Latest / Azadeh and Guy on AI / Unlocking Research with AI

Transcript

Unlocking Research with AI · Azadeh and Guy on AI

• 0:00
  SPEAKER_02: All right.
• 0:01
  SPEAKER_02: Welcome.
• 0:02
  SPEAKER_02: We're in episode six.
• 0:03
  SPEAKER_02: Can you believe it's episode six?
• 0:05
  SPEAKER_00: I know.
• 0:06
  SPEAKER_00: I know.
• 0:06
  SPEAKER_00: It's crazy.
• 0:07
  SPEAKER_00: But I'm really excited.
• 0:09
  SPEAKER_02: All right.
• 0:10
  So with all of this excitement,
• 0:11
  we have a few really interesting things for you today.
• 0:14
  SPEAKER_02: The first thing is we're going to talk about... Research Rabbit.
• 0:18
  SPEAKER_02: Research Rabbit.
• 0:19
  SPEAKER_00: Another AI tool.
• 0:20
  SPEAKER_02: Another AI.
• 0:21
  SPEAKER_02: Well, it is a show about AI.
• 0:23
  SPEAKER_02: So I think we need AI tools.
• 0:26
  And then we're going to talk a little bit about the paper that
• 0:28
  looked at AI detectors.
• 0:31
  SPEAKER_02: That's always a subject that people are interested in.
• 0:35
  SPEAKER_02: And we have opinions, but there's also some empirical data.
• 0:39
  SPEAKER_02: So let's start with the research rabbit.
• 0:41
  SPEAKER_00: And before that, maybe we want to introduce ourselves.
• 0:45
  SPEAKER_00: So I'm Azadeh,
• 0:46
  SPEAKER_00: a third-year PhD student here at the University of Nebraska-Lincoln,
• 0:51
  SPEAKER_00: and I'm co-hosting this podcast with...
• 0:54
  SPEAKER_02: Guy Trenin.
• 0:55
  SPEAKER_02: I'm a professor here at UNL, and I'm also excited.
• 0:59
  SPEAKER_02: So let's go for it.
• 1:00
  SPEAKER_02: Research Rabbit.
• 1:01
  SPEAKER_00: Research Rabbit.
• 1:02
  SPEAKER_00: Research Rabbit is a tool or AI tool,
• 1:05
  SPEAKER_00: more specifically,
• 1:06
  SPEAKER_00: that you can use for your literature review.
• 1:10
  SPEAKER_00: Many students,
• 1:11
  especially those who are at the initial stages of their
• 1:15
  academic journeys,
• 1:16
  SPEAKER_00: struggle kind of with literature reviews and they need help,
• 1:21
  SPEAKER_00: especially when an advisor decides that they need to write five pages.
• 1:28
  SPEAKER_00: in a week.
• 1:29
  SPEAKER_00: So I've heard that struggle from many MA students specifically.
• 1:34
  SPEAKER_00: Research Rabbit is a tool that helps you create such kind of a thing.
• 1:40
  SPEAKER_00: And compared with Google Scholar,
• 1:42
  it provides you with different features that make the whole
• 1:47
  process of writing a
• 1:48
  SPEAKER_00: let's review section much easier.
• 1:50
  SPEAKER_02: And what I love about it is that it is free.
• 1:54
  SPEAKER_00: It is free, yes.
• 1:55
  So if you're looking for something free and that interfaces
• 1:59
  with some of the ways
• 2:00
  SPEAKER_02: you're working,
• 2:01
  SPEAKER_02: so for example,
• 2:02
  SPEAKER_02: Azadeh and I are working with Zotero,
• 2:04
  SPEAKER_02: it will connect to your Zotero.
• 2:05
  That's really important because you don't want to work in three
• 2:08
  different places to
• 2:09
  SPEAKER_02: create your literature list.
• 2:11
  SPEAKER_02: You want to bring it to one place where everything gets saved.
• 2:13
  SPEAKER_00: Yes, yes.
• 2:15
  The ability to connect Research Rabbit to Zotero is a very,
• 2:19
  I think,
• 2:20
  practical feature.
