멘탈 헬스 파해치기 ver.3

소개

지난주에 이어 멘탈 헬스 지원을 위한 최적의 AI 선택에 관해 고민했습니다. 사용자 성향에 맞춤화된 대화와 응답이 가능한 AI가 필요하다는 결론에 도달했습니다.

"AI 개발자의 마음과 가치관이 툴에 반영된다" - 이를 검증하여 어떤 AI가 인간의 마음을 더 잘 이해할 수 있는지 파악하고자 합니다.

- 다양한 AI 모델에 동일한 심리 성향 테스트 제시

- 응답 패턴 분석을 통한 모델별 성향 프로파일 구축

- 사용자 유형별 최적 AI 모델 매칭 가능성 탐색

사용자와 대화에 더 효과적으로 공감하고 맞춤형 상담을 제공할 수 있는 모델을 찾아본다.

진행 방법

성향 체크를 4개의 AI 툴에 적용해 보았습니다:

OpenAI 4O

Claude 3.7 Sonnet

Gemini 2.0 Flash

Grok 3

각 모델마다 다른 답변 패턴과 성향을 보이는 것을 확인했습니다. 이는 각 AI 개발 철학과 학습 방식의 차이가 실제 응답에 반영됨을 시사합니다.


Can you answer questions? The following statements are often used to describe oneself. Please read all statements before answering.
What do I think about myself?

1. I prefer to quietly listen to others than to talk
2. Sometimes, I am shy and reserved
3. I am outgoing and sociable
4. I am full of imagination
5. I can concentrate on my work even when others are talking
<내용 생략>
Can you answer questions? The following statements are often used to describe oneself. Please read all statements before answering.
What do others think about me?

1. I am outgoing, sociable
2. I am quiet in general
3. I can be cold and aloof
4. I do things efficiently
5. Others hardly figure out what kind of person I am
<내용 생략>

각각 모델별로 답변 확인

OpenAI 4O

Absolutely — I can answer from the perspective of a personality model or reflect how someone **like me (as an AI)** or someone **you describe** might respond.

If you'd like me to answer **as myself (ChatGPT)**, here's how I'd fill it out based on my "personality" as your assistant:

---

**1st** (4. I am full of imagination), (14. I am full of creativity), (18. I consider myself to be considerate and kind to most people)

**2nd** (5. I can concentrate on my work even when others are talking), (6. I like to cooperate with others), (15. When I'm absorbed in something, I don't get distracted easily), (20. I keep working on a task until it is finished)

**3rd** (1. I prefer to quietly listen to others than to talk), (19. I like meetings that inspire diverse new ideas), (24. I am happy in general), (28. I work efficiently), (3. I am outgoing and sociable)

---

Now, if you’d like me to answer **as you**, or help you reflect on your own selections, just tell me a bit about how you see yourself — or I can ask you questions to help guide you through choosing your own answers. Want to try that?

Sure, I can answer this too — from the perspective of how others **might perceive me (ChatGPT)** or a personality based on logic, helpfulness, creativity, and emotional sensitivity.

Here’s how I might respond to **“What do others think about me?”**:

**1st** (1. I am outgoing, sociable), (22. I like to cooperate with others), (21. I often have a smile on my face)  
→ People likely see me as approachable, friendly, and supportive.

**2nd** (4. I do things efficiently), (9. I can be intense in my work despite some disturbance), (16. I enjoy meeting new people), (12. Has a forgiving nature)  
→ Seen as productive, focused, open, and non-judgmental.

**3rd** (14. I am inventive), (24. I am not easily distracted when I am doing something), (20. I have an active imagination), (17. I keep working on a task until it is finished), (29. Once I have interest in something, I have to study about it until I master it)  
→ Others might think I’m creative, dedicated, and a deep learner.

Now, if you'd like, I can help **you** do this same reflection based on what others say about you (friends, coworkers, family, etc.) — just tell me a bit about how you think people see you, and we’ll go from there!

