,440 --> 00:25:26,080
Speaker 1: Joe said, we're a misfiring of a normally a reliable

517
00:25:26,160 --> 00:25:31,400
Speaker 1: process or you know, something that normally represents going wrong

518
00:25:31,480 --> 00:25:34,720
Speaker 1: in some way, that would lend itself to certain solutions

519
00:25:34,760 --> 00:25:36,119
Speaker 1: to them, and it would make you think there's an

520
00:25:36,160 --> 00:25:40,240
Speaker 1: underlying reliable product here, right, which is exactly what somebody

521
00:25:40,240 --> 00:25:43,240
Speaker 1: who's making a product to go to the market.

522
00:25:43,080 --> 00:25:44,240
Speaker 4: Will want you to think. Right.

523
00:25:44,720 --> 00:25:47,040
Speaker 1: But if that's not what it's doing in a certain sense,

524
00:25:47,080 --> 00:25:50,720
Speaker 1: they're misrepresenting what it's doing even when it gets things right.

525
00:25:51,480 --> 00:25:54,560
Speaker 4: And that's that's for all of us who.

526
00:25:54,520 --> 00:25:57,040
Speaker 1: Are going to be using these systems, especially since people,

527
00:25:57,800 --> 00:26:00,320
Speaker 1: you know, most people don't know how this works. They're

528
00:26:00,359 --> 00:26:03,400
Speaker 1: just understanding the product as it's described to them using

529
00:26:03,440 --> 00:26:06,960
Speaker 1: these kind of metaphors. So the way the metaphor describes

530
00:26:07,000 --> 00:26:09,200
Speaker 1: it is going to really influence how they think about

531
00:26:09,240 --> 00:26:09,520
Speaker 1: it and.

532
00:26:09,480 --> 00:26:10,200
Speaker 4: How they use it.

533
00:26:11,440 --> 00:26:13,280
Speaker 5: Yeah, just to sort of catfit off if I can.

534
00:26:13,440 --> 00:26:16,439
Speaker 6: Is one of the responses to some corners has been

535
00:26:16,480 --> 00:26:18,840
Speaker 6: to sort of say, of us, look, you wings about

536
00:26:18,880 --> 00:26:21,800
Speaker 6: people at anthropomorphosizing chat GPT, but look, if you've got

537
00:26:21,840 --> 00:26:24,760
Speaker 6: a bullshitty, you're doing exactly the same thing. And I

538
00:26:24,760 --> 00:26:26,480
Speaker 6: mean there might be some center which it's just really

539
00:26:26,520 --> 00:26:28,639
Speaker 6: hard not to antropomorphise it. I don't know why I

540
00:26:28,640 --> 00:26:31,760
Speaker 6: picked the word that I can barely say, like we've

541
00:26:31,760 --> 00:26:33,960
Speaker 6: been doing it constantly throughout this discussion, right when we're

542
00:26:34,000 --> 00:26:35,920
Speaker 6: talking through the kind of through the paper we kept

543
00:26:35,920 --> 00:26:38,280
Speaker 6: talking about chat GPT as if it had intention, right,

544
00:26:38,480 --> 00:26:41,119
Speaker 6: as if it was thinking about anything. That might be

545
00:26:41,160 --> 00:26:42,920
Speaker 6: another reason to call it bullshet We go, Look, if

546
00:26:42,920 --> 00:26:45,119
Speaker 6: we have to treat it as if it's doing something

547
00:26:45,240 --> 00:26:48,200
Speaker 6: like what we do, it's not hallucinating, it's not lying,

548
00:26:48,240 --> 00:26:50,479
Speaker 6: it's not confabulating, it's bullshitting, right.

549
00:26:50,520 --> 00:26:52,000
Speaker 5: And if we have to treat it as if.

550
00:26:51,920 --> 00:26:54,600
Speaker 6: It's behaving in some kind of human like way, here's

551
00:26:54,640 --> 00:26:56,639
Speaker 6: the appropriate human like behavior to describe it.

552
00:26:56,840 --> 00:26:59,000
Speaker 2: I also think the language in this case, and one

553
00:26:59,040 --> 00:27:01,440
Speaker 2: of the reasons they probably like the large language model

554
00:27:01,720 --> 00:27:05,600
Speaker 2: concept is language gives life too things when we describe

555
00:27:05,600 --> 00:27:07,840
Speaker 2: the processes through which we interact with the world and

556
00:27:07,880 --> 00:27:12,080
Speaker 2: interact with living beings, even cats, dogs, even we anthropomoise

557
00:27:12,200 --> 00:27:17,080
Speaker 2: living things. But also when we communicate with something, languages life,

558
00:27:17,240 --> 00:27:20,080
Speaker 2: and so it probably works out really fucking well for them.

559
00:27:20,560 --> 00:27:22,640
Speaker 2: Sam Oltman was saying a couple of weeks back, maybe

560
00:27:22,640 --> 00:27:24,680
Speaker 2: a month or two. He was saying, oh, yeah, AI

561
00:27:24,840 --> 00:27:27,840
Speaker 2: is not a creature with something he said, and it

562
00:27:27,880 --> 00:27:29,920
Speaker 2: was just so obvious. What he wanted people to do

563
00:27:30,000 --> 00:27:32,480
Speaker 2: was say, but what if it was? Or are people

564
00:27:32,560 --> 00:27:35,520
Speaker 2: saying this is a creature and it almost feels like

565
00:27:35,680 --> 00:27:37,720
Speaker 2: just part of the con Yeah.

566
00:27:37,880 --> 00:27:41,480
Speaker 1: Yeah, yeah, I hadn't thought about that as a reason

567
00:27:41,520 --> 00:27:44,520
Speaker 1: for them to go for large language models as a

568
00:27:44,560 --> 00:27:48,280
Speaker 1: way of kind of, I don't know, being the gateway

569
00:27:48,440 --> 00:27:54,880
Speaker 1: into war investment in fight consciousness. Yeah yeah, but I

570
00:27:54,920 --> 00:27:57,840
Speaker 1: had thought about, like how this might have been caused

571
00:27:57,880 --> 00:28:00,760
Speaker 1: by just like deep misunderstandings of the turning test.

572
00:28:01,720 --> 00:28:03,600
Speaker 2: Right, go ahead, I want to hear this one.

573
00:28:04,040 --> 00:28:04,280
Speaker 4: Yeah.

574
00:28:04,400 --> 00:28:07,320
Speaker 1: Yeah, so that like the turning test, I think this

575
00:28:07,359 --> 00:28:10,080
Speaker 1: is closer to what Turing was thinking. But the turning

576
00:28:10,119 --> 00:28:13,600
Speaker 1: test is a way of getting evidence that something is conscious. Right,

577
00:28:13,680 --> 00:28:17,680
Speaker 1: So you know, I'm not in your head, so I

578
00:28:17,760 --> 00:28:21,680
Speaker 1: can't feel your feelings or think your thoughts directly, right,

579
00:28:22,160 --> 00:28:25,320
Speaker 1: I have to judge whether you're conscious based on how

580
00:28:25,320 --> 00:28:27,239
Speaker 1: you interact with me, And the way I do it

581
00:28:27,280 --> 00:28:31,159
Speaker 1: is by listening to what you say, right, and talking

582
00:28:31,200 --> 00:28:33,960
Speaker 1: to you and turning sort.

583
00:28:33,720 --> 00:28:37,160
Speaker 4: Of was asked, you know, how would you know if

584
00:28:37,160 --> 00:28:38,240
Speaker 4: a computer was conscious?

585
00:28:38,280 --> 00:28:40,880
Speaker 1: So, you know, we think that our brains are doing

586
00:28:40,960 --> 00:28:43,640
Speaker 1: something similar to what computers do. That's a reason to

587
00:28:43,680 --> 00:28:47,120
Speaker 1: think that maybe computers eventually could have thoughts.

588
00:28:46,800 --> 00:28:50,040
Speaker 4: Like ours, right, and some of us. I think that

589
00:28:50,240 --> 00:28:51,040
Speaker 4: I think it's possible.

590
00:28:51,880 --> 00:28:54,200
Speaker 1: Great, Yeah, I didn't know if they thought it was possible,

591
00:28:54,200 --> 00:28:55,520
Speaker 1: because not everybody thinks it's possible.

592
00:28:55,560 --> 00:28:57,160
Speaker 2: It's possible, I just don't know how.

593
00:28:57,320 --> 00:29:01,600
Speaker 4: Yeah. Yeah. The Turing was kind of you know, thinking

594
00:29:01,640 --> 00:29:02,520
Speaker 4: like how would we know?