• 2:22
  The thing that came to my mind when I was trying out this tool
• 2:27
  was that it's not
• 2:29
  that much easy to navigate compared with ChatGPT or other tools
• 2:35
  or Asta or
• 2:37
  SPEAKER_00: Perplexity that we already talked about.
• 2:41
  SPEAKER_00: it may kind of look a little bit complicated.
• 2:47
  SPEAKER_02: And sometimes it's complicated just because they have a bad interface,
• 2:51
  SPEAKER_02: and sometimes it is because they're trying something new.
• 2:55
  So what are they doing that others aren't doing that may be
• 2:59
  helpful for you as a
• 3:02
  graduate student or even an undergraduate student who's looking
• 3:04
  for some resources?
• 3:05
  SPEAKER_00: Sure, yeah.
• 3:06
  The other prominent feature of that AI tool is that you don't
• 3:12
  interact with that
• 3:13
  SPEAKER_00: tool using prompting.
• 3:15
  SPEAKER_02: Okay.
• 3:16
  SPEAKER_00: So you feel that you do not have that kind of conversation.
• 3:19
  You are provided with a couple of useful features that you can
• 3:23
  easily click and
• 3:24
  SPEAKER_00: see,
• 3:25
  SPEAKER_00: and they have a kind of visualization feature that is very helpful,
• 3:29
  SPEAKER_00: but you do not interact with that tool like the other kind of tools.
• 3:35
  SPEAKER_02: So let's say I want to start a search inside Research Rabbit.
• 3:39
  SPEAKER_02: Where do I start?
• 3:40
  SPEAKER_02: So I have a collection.
• 3:44
  SPEAKER_02: What do I do?
• 3:44
  The first thing that you need to do is that you need to click
• 3:48
  on new collection on
• 3:50
  SPEAKER_00: the left side menu,
• 3:52
  SPEAKER_00: navigation menu.
• 3:53
  SPEAKER_02: All right.
• 3:54
  That's why I said that we had a conversation today and we
• 3:56
  talked about the fact
• 3:57
  that we may switch to video if we keep doing the it's on the
• 4:01
  left side kind of
• 4:02
  SPEAKER_02: thing.
• 4:02
  SPEAKER_02: We're like, well.
• 4:03
  SPEAKER_00: Maybe we need to do that.
• 4:05
  SPEAKER_00: I totally understand.
• 4:06
  SPEAKER_00: Sometimes hard to imagine.
• 4:07
  SPEAKER_02: Yes.
• 4:08
  And when you click on that new collection,
• 4:11
  then you start adding some seed papers.
• 4:15
  And seed papers are those papers that contain the most amount
• 4:19
  of information that
• 4:20
  SPEAKER_00: is really related to your research topic.
• 4:23
  So this feels to me a little bit more like using Google Scholar
• 4:28
  in that you have a
• 4:29
  SPEAKER_02: bar where you do a search and you start adding papers.
• 4:32
  SPEAKER_00: Yep, yep, yep, yep.
• 4:33
  SPEAKER_00: Very true, yes.
• 4:35
  SPEAKER_00: And when you add those papers to your collection, then you can...
• 4:40
  SPEAKER_00: Oh, I need to also mention that because I tried a few papers.
• 4:44
  SPEAKER_00: There's a glitch in the system.
• 4:46
  When you add a paper,
• 4:48
  a window,
• 4:48
  a small green window,
• 4:50
  if I'm not mistaken,
• 4:51
  pops up.
• 4:51
  So you need to wait until the progress bar just shows that the
• 4:56
  loading is finished.
• 4:58
  SPEAKER_00: Other than that, you cannot use the other features.
• 5:01
  SPEAKER_00: So sometimes there's a kind of glitch.
• 5:04
  It shows that it's adding the paper,
• 5:06
  but it seems that the process is not complete.
• 5:10
  SPEAKER_02: All right.
• 5:11
  SPEAKER_02: So we have some challenges,
• 5:12
  but what we really want to say here is that this is a good way
• 5:17
  to start a search.
