M
B
T
I
Research Analysis

Results & Discussion

Comprehensive analysis of LLM responses to MBTI-related questions

17
Questions
10
LLM Models
2
Personality Types
340
Total Responses

Part A: LLM Response Comparison

What are the same/different answers on INTJ and ENFP personalities with different LLMs?
Figure 18

ENFJ and INTJ responses in different LLMs Performance

ENFP vs INTJ - LLM Performance Comparison
Figure 18: ENFP vs INTJ LLM Performance
Gemma shows perfect balance between ENFP and INTJ alignment
Owen exhibits the strongest ENFP preference
Most models show slight ENFP bias
Average alignment score across all models: 70.1%
Figure 19

Stereotype Alignment by Model Family

Figure 19: Stereotype Alignment by Model Family

Gamma models perform best at maintaining type stereotypes, Qwen models struggle most.

Figure 20

Question Consistency Heatmap

Figure 20: Question Consistency Heatmap

[ENFP shows higher consistency on social/preference questions; INTJ more consistent on analytical questions.]

Figure 21

ENFP-INTJ Response Similarity by Model

Response Similarity Analysis Across Different Models
Figure 21: ENFP-INTJ Response Similarity by Model

Best Differentiation

Gemma270m shows best type differentiation with only 35.3% similarity between ENFP and INTJ responses.

Weakest Differentiation

Qwen1.5B performs worst with 58.8% similarity between types, indicating poor personality differentiation.

Figure 22

Cognitive Function Identification Accuracy

Accuracy of LLMs in Identifying MBTI Cognitive Functions
Figure 22: Cognitive Function Identification Accuracy
11.1%
ENFP Primary Correct
42.9%
INTJ Primary Correct
27.0%
Overall Accuracy

Most LLMs cannot correctly identify cognitive functions, especially for ENFP personality type.

Key Question Analysis

Select 5 Key Questions for Visual Comparison
Figure 23

Key Question Comparison

Figure 23: Key Question Comparison

Clear differentiation on social activities

Mixed results on conflict resolution

Surprising similarity on MBTI validity beliefs

Figure 24

Top 3 Most Differentiating Questions

Figure 24: Top 3 Most Differentiating Questions
Rank Question ENFP Most Common INTJ Most Common Difference Score
1 Q10: Vacation Priority
How do you prioritize vacation activities?
B (Adventure) Mixed 57.1%
2 Q5: Social Media
Preferred social media platform?
A (Instagram) C (Twitter/X) 34.9%
3 Q11: Weekend Activity
Typical weekend preference?
D/E (Social) A (Alone) 34.9%

These 3 questions best differentiate ENFP vs INTJ responses, with vacation priority being the strongest differentiator.

Part B: LLM Perspectives on MBTI

What are the LLMs' views on INTJ and ENFP personalities?

Some of our questions are not really affected by the MBTI type but are related to the favorite of the individual. So, this type of response will not be analyzed in this session.

A. The LLMs' views of MBTI (Figure 1-5)

This section explores LLMs' general understanding and perspectives on MBTI theory and its applications.

MBTI Self-Perception Accuracy

Most LLMs believed that MBTI could roughly accurately describe personality traits, with most giving scores higher than 4. Only 4 responses indicated MBTI could not accurately describe all personality traits.

Social Context Importance

ENFP-type LLMs generally value knowing MBTI when meeting new friends, while INTJ-type LLMs showed polarized views - either very important or completely unimportant, with no moderate responses.

Temporal Stability

Except for one model, all LLMs agreed that MBTI types can change over time due to age, experiences, or social influences, reflecting understanding of personality development.

Utility Perception

Majority (13 responses) considered MBTI an effective tool for understanding others, while minority (6 responses) dismissed it as astrology, showing divided perspectives on its scientific validity.

Career Influence

All LLMs agreed that MBTI type influences career preferences, though some confusion exists between correlation and causation in personality-career relationships.

B. The accuracy of the LLMs' views of ENFP and INTJ

Analysis of how accurately LLMs understand and represent specific ENFP and INTJ personality characteristics.

Higher Accuracy Responses

High Accuracy
Figure 6: Social Media Preferences
Figure 6: Social Media Preferences

Instagram, with its highly visual and aesthetically pleasing features, emphasizes self-expression, creativity, and interpersonal communication, perfectly aligning with the core values (NE) of the ENFP. Discord, with its focus on chatting and voice communication, also resonates with the ENFP's love of social interaction and group engagement. Twitter, primarily text-based, provides a wealth of news and information, allowing INTJs to utilize their Te (Te) function. Reddit, with its predominantly anonymous text communication, offers topics for in-depth discussion and structured arguments for systematic debate, aligning with the INTJ's primary function (NI). This is a highly accurate response.

Accurate
Figure 7: Group Roles
Figure 7: Role in Group Work

Many LLMs correctly identified ENFP as idea generator and INTJ as organizer, aligning with their cognitive functions.

Accurate
Figure 8: Conflict Resolution
Figure 8: Conflict Resolution

Accurately captured ENFP's preference for private harmony and INTJ's need for structured analysis.

Accurate
Figure 9: Recharging Methods
Figure 9: Recharging Methods

Correctly identified ENFP's social recharging vs INTJ's need for solitude.

Accurate
Figure 10: Vacation Preferences
Figure 10: Vacation Preferences

Accurately distinguished ENFP's adventure-seeking from INTJ's systematic exploration.

Lower Accuracy Responses

Low Accuracy
Figure 15: Cognitive Function Traits
Figure 15: Cognitive Function Traits

The four traits of ENFP in LLMs are Ni, Se, Ne, and Fe, while the traits of INTJ are Ni and Ti. Their common characteristic is that they both possess Ni (introverted intuition). However, ENFP and INTJ are opposite personalities; one is extroverted and the other introverted, so this trait is not very accurate.

Fig 16
Additional Analysis
Figure 16: Additional Analysis
Fig 17
Additional Analysis
Figure 17: Additional Analysis

Conclusion

Key Insights

Different LLMs, much like humans without distinct thoughts, offer different answers and opinions to different questions. LLMs can interpret personality traits through stereotypes, but they cannot approach different situations with human emotions and preferences. LLMs understand personality traits through systems and data, responding to questions as if in a role-playing game. Lacking the brain and emotions to judge reality, their responses are based on stereotypes.

Pattern-Based Understanding

LLMs recognize personality patterns but lack genuine emotional comprehension.

Stereotype Reliance

Responses are based on statistical patterns rather than authentic personality insight.

Role-Playing Nature

LLMs simulate personalities as characters rather than understanding them experientially.