쉐도잉 연습: Using AI to read research articles - 영상으로 영어 말하기 배우기

When we start reading an article, I normally recommend students that they focus on first identifying the key concepts

and finding the definitions of those key concepts from the article, and only after that start reading the study.

But there is also another way of how you can read and understand research article with the help of AI.

And I must emphasize that what I present now is likely to be outdated sometime in the future.

So this has been recorded in the fall 2024

and this is the second time that I record a video on how we use AI to read articles.

I did the first one about a year or year

and a half ago and it is now completely outdated

because there are so many new tools and this process has become

so much more simpler than it was when ChatGPT first came out.

So let's take a look at how we understand this article by sapiens

and co-authors on capabilities perspective on the effects of early internationalization on firm survival and growth.

And the traditional way of reading the article would be that we try to first identify the concepts.

This article provides a figure so we would know that from this figure every box is concept.

We would have to understand what those concepts are, how they're defined before we start reading it

and maybe look a bit about these arrows like what kind of theories are used to justify those arrows.

You often find the definitions in the article, sometimes you can google the definition, sometimes you need to go and read some of the sources that the article writes.

So how can we make this more efficient?

We can use a large language model for the task.

ChatGPT is probably the most well-known one

but I'll talk about a number of other services or products in this talk as well.

Before we get started there is an issue about copyright.

So how about copyright law?

Are we allowed to upload an article that is behind paywall that is copyrighted by somebody else to these large language models?

This is something that we have been discussing at the university and the law is not super clear on this.

So the use conditions or the instructions of

OpenAI simply say that you need to make sure that you're not violating any law when you are using OpenAI's ChatGPT.

So they don't take a stand on the issue of copyright.

And the law is generally not super clear here.

There is something called personal use exemption, which means that you can, for example, copy books from a library to your own use,

but you're not allowed to publish them.

So it's completely okay to copy copyrighted material as long as it is for your own use.

And it is thought

that using these tools are covered by the personal use exemption as long as the data

that you provide to the large language model are not included in the training data of the next model.

So before you upload any copyrighted material to these services, make sure that you opt out from training.

So this is available in the settings of all these products.

So opt out giving your data to train the next model and you should be fine.

Of course the law might change then there is a EU level AI act coming soon

but it's still too early to say anything about that.

All right now let's get into actual tools.

The university recommends and I have a screencast on how to do this using Microsoft Edge.

So how you would use Microsoft Edge and Co-Pilot is

that you open the PDF of the article in Microsoft Edge and then you start asking questions about the PDF.

Before you do this make sure that you're logged in and you have this green icon here.

So it means that it's using open IE servers that reside within the EU area and it also indicates that your

data are not being used to train new models.

So if you have this green symbol here, then you should be okay with copyrighted materials.

We can ask Microsoft Edge and Co-Pilot to summarize the article for you.

So we are asking summarize the article and then provide the summary.

So this is like a short summary and it's pretty good.

So it mentions early internalization, dynamic capabilities, moderating factors, growth and survival and looks decent.

Yeah, that should get us started.

We can also ask Copilot to explain the concepts in the article to us.

So for example, we can ask what imprinting means and it gives us a pretty good answer.

We can do the same for research fungibility and it gets it correctly.

So this is a pretty good way of reading an article.

So you can read it here and when you see something

that you don't understand you ask Copilot to explain it to you.

There is one caveat with using a Copilot within Microsoft Edge and it is that it produces you answers from the website,

from internet and from the article that you have opened.

Make sure that you are asking it to summarize the article

that you have open and ask questions like based on the article how is imprinting defined.

Otherwise it might do a web search for you and imprinting is used in different contexts for different reading.

So unless you are sure that the answer comes from the pdf that you have open it might be incorrect.

So this is one one caveat about this particular tool.

And then you can also ask something like questions that are not specifically about the content of this article,

but how this article might apply to you.

So, for example, if we ask, we tell co-pilot that we are doing a master's thesis on venture capital

and we want to know if this article is directly relevant to our thesis.

And it says that the article is relevant.

I don't think that it is.

So these models tend to have a bias toward being positive.

So when you ask a question, they are biased toward agreeing.

If we ask something that is completely unrelated, for example, we ask that we are doing a thesis on consumer behavior

and particularly on the willingness to pay for healthcare services,

then GPT or co-pilot tells us that this article is probably not very relevant

but then nevertheless it tries to find us some kind of relevance.

So even if this would be completely irrelevant it tries to find like an angle from which this might be relevant.

And this is a general bias of these tools.

So these tools are programmed to be helpful and this is a demonstration of what that means.

So ChatGPT is programmed to be helpful to answer your questions even if the questions were really really stupid.

So we are asking here to use a general theory of relativity which is from physics,

that's Einstein's theory, to explain how entrepreneurs identify opportunities.