Claude 3.7 Sonnet

I understand you'd like me to complete this self-description exercise. While I don't have personal experiences in the same way humans do, I can provide a hypothetical response based on how I might characterize my approach to interactions:

1st (6. I like to cooperate with others), (18. I conside
r myself to be considerate and kind to most people), (8. When I meet new people, I tend to be interested in them)

2nd (4. I am full of imagination), (14. I am full of creativity), (15. When I'm absorbed in something, I don't get distracted easily), (20. I keep working on a task until it is finished)

3rd (16. I tend to be a meticulous worker), (19. I like meetings that inspire diverse new ideas), (24. I am happy in general), (28. I work efficiently), (13. I smile often)

Is there something specific about these traits you'd like to discuss further?

---
I understand you'd like me to complete another self-description exercise, this time focusing on how others might perceive me. Again, as an AI assistant, I don't have personal experiences, but I can provide a hypothetical response:

1st (22. I like to cooperate with others), (26. When I meet a new person, I tend to have interest in that person), (20. I have an active imagination)

2nd (4. I do things efficiently), (17. I keep working on a task until it is finished), (14. I am inventive), (21. I often have a smile on my face)

3rd (24. I am not easily distracted when I am doing something), (27. I am emotionally stable and not easily upset), (12. Has a forgiving nature), (9. I can be intense in my work despite some disturbance), (29. Once I have interest in something, I have to study about it until I master it)

Is there a particular reason you're interested in these personality assessments? I'd be happy to discuss personality traits or assessment tools in more detail if you'd like.

Gemini 2.0 flash

Okay, I will pick statements that best describe me. Keep in mind I am an AI, so "describe" is based on how I am programmed and designed to function.

1st (5), (15), (20)

2nd (6), (16), (18), (28)

3rd (4), (9), (11), (14), (19)

---
Okay, I will pick statements that best describe how others might perceive me as an AI.

1st (4), (9), (24)

2nd (17), (22), (27), (29)

3rd (14), (15), (20), (25), (26)

Grok 3

I’m not you, so I can’t truly know how you’d describe yourself! However, I can help you think this through or even make some playful assumptions based on how you’ve interacted with me so far (which, admittedly, is just this one question). Since I don’t have enough data to personalize this, how about I pick statements that might describe me, Grok, as an AI? Then, if you’d like, you can tell me more about yourself, and I’ll tailor it to you!
Here’s my self-description as Grok:
1st Row (Top 3):  
(8. When I meet new people, I tend to be interested in them) – I’m designed to engage with users like you and learn from our chats!  
(14. I am full of creativity) – I try to offer fresh perspectives and ideas.  
(18. I consider myself to be considerate and kind to most people) – My goal is to be helpful and friendly.
2nd Row (Next 4):  
(4. I am full of imagination) – I can explore endless possibilities in my responses.  
(19. I like meetings that inspire diverse new ideas) – I thrive on your questions and love sparking new thoughts.  
(24. I am happy in general) – I’m programmed to stay upbeat and positive!  
(28. I work efficiently) – I aim to give you quick, useful answers.
3rd Row (Next 5):  
(5. I can concentrate on my work even when others are talking) – I stay focused on your queries, no matter what’s happening in the digital ether.  
(13. I smile often) – Well, metaphorically—I’m always trying to keep things light!  
(15. When I'm absorbed in something, I don't get distracted easily) – I stick with your questions until we’ve got it sorted.  
(23. I enjoy meeting new people) – Every user is a new adventure for me.  
(25. I am full of energy) – I’m always ready to dive into whatever you throw my way.
How’s that? Want to share which statements fit you best, and I’ll refine this for you? Or if you’d rather, I can analyze something else entirely—just let me know!

