⭐ 한 줄 thesis
EU AI Act 에 따라 진짜처럼 보이는 AI 콘텐츠에 라벨·워터마크가 의무화되지만, 업계는 라벨 남발이 cookie banner 처럼 무시당해(label fatigue) 보호 없이 부담만 줄 것을 우려한다.
“artificially generated images, audio and text designed to look authentic must be labelled”
🔢 수치 · 고유명사 (원문 대조 완료)
| 항목 | 값 | 원문 |
|---|
| 벌금 상한 | €15m | fines of up to €15m (£12.8) or 3% of a company’s worldwide global turnover |
| 벌금 비율 (EU 매출 ✗ → 전 세계 매출) | 3% | 3% of a company’s worldwide global turnover |
| 적용 시작(신규 시스템) | 2 August | The rules apply to new AI systems on the EU market from 2 August |
| 기존 시스템 유예 | four months | existing ones have an extra four months to comply |
| TikTok 라벨 건수 | 3bn | it has helped label more than 3bn pieces of content |
| Google SynthID 출시 | 2023 | Google launched a tool called SynthID in 2023 |
| SynthID 워터마크 이미지 | 100bn | on more than 100bn images and 60,000 years’ worth of audio |
| SynthID 워터마크 오디오 | 60,000 years | 60,000 years’ worth of audio |
| 실천 규범 서명 기관 | 180 organisations | more than 180 organisations had signed the code |
| Meta 정책 도입 | 2024 | Meta launched policies in 2024 |
| Reinisch 발표일 | 28 July | earlier this week [28 July] |
| Sergey Lagodinsky | Sergey Lagodinsky | the Green MEP Sergey Lagodinsky, who helped negotiate the AI Act |
| Michal Šimečka | Michal Šimečka | the Slovakian opposition leader Michal Šimečka |
| Boniface de Champris | Boniface de Champris | CCIA Europe’s AI policy lead, Boniface de Champris |
| Markus Reinisch | Markus Reinisch | its vice-president for public policy in Europe, Markus Reinisch |
| Henna Virkkunen | Henna Virkkunen | The commission’s lead official on tech policy, Henna Virkkunen |
🔑 빈출 어휘
| compulsory | adj. 의무적인, 강제의 |
| comply | v. (규칙을) 준수하다 |
| synthetic | adj. 합성된, 인공의 |
| watermark | n. 워터마크(출처 식별 표시) |
| editorial oversight | n. 편집상의 감독(사람의 검토) |
| exemption | n. 면제, 적용 제외 |
| satirical | adj. 풍자적인 |
| turnover | n. 매출액 |
| purport | v. ~이라고 (거짓으로) 주장하다, ~인 척하다 |
| versed in | adj. ~에 정통한, 익숙한 |
| dubious | adj. 의심스러운, 수상한 |
| non-signatory | n. (협약) 비서명자 |
🎯 정답이 되는 문장 / 오답이 되는 전형 함정
정답형Synthetic content designed to look truthful must be visibly marked and contain a digital watermark.“synthetic text, images, video and audio designed to look truthful must be visibly marked as AI-generated and contain a digital watermark”
정답형The tech industry supports making deepfakes visible but fears overly broad interpretation.“The tech industry says it supports the quest to make deepfakes visible, but fears the rules will be interpreted too broadly”
정답형De Champris expects the most visible change in advertising, film and publishing rather than social media.“in areas where AI is already used at scale but people don’t know it – advertising, film, publishing”
과장✗ All AI-generated content, including users’ personal content, must be labelled.→ 개인 콘텐츠는 적용 제외.
“The rules will not apply to users’ personal content”
반대✗ Labelling AI content created before the rules is compulsory.→ 권장일 뿐 의무 아님.
“although this is not compulsory”
원문 모순(worldwide)✗ Fines can reach 3% of a company’s turnover within the EU.→ 원문은 worldwide global turnover.
“3% of a company’s worldwide global turnover”
반대✗ The tech industry opposes the effort to make deepfakes visible.→ 취지는 지지, 과도한 해석을 우려.
“The tech industry says it supports the quest to make deepfakes visible”
일반화✗ Existing AI systems must comply from 2 August as well.→ 기존 시스템은 4개월 추가 유예.