• 5:18
  SPEAKER_02: This is not the ultimate way, and this will not get you everything.
• 5:21
  SPEAKER_02: So,
• 5:22
  SPEAKER_02: for example,
• 5:22
  if you're doing a meta-analysis and you need to capture
• 5:25
  everything in a field,
• 5:26
  SPEAKER_02: this is probably not the right tool for that.
• 5:29
  SPEAKER_00: Yeah, yeah.
• 5:30
  SPEAKER_02: But if you're looking for some hints and some paths through,
• 5:38
  SPEAKER_02: then that might be very,
• 5:41
  SPEAKER_02: very useful.
• 5:41
  SPEAKER_00: Yeah.
• 5:42
  So when you add a paper,
• 5:43
  then you can go and it shows you the title of the paper.
• 5:47
  SPEAKER_01: Mm-hmm.
• 5:49
  SPEAKER_00: And also the name of the journal just under that title.
• 5:53
  So if you select that paper,
• 5:57
  and now I understand the value of a video,
• 5:59
  because yeah.
• 6:00
  SPEAKER_02: Yeah, but in general, what you get is you get a network.
• 6:05
  SPEAKER_02: of the papers and how they're connected to each other.
• 6:07
  SPEAKER_02: So who are they citing, who cites them, and all of that.
• 6:10
  So you can start seeing how the field is organized and not just
• 6:15
  individual studies.
• 6:16
  SPEAKER_02: And this is where the difference,
• 6:18
  SPEAKER_02: I think,
• 6:18
  from Google Scholar to this is really helpful because here you
• 6:22
  can see the
• 6:22
  connections between the papers and who are the authors that are
• 6:26
  more prominent and
• 6:26
  SPEAKER_02: less prominent.
• 6:27
  SPEAKER_00: Yeah,
• 6:28
  and the other good thing about Research Rabbit is that when you
• 6:32
  select a paper,
• 6:33
  SPEAKER_00: you can also add comments.
• 6:35
  SPEAKER_00: There is a very small icon.
• 6:37
  SPEAKER_00: I'm saying again, they are not very good at their user interface.
• 6:40
  SPEAKER_00: So for some of the features,
• 6:42
  SPEAKER_00: they are useful,
• 6:43
  SPEAKER_00: but you need to spend some time to explore and see how everything works.
• 6:47
  SPEAKER_00: It's not very intuitive compared with others.
• 6:50
  SPEAKER_00: Yes.
• 6:51
  SPEAKER_00: So when you click on a paper,
• 6:54
  SPEAKER_00: then you can see the title of the paper,
• 6:56
  SPEAKER_00: the journal name,
• 6:57
  SPEAKER_00: and also the abstract of the paper.
• 6:59
  And what is very helpful is that it gives you another menu for
• 7:05
  each paper that you select.
• 7:06
  It gives you another menu,
• 7:08
  and there are some very useful options on that.
• 7:10
  SPEAKER_00: For example, similar work.
• 7:13
  SPEAKER_00: which makes your search much easier.
• 7:16
  SPEAKER_00: So it gives you a couple of papers that are related,
• 7:20
  SPEAKER_00: thematically related to the paper that you selected initially.
• 7:25
  So what I'm saying is it's really important to have a narrow
• 7:30
  enough definition
• 7:31
  SPEAKER_02: without being too narrow.
• 7:33
  This is always the magic of finding the right literature is
• 7:37
  narrowing your search
• 7:38
  SPEAKER_02: so you're not getting 100,000 hits.
• 7:41
  SPEAKER_02: but also not narrowing it too much so you're getting two.
• 7:44
  SPEAKER_02: And you need to be somewhere in the middle.
• 7:46
  SPEAKER_02: I did a search on computational thinking,
• 7:48
  SPEAKER_02: something we're interested in,
• 7:49
  SPEAKER_02: and I'm getting 2,417 similar works.
• 7:52
  SPEAKER_02: I'm obviously not going to look at all 2,417.