And the model says that well this is an unusual request

but yeah we can use it as a metaphor and then it starts to explain about curvature, space, time, and market landscape, and gravity while the market caps and needs.

Just connecting completely unrelated things in a way that might look logical.

So this answer is complete nonsense, but because the pro the tpt is programmed to be helpful it tries to help even if it doesn't make sense.

A clawed opening the second tab does a bit better.

It just tells that these are two completely different things and they should be combined.

So whenever you use these chatbots to evaluate something, make sure that you understand that they're biased toward agreeing.

So if you're saying that should I read this study, the likely response is that yes you should read the study even

if the study is only trivial relevant for you or completely irrelevant.

All right so this is the Microsoft Edge co-pilot combination

that the university recommends but there are other alternatives and some of these are upcoming.

So for example, PDF readers have integrated AI assistants.

This is the latest Adobe Acrobat and this is a paid service.

You can use a few few queries per month I think

or day and then you have to start paying

100

and we can open the article in the PDF viewer

101

and we can ask it to give us an overview of the article

102

and what is the main idea and it starts to explain the main idea to us.

103

It correctly explains that this is about early internalization, it correctly identifies the authors but then it starts to explain something

104

that they likely discuss how they develop specific capabilities such as market knowledge,

105

network building, adaptability can influence the outcome of firms venturing into the international markets.

106

Well that's not, maybe the article says something about that

107

or something about those concepts like network building but that's not really the key point of the article.

108

So the LLM or the chatbot or whatever you like to call it AI goes a bit wrong here.

109

So reading the AI summary of the article is not the substitute for reading the article

110

because at least at this point of time the AI still are doing kind of like 80-90% correct job

111

and then 10 or 20% incorrect.

112

We can ask follow-up questions, we can explain or ask you to explain the concepts and it gives us explanations.

113

This looks pretty good but it's pretty basic so it didn't really talk about the more challenging concepts like imprinting or fungibility.

114

It just tells us that the firm growth which is self-evident and then survival which is also pretty obvious what it means.

115

So this is the PDF readers and Apple is releasing the previous software on Mac

116

that is AI powered as well

117

and this is likely to be integrated in pretty much every PDF reader from major major providers in the future.

118

Then the third way of using AI chatbots to help you understand articles is to use the web interface

119

and I'm using the chat GPT.

120

This is the GPT-4.0

121

and how it works is that you upload the article it's just drag and drop it here to the chat box

122

and then you ask it to explain the article to you.

123

It's very useful to give it a bit more context, like what is your level of understanding.

124

If you're a master's student, then tell that you're a master's student.

125

If you are a doctoral student in entrepreneurship or international business, then provide that information

126

because how the article is explained should depend on the level under previous understanding by the person listening to the explanation.

127

And it does a pretty good job of explaining.

128

It gives us what is the research problem, what's the theoretical framework, what are the main arguments, and then you can ask follow-up questions like what is imprinting,

129

what is fungibility, what role does imprinting play in the theory, that kind of questions.

130

Or you can ask questions

131

that are outside of the article like where does the idea of dynamic capabilities come from

132

and then the model responds based on its training data.

133

This is an example using Claude.

134

So this is another free software.

135

Of course, there are paid versions of this as well, but this is the free one.

136

And we can ask it to provide us a table of the key concepts in the article.

137

And this table is provided as marked down.

138

It's a bit hard to read here, but here it's shown in full.

139

So we have these concepts.

140

Some of these concepts are...

141

or most of these concepts are central to the article

142

so there's no nothing unnecessary and then I asked for a definition

143

and then I asked for an explanation for undergraduates

144

but the definitions are already pretty clear so the explanation for undergraduate is not that

145

that useful in this context but this just shows

146

that you can do all kinds of things with these these chat bots

147

and you can ask for let's say you could ask an explanation for a five-year-old

148

and then you would get a really simple explanation so

149

if the ai model explains things on a level

150

that is too challenging for you just tell

151

that english is not my main language explain in finnish explain

152

in swedish explain it like you would explain it to a kindergarten student

153

and that so and it tailors the level of sophistication of the output to whatever you prompt.

154

But you have to tell the context, you have to explain what kind of summary you need.

155

Then the fourth way of reading articles is various services and this is illicit.

156

I'm just using this as an example.

157

There are probably a dozen of these services already

158

that are well known and we don't know

159

which one is going to be the standard

160

which one is going to be the winner of the market at this point

161

but Elicit is made by a non-profit and it's kind of like a search engine, an article summarization tool aimed for researchers

162

so you can use this to ask questions about literature

163

or you can upload your own pdfs

164

and then ask it to write different like synthesize various pdfs together what do these articles say together.

165

The article, the PDF processing is a paid feature and I paid for this to show you.

166

So we can upload our document here and we can ask it to explain the article and it does an okay job.