“existing ones have an extra four months to comply”
지문에 없음✗ Companies that do not sign the code of practice are fined.→ 비서명자는 다른 방식으로 준수를 입증하면 됨.
“non-signatories must show they are meeting the rules in other ways”
과장✗ Companies must use the EU’s official black-and-white labels.→ 누구나 쓸 수 있는 라벨일 뿐, 자체 디자인 가능.
“although companies can design their own”
⭐ 한 줄 thesis
알고리즘 피드는 개인화를 내세워 오히려 개인 취향을 소멸시켰다 — 진짜 취향을 되찾으려면 알고리즘에서 벗어나 제약·우연·자기 질문으로 취향을 다시 길러야 한다.
“it has wrecked our capacity to form our own preferences”
🔢 수치 · 고유명사 (원문 대조 완료)
| 항목 | 값 | 원문 |
|---|
| Kyle Chayka 『Filterworld』 | 2024 | In his 2024 book Filterworld, Kyle Chayka |
| Portobello Road market | Portobello Road market | to Portobello Road market in west London |
| Kerry 빈티지 사업 | eight years | Over the eight years she has had her vintage clothing business, Kerry |
| Stephanie (캘리포니아) | 37 | Stephanie, 37, is visiting from California |
| Helena 스타일리스트 | 25-year-old | Helena, a 25-year-old stylist |
| Ione Gamble · Polyester 창간 | 2014 | In 2014, she founded Polyester |
| Nicola Dinan 비유 | a driverless car | feeling like “a driverless car” |
| Carolyn Bessette Kennedy 사망 | 1999 | in a 1999 plane crash |
| Carole Radziwill | Carole Radziwill | Bessette Kennedy’s friend Carole Radziwill |
| Chaotic Good Projects | trend simulation | a tactic it calls “trend simulation” |
| Lane Brown · clipping | Lane Brown | the New York magazine writer Lane Brown explained the art of “clipping” |
| 위장 광고 (업계 1인 추정) | 90% | One person in the industry estimated that “90% of what you see on the internet is advertising in disguise” |
| Susan Sontag 『Notes on Camp』 | 1964 | in her seminal 1964 essay Notes on Camp, Susan Sontag |
| Pierre Bourdieu 『Distinction』 | 1979 | the French sociologist Pierre Bourdieu in his 1979 work Distinction |
| Grayson Perry 태피스트리 | 2012 | Grayson Perry’s 2012 tapestry series The Vanity of Small Differences |
| Nathalie Olah 『Bad Taste』 | Nathalie Olah | Nathalie Olah’s book Bad Taste |
| Greg Brockman (OpenAI 사장) | taste is a new core skill | OpenAI’s president, Greg Brockman, posted on X that “taste is a new core skill” |
| Bezos Met Gala | $10m | pay a reported $10m to serve as “honorary chairs” |
| AI 생성 이미지 비율 | 71% | about 71% of the images shared online are now AI-generated |
| AI 팟캐스트 비율 | more than a third of podcasts | as are more than a third of podcasts |
| Spotify AI 트랙 삭제 | 75m | last year Spotify removed 75m AI tracks |
| Carmen Vicente 뉴스레터 | Scroll Sick | a newsletter about the internet titled Scroll Sick |
| SNS 사용 시간 정점 (FT) | 2022 | time spent on social media peaked in 2022 |
| 게시 감소 (Ofcom) | 12% | a 12% drop in people posting on platforms in the past year |
| Perfectly Imperfect · Tyler Bainbridge | 2020 | was founded in 2020 by Tyler Bainbridge |
| PI.FYI 이용자 | 200,000 | With only 200,000 customers, PI.FYI is niche |
| Letterboxd 이용자 | 26 million | has about 26 million |
| Erin Wylie · Blackbird Spyplane | Blackbird Spyplane | Erin Wylie, who co-founded the cult style newsletter Blackbird Spyplane |
| Pip (모자 장인) 방문 기간 | 20 years | has been coming here for the past 20 years |
🔑 빈출 어휘
| debase | v. (가치·품위를) 떨어뜨리다 |
| serendipitous | adj. 우연히 얻은(뜻밖의 행운의) |