• 7:54
  So the thinking is who do we want to identify and what are the
• 8:00
  key works in the
• 8:01
  SPEAKER_02: field that would help?
• 8:03
  SPEAKER_01: Mm-hmm.
• 8:04
  SPEAKER_02: That's where my brain is going to because this is just too much,
• 8:09
  SPEAKER_02: definitely,
• 8:10
  SPEAKER_02: if you're a graduate student,
• 8:11
  SPEAKER_02: but for anybody.
• 8:12
  SPEAKER_02: We have no capacity to process that many studies.
• 8:15
  SPEAKER_02: So having a strategy.
• 8:17
  So let's kind of bring it to a short summary because this is
• 8:22
  part of – we talked about SciSpace.
• 8:25
  SPEAKER_02: We talked about –
• 8:28
  SPEAKER_02: Which one?
• 8:29
  SPEAKER_00: We talked about perplexity.
• 8:31
  SPEAKER_02: We talked about perplexity.
• 8:32
  SPEAKER_02: We obviously talk about deep research often, which can do the same thing.
• 8:39
  SPEAKER_02: And so there are lots of tools out there.
• 8:41
  SPEAKER_02: For you, what are the benefits of using this specific tool?
• 8:45
  SPEAKER_00: Research Rabbit.
• 8:47
  SPEAKER_00: I really like the idea that I can explore references and also citations.
• 8:51
  By references,
• 8:52
  I mean all of the papers that the first paper cited in their
• 8:56
  thing.
• 8:57
  SPEAKER_00: And also citations are the works that refer to this paper.
• 9:01
  SPEAKER_00: So I find that feature very useful.
• 9:03
  SPEAKER_00: Because you cannot easily navigate those things in Google Scholar.
• 9:07
  SPEAKER_00: It's much as straightforward.
• 9:08
  SPEAKER_00: Okay.
• 9:08
  And what I love about this is the networks allow you to do this
• 9:13
  visual analysis
• 9:14
  SPEAKER_02: that kind of reminds me of the way we used to do research,
• 9:18
  which is look in the card catalog and then go to the shelf at
• 9:23
  the library and find
• 9:24
  SPEAKER_02: what's next to it and all of that.
• 9:25
  That happenstance that when you're doing a regular search may
• 9:29
  not happen quite as
• 9:30
  SPEAKER_02: effectively because of the network effect.
• 9:34
  SPEAKER_02: you are able to browse very quickly and see if it's a good fit without,
• 9:39
  SPEAKER_02: like you do in Google Scholar,
• 9:41
  SPEAKER_02: for example,
• 9:41
  SPEAKER_02: opening each one separately.
• 9:42
  SPEAKER_02: So that's definitely an advantage for me.
• 9:45
  The one thing that I want you all to remember is just like
• 9:49
  anything else on the
• 9:50
  SPEAKER_02: Internet,
• 9:51
  it can become a rabbit hole where you keep pushing on the next
• 9:54
  paper and the next
• 9:55
  SPEAKER_02: paper and the next paper.
• 9:57
  SPEAKER_02: And then you have to remember why you started this whole thing.
• 10:01
  SPEAKER_00: Yeah.
• 10:01
  SPEAKER_02: And what are some of the challenges you're seeing with this one?
• 10:05
  SPEAKER_00: As I told you before, the interface is kind of challenging.
• 10:10
  SPEAKER_00: I'm not saying that it's not beneficial.
• 10:12
  SPEAKER_00: It has very good features, but you need to be patient.
• 10:15
  SPEAKER_00: I mean, if you want to explore that tool, you need to be very patient.
• 10:20
  SPEAKER_00: There is also another thing.
• 10:22
  SPEAKER_00: It's not challenging, but it's helpful.
• 10:23
  SPEAKER_00: When you're talking about the visualization, it gives you a network.
• 10:27
  SPEAKER_00: The other feature is that it gives you a timeline.
• 10:30
  SPEAKER_00: which I find interesting.