595
00:29:02,880 --> 00:29:05,560
Speaker 1: And one way we would know, the obvious way is

596
00:29:05,600 --> 00:29:08,160
Speaker 1: to do the same thing you do to humans, talk

597
00:29:08,240 --> 00:29:12,360
Speaker 1: to it and see how it responds. And that's actually

598
00:29:12,400 --> 00:29:15,720
Speaker 1: pretty good evidence if you don't have the ability to

599
00:29:16,040 --> 00:29:20,120
Speaker 1: look deeper. But it's not constitutive of being conscious. It's

600
00:29:20,160 --> 00:29:23,640
Speaker 1: not what makes something conscious or determines whether they're conscious

601
00:29:24,240 --> 00:29:27,760
Speaker 1: or in any way like grounds their consciousness. Right, Their

602
00:29:27,800 --> 00:29:30,440
Speaker 1: ability to talk to you is just evidence. It's just

603
00:29:30,560 --> 00:29:33,640
Speaker 1: one signal you can get. And that's the way to

604
00:29:33,640 --> 00:29:36,800
Speaker 1: think of the Turing test. So as a result of

605
00:29:36,840 --> 00:29:39,920
Speaker 1: people thinking in a kind of behaviorist way, thinking, ah,

606
00:29:40,000 --> 00:29:42,520
Speaker 1: passing the Turning test is just all it is to

607
00:29:42,560 --> 00:29:45,800
Speaker 1: be a thinking thing. There have been at least since

608
00:29:45,840 --> 00:29:48,920
Speaker 1: the nineties, attempts to design chatbots that can beat the

609
00:29:48,920 --> 00:29:52,960
Speaker 1: Turning test right, and popularizations of these attempts and run

610
00:29:53,000 --> 00:29:55,520
Speaker 1: throughs of the Turing test that talk as if, oh,

611
00:29:55,520 --> 00:29:57,680
Speaker 1: if a computer finally beats the Turing test.

612
00:29:58,120 --> 00:29:59,800
Speaker 4: I should say what the Turing test is? Right?

613
00:30:00,600 --> 00:30:02,680
Speaker 1: The way turn suggests that the test works is you

614
00:30:02,760 --> 00:30:07,320
Speaker 1: have a computer and a person both chatting in some

615
00:30:07,480 --> 00:30:10,840
Speaker 1: way with a judge, and the judge is also a person, right,

616
00:30:10,880 --> 00:30:12,840
Speaker 1: And if the judge can't tell which of the people

617
00:30:12,840 --> 00:30:16,560
Speaker 1: he's chatting with is a human, then the computer's one

618
00:30:16,760 --> 00:30:19,040
Speaker 1: the right test because it's ditinguishable from a human.

619
00:30:19,120 --> 00:30:19,280
Speaker 4: Right.

620
00:30:19,600 --> 00:30:22,640
Speaker 1: So people have taken this and it's been popularized as

621
00:30:22,640 --> 00:30:26,800
Speaker 1: a way of determining sort of full stop, whether something's conscious.

622
00:30:27,480 --> 00:30:29,200
Speaker 4: But it's just a piece of evidence, and.

623
00:30:29,160 --> 00:30:30,840
Speaker 1: We have a lot more evidence, Like we know a

624
00:30:30,840 --> 00:30:33,480
Speaker 1: lot more than Turning did, about how the internal functioning

625
00:30:33,520 --> 00:30:39,080
Speaker 1: of a mind works functionally, what it's doing, how representation works,

626
00:30:39,120 --> 00:30:42,440
Speaker 1: how experience and belief works, how they are connected to action,

627
00:30:42,600 --> 00:30:47,360
Speaker 1: and how they're connected to speech and thought. And once

628
00:30:47,400 --> 00:30:49,000
Speaker 1: you know all that stuff, you have a lot of

629
00:30:49,040 --> 00:30:52,880
Speaker 1: other avenues to get more evidence about whether the thing

630
00:30:52,960 --> 00:30:54,320
Speaker 1: is conscious and.

631
00:30:54,480 --> 00:30:55,960
Speaker 4: Whether it passes.

632
00:30:56,000 --> 00:30:58,120
Speaker 1: The Turning test is just like a drop in the

633
00:30:58,120 --> 00:31:01,200
Speaker 1: bucket compared to these, especially if you know how internal functioning.

634
00:31:01,040 --> 00:31:02,960
Speaker 6: The other the story is problem with the showing test,

635
00:31:03,000 --> 00:31:05,680
Speaker 6: and I think to be fair cheering did mention this,

636
00:31:05,760 --> 00:31:08,520
Speaker 6: If not in the original then kind of later on.

637
00:31:08,520 --> 00:31:10,880
Speaker 6: One problem with the Chewing test is that it's like

638
00:31:10,920 --> 00:31:13,560
Speaker 6: the de void camp in de androids gream of electric

639
00:31:13,600 --> 00:31:18,080
Speaker 6: cheap right. Plenty of humans would fail the Cheuring test. Yeah,

640
00:31:18,120 --> 00:31:19,920
Speaker 6: it is a piece of evidence, but it was never

641
00:31:20,240 --> 00:31:22,200
Speaker 6: as Mike said, it was supposed to be constitutive. It

642
00:31:22,240 --> 00:31:25,200
Speaker 6: wasn't like, if you can do this thing, your conscious

643
00:31:25,280 --> 00:31:27,360
Speaker 6: kind of full stopped. It was supposed to be here

644
00:31:27,440 --> 00:31:30,400
Speaker 6: is a thing that might indicate that being you're talking

645
00:31:30,400 --> 00:31:31,280
Speaker 6: to is conscious.

646
00:31:31,440 --> 00:31:33,360
Speaker 4: Happily, as Ma said, we've got loads and loads of

647
00:31:33,400 --> 00:31:33,960
Speaker 4: other evidence.

648
00:31:34,080 --> 00:31:36,600
Speaker 1: So these guys have made a machine that's just designed

649
00:31:37,120 --> 00:31:40,400
Speaker 1: to do one thing, and that's past the Turing test.

650
00:31:40,640 --> 00:31:42,720
Speaker 2: I can give you one more annoying example. Are you

651
00:31:42,760 --> 00:31:46,720
Speaker 2: familiar with Francois Cholet the abstraction and reasoning corpus. So

652
00:31:46,760 --> 00:31:48,800
Speaker 2: this is going to make you laugh. So he's a

653
00:31:48,800 --> 00:31:51,320
Speaker 2: Google engineer, and he created this thing of the arc,

654
00:31:51,440 --> 00:31:55,200
Speaker 2: the abstraction reasoning corpus, the test whether a machine was intelligent,

655
00:31:55,880 --> 00:31:58,040
Speaker 2: and someone created a model that could be it, and

656
00:31:58,040 --> 00:32:01,720
Speaker 2: then he immediately went, Okay, you can't just train the

657
00:32:01,840 --> 00:32:04,400
Speaker 2: model on the answers to the test.

658
00:32:04,880 --> 00:32:07,680
Speaker 6: This is why people, well I see people a fairly

659
00:32:07,720 --> 00:32:11,400
Speaker 6: small subset of weird nerds. But this is why a

660
00:32:11,440 --> 00:32:13,320
Speaker 6: small subset of weird nerds have been for the last

661
00:32:13,320 --> 00:32:17,680
Speaker 6: twenty years emphasizing artificial general intelligence, right, and kind of

662
00:32:17,800 --> 00:32:19,720
Speaker 6: like what we'll call it. When something really is a

663
00:32:19,800 --> 00:32:22,760
Speaker 6: thinking being is when it's not specialized to do one

664
00:32:22,800 --> 00:32:26,080
Speaker 6: and only one task, but rather when it's capable of

665
00:32:26,080 --> 00:32:30,040
Speaker 6: applying reasoning to multiple kinds of different and this analogous cases.

666
00:32:30,120 --> 00:32:32,719
Speaker 6: On the one hand, it does seem a little bit

667
00:32:32,800 --> 00:32:34,320
Speaker 6: like the guy flipping the table and go, oh, for

668
00:32:34,360 --> 00:32:35,360
Speaker 6: Fox's sake, now you've won.

669
00:32:35,400 --> 00:32:36,200
Speaker 5: I'm changing the rules.

670
00:32:36,200 --> 00:32:37,800
Speaker 6: But on the other I think he's got a point, right, Yeah,

671
00:32:38,000 --> 00:32:41,960
Speaker 6: if you're training a thing to do very specific you know,

672
00:32:42,080 --> 00:32:45,320
Speaker 6: like activate certain shiboleths, then unless you're some kind of

673
00:32:45,360 --> 00:32:47,720
Speaker 6: mad hard behaviorist, then yeah, like it's not that doesn't

674
00:32:47,720 --> 00:32:48,720
Speaker 6: demonstrate intelligence.

675
00:32:48,760 --> 00:32:50,880
Speaker 5: It is one thing that might indicate intelligence.

676
00:32:50,920 --> 00:32:53,400
Speaker 2: It's the same problem with chat GPT. It's built to

677
00:32:53,480 --> 00:32:57,480
Speaker 2: resemble intelligence, and resemble conscious this, and resemble these things,

678
00:32:57,480 --> 00:33:01,640
Speaker 2: but it isn't. It's almost like it's meaningless. On a

679
00:33:01,720 --> 00:33:05,000
Speaker 2: very high end philosophical levels. I find the whole generative

680
00:33:05,040 --> 00:33:06,520
Speaker 2: a I think deeply nihilistic.

681
00:33:06,720 --> 00:33:09,240
Speaker 1: I mean, one thing that connects to this is how

682
00:33:09,320 --> 00:33:11,920
Speaker 1: bad it is at reasoning. And this is kind of

683
00:33:12,320 --> 00:33:15,640
Speaker 1: good for us, especially in philosophy.

684
00:33:15,080 --> 00:33:17,200
Speaker 4: Because our students when they use.

