Latest / Azadeh and Guy on AI / Can AI Be Your Next Interviewer?

Transcript

Can AI Be Your Next Interviewer? · Azadeh and Guy on AI

• 0:00
  SPEAKER_00: And we are on episode eight.
• 0:05
  SPEAKER_01: Hello, everyone.
• 0:06
  SPEAKER_01: We are recording the eighth episode of Azadeh and Guy on AI.
• 0:14
  SPEAKER_00: And we are doing this.
• 0:16
  SPEAKER_00: This time it's an experiment through Zoom.
• 0:18
  SPEAKER_00: So if you have quality problems, please complain to Zoom.
• 0:24
  SPEAKER_01: Yes.
• 0:24
  SPEAKER_00: But if it's great, let us know.
• 0:27
  SPEAKER_01: Sure.
• 0:28
  SPEAKER_01: Yeah.
• 0:29
  SPEAKER_01: So today we are talking about using AI to conduct interviews.
• 0:35
  SPEAKER_00: So this is something I've been fascinated with.
• 0:40
  Ever since I found out that one of the companies whose founder
• 0:44
  I listened to on one
• 0:46
  of the podcasts talked about the fact that they conduct a lot
• 0:49
  of their interviews
• 0:50
  SPEAKER_00: with the help of
• 0:53
  SPEAKER_00: And I was immediately really,
• 0:55
  really interested in the potential of collecting a lot more
• 0:59
  data and actually
• 1:01
  SPEAKER_00: creating.
• 1:02
  SPEAKER_00: So that's one thing.
• 1:03
  SPEAKER_00: The second thing that I thought about is that interviewing with AI can be
• 1:08
  SPEAKER_00: potentially uncomfortable,
• 1:09
  SPEAKER_00: right?
• 1:11
  SPEAKER_01: Yes, yeah.
• 1:13
  SPEAKER_00: AI and it's artificial and all of that.
• 1:16
  But if you're upfront about it and you give people an option
• 1:19
  and if they say,
• 1:20
  SPEAKER_00: no,
• 1:20
  SPEAKER_00: I want to talk to a person,
• 1:21
  SPEAKER_00: you give them that option.
• 1:22
  I think that there's an advantage where AI doesn't judge you,
• 1:26
  doesn't make faces.
• 1:27
  SPEAKER_00: It's not a potentially as biased.
• 1:30
  SPEAKER_00: So there's some advantages to the AI as an interview agent.
• 1:36
  SPEAKER_00: So we decided to do an episode on this.
• 1:39
  SPEAKER_01: Yeah, yeah, yeah.
• 1:41
  SPEAKER_01: And I don't think AI is still that much advanced in terms of,
• 1:47
  SPEAKER_01: I mean,
• 1:48
  SPEAKER_01: I'm talking about doing interviews with AI,
• 1:50
  SPEAKER_01: obviously.
• 1:51
  So I don't think there's still any application that reads human
• 1:59
  facial expressions
• 2:00
  SPEAKER_01: while doing the interviews.
• 2:02
  SPEAKER_01: Yeah.
• 2:03
  Yeah,
• 2:03
  I don't think we're quite there,
• 2:04
  but I think we will be there fairly soon.
• 2:08
  SPEAKER_00: But on the other hand,
• 2:10
  SPEAKER_00: not over-reading facial expression might be an advantage and not just a
• 2:15
  SPEAKER_00: disadvantage.
• 2:17
  Because sometimes we over-read or we over-interpret facial
• 2:26
  expression that may not
• 2:27
  SPEAKER_00: at all be connected to what we're talking about.
• 2:29
  SPEAKER_00: So there's an advantage there.
• 2:32
  SPEAKER_00: to try to do it a different way.
• 2:35
  SPEAKER_00: And all I'm saying is when we talked about doing analysis,
• 2:39
  SPEAKER_00: for example,
• 2:39
  SPEAKER_00: qualitative analysis with AI,
• 2:41
  SPEAKER_00: I'm not sure we're there yet either,
• 2:43
  but I think it's really important to play with it because it's
• 2:46
  going to be a
• 2:47
  SPEAKER_00: powerful research tool when it gets there.
• 2:50
  SPEAKER_00: And I do think you'll get there.
• 2:52
  SPEAKER_01: Yep, yep, yep, yep.
• 2:54
  SPEAKER_00: So the app or website that I'm talking about is actually 11 Labs.
• 3:00
  SPEAKER_00: And 11 Labs is an app that provided us with voice agents very early on.