• 10:32
  SPEAKER_00: It kind of shows all of those papers in a kind of chronological,
• 10:37
  SPEAKER_00: if I can pronounce that,
• 10:38
  SPEAKER_00: chronological order,
• 10:39
  SPEAKER_00: yeah.
• 10:41
  SPEAKER_00: And let me see.
• 10:43
  SPEAKER_00: I also, oh, fact-checking.
• 10:47
  So we talked about it before when we were looking at Research
• 10:51
  Rabbit together.
• 10:52
  So when it comes to all references,
• 10:54
  it doesn't show you all of the references,
• 10:56
  right?
• 10:57
  SPEAKER_00: That's a paper use.
• 10:58
  SPEAKER_00: There is a kind of discrepancy.
• 11:01
  SPEAKER_00: And you need to,
• 11:02
  SPEAKER_00: like all of those AI tools that are available,
• 11:06
  SPEAKER_00: you need to do the fact checking.
• 11:08
  SPEAKER_02: Yeah,
• 11:08
  SPEAKER_02: and especially if the numbers and the references need to be exact,
• 11:14
  you want to make sure that you verify everything and you don't
• 11:17
  take anything for
• 11:19
  SPEAKER_02: granted.
• 11:20
  SPEAKER_02: Kind of buyers beware,
• 11:22
  SPEAKER_02: especially when they're giving you something for free,
• 11:24
  SPEAKER_02: make sure that what you're getting is good because it's not always clear.
• 11:30
  SPEAKER_02: All right, I think it's time to switch over, right?
• 11:32
  SPEAKER_02: Yes, yes.
• 11:33
  SPEAKER_02: I'm really excited to talk about this paper.
• 11:35
  SPEAKER_02: This is a paper by Perkins, Rowe, Vu, Postma, Hickerson.
• 11:41
  SPEAKER_02: And it's about... McGoggerhan and what?
• 11:45
  SPEAKER_02: Oh, okay.
• 11:46
  SPEAKER_02: And what I love about this,
• 11:48
  SPEAKER_02: it's called Simple Techniques to Bypass Gen AI Text Detectors,
• 11:51
  SPEAKER_02: Implications for Inclusive Education.
• 11:54
  What I love about this,
• 11:55
  this is something I wrote about just recently,
• 11:57
  and that is...
• 11:58
  This is produced from the British University of Vietnam and the
• 12:04
  James Cook
• 12:05
  SPEAKER_02: University in Singapore.
• 12:07
  And this is the ascendancy that I'm both delighted by,
• 12:11
  but also concerned about.
• 12:12
  SPEAKER_02: And that is the impact of research conducted in East Asia,
• 12:20
  SPEAKER_02: especially China,
• 12:20
  SPEAKER_02: but beyond China,
• 12:21
  SPEAKER_02: Singapore is becoming an AI giant and all of that,
• 12:24
  SPEAKER_02: that
• 12:27
  What used to come mostly from the United States and Europe is
• 12:30
  now coming from all
• 12:32
  SPEAKER_02: around the world with two spots that are really challenging.
• 12:35
  So we're seeing a lot less from Africa and a lot less from some
• 12:38
  from America.
• 12:39
  SPEAKER_02: This is something to worry about.
• 12:41
  SPEAKER_02: But this is a great piece because I'm really interested.
• 12:44
  Every time we talk about AI,
• 12:46
  people who are not familiar talk about detectors.
• 12:49
  So it's really nice to see some empirical work and not just the
• 12:53
  discussion about the detectors.
• 12:54
  SPEAKER_00: Yeah.
• 12:55
  SPEAKER_00: So before you start talking about the details of that paper,
• 12:58
  SPEAKER_00: could you tell me what was the main takeaway?
• 13:02
  SPEAKER_02: So the main takeaway of this is very simple.
• 13:04
  SPEAKER_02: It's fairly simple to use strategies to make the AI undetectable.
• 13:12
  SPEAKER_02: And they actually outline those.
• 13:15
  I'm not going to outline all the strategies,
• 13:17
  but people know some of them.
• 13:19
  SPEAKER_02: Varying sentence lengths is one feature that they're saying.