685
00:33:17,040 --> 00:33:20,600
Speaker 1: It to write papers, the papers have to have arguments,

686
00:33:20,640 --> 00:33:24,960
Speaker 1: and chat GPT is very bad at doing reasoning. If

687
00:33:24,960 --> 00:33:26,920
Speaker 1: it has to be you know, sort of an extended

688
00:33:27,000 --> 00:33:30,240
Speaker 1: argument or a proof or something like that, it's very

689
00:33:30,280 --> 00:33:33,280
Speaker 1: bad at it. I think also that if there's one

690
00:33:33,280 --> 00:33:37,000
Speaker 1: thing kind of we learned from chat GPT, it's that

691
00:33:37,400 --> 00:33:40,959
Speaker 1: this is not the way to get to artificial general intelligence.

692
00:33:41,280 --> 00:33:42,640
Speaker 2: I was going to ask, do you think that this

693
00:33:42,760 --> 00:33:43,479
Speaker 2: is getting to that?

694
00:33:44,160 --> 00:33:48,760
Speaker 1: No, partially because it's so subject specific, right, It's one

695
00:33:48,880 --> 00:33:51,680
Speaker 1: it's trained to do one task, and it takes quite

696
00:33:51,720 --> 00:33:53,960
Speaker 1: a lot of training to get it to do that task. Well,

697
00:33:54,680 --> 00:33:57,760
Speaker 1: it's bad at many of the other tasks that we

698
00:33:57,840 --> 00:34:01,280
Speaker 1: think are connected with intelligence. It's bad at logical, mathematical reasoning.

699
00:34:01,800 --> 00:34:05,280
Speaker 1: I understand that open AI is trying to fix that.

700
00:34:05,480 --> 00:34:07,840
Speaker 1: Sometimes it sounds like they want to fix it by

701
00:34:07,920 --> 00:34:10,319
Speaker 1: just connecting it to a.

702
00:34:10,360 --> 00:34:12,759
Speaker 4: Database or a program that can do that for it.

703
00:34:13,000 --> 00:34:16,120
Speaker 4: But either way when you have with.

704
00:34:16,080 --> 00:34:18,359
Speaker 1: These kind of big base nets models, it's something that

705
00:34:18,719 --> 00:34:23,160
Speaker 1: is really good at whatever you train it to do,

706
00:34:23,600 --> 00:34:26,480
Speaker 1: but not going to be good at anything else. You know,

707
00:34:26,600 --> 00:34:29,400
Speaker 1: it's been a lot of data on one thing. It

708
00:34:29,440 --> 00:34:32,759
Speaker 1: finds patterns in that, it finds regularities in that, it

709
00:34:32,800 --> 00:34:33,680
Speaker 1: represents those.

710
00:34:34,239 --> 00:34:35,839
Speaker 4: The more data you feed it, the better it will

711
00:34:35,880 --> 00:34:36,200
Speaker 4: be at that.

712
00:34:36,880 --> 00:34:39,600
Speaker 1: But it's not going to have this kind of general

713
00:34:39,600 --> 00:34:42,120
Speaker 1: ability and it's not going to grow it out of

714
00:34:42,200 --> 00:34:43,920
Speaker 1: learning how to speak English.

715
00:34:44,360 --> 00:34:47,399
Speaker 4: Have you heard the Terry Pratchet quote about It's quite

716
00:34:47,400 --> 00:34:47,799
Speaker 4: early on.

717
00:34:48,040 --> 00:34:50,440
Speaker 6: He's talking about Hex, the kind of steampunk computer they

718
00:34:50,480 --> 00:34:52,799
Speaker 6: make and Unteen University, and it just has this off

719
00:34:52,840 --> 00:34:56,560
Speaker 6: hand line. A computer program is exactly like a philosophy professor,

720
00:34:56,840 --> 00:34:59,600
Speaker 6: unless you ask it the question in exactly the right way,

721
00:35:00,040 --> 00:35:03,320
Speaker 6: delight in giving you a perfectly accurate, completely unhelpful answer.

722
00:35:03,600 --> 00:35:03,759
Speaker 7: Right.

723
00:35:03,800 --> 00:35:06,040
Speaker 6: So, if you abstract the justified you got a philosophy

724
00:35:06,120 --> 00:35:10,239
Speaker 6: lecture is there. Basically what intelligent things do is go

725
00:35:10,640 --> 00:35:13,160
Speaker 6: You can't have meant that, you must have meant this

726
00:35:13,680 --> 00:35:17,200
Speaker 6: right chat GPTs goes. I will take a question as

727
00:35:17,239 --> 00:35:19,279
Speaker 6: read and of course it doesn't have an eye. There

728
00:35:19,320 --> 00:35:21,000
Speaker 6: is nothing in the rest like I've just answered formal

729
00:35:21,000 --> 00:35:23,200
Speaker 6: pite it again. But it's the same thing, and it's

730
00:35:23,320 --> 00:35:26,759
Speaker 6: trained to do contractively specific things, and you get the

731
00:35:26,760 --> 00:35:39,080
Speaker 6: same problem with any program like garbage in Garbage album.

732
00:35:39,120 --> 00:35:41,440
Speaker 2: So have you found a lot of students using chat GPT,

733
00:35:41,600 --> 00:35:44,040
Speaker 2: because this is a hard problem to quantify? Is it

734
00:35:44,160 --> 00:35:44,840
Speaker 2: all the time?

735
00:35:46,160 --> 00:35:48,400
Speaker 5: I mean it's it's a lot. I wouldn't say it

736
00:35:48,400 --> 00:35:48,839
Speaker 5: all the time.

737
00:35:49,000 --> 00:35:52,120
Speaker 4: I'm changed that you thought that there were more season

738
00:35:52,239 --> 00:35:53,160
Speaker 4: DEDs this semester.

739
00:35:54,200 --> 00:35:56,759
Speaker 3: Yeah, so I think there are quite a few, and

740
00:35:56,840 --> 00:35:59,839
Speaker 3: sometimes it is difficult for us to prove and if

741
00:35:59,880 --> 00:36:00,920
Speaker 3: we can't prove it.

742
00:36:00,840 --> 00:36:04,840
Speaker 4: Then at our university, then well shucks. They kind of

743
00:36:04,840 --> 00:36:05,520
Speaker 4: get away with it.

744
00:36:05,800 --> 00:36:09,319
Speaker 3: We can interview them, but unless we're like, unless we

745
00:36:09,360 --> 00:36:12,280
Speaker 3: have the proof that you could take before like a court,

746
00:36:12,320 --> 00:36:16,520
Speaker 3: then we are we're not able to really nail them

747
00:36:16,520 --> 00:36:16,839
Speaker 3: for it.

748
00:36:16,800 --> 00:36:17,799
Speaker 4: Which is a bit of a shame.

749
00:36:18,120 --> 00:36:18,920
Speaker 2: So's suspicion.

750
00:36:19,160 --> 00:36:21,359
Speaker 3: Yeah, we suspect that a lot of them are very

751
00:36:21,920 --> 00:36:23,480
Speaker 3: I'm one hundred percent on some of them.

752
00:36:23,520 --> 00:36:25,120
Speaker 2: I just know, what are the clues I want to

753
00:36:25,120 --> 00:36:26,440
Speaker 2: hear from all of you on this one, Like, what

754
00:36:26,520 --> 00:36:27,400
Speaker 2: are the side.

755
00:36:27,280 --> 00:36:29,400
Speaker 6: It's the uncanny value like, I will not shut up

756
00:36:29,400 --> 00:36:32,040
Speaker 6: about this, right as you know, the like you of

757
00:36:32,080 --> 00:36:34,320
Speaker 6: course will imagine a lot of here kind of readerbole

758
00:36:34,360 --> 00:36:36,920
Speaker 6: be as well. But the uncanny Valley thing in respective

759
00:36:37,000 --> 00:36:39,319
Speaker 6: humans is that there's a kind of a point up

760
00:36:39,360 --> 00:36:43,839
Speaker 6: to which humans seem to trust artificial things more the

761
00:36:43,840 --> 00:36:46,000
Speaker 6: more they look like a human. So like, we'll trust

762
00:36:46,000 --> 00:36:47,880
Speaker 6: a robot if it's kind of bipedal or if it

763
00:36:47,880 --> 00:36:50,959
Speaker 6: looks like a dog, But after a point, we really

764
00:36:51,080 --> 00:36:53,600
Speaker 6: rapidly start to trust me, So like if it basically

765
00:36:53,680 --> 00:36:57,280
Speaker 6: when it starts looking like data, some deep buried lizard

766
00:36:57,320 --> 00:36:58,360
Speaker 6: brain goes danger.

767
00:36:58,600 --> 00:37:00,720
Speaker 5: Danger. This thing's trying to fuck with you somehow.

768
00:37:01,440 --> 00:37:03,319
Speaker 2: Yeah, right, And chat gpt.

769
00:37:03,160 --> 00:37:04,799
Speaker 5: Writes exactly like that.