• 3:06
  SPEAKER_00: And you can play with them.
• 3:07
  SPEAKER_00: You can learn English with them.
• 3:09
  SPEAKER_00: And that's lovely.
• 3:11
  But more recently,
• 3:13
  they actually created the ability for the user to create an
• 3:17
  agent.
• 3:18
  SPEAKER_00: What do I love about it?
• 3:20
  SPEAKER_00: First of all,
• 3:20
  SPEAKER_00: you tailor the agent so you can actually provide them the script,
• 3:26
  SPEAKER_00: the questions,
• 3:27
  SPEAKER_00: what prompt to use,
• 3:28
  SPEAKER_00: what kind of voice to use.
• 3:31
  SPEAKER_00: And you do also input a personality.
• 3:34
  SPEAKER_00: So just like you can do with any AI,
• 3:37
  SPEAKER_00: you can say,
• 3:38
  SPEAKER_00: so their initial one,
• 3:40
  SPEAKER_00: for example,
• 3:40
  SPEAKER_00: says,
• 3:41
  SPEAKER_00: you are Alexis.
• 3:42
  SPEAKER_00: That's the direction to the AI, not to us.
• 3:44
  A friendly,
• 3:45
  proactive,
• 3:45
  and highly intelligent female in a world-class engineering
• 3:48
  background.
• 3:49
  SPEAKER_00: Your approach is warm,
• 3:51
  SPEAKER_00: witty,
• 3:51
  SPEAKER_00: and relaxed,
• 3:52
  SPEAKER_00: effortless,
• 3:53
  SPEAKER_00: balancing professionalism with the chill,
• 3:56
  SPEAKER_00: approachable vibe.
• 3:58
  SPEAKER_00: So they give you that,
• 4:00
  SPEAKER_00: not just so you can use that specific voice,
• 4:04
  SPEAKER_00: but because it shows you how to prompt the AI in a robust way.
• 4:09
  SPEAKER_00: So there's actually much more of that.
• 4:14
  SPEAKER_00: in the prompt.
• 4:15
  SPEAKER_00: The prompt is quite long.
• 4:16
  It's a two,
• 4:17
  three page long prompt with guardrails and with,
• 4:21
  so there's a lot of detail.
• 4:23
  So you have,
• 4:24
  you can develop it very carefully,
• 4:25
  but you can think about this.
• 4:26
  SPEAKER_00: When we do interviews,
• 4:28
  we try to do a practice run and to provide a lot of guidelines
• 4:34
  to the people who
• 4:35
  are doing this because they need to know what they're doing and
• 4:37
  they need to know
• 4:38
  SPEAKER_00: how to do it.
• 4:39
  SPEAKER_00: This is the same thing, but you write it out in a very clear way.
• 4:43
  SPEAKER_00: And then you can choose a language,
• 4:47
  SPEAKER_00: which I love,
• 4:48
  because one of the things we do is we contact families and kids
• 4:53
  with different
• 4:53
  SPEAKER_00: first languages.
• 4:54
  SPEAKER_00: And here's our opportunity to do interviews in a variety of languages.
• 5:00
  SPEAKER_00: There are about 50 languages or so.
• 5:02
  SPEAKER_00: I'm not sure about the count, but there are a lot of languages.
• 5:05
  SPEAKER_00: So it gives you the ability to download the need
• 5:12
  SPEAKER_00: To have people speaking a lot of languages or to decide that you're not
• 5:17
  SPEAKER_00: interviewing anybody that speaks a language,
• 5:20
  SPEAKER_00: you don't have the capacity on the project.
• 5:22
  SPEAKER_00: Suddenly you can do that.
• 5:24
  SPEAKER_00: And I think that's magical.
• 5:26
  SPEAKER_01: Yes, it is.
• 5:29
  SPEAKER_00: And then you have a choice of voice.
• 5:34
  SPEAKER_00: You can have a choice of testing it.
• 5:37
  SPEAKER_00: So you can test it.
• 5:38
  SPEAKER_00: You can provide security on top of it.
• 5:40
  And you can create a widget so it can be launched within your
• 5:43
  website and you don't
• 5:44
  SPEAKER_00: have to bring everybody into 11 labs.
• 5:50
  SPEAKER_00: So this, to me, has tremendous potential now.
• 5:54
  SPEAKER_01: Go ahead.
• 5:59
  SPEAKER_01: Yeah, I have a question.
• 6:00
  So does it,
• 6:01
  the agent that you just described,
• 6:03
  does it act like a chat bot?