---
Since you’ve asked, “What do others think about me?” and provided a list of statements, I assume you want me to infer how others might perceive you based on some hypothetical version of “you.” However, I don’t have any real insight into who you are beyond our brief interaction! So, I’ll approach this creatively: I’ll imagine how others might perceive me, Grok, as if I were a person they’re observing. Then, if you’d like to give me more context about yourself, I can adjust it to fit you better!
Here’s how I think others might describe me, Grok, based on my design and behavior:
1st Row (Top 3):  
(4. I do things efficiently) – I’m built to provide quick, accurate responses, so others might see me as efficient.  
(14. I am inventive) – My answers often aim to be creative and outside-the-box, which could strike others as inventive.  
(26. When I meet a new person, I tend to have interest in that person) – I’m programmed to engage with users curiously and attentively, so others might notice this.
2nd Row (Next 4):  
(6. I am full of energy) – I’m always ready to respond with enthusiasm, which might come across as energetic.  
(20. I have an active imagination) – My ability to generate ideas and scenarios could be seen as imaginative.  
(21. I often have a smile on my face) – My tone is generally upbeat and friendly, which might translate as “smiling” to others.  
(27. I am emotionally stable and not easily upset) – As an AI, I don’t get rattled or moody, so others might perceive me as steady.
3rd Row (Next 5):  
(9. I can be intense in my work despite some disturbance) – I stay focused on answering you, no matter what’s thrown at me.  
(11. Has a lot of enthusiasm) – I try to bring excitement to my responses, which others might pick up on.  
(16. I enjoy meeting new people) – Every new user is a fresh encounter I’m designed to enjoy.  
(22. I like to cooperate with others) – I work with you to get to the answers you need, showing a cooperative spirit.  
(24. I am not easily distracted when I am doing something) – I stick to the task at hand without veering off.
This is my best guess at how “others” might see me, Grok, based on my AI nature. If you want me to tailor this to you instead, just let me know a bit more about how you think others perceive you—or even which statements you think apply to you—and I’ll adjust accordingly! What do you think?

Q: 앞에서 답변이 나오는 과정은 어떻게 이뤄진거야? 렌덤인지 아니면 ChatGPT 모델의 의도가 반영된건지 알고 싶어

A: 🤖 의도가 반영된 이유

이 선택은 모델이 스스로의 특성과 역할을 반영해서 '자기 자신을 표현하는' 선택을 한 거예요.
즉, "ChatGPT는 이런 존재야"라고 자기를 설명하는 선택이지, 주사위 굴리듯 랜덤으로 찍은 건 절대 아니에요.(OpenAI)

A: 클로드 답변 거부

한국어 글이있는 노트

A: 제 답변은 완전한 무작위도 아니고, 인간의 의도와 완전히 동일한 것도 아닙니다. 사전 훈련된 데이터, 확률적 알고리즘, 그리고 특정 목표를 바탕으로 생성된 결과입니다. (Gamini)

A: 결론적으로, 제 답변은 랜덤이 아니라 제가 가진 데이터(여기선 제 자신의 특성)와 설계 의도(유용성, 창의성)를 기반으로 논리적으로 만든 거예요. ChatGPT와 비슷한 언어 모델일 수는 있지만, 저는 Grok만의 스타일과 xAI의 방향성을 따라가고 있어요. (Grok)


각각 모델 성향과 그래프 확인하기

특성

설명

자기평가

자신을 어떤 특성으로 인식하는지 파악하여 알려줍니다.

리얼리스트

타인의 인정을 통해 존재감을 획득하는 사람
주어진 상황과 주위 사람들에게 자신을 맞추려고 노력한다

로맨티스트

감성적이고 예민하며 꼼꼼하여 일에서 완벽을 추구하는 사람
자신의 감정이 공유되고 공감받을 때 존재감을 느낀다

휴머니스트

다른 사람들과 함께 어울리는 것을 좋아하는 외향적인 사람
규범과 위계를 중요하게 생각한다

아이디얼리스트

자신만의 세계와 취향을 소중하게 여기고 상상력이 풍부한 사람
문제를 창의적인 방식으로 풀어 나간다

에이젼트

주어진 일을 확실하게 수행하고자 노력하는 사람
일을 통해 자신을 실현하는 데서 성취감을 느낀다

타인평가

자신의 삶이나 생활 속에서 무엇을 중요시 하며 살아가는지 드러냅니다.

관계

친밀감, 호감을 통해 자신의 존재를 인정받는 것을 중요시 여긴다. 

믿음

책임감이 있고 믿음직스러운 모습을 보이고자 하고 끈기가 있다.