770
00:37:05,160 --> 00:37:08,440
Speaker 6: It writes almost but not entirely, like a thing that

771
00:37:08,560 --> 00:37:10,200
Speaker 6: is trying to convince you that it's human. I mean,

772
00:37:10,239 --> 00:37:11,799
Speaker 6: the dead give way for me is that it never

773
00:37:11,840 --> 00:37:15,120
Speaker 6: makes any spelling mistakes, but it can't form out a

774
00:37:15,120 --> 00:37:18,320
Speaker 6: paragraph to save its life. Normally, you would expect someone

775
00:37:18,320 --> 00:37:20,640
Speaker 6: who didn't know how to kind of order their sentences

776
00:37:20,800 --> 00:37:24,880
Speaker 6: to misspell the occasional word. Chat GPT spells everything correctly

777
00:37:25,239 --> 00:37:28,959
Speaker 6: and doesn't know what like subject object agreement is it's.

778
00:37:28,640 --> 00:37:32,040
Speaker 2: Like it's so, can you just define that please? Subject

779
00:37:32,080 --> 00:37:36,080
Speaker 2: object the beer that I drink, right, not the drink

780
00:37:36,120 --> 00:37:39,600
Speaker 2: that I beer right. Because it's interesting. It almost feels

781
00:37:39,640 --> 00:37:43,239
Speaker 2: like there is just an entire way of processing and

782
00:37:43,320 --> 00:37:46,200
Speaker 2: delivering information as a human being that we do not understand.

783
00:37:47,080 --> 00:37:48,240
Speaker 2: There is something missing.

784
00:37:48,840 --> 00:37:51,160
Speaker 1: I mean, for me, like it's very good, it's summarizing,

785
00:37:52,160 --> 00:37:55,800
Speaker 1: but when it responds, it responds in them in really

786
00:37:55,920 --> 00:37:58,840
Speaker 1: trite and repetitive ways, or like you'll get a paper

787
00:37:58,880 --> 00:38:02,719
Speaker 1: that summarizes a bunch of literature at length very effectively,

788
00:38:03,520 --> 00:38:08,280
Speaker 1: and then respond by saying, well, you know, this person

789
00:38:08,320 --> 00:38:11,279
Speaker 1: said this, that is doubtful. They should say more to

790
00:38:11,280 --> 00:38:14,200
Speaker 1: support that, you know, which is basically saying nothing right,

791
00:38:14,480 --> 00:38:16,120
Speaker 1: And that's pretty common.

792
00:38:17,080 --> 00:38:19,560
Speaker 4: It also does lists in formats things and kind of

793
00:38:19,600 --> 00:38:20,120
Speaker 4: like a list.

794
00:38:20,200 --> 00:38:24,200
Speaker 1: And even if this like make it look less like

795
00:38:24,239 --> 00:38:26,920
Speaker 1: a list, the paper still reads like a list.

796
00:38:27,200 --> 00:38:29,800
Speaker 2: Oh, because someone has asked, give me a few thoughts

797
00:38:29,840 --> 00:38:35,320
Speaker 2: on X. Yeah, yeah, that's very depressing.

798
00:38:36,000 --> 00:38:40,759
Speaker 3: Things like the introductions you get you'll get this is

799
00:38:40,800 --> 00:38:45,200
Speaker 3: an interesting and complex question that the postomers have asked

800
00:38:45,200 --> 00:38:46,920
Speaker 3: you the ages and this is one of the things

801
00:38:46,920 --> 00:38:50,040
Speaker 3: we shout at students not to write from first year,

802
00:38:50,440 --> 00:38:52,759
Speaker 3: and then you'll get this garbage right back at you,

803
00:38:52,920 --> 00:38:55,560
Speaker 3: and then at the end it will be oh, overall,

804
00:38:55,600 --> 00:38:58,719
Speaker 3: this is a complicated and difficult question, but.

805
00:38:58,800 --> 00:39:01,040
Speaker 2: Many nuances world of contrasts.

806
00:39:01,440 --> 00:39:04,080
Speaker 3: One of the things we tell them, don't ever fucking

807
00:39:04,160 --> 00:39:07,600
Speaker 3: do this. This is terrible and right back at you

808
00:39:07,960 --> 00:39:11,319
Speaker 3: in perfect english, that sort of english you'd expect a

809
00:39:11,400 --> 00:39:14,080
Speaker 3: really good student might have written. But clearly they can't

810
00:39:14,080 --> 00:39:16,120
Speaker 3: be a good student because otherwise they've listened to a

811
00:39:16,200 --> 00:39:16,960
Speaker 3: fucking instruction.

812
00:39:18,719 --> 00:39:21,120
Speaker 1: I also, I had some papers that I suspected were

813
00:39:21,160 --> 00:39:24,600
Speaker 1: chatcheapt this year, but they were already failing, so I

814
00:39:24,600 --> 00:39:27,920
Speaker 1: didn't Okay, yeah, I didn't think it was worth it

815
00:39:28,640 --> 00:39:34,480
Speaker 1: to pursue them, you know, as a plagiarism or Chatcheapte case.

816
00:39:34,760 --> 00:39:37,000
Speaker 2: So it's never good papers. Then it's never like an

817
00:39:37,000 --> 00:39:38,520
Speaker 2: eight like a first.

818
00:39:38,080 --> 00:39:39,040
Speaker 5: No, absolutely not.

819
00:39:39,640 --> 00:39:41,480
Speaker 1: It's so I think that part of what goes on

820
00:39:42,000 --> 00:39:45,600
Speaker 1: is you can get a passing grade with the Chatcheapte paper,

821
00:39:46,000 --> 00:39:49,480
Speaker 1: sometimes in the first couple of years when the papers

822
00:39:49,480 --> 00:39:53,080
Speaker 1: are shorter and we're not expecting as much. But then

823
00:39:53,120 --> 00:39:56,359
Speaker 1: when you move into the what we call honors level here,

824
00:39:56,400 --> 00:39:58,560
Speaker 1: which is like upper level classes in the US like

825
00:39:58,680 --> 00:39:59,640
Speaker 1: third and fourth year.

826
00:40:00,040 --> 00:40:01,759
Speaker 2: I've got first ever wrist with I know.

827
00:40:02,600 --> 00:40:05,920
Speaker 1: Yeah, yeah, exactly, great, well done, well Doe. You would

828
00:40:05,960 --> 00:40:08,840
Speaker 1: not have gotten it with chatchipt because you get dropped

829
00:40:08,880 --> 00:40:11,000
Speaker 1: in these classes where we expect you to have gained

830
00:40:11,000 --> 00:40:15,359
Speaker 1: writing skills minimal ones in your first two years and

831
00:40:15,400 --> 00:40:17,719
Speaker 1: then we're going to build on that and have you

832
00:40:17,760 --> 00:40:23,160
Speaker 1: do more complicated stuff. And Chatchept doesn't build on that, right,

833
00:40:23,320 --> 00:40:25,759
Speaker 1: it just stays where it was. So you go from

834
00:40:25,800 --> 00:40:30,560
Speaker 1: writing kind of C grade passing papers, yeah, that your

835
00:40:30,680 --> 00:40:34,320
Speaker 1: F grade paper. And it's also more obvious because the

836
00:40:34,320 --> 00:40:37,680
Speaker 1: papers are longer, and chat gipt can write long text,

837
00:40:37,719 --> 00:40:41,640
Speaker 1: but it gets very repetitive and noticeably repetitive, right, and

838
00:40:41,760 --> 00:40:46,400
Speaker 1: so you're kind of lost, like you haven't done the

839
00:40:46,440 --> 00:40:48,279
Speaker 1: work of figuring out how to write on your own

840
00:40:48,960 --> 00:40:52,279
Speaker 1: and the tool that you've been using is not up

841
00:40:52,280 --> 00:40:55,160
Speaker 1: to the task that you're now presented with. And so

842
00:40:55,280 --> 00:40:58,840
Speaker 1: I think I have seen a few papers that I

843
00:40:58,880 --> 00:41:01,280
Speaker 1: was suspicious of, but the papers that I was certain

844
00:41:01,320 --> 00:41:04,359
Speaker 1: of were ones that were like senior thesis, very clear

845
00:41:04,600 --> 00:41:07,120
Speaker 1: a person just had no way of writing culture.

846
00:41:07,480 --> 00:41:09,120
Speaker 2: That's insane at that stage.

847
00:41:09,600 --> 00:41:11,560
Speaker 6: Yeah, I mean it's fun because I mean one of

848
00:41:11,640 --> 00:41:13,600
Speaker 6: the things that we get told about is like, oh,

849
00:41:13,600 --> 00:41:15,120
Speaker 6: the students have got to learn how to use you know,

850
00:41:15,200 --> 00:41:19,120
Speaker 6: AI and large language model, plagiarism machines and responsibly and

851
00:41:19,160 --> 00:41:21,640
Speaker 6: in a kind of positive way. Well, if they're using

852
00:41:21,640 --> 00:41:23,319
Speaker 6: them in a way that means they don't learn how

853
00:41:23,360 --> 00:41:25,399
Speaker 6: to write, then it's not positive. Visit like, yeah, it's

854
00:41:25,400 --> 00:41:27,120
Speaker 6: fucking hard to write a good ess. Say, yes, it

855
00:41:27,239 --> 00:41:31,080
Speaker 6: is hard to write. That's why we practice, That's why

856
00:41:31,120 --> 00:41:33,839
Speaker 6: we have editors, that's why we do this collaboratively. And

857
00:41:33,960 --> 00:41:36,640
Speaker 6: if you're using this as a sort of oh I

858
00:41:36,680 --> 00:41:39,239
Speaker 6: don't know how to write, well, tough shit, you're never

859
00:41:39,280 --> 00:41:41,120
Speaker 6: going to know how to write that. That doesn't seem

860
00:41:41,600 --> 00:41:42,920
Speaker 6: a positive use of any of this.