• 13:24
  SPEAKER_02: Spelling mistakes is another one.
• 13:26
  SPEAKER_02: And obviously detectors get better with time.
• 13:29
  SPEAKER_02: But what they're saying is, yes, the detectors will get better.
• 13:32
  but also the AIs and their ability to create texts that are not
• 13:39
  easily detectable
• 13:40
  SPEAKER_02: is going to climb up.
• 13:42
  SPEAKER_02: And we actually know that there are,
• 13:44
  SPEAKER_02: because I go on social media and I get this,
• 13:47
  SPEAKER_02: these people who say,
• 13:48
  SPEAKER_02: hey,
• 13:49
  SPEAKER_02: if you use our engine,
• 13:51
  SPEAKER_02: it'll take your AI text and make it undetectable.
• 13:54
  SPEAKER_02: So it was nice to see this.
• 13:57
  SPEAKER_02: It's part of the discussion about detectors.
• 14:00
  There are some detectors that are emerging that seem to be
• 14:02
  doing a little bit
• 14:03
  SPEAKER_02: better or a lot better.
• 14:05
  SPEAKER_02: But what is obvious is there's going to be the constant escalation.
• 14:09
  SPEAKER_02: The more you have AI detectors that are better,
• 14:11
  SPEAKER_02: you're going to have AI engines that are better at...
• 14:14
  SPEAKER_02: Being undetectable.
• 14:16
  SPEAKER_02: So that's part of it.
• 14:18
  What I loved about this paper is that it was an empirical
• 14:21
  paper,
• 14:22
  which I love.
• 14:23
  SPEAKER_02: I know not everybody does.
• 14:24
  SPEAKER_02: I do.
• 14:25
  SPEAKER_02: They produced a lot of texts,
• 14:28
  and then they changed them based on different algorithms that
• 14:32
  change both AI text
• 14:34
  SPEAKER_02: and human text to see what happens.
• 14:36
  SPEAKER_00: And I think they had three type of texts, right?
• 14:40
  SPEAKER_00: Original AI text, human text, and also manipulated AI text.
• 14:44
  SPEAKER_02: Yes.
• 14:45
  And so they're really trying to see how good are the detectors
• 14:49
  at detecting the AI text.
• 14:51
  SPEAKER_02: And they're fairly good.
• 14:52
  SPEAKER_02: And when I say fairly good, it's not very good.
• 14:55
  SPEAKER_02: At the most, 70% when it's just AI.
• 14:57
  SPEAKER_02: Again, this goes back.
• 14:59
  SPEAKER_02: Remember, there's a publishing gap.
• 15:02
  SPEAKER_02: It always takes time to publish.
• 15:04
  SPEAKER_02: So this is from a while back.
• 15:06
  SPEAKER_02: The paper is from 24.
• 15:08
  SPEAKER_02: The data is most likely from 23.
• 15:09
  SPEAKER_02: But...
• 15:12
  SPEAKER_02: It was clear that they were not very good to begin with.
• 15:15
  SPEAKER_02: And once they started to manipulate the text,
• 15:19
  the detectors kind of fell off a cliff and were really bad at
• 15:25
  detecting whether a
• 15:27
  SPEAKER_02: person wrote it or AI wrote it.
• 15:30
  SPEAKER_02: And those are things that we need to consider.
• 15:34
  SPEAKER_02: My personal take is...
• 15:38
  SPEAKER_02: I feel like it's futile to chase this,
• 15:41
  SPEAKER_02: taking from the previous segment,
• 15:43
  SPEAKER_02: this rabbit that runs around.
• 15:45
  SPEAKER_02: Who's going to be better?
• 15:46
  Am I going to have a better detector or are you going to have a
• 15:49
  better AI that is undetectable?
• 15:52
  SPEAKER_02: I don't want to have that race.
• 15:54
  SPEAKER_02: I want to have a conversation about AI.
• 15:56
  SPEAKER_02: I want to teach my students to use AI.
• 15:58
  SPEAKER_02: And there's one more thing.