167

Some of these key concepts here like for example positional advantage,

168

it is mentioned in the article but I wouldn't list that as one of the main concepts in the theory.

169

So it talks about positional advantage

170

but that's not one of the boxes in the diagram

171

and it's also not one of the main theories that explain the arrows in the boxes and arrows diagram.

172

So it's a bit of a miss here.

173

You can use the same service for asking questions like what do we know about the effects of early internalization

174

and then it tries it identifies it pulls papers from from a citation database

175

and then it writes a summary of those papers for you.

176

So if you are new to the top, new to a topic like for example if you're working on a master's thesis then this is a free service.

177

The PDF process is paid so you can use this free service to get started.

178

Like just have an open-ended question like what do we know about dynamic capabilities, what do we know about employee motivation and so on and then it gives you like a short summary

179

based on some recent research articles.

180

Okay, to wrap up, these AR tools are useful for getting the big picture in simpler terms.

181

So you can ask summaries, you can ask explanations of an article,

182

but they are not a substitute for reading the full article for two reasons.

183

One is that typically there is nuance in the article that the summary doesn't capture, but more importantly the AI models do mistakes.

184

Like we saw several mistakes here, but if you have like an 80% correct understanding of what the article is about, then you are in a lot better position to actually read

185

that article than if you started reading the article without knowing anything about the topic.

186

And finally, these tools are likely to improve in the future.

187

So this is the fall 2024 version.

188

If you're listening this in in 2028 for example then this is probably very outdated

189

and finally make sure that if you upload something to llms opt out from training if you're uploading copyrighted materials

📺 같은 채널

Introduction to statistical prediction models

Business EnglishUniversity StudentsWorking Adults

✨ 추천 영상

79 문장

Can you learn English with AI?

Business EnglishDaily ConversationUniversity Students

126 문장

How I use AI as a senior software engineer and 5 principles of AI explained in 10 minutes

Business EnglishUniversity StudentsWorking Adults

182 문장

We Let $200 GPT-Pro Think for 7 Days on Unsolved Math

Business EnglishUniversity StudentsWorking Adults

145 문장

13k

Business EnglishUniversity StudentsWorking Adults

103 문장

Amazon Bedrock Tutorial | Getting Started with Amazon Bedrock | AI Model Access & Testing

Business EnglishUniversity StudentsWorking Adults

408 문장

Understand search engine optimization (SEO) | Google Digital Marketing & E-commerce Certificate

Business EnglishUniversity StudentsWorking Adults

55 문장

When AIs act emotional

Business EnglishUniversity StudentsWorking Adults

254 문장

AI Will Create New Wealth, But Not Where You Think | Carnegie Mellon University, Po-Shen Loh

Business EnglishDaily ConversationUniversity Students

407 문장

이 레슨의 어휘와 말하기 포인트

이 영상에는 섀도잉할 문장 189개와 단어 2672개가 있습니다. 말하는 구간의 길이는 16:32입니다. 화자는 분당 약 162단어로, 일상 대화에 가까운 자연스러운 속도로 말합니다. 단어의 86%가 영어에서 가장 많이 쓰이는 3,000단어에 속합니다. 나머지는 연습 전에 미리 확인해 두세요.

이 영상의 핵심 어휘

영상에 나오는 덜 흔한 단어 15개를 발음, 뜻과 함께 정리했습니다.

summary /ˈsʌm.ə.ɹi/ (명사) — 요약, 개요. An abstract or a condensed presentation of the substance of a body of material.
explanation /ˌɛkspləˈneɪʃən/ (명사) — 설명. The act or process of explaining.
upload /ˈʌpˌloʊd/ (동사) — 업로드하다, 올리다. To transfer data to a computer on a network, especially to a server on the Internet.
bias /ˈbaɪ.əs/ (명사) — 편견. Inclination towards something.
arrow /ˈæɹəʊ/ (명사) — 화살. A projectile consisting of a shaft, a point and a tail with stabilizing fins that is shot from a bow.
thesis /ˈθisɪs/ (명사) — 테제, 논제. A proposition or statement supported by arguments.
survival /sɚˈvaɪ.vəl/ (명사) — 생존. The fact or act of surviving; continued existence or life.
reader /ˈɹi.dɚ/ (명사) — 독자, 읽는 사람. A person who reads.
correctly /kəˈɹɛk(t).li/ (부사) — 정확(正確)히. In a correct manner.
outdated /aʊtˈdeɪtɪd/ (형용사) — 구식이다, 철 지나다. Out of date, old-fashioned, antiquated.
topic /ˈtɑpɪk/ (명사) — 화제, 주제. A subject; a theme; a category or general area of interest.
venture /ˈvɛn.t͡ʃɚ/ (동사) — 감행하다. To undertake a risky or daring journey.
incorrect /ˌɪn.kəˈɹɛkt/ (형용사) — 그른, 잘못된. Not correct; erroneous or wrong.
diagram /ˈdaɪ.ə.ɡɹæm/ (명사) — 다이어그램, 도표. A plan, drawing, sketch or outline to show the function or operation of something, or to show the relationships between the parts of a whole.