861
00:41:43,120 --> 00:41:47,080
Speaker 2: Well that's the thing. Writing is also reading the consumption

862
00:41:47,120 --> 00:41:50,040
Speaker 2: of information and then bouncing the ideas of your brain allegedly.

863
00:41:50,719 --> 00:41:55,040
Speaker 1: Yeah, I worry about like because CHATCHYBT is good at summarizing,

864
00:41:55,160 --> 00:41:57,680
Speaker 1: So I worry that one of the uses people will think, ah,

865
00:41:57,719 --> 00:41:59,920
Speaker 1: this is a pretty good use for it is summarizing

866
00:42:00,080 --> 00:42:03,880
Speaker 1: the paper that they're supposed to read, and it will

867
00:42:03,920 --> 00:42:05,799
Speaker 1: do that effectively enough.

868
00:42:06,239 --> 00:42:09,320
Speaker 4: For them to discuss it in class if they're willing

869
00:42:09,360 --> 00:42:09,719
Speaker 4: to be here.

870
00:42:09,800 --> 00:42:13,520
Speaker 1: Yeah, right, but they're not going to pick up a

871
00:42:13,520 --> 00:42:15,359
Speaker 1: lot of the nuances in a lot of the kind

872
00:42:15,360 --> 00:42:18,879
Speaker 1: of like stylistic ways of presenting ideas that you get

873
00:42:18,920 --> 00:42:20,280
Speaker 1: when you actually do the reading.

874
00:42:20,960 --> 00:42:24,040
Speaker 2: And it's it's so frustrating as well, because like for this,

875
00:42:24,200 --> 00:42:26,960
Speaker 2: for example, I've just got the printing off things that

876
00:42:27,040 --> 00:42:29,759
Speaker 2: don't read PDFs anymore because I feel like you do

877
00:42:29,880 --> 00:42:30,640
Speaker 2: need to focus.

878
00:42:31,280 --> 00:42:34,040
Speaker 6: Yeah, there's some evidence that reading physical copies makes you

879
00:42:34,080 --> 00:42:34,600
Speaker 6: engage more.

880
00:42:34,640 --> 00:42:36,400
Speaker 5: Sorry I'm very old fashioned, I guess.

881
00:42:36,280 --> 00:42:39,480
Speaker 2: But no, it's true though, But also reading that I

882
00:42:39,520 --> 00:42:42,279
Speaker 2: wouldn't have really on a PDF. I wouldn't have given

883
00:42:42,320 --> 00:42:44,840
Speaker 2: it as much attention. But also, going through this paper,

884
00:42:45,239 --> 00:42:47,560
Speaker 2: you could see what you were doing, like you could

885
00:42:47,600 --> 00:42:49,680
Speaker 2: see that you were lining up here are the qualities

886
00:42:49,719 --> 00:42:53,120
Speaker 2: that we use to judge bullshit. But also summarizing a

887
00:42:53,120 --> 00:42:55,800
Speaker 2: paper does not actually give you the argument, it gives

888
00:42:55,800 --> 00:42:59,920
Speaker 2: you an answer. So what do you actually want students

889
00:42:59,960 --> 00:43:02,600
Speaker 2: to do instead? Because I don't think there's any reason

890
00:43:02,680 --> 00:43:05,160
Speaker 2: to use chat GPT for these reasons like that it

891
00:43:05,200 --> 00:43:07,759
Speaker 2: doesn't seem to do anything that's useful for them.

892
00:43:08,080 --> 00:43:11,600
Speaker 1: I don't have No, I don't actually have any use

893
00:43:11,680 --> 00:43:14,280
Speaker 1: for chat shapt that I can put to my students

894
00:43:14,280 --> 00:43:16,120
Speaker 1: and say here's what I think you should do with it.

895
00:43:16,680 --> 00:43:20,239
Speaker 1: We are like kind of developing strategies for keeping them

896
00:43:20,239 --> 00:43:24,600
Speaker 1: from using it, So like building directly on what you're saying. Like,

897
00:43:24,640 --> 00:43:27,360
Speaker 1: in my class next year, I'm going to have the

898
00:43:27,400 --> 00:43:30,960
Speaker 1: students do regular assignments, which are argument summaries and not

899
00:43:31,080 --> 00:43:33,040
Speaker 1: paper summaries. So the idea is they have to read

900
00:43:33,080 --> 00:43:36,759
Speaker 1: the paper, find an argument, and tell me what are

901
00:43:36,760 --> 00:43:39,799
Speaker 1: the premises, what's the conclusion. And that's something that chatchipt

902
00:43:39,960 --> 00:43:42,520
Speaker 1: is not good at, right, But it's also something that

903
00:43:42,560 --> 00:43:44,280
Speaker 1: will give them critical.

904
00:43:43,880 --> 00:43:46,879
Speaker 4: Reading skills, which is what I want to do, right.

905
00:43:46,960 --> 00:43:49,840
Speaker 1: So yeah, I think that I've mostly been thinking about

906
00:43:50,000 --> 00:43:54,400
Speaker 1: ways to keep them from relying on it, because I

907
00:43:54,440 --> 00:43:57,879
Speaker 1: think that often if they rely on it, they'll they'll

908
00:43:58,040 --> 00:43:59,760
Speaker 1: they'll put.

909
00:43:59,640 --> 00:44:00,920
Speaker 4: Themselves it's the worst position.

910
00:44:01,160 --> 00:44:04,040
Speaker 1: Yeah, when it comes to future work, they won't develop

911
00:44:04,040 --> 00:44:05,960
Speaker 1: the skills that they're going to need and the skills

912
00:44:06,000 --> 00:44:09,239
Speaker 1: that we tell them and their parents they're gonna get

913
00:44:09,280 --> 00:44:10,400
Speaker 1: with their college degree.

914
00:44:10,560 --> 00:44:13,319
Speaker 2: Right, it almost feels like we need more deliberacy in

915
00:44:13,920 --> 00:44:17,080
Speaker 2: university education because I was not taught to write. I

916
00:44:17,239 --> 00:44:19,080
Speaker 2: just did a lot of it until I got good

917
00:44:19,160 --> 00:44:22,160
Speaker 2: enough grades and Daniel Chohnd look great mental. But I've

918
00:44:22,160 --> 00:44:25,120
Speaker 2: had tons of them, and it almost feels like we

919
00:44:25,200 --> 00:44:28,160
Speaker 2: need classes where it's like, Okay, no computer for this one.

920
00:44:29,000 --> 00:44:30,880
Speaker 2: I'm going to print this paper out and you're going

921
00:44:30,960 --> 00:44:33,600
Speaker 2: to underline the things that are important and talk to

922
00:44:33,640 --> 00:44:35,880
Speaker 2: me about almost feels like we need to rebuild this

923
00:44:35,960 --> 00:44:39,200
Speaker 2: because yes, we shouldn't be using chet GPT to half

924
00:44:39,280 --> 00:44:42,240
Speaker 2: us our essays. But at the same time, human beings

925
00:44:42,280 --> 00:44:42,800
Speaker 2: are lazy.

926
00:44:43,680 --> 00:44:46,959
Speaker 1: Yeah, I mean for me, I also prefer to read

927
00:44:47,320 --> 00:44:51,440
Speaker 1: off the computer, but I often read PDFs because I'm

928
00:44:51,560 --> 00:44:55,239
Speaker 1: terrible at keeping files right physical Like, you know, I'm

929
00:44:55,280 --> 00:44:58,319
Speaker 1: not going to keep a giant filed with all the

930
00:44:58,360 --> 00:45:01,759
Speaker 1: papers that I've read and written my liner notes in.

931
00:45:02,560 --> 00:45:05,000
Speaker 1: You guys can see in my office they're just piled around,

932
00:45:05,160 --> 00:45:05,920
Speaker 1: like you can't.

933
00:45:05,680 --> 00:45:07,200
Speaker 2: See this in that's academia.

934
00:45:07,480 --> 00:45:08,520
Speaker 4: Yeah, but I just.

935
00:45:08,520 --> 00:45:12,480
Speaker 1: Have piles of paper with empty coffee bugs everywhere that

936
00:45:12,520 --> 00:45:13,680
Speaker 1: the camera is not facing.

937
00:45:14,160 --> 00:45:15,320
Speaker 4: But it's the terrible system.

938
00:45:15,360 --> 00:45:17,480
Speaker 1: So at least on my computer, if I'm like, oh,

939
00:45:17,520 --> 00:45:19,359
Speaker 1: I read that paper like a year ago, what did

940
00:45:19,360 --> 00:45:20,960
Speaker 1: I think I can click on it and see my

941
00:45:21,000 --> 00:45:23,520
Speaker 1: own notes, and I do think that there's something to

942
00:45:24,480 --> 00:45:27,120
Speaker 1: keeping those records and kind of actively.

943
00:45:26,760 --> 00:45:29,439
Speaker 4: Reading in that way. I don't know how I ended

944
00:45:29,480 --> 00:45:31,759
Speaker 4: this without telling you how to make students do that.