• 15:59
  SPEAKER_02: I thought about this last night.
• 16:01
  SPEAKER_01: Uh-huh.
• 16:02
  SPEAKER_02: Because that's all I do is think about AI.
• 16:05
  SPEAKER_02: And what I thought about this,
• 16:07
  SPEAKER_02: if kids learn to...
• 16:08
  SPEAKER_02: I think about my history with Grammarly,
• 16:10
  SPEAKER_02: right?
• 16:10
  SPEAKER_02: Grammarly taught me where to put commas.
• 16:13
  SPEAKER_02: And I'm much better than I used to be at commas.
• 16:17
  SPEAKER_02: But that means that I put commas the way Grammarly does.
• 16:20
  SPEAKER_02: So now I think I kind of project this.
• 16:23
  SPEAKER_02: If you're learning to write with the help of AI,
• 16:27
  SPEAKER_02: so you write yourself and then AI gives you feedback,
• 16:30
  SPEAKER_02: you're...
• 16:31
  SPEAKER_02: writing will start resembling the AI.
• 16:34
  SPEAKER_00: That's very true.
• 16:36
  SPEAKER_02: So the question then becomes,
• 16:38
  is this AI or is this somebody who learned how to write with
• 16:42
  the help of AI?
• 16:43
  SPEAKER_02: So those patterns are starting to show up.
• 16:47
  And I'm not going to stop using AI for feedback because I think
• 16:51
  it does improve
• 16:53
  SPEAKER_02: some qualities of writing.
• 16:55
  SPEAKER_02: So I'm going to encourage my students to at least use it for feedback.
• 17:00
  SPEAKER_02: which raises the question of convergence around those issues.
• 17:04
  SPEAKER_02: So that's another thing to think about.
• 17:05
  SPEAKER_00: Yeah,
• 17:05
  SPEAKER_00: I'm thinking about at the same time,
• 17:07
  and I think that's where it's very important to decide when you
• 17:14
  want to use AI.
• 17:15
  SPEAKER_00: Mm-hmm.
• 17:16
  When you have enough foundational knowledge that you know where
• 17:19
  you are going to,
• 17:20
  SPEAKER_00: and then you use AI just as a support,
• 17:23
  which emerged a lot as a kind of theme during the many
• 17:27
  interviews that I had with
• 17:28
  SPEAKER_00: graduate students.
• 17:30
  SPEAKER_00: So many of them believe that when you have enough knowledge,
• 17:33
  SPEAKER_00: when you know what you're going to know,
• 17:35
  SPEAKER_00: when you know about your destination,
• 17:37
  SPEAKER_00: then you can use AI.
• 17:39
  SPEAKER_00: Okay.
• 17:40
  Yeah,
• 17:40
  and that goes back to that general idea of the novice problem,
• 17:44
  right?
• 17:44
  If you're a novice and if you're a total expert,
• 17:47
  AI is going to help you a lot less.
• 17:49
  It's in the areas where you have some knowledge or you know
• 17:52
  where you want to go,
• 17:52
  SPEAKER_02: but you need that extra oomph.
• 17:54
  SPEAKER_02: That's where it's most helpful.
• 17:56
  SPEAKER_00: Yeah, yeah.
• 17:57
  And that's when you understand that you are not just simply
• 18:01
  affected by what AI told you.
• 18:03
  SPEAKER_00: Yeah, yeah, yeah.
• 18:04
  SPEAKER_00: Yes.
• 18:06
  SPEAKER_02: All right.
• 18:06
  SPEAKER_02: So today on...
• 18:09
  SPEAKER_00: And Guy on AI.
• 18:13
  SPEAKER_02: We talked about... This is good.
• 18:18
  SPEAKER_02: We did remember what the name of the show is.
• 18:20
  We talked about the AI detectors and we also talked about our
• 18:27
  rabbit that helps us
• 18:29
  SPEAKER_02: find the research rabbit that helps us find the things we need to know.
• 18:35
  SPEAKER_02: And we'll see you next time on episode 7.
• 18:37
  SPEAKER_02: Sure.