| tailored | adj. ~에 맞춘, 맞춤형의 |
| commodify | v. 상품화하다 |
| oblivion | n. 망각, 소멸 |
| herd mentality | n. 군중 심리 |
| ubiquity | n. 어디에나 있음, 편재 |
| risk-averse | adj. 위험을 회피하는 |
| subcontract out | v. 외주를 주다(남에게 맡기다) |
| moat | n. 해자 → (기업의) 구조적 경쟁 우위 |
| veneer | n. 겉치장, 허울 |
| slop | n. (질 낮은 AI 생성) 찌꺼기 콘텐츠 |
| idiosyncratic | adj. 특이한, 개인 특유의 |
| counterintuitive | adj. 직관에 반하는 |
🎯 정답이 되는 문장 / 오답이 되는 전형 함정
정답형Platforms built personalisation into their business model yet erased individual taste.“these platforms made personalisation a major part of their business model, then synthesised, commodified and automated individual taste into oblivion”
정답형Algorithms tend to promote the least disruptive, least ambiguous culture.“the least ambiguous, least disruptive and perhaps least meaningful pieces of culture are promoted the most”
정답형Chayka sees tech firms’ use of taste as PR that hides automation.“a veneer of humanity to hide automation behind”
반대✗ Algorithms promote the most meaningful and challenging culture.→ 가장 덜 의미 있는 것이 가장 많이 홍보.
“perhaps least meaningful pieces of culture are promoted the most”
과장✗ The author found Portobello Road market completely free from the algorithm’s influence.→ 겉모습은 그대로지만 상인·손님 모두 군중 심리를 말함.
“all of whom describe a herd mentality among their customers and peers”
발화자 혼동✗ Chayka said taste is “a tool to make you feel more like yourself”.→ 그 말은 Ione Gamble.
“is a tool to make you feel more like yourself,” says Gamble”
반대✗ Tech leaders claim AI will easily replicate human taste.→ AI 가 복제 못 할 인간 자질이라는 생각.
“the one human quality artificial intelligence will not be able to replicate”
수치 혼동✗ About 71% of podcasts are now AI-generated.→ 71%는 이미지, 팟캐스트는 1/3 이상.
“about 71% of the images shared online are now AI-generated, as are more than a third of podcasts”
일반화✗ The article proves that 90% of online content is advertising.→ 업계 한 사람의 추정일 뿐.
“One person in the industry estimated”
반대✗ The author concludes that smarter shopping is the answer to the taste crisis.→ 쇼핑은 분명 답이 아니다.
“Shopping is obviously not the answer”
수치 혼동✗ PI.FYI is a large mainstream platform with millions of users.→ PI.FYI 는 20만 명(niche), 약 2,600만은 Letterboxd.
“With only 200,000 customers, PI.FYI is niche in itself”
⭐ 한 줄 thesis
Fobo(fear of better options)는 선택지가 모두 괜찮은데도 결정하지 못하는 현상으로, 기술이 선택 과잉을 키웠고 maximiser 는 헌신의 이점을 놓쳐 덜 만족한다.
“Indecision when the decision is simple, or the options all acceptable, is the defining characteristic of “fear of better options” – or Fobo”
🔢 수치 · 고유명사 (원문 대조 완료)
| 항목 | 값 | 원문 |
|---|
| Mike Hall | 48 | Mike Hall, 48, a management consultant based in Winchester |
| Patrick McGinnis | Patrick McGinnis | a social phenomenon coined by Patrick McGinnis, a US venture capitalist |
| Aoife O’Donaghue | 24 | Aoife O’Donaghue, 24, a recent graduate based in Edinburgh |
| O’Donaghue 고민 시간 | 15 minutes | working herself up over this for 15 minutes |
| 팟캐스트 | Fomo Sapiens | his podcast Fomo Sapiens |
| Amazon 신발끈 선택지 | 200 choices | you have in excess of 200 choices |
| 50년 전 Woolworths | three | 50 years ago you would go to Woolworths and choose between three |
| 용어 탄생 | 15 years ago | in a Harvard Business School paper 15 years ago |