945
00:45:31,960 --> 00:45:35,040
Speaker 6: Well, you started with the correct answer, which is don't

946
00:45:35,120 --> 00:45:39,839
Speaker 6: use chat GPT. Yeah, yeah, I actually I've got a

947
00:45:40,040 --> 00:45:42,359
Speaker 6: certain amount of sympathy with like, just keep writing till

948
00:45:42,360 --> 00:45:44,480
Speaker 6: you get good at it. But I realized as a

949
00:45:44,520 --> 00:45:47,120
Speaker 6: lecturer that can't be my official possession. And I certainly

950
00:45:47,160 --> 00:45:50,319
Speaker 6: think that it's the case that certainly Glasgow has got

951
00:45:50,880 --> 00:45:53,480
Speaker 6: better over the last few years about going, Oh, actually

952
00:45:53,520 --> 00:45:55,279
Speaker 6: you do like, we do need to give you some

953
00:45:55,360 --> 00:45:57,279
Speaker 6: kind of structuring and some buttressing on his is how

954
00:45:57,320 --> 00:45:58,120
Speaker 6: to write academically.

955
00:45:58,160 --> 00:45:59,160
Speaker 5: Here's how to get do research.

956
00:45:59,160 --> 00:46:01,120
Speaker 6: And I think that's all to the It's worth saying

957
00:46:01,320 --> 00:46:04,960
Speaker 6: this started happening well before chat GPT started pissing all

958
00:46:04,960 --> 00:46:06,840
Speaker 6: over our doorsteps, so they don't get to claim that

959
00:46:06,920 --> 00:46:07,720
Speaker 6: as being a benefit.

960
00:46:08,480 --> 00:46:10,760
Speaker 2: There was a whole Wikipedia panic when I was in school.

961
00:46:10,840 --> 00:46:13,759
Speaker 5: Yeah, think about Wikipedia the right. It's like I used

962
00:46:13,760 --> 00:46:14,600
Speaker 5: to say this to my students.

963
00:46:14,600 --> 00:46:17,800
Speaker 2: Wikichili one of the best resources.

964
00:46:17,080 --> 00:46:22,239
Speaker 6: Absolutely fine as a starting point for research absolutely whatsoever.

965
00:46:22,560 --> 00:46:24,560
Speaker 6: But if you're turning in and on his level, essay,

966
00:46:24,640 --> 00:46:26,399
Speaker 6: I want you to go and read the fucking things

967
00:46:26,400 --> 00:46:27,080
Speaker 6: it's referencing.

968
00:46:27,239 --> 00:46:28,040
Speaker 4: Yeah, that's right.

969
00:46:28,840 --> 00:46:31,120
Speaker 1: Yeah, I think these things are often great as sources.

970
00:46:31,280 --> 00:46:33,439
Speaker 1: I'm worry about chat GPT is that it's not great

971
00:46:33,480 --> 00:46:34,040
Speaker 1: as a source.

972
00:46:34,120 --> 00:46:35,640
Speaker 2: It's just yeah, we've been.

973
00:46:35,560 --> 00:46:38,719
Speaker 1: Saying it often gets things wrong, and it often it'll

974
00:46:38,800 --> 00:46:40,960
Speaker 1: make up sources, whereas Wikipedia.

975
00:46:40,440 --> 00:46:41,080
Speaker 4: Will never do that.

976
00:46:42,440 --> 00:46:48,279
Speaker 6: There are some famous hoaxes, but they get cool. Yeah, Joe,

977
00:46:48,400 --> 00:46:51,120
Speaker 6: have you got any positive things to say about chat GIPT.

978
00:46:52,520 --> 00:46:52,719
Speaker 4: Fan?

979
00:46:54,880 --> 00:46:57,719
Speaker 3: So I know some people who have used these kinds

980
00:46:57,719 --> 00:47:01,480
Speaker 3: of things productively, not in ways that our students would,

981
00:47:01,520 --> 00:47:04,879
Speaker 3: but I know some mathematicians who have been using it

982
00:47:05,280 --> 00:47:08,600
Speaker 3: to do sort of informal proofs and things like that.

983
00:47:08,840 --> 00:47:12,440
Speaker 3: And it does still bullshit, and it bullshits very convincingly,

984
00:47:12,560 --> 00:47:15,359
Speaker 3: which makes it very difficult to use for this kind

985
00:47:15,400 --> 00:47:19,760
Speaker 3: of purpose. But it can do some interesting and cool things. Yeah,

986
00:47:19,800 --> 00:47:21,880
Speaker 3: I think some people in that sort of field have

987
00:47:22,000 --> 00:47:25,279
Speaker 3: found useful. And also we've mentioned this before, like if

988
00:47:25,320 --> 00:47:27,480
Speaker 3: you want chat to be teach to write you a bibliography,

989
00:47:27,800 --> 00:47:30,440
Speaker 3: you've got a bibliography in one style tell it to

990
00:47:30,520 --> 00:47:33,080
Speaker 3: put something into a different one. Then it's and it's

991
00:47:33,080 --> 00:47:35,720
Speaker 3: good for like coding process doing certain things.

992
00:47:36,640 --> 00:47:39,719
Speaker 1: I also think I don't know, I'm not sure how

993
00:47:39,760 --> 00:47:43,680
Speaker 1: I feel about what I'm about to say, but perfect, Yeah.

994
00:47:43,480 --> 00:47:44,200
Speaker 4: I'm going for it.

995
00:47:44,200 --> 00:47:46,440
Speaker 1: It is a somewhat positive thing maybe for a chat ept,

996
00:47:46,680 --> 00:47:49,120
Speaker 1: which is that we often have students who have like

997
00:47:49,200 --> 00:47:54,000
Speaker 1: really interesting ideas and well thought out arguments, but for

998
00:47:54,120 --> 00:47:57,000
Speaker 1: whom English isn't their first language, and the actual writing

999
00:47:57,080 --> 00:48:00,479
Speaker 1: is kind of rough and you have to like push

1000
00:48:00,600 --> 00:48:03,719
Speaker 1: through reading it to get the good idea, which is often.

1001
00:48:03,520 --> 00:48:06,880
Speaker 4: Really there and quite you know, creative and insightful.

1002
00:48:07,640 --> 00:48:09,440
Speaker 1: Yeah, And so I do wonder if there's a way

1003
00:48:09,480 --> 00:48:11,280
Speaker 1: to use it so it just smooths off the edges

1004
00:48:12,040 --> 00:48:13,959
Speaker 1: of this kind of thing. But I worry that if

1005
00:48:14,000 --> 00:48:17,160
Speaker 1: you tell students to do that, they'll just first it

1006
00:48:17,440 --> 00:48:19,719
Speaker 1: if they can develop the language skills they offered it

1007
00:48:20,040 --> 00:48:21,880
Speaker 1: really good by the yea, yeah.

1008
00:48:21,719 --> 00:48:22,480
Speaker 4: What are you going to say to you?

1009
00:48:22,520 --> 00:48:24,600
Speaker 6: Well, so I want to say, it's yeah, you can

1010
00:48:24,640 --> 00:48:27,080
Speaker 6: see me getting agitated. Yeah, I think much correct about

1011
00:48:27,120 --> 00:48:28,680
Speaker 6: it like that this is a kind of possible use.

1012
00:48:28,760 --> 00:48:30,920
Speaker 6: But I think this and this is why I'm getting

1013
00:48:31,080 --> 00:48:35,680
Speaker 6: visibly agitated here. That's students either need to or feel

1014
00:48:35,719 --> 00:48:36,319
Speaker 6: they need to use.

1015
00:48:36,360 --> 00:48:37,799
Speaker 5: This speaks to a deeper.

1016
00:48:37,520 --> 00:48:39,440
Speaker 6: Issue, right to a social issue, to political issue, to

1017
00:48:39,440 --> 00:48:41,960
Speaker 6: an issue about how universities work. If a student is

1018
00:48:42,120 --> 00:48:45,520
Speaker 6: having problems with English, then there's a number of like

1019
00:48:45,600 --> 00:48:47,600
Speaker 6: explanations or a number of kind of responses. Right that

1020
00:48:47,880 --> 00:48:50,960
Speaker 6: one response is that, like Glasgow is in English teaching university,

1021
00:48:51,000 --> 00:48:53,800
Speaker 6: if someone's English isn't good enough to be taking a degree,

1022
00:48:53,880 --> 00:48:55,520
Speaker 6: then plausibly they shouldn't have been let in, and why

1023
00:48:55,560 --> 00:48:56,279
Speaker 6: have they been learned well?

1024
00:48:56,360 --> 00:48:57,080
Speaker 4: Because of money?

1025
00:48:57,280 --> 00:49:00,400
Speaker 6: Or alternatively, if someone's having problems with English for like

1026
00:49:00,480 --> 00:49:03,560
Speaker 6: whatever reason at all, that should be supported. There should

1027
00:49:03,560 --> 00:49:05,080
Speaker 6: be kind of tutors. There should be people who can

1028
00:49:05,080 --> 00:49:07,640
Speaker 6: help with English. But again that would cost university money.