영상에 나오는 구동사

come out /ˌkʌm ˈaʊt/ (동사) — 나오다. To be discovered; to be revealed.
log in (동사) — 로그인하다. To gain access to a computer system, usually by providing a previously registered username and password.
wrap up (동사) — 시마이치다. To cover or enclose (something) by folding and securing a covering entirely around it.

주의할 발음

화자는 you're, don't, doesn't 같은 축약형과 약화된 형태를 22번 사용합니다. 들리는 대로 짧게 발음하세요.

“th” 소리: thesis /ˈθisɪs/, nevertheless /ˌnɛvɚðəˈlɛs/, theoretical /ˌθi.əˈɹɛt.ɪ.kəl/, synthesize /ˈsɪnθəsaɪz/
“sh”와 “zh” 소리: explanation /ˌɛkspləˈneɪʃən/, exemption /ɪɡˈzɛm(p).ʃən/, unusual /ʌnˈjuːʒ(u)əl/, capture /ˈkæp.(t)ʃɚ/, efficient /ɪˈfɪʃənt/
긴 단어 — 강세 위치에 주의: explanation /ˌɛkspləˈneɪʃən/, capability /ˌkeɪ.pəˈbɪl.ɪ.ti/, irrelevant /ɪˈɹɛləvənt/, unusual /ʌnˈjuːʒ(u)əl/, nevertheless /ˌnɛvɚðəˈlɛs/

이 영상으로 연습하는 방법

먼저 말하지 않고 영상을 끝까지 듣고 모르는 단어를 적어 둡니다.
0.75배속으로 한 문장씩 섀도잉을 시작하고, 익숙해지면 보통 속도로 돌아갑니다.
자신의 목소리를 녹음해 원본과 비교하고, summary, copyright, explanation 같은 단어에 특히 주의합니다.

이 영상의 문법

화자가 가장 많이 쓰는 문형을 영상 속 실제 표현과 함께 정리했습니다.

문형	영상 속 표현
수동태 be + 과거분사 — 누가 하는지보다 무슨 일이 일어나는지에 초점	be outdated · has been recorded · is copyrighted
현재완료 have/has + 과거분사 — 과거의 일이 지금도 관련이 있을 때	has been recorded · has become · have opened
관계절 who / which + 절 — 사람이나 사물에 대한 추가 정보	exemption, which means · relativity which is · growth which is

쉐도잉이란? 영어 실력을 빠르게 키우는 과학적 방법

쉐도잉(Shadowing)은 원래 전문 통역사 훈련을 위해 개발된 언어 학습 기법으로, 다언어 학자인 Dr. Alexander Arguelles에 의해 대중화된 방법입니다. 핵심 원리는 간단하지만 매우 강력합니다: 원어민의 영어를 들으면서 1~2초의 짧은 지연으로 즉시 소리 내어 따라 말하는 것——마치 '그림자(shadow)'처럼 화자를 따라가는 것입니다. 문법 공부나 수동적인 청취와 달리, 쉐도잉은 뇌와 입 근육이 동시에 실시간으로 영어를 처리하고 재현하도록 훈련합니다. 연구에 따르면 이 방법은 발음 정확도, 억양, 리듬, 연음, 청취력, 말하기 유창성을 크게 향상시킵니다. IELTS 스피킹 준비와 자연스러운 영어 소통을 원하는 분들에게 특히 효과적입니다.

섀도잉 방법: 단계별 전체 가이드 읽기 →

쉐도잉 연습: Using AI to read research articles - 영상으로 영어 말하기 배우기

Introduction to statistical prediction models

Can you learn English with AI?

How I use AI as a senior software engineer and 5 principles of AI explained in 10 minutes

We Let $200 GPT-Pro Think for 7 Days on Unsolved Math

13k

Amazon Bedrock Tutorial | Getting Started with Amazon Bedrock | AI Model Access & Testing

Understand search engine optimization (SEO) | Google Digital Marketing & E-commerce Certificate

When AIs act emotional

AI Will Create New Wealth, But Not Where You Think | Carnegie Mellon University, Po-Shen Loh

Why AI Tokens are so Expensive - Computerphile

MIT Introduction to Deep Learning | 6.S191

How to Use Google Gemini Al (Full Tutorial)

이 레슨의 어휘와 말하기 포인트

이 영상의 핵심 어휘

영상에 나오는 구동사

주의할 발음

이 영상으로 연습하는 방법

이 영상의 문법

쉐도잉이란? 영어 실력을 빠르게 키우는 과학적 방법