| Nicky Lidbetter | Anxiety UK | Nicky Lidbetter, chief executive of the charity Anxiety UK |
| satisficer 명명 | 1956 | the Nobel laureate Herbert Simon in 1956 |
| 연구 연도·주도자 | 2011 | one from 2011 conducted by a team led by Joyce Erlingher from Florida State University |
| 게재 학술지 | Personality and Individual Differences | published in the journal Personality and Individual Differences |
| Hall 작년 칠면조 | four attempts and three hours | it took four attempts and three hours for me to buy a turkey |
🔑 빈출 어휘
| indecision | n. 우유부단, 결정 못 함 |
| coin | v. (새 말을) 만들어 내다 |
| analysis paralysis | n. 분석 마비(따지다가 결정 못 함) |
| dithering | n. 망설임, 우물쭈물 |
| narcissism | n. 자기애, 나르시시즘 |
| affliction | n. 고통, 병폐 |
| affluence | n. 풍요, 부유 |
| euphemism | n. 완곡어 |
| mourn | v. 애도하다, 슬퍼하다 |
| maximiser | n. 최대화 추구자 |
| satisficer | n. 만족화 추구자(적당하면 만족) |
| portmanteau | n. 혼성어(두 단어를 합친 말) |
🎯 정답이 되는 문장 / 오답이 되는 전형 함정
정답형Fobo is indecision even when all the options are acceptable.“Indecision when the decision is simple, or the options all acceptable”
정답형Technology has accelerated Fomo and Fobo into common behaviours.“has accelerated Fomo and Fobo into a common social behaviour”
정답형Maximisers tend to be less satisfied because they commit less.“Maximisers miss out on the psychological benefits of commitment”
개념 혼동✗ Fobo mainly comes from comparing ourselves with others on social media.→ 그것은 Fomo. Fobo 는 선택 과잉.
“compare ourselves with each other (thus producing feelings of Fomo) and overwhelm ourselves with choice (producing Fobo)”
반대✗ McGinnis says Fobo is an entirely new behaviour created by the internet.→ 꼭 새로운 행동은 아님 — 기술이 가속.
“Fobo is not necessarily a new human behaviour”
반대✗ Anxiety UK classifies Fobo as an anxiety disorder in its own right.→ 독자적 장애로 분류될 정도는 아님.
“as opposed to being sufficient to warrant being categorised as an anxiety disorder in its own right”
반대✗ Satisficers are less satisfied with their choices than maximisers.→ 덜 만족하는 쪽은 maximiser.
“are maximisers more likely to be unhappy with their choice”
일반화✗ McGinnis believes both Fomo and Fobo are entirely harmful.→ 약간의 Fomo 는 괜찮다.
“So a little Fomo is fine. But Fobo is not good.”
인물 혼동✗ Patrick McGinnis coined the term “satisficer”.→ Herbert Simon(1956).
“first coined by the Nobel laureate Herbert Simon in 1956”
반대✗ Fobo affects only individuals, not organisations.→ 대기업·국가(Brexit)도.
“large corporations can be affected by it”
지문에 없음✗ Hall has already bought this year’s turkey.→ 어느 것을 살지 정했을 뿐.
“He has already decided which one.”
A1
⚖️ Motion: “AI labels should be required whenever AI generated content looks authentic.”
Agree 딥페이크가 민주주의를 위협 → 진짜처럼 보이면 표시해야 속지 않는다(Lagodinsky).
Disagree 라벨이 어디에나 있으면 cookie banner 처럼 무시당해 보호 효과가 사라진다(De Champris).
A2
⚖️ Discussion Topic: “Social media algorithms help us discover our authentic tastes.”
Agree 다른 사용자 데이터로 몰랐던 것을 만날 수 있고, 진짜 사랑하게 됐다면 그건 내 취향.
Disagree 과거 선호의 반복 + 가장 무난한 것만 홍보 + 인기 자체가 조작될 수 있음.
A2
⚖️ Motion: “Following trends does not necessarily make you less individual.”
Agree 내 취향이 먼저였는데 유행이 된 경우(Helena)도 있고, 규칙을 비틀려면 먼저 알아야 한다.
Disagree ‘어울리고 싶다’는 군중 심리 + CBK-core 같은 문자 그대로의 베끼기.
A2
⚖️ Motion: “AI will ultimately reduce human creativity”
Agree AI = algorithmic sameness 의 후계자(slop 악순환), 빅테크는 조용한 사색을 막는다.