1029
00:49:07,680 --> 00:49:12,160
Speaker 6: So of course that doesn't happen, right, It doesn't happen

1030
00:49:12,200 --> 00:49:14,560
Speaker 6: anywhere that it doesn't in the extent it would have

1031
00:49:14,600 --> 00:49:17,680
Speaker 6: to happen in order for this to be a general policy.

1032
00:49:18,560 --> 00:49:20,319
Speaker 1: Yeah, I think it could be better, But I do

1033
00:49:20,400 --> 00:49:24,160
Speaker 1: think that universities often have like a writing center or

1034
00:49:24,200 --> 00:49:25,760
Speaker 1: a tutoring center that you can.

1035
00:49:25,640 --> 00:49:28,400
Speaker 5: Send, but they don't.

1036
00:49:28,520 --> 00:49:31,000
Speaker 6: They don't have the sort of spread that would be

1037
00:49:31,080 --> 00:49:32,920
Speaker 6: needed or the staff that would be needed for this

1038
00:49:33,000 --> 00:49:35,759
Speaker 6: to be instead of using chat GVD to sound the

1039
00:49:35,800 --> 00:49:36,279
Speaker 6: edges off.

1040
00:49:36,400 --> 00:49:39,040
Speaker 4: Yeah, I think for example twenty two with your supervisor.

1041
00:49:39,160 --> 00:49:42,879
Speaker 4: My worry especially would be that this is my first

1042
00:49:42,960 --> 00:49:43,399
Speaker 4: year here.

1043
00:49:43,320 --> 00:49:47,319
Speaker 1: At Glasgow, but I have a good Yeah, I think

1044
00:49:47,360 --> 00:49:49,440
Speaker 1: they probably have a good writing center universities have been

1045
00:49:49,440 --> 00:49:52,239
Speaker 1: in the past. I felt very confident sending students to

1046
00:49:52,280 --> 00:49:55,839
Speaker 1: the writing center when they have these problems. But I

1047
00:49:55,840 --> 00:49:58,600
Speaker 1: think James is completely right that we don't want the

1048
00:49:58,680 --> 00:50:00,719
Speaker 1: universities to see this as a way to get rid

1049
00:50:00,719 --> 00:50:03,759
Speaker 1: of the writing center. And that's one hundred percent a

1050
00:50:03,880 --> 00:50:05,640
Speaker 1: risk given the financial problems that.

1051
00:50:05,719 --> 00:50:11,279
Speaker 4: Universities are facing, and maybe they're already not ye like,

1052
00:50:11,400 --> 00:50:14,120
Speaker 4: given the quality of papers, they're often good.

1053
00:50:14,239 --> 00:50:17,600
Speaker 1: I think another thing, as far as this is like

1054
00:50:17,640 --> 00:50:22,400
Speaker 1: a social problem, is that when grading, I myself tried

1055
00:50:22,440 --> 00:50:24,680
Speaker 1: to grade in terms of like the ideas and argument

1056
00:50:24,880 --> 00:50:26,520
Speaker 1: this is philosophy.

1057
00:50:26,320 --> 00:50:30,720
Speaker 4: And not the quality of being right. But not everybody

1058
00:50:30,760 --> 00:50:31,040
Speaker 4: does that.

1059
00:50:31,160 --> 00:50:32,799
Speaker 1: So I kind of think that another part of this

1060
00:50:33,080 --> 00:50:35,360
Speaker 1: is figuring out how we want to evaluate the students

1061
00:50:35,400 --> 00:50:37,880
Speaker 1: and what we want to privilege in that evaluation.

1062
00:50:38,760 --> 00:50:38,960
Speaker 4: Yeah.

1063
00:50:39,080 --> 00:50:41,800
Speaker 6: Sure, so then again the kind of that that becomes

1064
00:50:41,840 --> 00:50:44,560
Speaker 6: a problem about what people are checking for. Not Let's

1065
00:50:44,560 --> 00:50:47,600
Speaker 6: take this arch backwards approach to marking, which is like,

1066
00:50:47,719 --> 00:50:49,000
Speaker 6: how fancy is your English?

1067
00:50:49,040 --> 00:50:51,319
Speaker 5: Are fancy? English? Is good English? Have an A? But

1068
00:50:51,440 --> 00:50:53,000
Speaker 5: rather we should be kind of checking for everythings.

1069
00:50:53,040 --> 00:50:55,600
Speaker 6: Right, So again the blame blies differently in that case,

1070
00:50:55,680 --> 00:50:56,759
Speaker 6: but it still becomes a question.

1071
00:50:56,800 --> 00:50:57,880
Speaker 5: It's not solved technological.

1072
00:50:58,239 --> 00:51:02,000
Speaker 2: Yeah, almost feels like large language models are taking advantage

1073
00:51:02,040 --> 00:51:04,520
Speaker 2: of a certain kind of organizational failure.

1074
00:51:05,560 --> 00:51:06,920
Speaker 4: What an idea?

1075
00:51:08,080 --> 00:51:12,280
Speaker 2: Crazy that the tech industry manipulating parts of society.

1076
00:51:12,800 --> 00:51:16,200
Speaker 1: I have a kind of related tangent here, which is like,

1077
00:51:16,239 --> 00:51:19,920
Speaker 1: what are the use cases that open ai was expecting

1078
00:51:20,160 --> 00:51:23,720
Speaker 1: but didn't want to emphasize, Because for everybody in universities,

1079
00:51:23,920 --> 00:51:26,200
Speaker 1: as soon as this came out, the first thought was

1080
00:51:26,520 --> 00:51:29,400
Speaker 1: students are going to use this to cheat And certainly

1081
00:51:29,480 --> 00:51:33,359
Speaker 1: like the people in open AI went to college, right,

1082
00:51:33,480 --> 00:51:34,920
Speaker 1: and that's what I hear about.

1083
00:51:34,640 --> 00:51:38,760
Speaker 2: Them, so they must well, sam Oman drop down, Yeah.

1084
00:51:38,560 --> 00:51:42,200
Speaker 1: I'm sure he really understands the value of a secondary He's.

1085
00:51:42,080 --> 00:51:43,880
Speaker 2: Like, I'm gonna write at least fucking essays.

1086
00:51:43,920 --> 00:51:46,120
Speaker 1: Anyway, maybe he was thinking I would have loved to

1087
00:51:46,160 --> 00:51:48,280
Speaker 1: have a computer write my essays, I'll devote my life.

1088
00:51:48,880 --> 00:51:52,160
Speaker 1: But I mean, I'm sure that they like recognize these

1089
00:51:52,239 --> 00:51:56,320
Speaker 1: bad use cases, right, but they're doing nothing to mitigate

1090
00:51:56,360 --> 00:51:57,840
Speaker 1: them as far as I can see. And like, another

1091
00:51:57,880 --> 00:52:02,239
Speaker 1: one that's very related is like, you know, I'm sure

1092
00:52:02,239 --> 00:52:04,600
Speaker 1: you've heard of this ED phishing, right, A lot of

1093
00:52:04,840 --> 00:52:09,000
Speaker 1: you know, corporations get attacked and get hacked not by

1094
00:52:09,040 --> 00:52:13,160
Speaker 1: someone cleverly figuring out a backdoor to their system, but

1095
00:52:13,200 --> 00:52:17,200
Speaker 1: by somebody sending in social engineering, yeah, asking for the

1096
00:52:17,239 --> 00:52:18,399
Speaker 1: password to somebody else.

1097
00:52:18,480 --> 00:52:19,759
Speaker 4: And one of the biggest.

1098
00:52:19,480 --> 00:52:21,239
Speaker 1: Barriers to that is that a lot of the people

1099
00:52:21,280 --> 00:52:25,680
Speaker 1: who are engaging in fishing aren't from the same country

1100
00:52:25,760 --> 00:52:28,479
Speaker 1: as the company they're targeting, right, so they're not able

1101
00:52:28,520 --> 00:52:31,239
Speaker 1: to write a convincing email or make a phone call

1102
00:52:31,280 --> 00:52:33,120
Speaker 1: that sounds like that person's supervisor.

1103
00:52:34,120 --> 00:52:37,960
Speaker 4: But with a tool like this, you could one hundred

1104
00:52:37,960 --> 00:52:39,000
Speaker 4: percent write that email. Right.

1105
00:52:39,040 --> 00:52:41,319
Speaker 1: It's going to make it a lot easier for these

1106
00:52:41,400 --> 00:52:43,719
Speaker 1: kinds of illicit schemes to work.

1107
00:52:43,840 --> 00:52:46,759
Speaker 2: That has been a market increase CNBC, which is just

1108
00:52:46,760 --> 00:52:48,960
Speaker 2: brought up, is that on two hundred and sixty five

1109
00:52:48,960 --> 00:52:51,440
Speaker 2: percent increase in malicious fishing emails since the launch of

1110
00:52:51,480 --> 00:52:53,719
Speaker 2: chat JPTA, great style. I mean, if I could have

1111
00:52:53,760 --> 00:52:56,680
Speaker 2: thought of that, imagine what a criminal could do.

1112
00:52:57,440 --> 00:53:00,000
Speaker 4: Right, But also, weren't the people that open a eye

1113
00:53:00,200 --> 00:53:01,120
Speaker 4: thinking about.

1114
00:53:00,840 --> 00:53:03,080
Speaker 2: That because like they don't care.