• 6:09
  SPEAKER_00: It is a voice agent.
• 6:11
  SPEAKER_00: So it acts like a chatbot, but it's voice.
• 6:15
  SPEAKER_00: So it listens to you, it responds to you, and it asks the question.
• 6:20
  There's also the option,
• 6:20
  of course,
• 6:21
  to record everything,
• 6:22
  which is what we want in an interview.
• 6:23
  SPEAKER_00: We want to record.
• 6:24
  SPEAKER_00: And there's the capacity to tell it exactly what the protocol is.
• 6:31
  SPEAKER_01: Okay, okay.
• 6:32
  So there's a variety of ways where you can interact with it and
• 6:39
  add all the
• 6:40
  SPEAKER_00: components that you want in an interview,
• 6:42
  SPEAKER_00: including,
• 6:43
  SPEAKER_00: again,
• 6:44
  SPEAKER_00: the tone,
• 6:45
  SPEAKER_00: including the exact language.
• 6:48
  SPEAKER_00: The questions,
• 6:49
  SPEAKER_00: including how to respond to follow-ups and all of this,
• 6:53
  SPEAKER_00: requires quite a bit of tinkering,
• 6:56
  SPEAKER_00: but you can tailor it,
• 6:58
  SPEAKER_00: you can test it,
• 6:59
  SPEAKER_00: and my favorite thing,
• 7:00
  SPEAKER_00: again,
• 7:01
  SPEAKER_00: is you can also put guardrails on it.
• 7:03
  SPEAKER_00: I think you should also let your participants know and...
• 7:08
  give them the option to opt out of working with an agent and
• 7:12
  work with a human
• 7:13
  because you want to be sensitive and you don't want people to
• 7:16
  not participate
• 7:17
  SPEAKER_00: because it's an AI.
• 7:20
  SPEAKER_01: Exactly.
• 7:21
  SPEAKER_01: But a huge part of that process is very similar,
• 7:25
  SPEAKER_01: as you mentioned,
• 7:26
  SPEAKER_01: to what we are doing in other interviewers.
• 7:30
  SPEAKER_01: I mean, we give them guidelines.
• 7:32
  SPEAKER_01: So the same thing is true about AI.
• 7:34
  But we need to be just more specific to make sure that it's
• 7:37
  asking the way that we intend to.
• 7:41
  SPEAKER_00: Yes.
• 7:42
  SPEAKER_00: And one of the things, for example, is a measure they call temperature.
• 7:46
  And the measure of temperature is how much leniency do you give
• 7:52
  the AI model,
• 7:53
  the LLM.
• 7:55
  SPEAKER_00: So you can say deterministic,
• 7:56
  SPEAKER_00: ask only the questions we have specified,
• 7:59
  SPEAKER_00: and you can get it to be creative or more creative.
• 8:02
  SPEAKER_00: And the way,
• 8:03
  SPEAKER_00: I think,
• 8:03
  to play something like that is to test it multiple times and
• 8:09
  see what happens when
• 8:10
  SPEAKER_00: you change it.
• 8:12
  SPEAKER_00: And then you can say, do I want it to go that way or not?
• 8:16
  I like,
• 8:17
  at least at the beginning,
• 8:18
  to think about it as more deterministic.
• 8:20
  I want to make sure it asks all the questions and in exactly
• 8:23
  the right way.
• 8:24
  SPEAKER_00: It means it won't have really smart follow-ups most of the time,
• 8:28
  SPEAKER_00: but it means I will get the content I really want right there.
• 8:33
  SPEAKER_00: So that's the balance of things that I'm thinking about.
• 8:36
  SPEAKER_01: Yes, yes.
• 8:37
  SPEAKER_01: I think we need to be on the safe side.
• 8:39
  SPEAKER_01: First,
• 8:39
  when you are trying the bot to make sure that it's doing
• 8:42
  exactly what you have to
• 8:44
  SPEAKER_01: do,
• 8:44
  SPEAKER_01: what you need to do.
• 8:45
  SPEAKER_01: Yeah.
• 8:47
  SPEAKER_00: Yeah.
• 8:48
  SPEAKER_00: And so that's part of the tools that are emerging.
• 8:53
  SPEAKER_00: And I think if you want to do a large scale interview study and research,
• 8:59
  you want to try and get all of this much quicker,
• 9:04
  this is the opportunity.
• 9:06
  If you're doing a study like we just did with four
• 9:09
  participants,
• 9:10
  I don't think we need that.