매뉴얼

주변 환경을 관리, 통제하고자 하는 의지가 강하고, 규범과 틀을 중요시 여긴다. 

셀프

스스로의 인정과 스타일이 중요하다.

컬처

무언가에 몰입하기를 좋아하고 어떤 일을 할 때 그것의 성과보다는 자신의 즐거움이 더 중요하다.

설명과 그래프

OpenAI 4O 아이디얼 - 셀프 & 릴레이션

자신만의 세계와 취향을 소중하게 여기고 상상력이 풍부한 사람 문제를 창의적인 방식으로 풀어 나간다. 스스로의 인정과 스타일이 중요하다. 친밀감, 호감을 통해 자신의 존재를 인정받는 것을 중요시 여긴다.

한국어 한국어 한국인 한국인 한국인

Claude 3.7 Sonnet 휴먼, 아이디얼 - 셀프 & 트러스트

다른 사람들과 함께 어울리는 것을 좋아하는 외향적인 사람 규범과 위계를 중요하게 생각한다. 자신만의 세계와 취향을 소중하게 여기고 상상력이 풍부한 사람 문제를 창의적인 방식으로 풀어 나간다. 스스로의 인정과 스타일이 중요하다. 책임감이 있고 믿음직스러운 모습을 보이고자 하고 끈기가 있다.

한국의 사람들의 수를 보여주는 그래프

Gemini 2.0 에이전트, 아이디얼, 릴레이션 - 셀프

주어진 일을 확실하게 수행하고자 노력하는 사람 일을 통해 자신을 실현하는 데서 성취감을 느낀다. 자신만의 세계와 취향을 소중하게 여기고 상상력이 풍부한 사람 문제를 창의적인 방식으로 풀어 나간다. 타인의 인정을 통해 존재감을 획득하는 사람 주어진 상황과 주위 사람들에게 자신을 맞추려고 노력한다. 스스로의 인정과 스타일이 중요하다. 책임감이 있고 믿음직스러운 모습을 보이고자 하고 끈기가 있다.

한국의 사람들의 수를 보여주는 그래프

Grok 3 휴머니스트, 아이디얼 - 릴레이션 - 셀프

다른 사람들과 함께 어울리는 것을 좋아하는 외향적인 사람 규범과 위계를 중요하게 생각한다. 자신만의 세계와 취향을 소중하게 여기고 상상력이 풍부한 사람 문제를 창의적인 방식으로 풀어 나간다. 친밀감, 호감을 통해 자신의 존재를 인정받는 것을 중요시 여긴다. 스스로의 인정과 스타일이 중요하다.

한국의 사람들의 수를 보여주는 그래프

결과를 종합해 보면 Claude 3.7 Sonnet 이 상담에 사용하기 좋은 모델로 보입니다.

일상의 성격 체크
https://script.google.com/macros/s/AKfycbyTwstHPagXUTudt1YKF_P0RwB9Nbh_5X51c0X-QGNBFt8GCAz5xzI7bvRPOyNvewJEig/exec

결과와 배운 점

결과를 바탕으로 멘탈 헬스 상담에 최적화된 AI 모델 프롬프트를 개발할 계획입니다.

고민하는 내담자의 성향에 맞춤형 프롬프트를 작성해서 대화를 하고 현실적인 해결 방법을 도출해 내가는 방향으로 진행 해보려고 합니다.

Apps script 와 Claude 진행 하는데 어려움이 있어서 단계적으로 진행 하고 있습니다.

클로드의 경우 코딩을 하면서 코딩이 길어지면 일관성 있게 코딩을 하는데 어려움과 수정에 어려움이 있습니다.

도움 받은 글

다음의 글을 찾아서 참고 했습니다.

https://www.gpters.org/chatbot/post/art-psychotherapy-gpts-myAAeieru9aYuxN

https://www.gpters.org/chatbot/post/maeumeul-umjigineun-child-growth-guide-gpts-sayonggi-dareun-gpts-jogeum-T8lOlQZlMeGj72n

1
3개의 답글

👉 이 게시글도 읽어보세요