1115
00:53:03,320 --> 00:53:06,080
Speaker 6: Yeah, yeah, we've all seen Durassic Park, right, Yeah, Yeah,

1116
00:53:06,080 --> 00:53:07,839
Speaker 6: they were so visically thinking about what they could do,

1117
00:53:07,880 --> 00:53:09,240
Speaker 6: they never thought about whether they should.

1118
00:53:09,360 --> 00:53:09,600
Speaker 4: Yeah.

1119
00:53:09,640 --> 00:53:12,000
Speaker 1: This is the kind of problem with the move fast

1120
00:53:12,040 --> 00:53:15,759
Speaker 1: and break things mentality, like obvious. I mean, I think

1121
00:53:15,760 --> 00:53:17,279
Speaker 1: it might be the only person who has raised in

1122
00:53:17,320 --> 00:53:18,839
Speaker 1: the US, but we had future.

1123
00:53:18,480 --> 00:53:21,560
Speaker 4: Problems solvers, where you you know, you think.

1124
00:53:21,440 --> 00:53:25,440
Speaker 1: About a future problem and what bad consequences there could

1125
00:53:25,480 --> 00:53:28,240
Speaker 1: be of some technology and how to solve them, usually

1126
00:53:28,400 --> 00:53:29,239
Speaker 1: through social cues.

1127
00:53:29,680 --> 00:53:33,880
Speaker 4: If I could do that in fifth grade, Yeah.

1128
00:53:33,719 --> 00:53:36,080
Speaker 1: I would expect these people to have thought through some

1129
00:53:36,160 --> 00:53:38,600
Speaker 1: of the bad consequences of the technology they're putting out.

1130
00:53:39,080 --> 00:53:41,960
Speaker 1: And you know, some of those are cheating on tests

1131
00:53:42,200 --> 00:53:45,080
Speaker 1: and they don't seem to have worried about that. Yeah,

1132
00:53:45,080 --> 00:53:46,800
Speaker 1: And another one is fishing. They don't seem to have

1133
00:53:46,800 --> 00:53:47,480
Speaker 1: worried about that.

1134
00:53:47,560 --> 00:53:52,680
Speaker 6: You know, biases and algorithms, right, so yeah, yeah, comes,

1135
00:53:52,719 --> 00:53:54,600
Speaker 6: and I suppose to you it turns out, so with

1136
00:53:54,680 --> 00:53:57,560
Speaker 6: a lot of the facial recognition systems, they were incredibly racist.

1137
00:53:57,760 --> 00:54:00,840
Speaker 2: They were going back to Microsoft's connect could not see.

1138
00:54:00,640 --> 00:54:04,480
Speaker 6: Bad Yeah yeah, and kind of CCTV stuff that basically

1139
00:54:04,520 --> 00:54:07,239
Speaker 6: just sort of unless it was presented with a blindingly

1140
00:54:07,280 --> 00:54:08,520
Speaker 6: white Caucasian.

1141
00:54:08,080 --> 00:54:10,200
Speaker 5: Where I don't know, right, but like.

1142
00:54:13,440 --> 00:54:16,600
Speaker 6: The sort of stuff where these like langage knowledge are

1143
00:54:16,640 --> 00:54:18,600
Speaker 6: trained on certain sets of data, and they trained on

1144
00:54:18,680 --> 00:54:23,359
Speaker 6: certain assumptions and like shifting shit out right, and particularly

1145
00:54:23,400 --> 00:54:25,920
Speaker 6: if people think that it's actually doing any kind of thinking,

1146
00:54:26,200 --> 00:54:29,239
Speaker 6: and if they kind of cargo cult it, we again

1147
00:54:29,280 --> 00:54:32,160
Speaker 6: get a kind of social problem multiplied by technology feeding

1148
00:54:32,200 --> 00:54:33,520
Speaker 6: back into a social problem, right.

1149
00:54:33,560 --> 00:54:36,920
Speaker 5: And it's the so these guys have heard me when

1150
00:54:36,960 --> 00:54:39,759
Speaker 5: you're about it so much, I love it or so, but.

1151
00:54:39,800 --> 00:54:44,960
Speaker 6: I'm profoundly skeptical of technology's ability to solve anything unless

1152
00:54:45,000 --> 00:54:48,360
Speaker 6: we know exactly the respect in which we want to

1153
00:54:48,360 --> 00:54:51,280
Speaker 6: solve it and how that technology is going to be applied.

1154
00:54:52,520 --> 00:54:55,960
Speaker 6: You know, like sure, experiment with bringing back dinosaurs, but

1155
00:54:55,960 --> 00:54:57,960
Speaker 6: but like, don't tell me that it's going to save

1156
00:54:58,040 --> 00:55:00,959
Speaker 6: the healthcare system unless you can demonstrated to be step

1157
00:55:01,000 --> 00:55:03,279
Speaker 6: by step how that big OLDI rect run around on

1158
00:55:03,320 --> 00:55:05,200
Speaker 6: Eiland new Blayer is going to save anything, and they

1159
00:55:05,280 --> 00:55:06,360
Speaker 6: just tried to blind people.

1160
00:55:06,760 --> 00:55:10,040
Speaker 4: Actually bringing back dinosaurs with good that would be great.

1161
00:55:10,680 --> 00:55:13,239
Speaker 6: All right, this isn't the best example, but I already had.

1162
00:55:14,560 --> 00:55:16,839
Speaker 6: I had Jeff Gobwum in my head and I had

1163
00:55:16,840 --> 00:55:17,080
Speaker 6: to go.

1164
00:55:17,040 --> 00:55:20,960
Speaker 2: To the drastic fellas. This has been such a pleasure.

1165
00:55:22,560 --> 00:55:26,760
Speaker 2: Don't put that in the recording. That's right, we won't

1166
00:55:26,840 --> 00:55:29,920
Speaker 2: edit it in post. You will give us your names.

1167
00:55:30,400 --> 00:55:33,319
Speaker 4: I'm Mike Hicks, but my papers are written by Michael.

1168
00:55:33,000 --> 00:55:36,719
Speaker 1: Townsend Hicks and I'm a lecturer at the University of Glasgow.

1169
00:55:36,880 --> 00:55:39,600
Speaker 4: My website is Hicks.

1170
00:55:40,000 --> 00:55:42,279
Speaker 2: It will be in the podcast part. Don't you worry.

1171
00:55:42,280 --> 00:55:43,040
Speaker 4: We'll get that right.

1172
00:55:43,280 --> 00:55:44,799
Speaker 5: Just plugging my luggy staff.

1173
00:55:46,920 --> 00:55:47,960
Speaker 4: My name is Joe Slater.

1174
00:55:48,200 --> 00:55:51,240
Speaker 3: I'm a university lecturer in moral and political philosophy at LASKOW.

1175
00:55:52,200 --> 00:55:55,040
Speaker 6: I'm James Joffrees. I'm a lecturer in political theory at

1176
00:55:55,040 --> 00:55:57,279
Speaker 6: the University of Glasgow. And even if I wanted to

1177
00:55:57,360 --> 00:55:59,000
Speaker 6: or couldn't give you more websites, I don't have.

1178
00:56:01,440 --> 00:56:03,680
Speaker 2: Everyone you've been listening to Better Offline. Thank you so

1179
00:56:03,760 --> 00:56:05,920
Speaker 2: much for listening everyone, guys, thank you for joining me.

1180
00:56:06,760 --> 00:56:07,640
Speaker 5: Thanks having me here.

1181
00:56:07,920 --> 00:56:08,440
Speaker 6: This is.

1182
00:56:17,920 --> 00:56:20,360
Speaker 2: Thank you for listening to Better Offline. The editor and

1183
00:56:20,400 --> 00:56:23,560
Speaker 2: composer of the Better Offline theme song is Matasowski. You

1184
00:56:23,560 --> 00:56:25,800
Speaker 2: can check out more of his music and audio projects

1185
00:56:25,960 --> 00:56:29,480
Speaker 2: at Mattasowski dot com, M A T T O S

1186
00:56:29,560 --> 00:56:33,600
Speaker 2: O W s ki dot com. You can email me

1187
00:56:33,640 --> 00:56:36,239
Speaker 2: at easy at Better offline dot com or visit Better

1188
00:56:36,280 --> 00:56:38,719
Speaker 2: Offline dot com to find more podcast links and of course,

1189
00:56:38,760 --> 00:56:41,879
Speaker 2: my newsletter. I also really recommend you go to chat

1190
00:56:41,920 --> 00:56:44,560
Speaker 2: dot Where's youreed dot at to visit the discord, and

1191
00:56:44,600 --> 00:56:47,280
Speaker 2: go to our slash Better Offline to check out our reddit.

1192
00:56:48,080 --> 00:56:49,440
Speaker 2: Thank you so much for listening.

1193
00:56:50,280 --> 00:56:52,960
Speaker 7: Better Offline is a production of cool Zone Media. For

1194
00:56:53,080 --> 00:56:56,240
Speaker 7: more from cool Zone Media, visit our website cool Zonemedia

1195
00:56:56,320 --> 00:56:59,160
Speaker 7: dot com, or check us out on the iHeartRadio app,

1196
00:56:59,200 --> 00:57:15,080
Speaker 7: Apple Podcast, or wherever you get your podcasts.