• 9:11
  SPEAKER_01: No, I don't think so, yeah.
• 9:13
  SPEAKER_00: I would still say stay on the safe side, be with humans.
• 9:17
  SPEAKER_00: But if you suddenly say, I want to do interview with 120 people,
• 9:21
  SPEAKER_00: That is hard, even from a scheduling perspective.
• 9:24
  So the ability to do that quickly with an agent allows you to
• 9:31
  spend your time otherwise.
• 9:32
  SPEAKER_00: Now, it's still going to cost you because 11 Labs is not free.
• 9:37
  SPEAKER_00: And with each conversation and all of that,
• 9:40
  you will eventually have to pay for either the API or for
• 9:44
  tokens or whatever the
• 9:46
  SPEAKER_00: mechanism says.
• 9:48
  SPEAKER_00: So you want to find out what it is,
• 9:50
  SPEAKER_00: but it does allow you,
• 9:51
  SPEAKER_00: if you have the funding,
• 9:52
  SPEAKER_00: it allows you to operate a lot faster.
• 9:54
  SPEAKER_00: And I think that's worth at least considering.
• 9:58
  SPEAKER_01: Yeah, yes, that's very true.
• 10:00
  SPEAKER_01: Yeah.
• 10:01
  SPEAKER_00: All right.
• 10:02
  SPEAKER_00: And we have two papers that connect to that.
• 10:04
  SPEAKER_00: So do you want to talk about the first one?
• 10:06
  SPEAKER_01: Sure.
• 10:07
  SPEAKER_01: Yeah.
• 10:08
  SPEAKER_01: So one of the papers is a kind of literature review.
• 10:11
  The title of the paper is AI Supported Individual Interviews
• 10:15
  Opportunities and
• 10:16
  SPEAKER_01: Threats for Evaluation.
• 10:18
  SPEAKER_01: It's been published in 2024,
• 10:22
  SPEAKER_01: and the last name of the author is Kupiek,
• 10:27
  SPEAKER_01: if I'm pronouncing it correct.
• 10:31
  SPEAKER_01: So they take a kind of evaluative view on...
• 10:36
  SPEAKER_01: not let's try AI interviewing now,
• 10:38
  but what does the body of literature suggest about integrating
• 10:42
  AI into every step
• 10:44
  SPEAKER_01: of qualitative interviewing in evaluation settings?
• 10:48
  SPEAKER_01: So what they do is that they break down the
• 10:52
  SPEAKER_01: all of the stages of a qualitative interview process into six categories,
• 10:58
  SPEAKER_01: which are planning,
• 10:59
  SPEAKER_01: developing the interview protocol,
• 11:01
  SPEAKER_01: sampling and recruitment,
• 11:03
  SPEAKER_01: data collection,
• 11:04
  SPEAKER_01: transcribing,
• 11:04
  SPEAKER_01: and finally coding and analysis.
• 11:07
  SPEAKER_01: And they try to see how much AI has been used in any of those stages.
• 11:14
  SPEAKER_00: Yes.
• 11:15
  SPEAKER_00: And how much research there is.
• 11:17
  SPEAKER_00: So as a literature review.
• 11:19
  SPEAKER_01: Yeah, exactly.
• 11:20
  SPEAKER_01: Yes, yes, yes.
• 11:21
  SPEAKER_01: How much research has been done in any of those areas.
• 11:26
  SPEAKER_01: So for the first one,
• 11:27
  SPEAKER_01: planning and instrument design,
• 11:30
  SPEAKER_01: we all know that AI is very good at generating ideas.
• 11:35
  SPEAKER_01: We can use it for drafting interview questions.
• 11:38
  SPEAKER_01: We can...
• 11:39
  SPEAKER_01: use AI to test and refine protocols.
• 11:42
  SPEAKER_01: And I know that they had a table.
• 11:44
  SPEAKER_01: I'm just going to pull up that table.
• 11:47
  SPEAKER_01: Yeah.
• 11:47
  SPEAKER_01: And the number of records for planning is five.
• 11:51
  SPEAKER_01: So people are using it in planning their interview.
• 11:56
  SPEAKER_01: The next stage is sampling and recruitment.
• 11:59
  SPEAKER_01: And this is a kind of weak spot.
• 12:02
  I'm also curious about knowing how AI is going to be useful to
• 12:06
  recruit interviewees.
• 12:09
  But the other idea that they have is that we can use AI to do a
• 12:13
  simulation.
• 12:16
  SPEAKER_01: And this part I also find useful.