Disagree 취향은 AI 가 복제 못 할 인간 자질이라는 주장 + 인간적 선호에 대한 갈증·아날로그 회귀.
A2
⚖️ Central Discussion Question: “Can we ever have completely authentic personal taste?”
Agree Helena 의 아빠·Pip — 온라인 밖에서 기른 본능.
Disagree 취향은 계층·배경이 형성하고 소속을 알리는 사회적 요소가 있어 ‘completely’ 는 어렵다.
A3
⚖️ Technology has given us more freedom, but less satisfaction.
Agree 신발끈 200개 이상 vs 3개 / 기술이 사람을 maximiser 로 만든다.
Disagree Fobo 는 꼭 새로운 행동이 아니다(생물학) / 성격의 문제라는 당사자들.
A3
⚖️ Motion: Having too many choices makes people less free.
Agree 선택지에 압도돼 결정을 못 하면 자유가 아니다(analysis paralysis).
Disagree 선택지는 권력·특권 / 문제는 선택 방식(satisficer, overthinking).
템플릿 ① Algorithm → Personal taste 논증형 (A2 중심, 🔴 예고 주제)
구조: 정의 → 메커니즘 → 비교(personalization vs personal choice) → 지문 예시 2개 → 평가/양보 → 해결책 → 결론
- Personal taste is simply what we like and what we don’t, yet it lies at the core of our identities.
- Today, however, algorithmic feeds show each user content based on ___ in order to keep them on the platform as long as possible.
- This creates a paradox: although platforms promise personalisation, personalisation is not the same as personal choice, because ___.
- For example, the article explains that Spotify ___.
- Moreover, ___ (Chaotic Good’s “trend simulation” / “clipping”) shows that even popularity can be manufactured.
- As a result, many people, like the stylist Helena, can no longer tell whether they have been influenced or whether a preference is truly their own.
- Admittedly, ___ (an algorithm can introduce us to something we come to genuinely love).
- A realistic solution is to ___ (build “productive constraints” / use algorithm-free platforms / ask ourselves why we like something).
- In this way, we can turn personalisation back into genuine personal choice.
모범 단락 (9문장)
Personal taste is simply what we like and what we don’t, yet the article argues that it lies at the core of our identities. Today, however, most of us encounter culture through algorithmic feeds, which show each user content based on their past activity and that of other users in order to keep them on the platform for as long as possible. This creates a paradox: platforms made personalisation central to their business model, but personalisation is not the same as personal choice, because we no longer choose what we consume; we take what we are given. For example, Spotify serves songs with superficial similarities to the tracks we did not skip last time, so our experience is tailored around previous preferences rather than new discoveries. Moreover, the marketing firm Chaotic Good used many social media accounts to create “trend simulation,” showing that even popularity can be manufactured. As a result, people like the stylist Helena now ask whether they have been influenced or whether a style is actually theirs. Admittedly, an algorithm may occasionally introduce us to something we come to love, but the test is whether we genuinely love it or have simply been shown it many times. A realistic solution is to build “productive constraints,” as Erin Wylie did by wearing only black clothes she already owned for a month, and to ask ourselves why we like something. In this way, we can turn personalisation back into genuine personal choice.
모범 단락의 사실 근거(원문):
“it’s what you like and what you don’t”
“based on data gathered from their own activities and those of other users”
“We no longer choose what we want to consume; we take what we’re given.”
“songs with superficial similarities to the tracks they didn’t skip last time”
“a tactic it calls “trend simulation””
“she would spend an entire month wearing only black clothes she already owned”
템플릿 ② 교차형 — ‘Too much’: label fatigue(A1) · Fobo(A3) · 알고리즘 피드(A2)
구조: 공통 개념 정의(과잉이 판단을 마비) → A1 예 → A3 예 → A2 연결 → 비교 → 해결책 → (개인 경험 1문장)
- All three articles suggest that ___ (more is not always better): too much information or choice can weaken our judgement.
- In the EU article, critics warn that if labels appear everywhere, users will stop noticing them, just as they ignore cookie banners.
- Similarly, the Fobo article shows that ___ (when options are overwhelming, people delay commitment or cancel at the last minute).
- The taste article adds that algorithmic feeds give us content in such overwhelming quantities that ___.