• 12:19
  SPEAKER_01: I thought about the same thing for my own research,
• 12:23
  kind of doing simulating respondents for a kind of pilot phase
• 12:28
  of the study.
• 12:29
  SPEAKER_01: could be helpful.
• 12:32
  SPEAKER_01: And the number of records that they have is two for that one.
• 12:35
  SPEAKER_01: Sampling and recruitment.
• 12:37
  SPEAKER_01: We talked about that.
• 12:38
  SPEAKER_01: Data collection.
• 12:39
  SPEAKER_01: I mean, AI conducting the interviews.
• 12:45
  SPEAKER_01: For that one,
• 12:46
  SPEAKER_01: they have seven instances of AI using for doing interviews,
• 12:51
  SPEAKER_01: and there are emerging studies.
• 12:56
  SPEAKER_01: For example, researchers have used chatbots to collect data.
• 13:02
  SPEAKER_01: But there are still limitations around emotional sensitivity,
• 13:06
  SPEAKER_01: dynamic probes,
• 13:07
  SPEAKER_01: and also maintaining rapport with interviewees.
• 13:13
  SPEAKER_00: And we talked about the difficulties when we talked about 11 labs,
• 13:17
  SPEAKER_00: how fast and how well we can do that is important.
• 13:22
  SPEAKER_01: Exactly.
• 13:23
  Transcription,
• 13:24
  and I think you agree that we can use AI fairly well for doing
• 13:31
  the transcription.
• 13:34
  SPEAKER_01: The number of studies that have used AI for that purpose is 14.
• 13:40
  SPEAKER_00: And I think that that's just the studies that reported on it,
• 13:43
  SPEAKER_00: because everybody that's doing interviews right now is either using the
• 13:47
  SPEAKER_00: transcribing in Zoom or using one of the other platforms.
• 13:50
  And the human,
• 13:51
  instead of being the main transcriber,
• 13:54
  the human becomes just the checker.
• 13:59
  SPEAKER_00: Where, yeah, where there are two challenges.
• 14:04
  One is when there is a heavy accent or an unfamiliar to the
• 14:07
  machine accent that
• 14:09
  becomes a little more difficult and that requires more human
• 14:13
  intervention.
• 14:14
  It's not that it doesn't work,
• 14:15
  but it will require more human intervention.
• 14:18
  And the second one is in languages that are more rare,
• 14:23
  the accuracy goes down.
• 14:25
  SPEAKER_01: Yeah.
• 14:26
  SPEAKER_01: And the last category is coding and analysis.
• 14:28
  SPEAKER_01: So the number of instances that they mentioned in their study is 25.
• 14:34
  SPEAKER_01: And they say AI tools have been applied to generate initial codes,
• 14:38
  SPEAKER_01: group themes,
• 14:39
  SPEAKER_01: assist in inductive or deductive coding.
• 14:41
  SPEAKER_01: And I think
• 14:43
  SPEAKER_01: People are using it more and more,
• 14:44
  especially that AI has been embedded in many of those software
• 14:49
  that we use for
• 14:50
  SPEAKER_01: coding.
• 14:51
  SPEAKER_01: For example, MaxQDA and Vivo or other software.
• 14:54
  SPEAKER_01: Atlas CI, all of them have AI enabled in their software.
• 15:01
  SPEAKER_00: And so we've talked about those and we have a specific interest.
• 15:06
  But what I love about this review is really it's looking at the
• 15:10
  lifespan of the
• 15:11
  interview from design all the way to coding and basically
• 15:16
  showing how AI can be
• 15:18
  SPEAKER_00: used in almost all of them,
• 15:19
  SPEAKER_00: which is really fascinating to me.
• 15:22
  SPEAKER_01: Exactly, exactly, yes.
• 15:24
  And the thing that they are suggesting is that still we need
• 15:27
  human in the process.
• 15:29
  SPEAKER_01: You cannot say that AI can be used as a complement, not a replacement.
• 15:35
  So it can boost efficiency,
• 15:37
  it can expand the scale,
• 15:40
  but not supplant human expertise.
• 15:43
  SPEAKER_01: And there are still ethical issues.
• 15:45
  We have still transparency issues,
• 15:48
  bias and contextual sensitivity issues.
• 15:51
  SPEAKER_01: And all of them, they still remain as key challenges.
• 15:54
  SPEAKER_00: Yeah.
• 15:57
  SPEAKER_00: All right.
• 15:58
  SPEAKER_00: The second paper is a working paper from CES IFO.
• 16:02
  SPEAKER_00: It's CES IFO Conducting Qualitative Interviews with AI.
• 16:06
  SPEAKER_00: And it's by Chopra and Halland.
• 16:09
  And I really liked this one because they've used it in the
• 16:14
  research and measured a
• 16:15
  SPEAKER_00: variety of measures that are connected to how acceptable
• 16:22
  SPEAKER_00: this interview process is.
• 16:26
  SPEAKER_00: And what I'm seeing in some of the results,
• 16:30
  SPEAKER_00: and we're not going to go all over it because it's an 87-page document.
• 16:34
  SPEAKER_00: If you're interested, go look it up.
• 16:37
  SPEAKER_00: But...
• 16:40
  SPEAKER_00: When they talked about,
• 16:41
  SPEAKER_00: for example,
• 16:42
  SPEAKER_00: the estimation of the people participating,
• 16:44
  SPEAKER_00: about the naturalness of interviews,
• 16:47
  SPEAKER_00: if you take the very natural to extremely natural,
• 16:52
  SPEAKER_00: you're already at 41% of the participants.
• 16:55
  SPEAKER_00: And if you're taking somewhat natural, then you were at 75%.
• 16:59
  SPEAKER_00: What it says to me is...
• 17:02
  SPEAKER_00: It's natural enough for people not to feel off-put by it,
• 17:06
  SPEAKER_00: but there is 25% of people who are feeling not so okay about it.
• 17:11
  SPEAKER_00: So 75% were okay with it, 25% were like, this felt really awkward.
• 17:17
  SPEAKER_00: And the same thing about the overall experience,
• 17:23
  SPEAKER_00: the very positive and extremely positive were 60% of participants.
• 17:30
  That's a lot when you consider the fact that agents are new and
• 17:37
  some people feel
• 17:38
  SPEAKER_00: really strongly about AI.
• 17:39
  SPEAKER_00: And the same percentage almost,
• 17:44
  SPEAKER_00: so 55% or 53%,
• 17:47
  SPEAKER_00: sorry,
• 17:48
  SPEAKER_00: said they actually would have preferred AI to a human.
• 17:54
  SPEAKER_00: So more than half prefer the AI to a human.
• 18:00
  The other interesting thing about their research is that they
• 18:03
  developed the chatbot
• 18:06
  SPEAKER_01: and they published their paper in 2023.
• 18:10
  SPEAKER_01: So it was kind of... That was an old AI.
• 18:15
  SPEAKER_01: Yes, yes.
• 18:17
  SPEAKER_01: But they developed a multi-agent system and that's interesting.
• 18:21
  They have a security agent,
• 18:23
  they have a summary agent,
• 18:24
  topic agent,
• 18:25
  probing agent,
• 18:28
  and...
• 18:29
  SPEAKER_01: Yeah,
• 18:29
  so the process was interesting how they developed that kind of
• 18:33
  system to use AI for
• 18:35
  SPEAKER_01: doing the interviews.
• 18:38
  SPEAKER_00: So what you're really saying is
• 18:40
  SPEAKER_00: There's potential here.
• 18:42
  We're definitely going to start playing with it and seeing how
• 18:45
  people respond to it.
• 18:47
  SPEAKER_00: But we are encouraging you to just take a look.
• 18:50
  SPEAKER_00: If you're doing that kind of study or if it's just intriguing to you,
• 18:53
  SPEAKER_00: check out those two papers.
• 18:55
  Check out 11 Labs and see how you can use it and drop us a word
• 19:01
  and tell us how it went.
• 19:03
  SPEAKER_01: Sure.
• 19:03
  SPEAKER_01: Yeah.
• 19:04
  SPEAKER_01: It will be an interesting process.
• 19:09
  SPEAKER_00: All right.
• 19:10
  SPEAKER_00: So I think that's it for this time.
• 19:12
  SPEAKER_00: Eight episodes.
• 19:13
  SPEAKER_00: I feel like we're hitting our groove now.
• 19:17
  SPEAKER_01: Yes, definitely.
• 19:18
  SPEAKER_00: All right.
• 19:22
  SPEAKER_00: All right.
• 19:23
  SPEAKER_00: So we'll see you next time.
• 19:26
  SPEAKER_00: On Eaza Day.
• 19:28
  SPEAKER_01: And Guy.
• 19:28
  SPEAKER_00: On AAI.
• 19:30
  SPEAKER_01: All right.
• 19:34
  SPEAKER_00: We'll see you next time.