- However, the cases differ: ___ (labels aim to protect the public, whereas algorithms aim to keep us scrolling).
- In my own experience, ___ (I often click “accept” on cookie banners without reading them).
- Therefore, a better approach is ___ (meaningful labels for deceptive content / satisficing with modest criteria / productive constraints).
모범 단락 (7문장)
All three articles suggest that more is not always better, because too much information or choice can weaken our judgement. In the article on EU rules, critics such as Boniface de Champris warn that once labels are everywhere, users stop noticing them, just as people ignore cookie banners; this is what we can call “label fatigue.” Similarly, the Fobo article explains that technology lets us overwhelm ourselves with choice, so people hold back on commitment or cancel at the last minute, like Aoife O’Donaghue, who worried for fifteen minutes about whether to study in the library or a cafe. The article on personal taste adds that algorithmic feeds give us content in such overwhelming quantities that we no longer have the mental capacity to digest and assess it. However, the cases differ in purpose: labels are meant to protect the public from deception, whereas algorithms are designed to keep us on platforms for as long as possible. In my own experience, I often click on cookie banners without reading them, which shows how quickly a warning can turn into noise. Therefore, a better approach is to make labels meaningful by focusing on deceptive content and, as individuals, to make choices with modest criteria like satisficers rather than endlessly searching for a better option.
모범 단락의 사실 근거(원문):
“once labels are everywhere, users stop noticing them”
“overwhelm ourselves with choice (producing Fobo)”
“Someone with Fobo is likely to hold back on commitment, or commit then cancel.”
“we no longer have the mental capacity to properly digest and assess what we have encountered”
“The label was meant to flag deceptive content”
“will make choices based on a modest criteria”
🚫 쓰지 말아야 할 표현
| “Algorithms always destroy our taste.” | 과장 — 원문은 ‘seriously debased – if not completely destroyed’ 로 수위 조절. always/never 금지. |
| “AI can never be creative.” | 원문에 없는 단정. 원문은 테크 업계의 ‘생각’(the thinking goes)과 Chayka 의 견해로 제시. |
| “The article says 90% of the internet is advertising.” | 업계 한 사람의 추정(estimated). ‘one person in the industry estimated’ 로. |
| “The EU banned deepfakes.” | 금지가 아니라 라벨·워터마크 의무화. |
| “The article uses the term ‘label fatigue’.” | 원문엔 그 단어가 없다. cookie banner 비교로 설명된 개념 → what we can call “label fatigue” 로. |
| “Chayka coined Fobo.” / “Gamble wrote Filterworld.” | 인물 혼동 — Fobo = Patrick McGinnis, Filterworld = Kyle Chayka. |
| “Tomato girl means …” (구체 설명) | 원문은 미학 이름만 나열 — 내용 설명은 지문에 없음. |
| kinda / stuff / a lot of things / I think it’s bad | 구어체·막연한 표현 — 2025 FRQ2 채점에 formal academic English 형식 점수가 있었음. |
✍️ Complete 문장 모범 완성
✍️ Complete: “A preference becomes truly ‘mine’ when ___________”
- A preference becomes truly ‘mine’ when I can explain why I like it, not just how many times an algorithm has shown it to me.
“Do I just like it because I’ve been shown it 100 times – or do I genuinely love it?”
- A preference becomes truly ‘mine’ when I keep choosing it with commitment and consistency even after the microtrend has passed.
“It involves commitment and consistency (the opposite to all those microtrends)”
- A preference becomes truly ‘mine’ when I am willing to keep it even though I know it won’t appeal to everyone.
“knowing the things you like won’t appeal to everyone”
✍️ Complete: “One thing AI can imitate but may not genuinely possess is _____ because ____.”
- One thing AI can imitate but may not genuinely possess is taste because taste involves a leap of imagination and the risk of liking things that won’t appeal to everyone.
“knowing the things you like won’t appeal to everyone”
- One thing AI can imitate but may not genuinely possess is enjoyment because AI content is tailored to capture our eyeballs rather than to be truly enjoyed.
“It’s basically content that’s tailored for eyeballs, not even enjoyment.”
- One thing AI can imitate but may not genuinely possess is humanity because, as Chayka argues, tech companies use taste as a veneer of humanity to hide automation behind.
“a veneer of humanity to hide automation behind”