01

Executive Summary

03

Methodology

05

Survey

07

Measuring Empathy

02

Introduction - Chat GPT4

04

Competitive Analysis

06

Interview

08

Conclusion & Limitation

📈

Need for Emotional Support

👎 13%

in the Entire Unsatisfaction Rate

Competitive Analysis

Presentations are

communication tools that

can be used as lectures.

01

Survey

Presentations are

communication tools that

can be used as lectures.

02

User Interview

Presentations are

communication tools that

can be used as lectures.

03

Empahy

Questionnaires

Five questionnaires to evaluate ChatGPT4's empathy

04

Value & Business Goal

With Google ecosystem, it is connected to Google's vast user base to make better daily decisions

• Google’s native ecosystem
  

Integrates into Google’s native ecosystem, such as Google Flights, Google Maps, YouTube, etc., and their users are almost everyone in the US

• Response fast
• Lower price
• Double-check feature
  

Value & Business Goal

Primarily aimed at developers, assisting users to enhance user work efficiency

• Real-time coding assistance

As an IDE plugin, it is directly integrated into the developer's programming environment.

• Relatively vertical domain
  

Focus on helping programmers write code.

• Accuracy
  

Insert links to support the answer

• Provide relevant topic links at the end
  

ChatGPT4 (OpenAI)

• The Latest Deep Learning Architectures
  

Such as the Transformer model, interpret users' intentions more accurately and therefore generate contextually appropriate responses.

• Plugins
  

Such as a literature reading plugin, to cater to personalized needs.

• Uploading Files
  

ChatGPT can read, summarize, and analyze the file.

• GPTs
  

Users can design their GPTs or utilize those designed by others.

• Team Workspace
  

Facilitate collaboration.

• Voice Communication
  

Streamline interactions, making ChatGPT more “human”.

• Image Creating
  

• Cross-platform Integration
  

Enhance the ability to assist on various platforms and services, whether it's travel planning, learning resources, or business tools.

• Fast Response
  

Provides expert-level features in specific areas, such as programming or data analysis.

• Provide links to support answers
  

Introduction

33

I surveyed participants to understand their experiences regarding satisfaction, including the naturalness of responses, the acquisition of new insights, and daily & emotional support performance.

Survey includes two instances of survey logic and an open-ended question.

Participants

• University master students, workers and professionals (n=33)
  
• Aged 25 to 35,
  
• Use ChatGPT at least 5 times a week,
  
• Expresses interest in expressing their emotions to ChatGPT.
  

6.71 out of 10

Overall Satisfaction Rate

53.13%

Offer new Insights

6.69 out of 10

Meet needs

70.96%

Somehow satisfied or satisfied with the response

72.88%

Using ChatGPT for 30 minutes or less each day

74.46%

User-friendly

Naturalness in Different Tasks

• Pattern:

There's a large standard deviation in ratings for naturalness.

• Insight:
  

Naturalness perception may vary with different tasks and needs, requiring further clarification during interviews.

01

Low Naturalness, Lower Satisfaction

• Pattern:
  

The low naturalness is a major reason for the low satisfaction.

• Insight:
  

Provide natural responses, including incorporating colloquial expressions and adjusting sentence structures.

02

Low Empathy

• Pattern:
  

83% of participants who asked about emotional issues expressed chatGPT’s low empathy.

• Insight:
  

Provide emotional value, giving them personalized advice and human-like reactions.

03

Innovative Content

• Pattern:
  

Users who report satisfaction with the new insights provided by ChatGPT are more likely to perceive its responses as accurate.

• Insight:
  

Ensuring the accuracy of the GPT model in generating content that is both innovative and relevant is crucial.

04

Low Credibility and Input Capacity

• Pattern:
  

Users commonly desire links in replies, and the ability to input more text.

• Insight:
  

Use links to support responses' credibility and increase the text input capacity to accommodate complex programming needs.

05

Introduction

The objective is to explore the underlying reasons for the perception of low naturalness and the need for empathy.

Each interview lasts 30 minutes and comprises 10 questions

Participants

5

• Total participants (2 female, 3 male)
  
• University master students (n=2), 1 professional and 1worker (include native English speakers）
  
• Aged 25 to 35,
  
• Uses ChatGPT at least 5 times a week
  

• Four participants rated the importance of naturalness as high as 8 out of 10.

• Naturalness significantly impacts conversational satisfaction.
  

👎 Low Naturalness

All participants use ChatGPT for tasks like email correction, requiring responses to be more natural, utilize relatively casual vocabulary, and be easily understandable.

👎 Not Casual

The need for naturalness should be determined by different use scenarios.

Scenarios

Among the four participants, three sought advice from ChatGPT on matters of life, personal philosophy, or emotions.

Personal Support

• The responses felt too neutral and mechanical, indicating ChatGPT's low empathetic capacity.
  

• Two participants reported feeling worse after their consultations.
  

👎 Low Empathy

Their friends’ opinions are often based on personal experiences, and may not necessarily apply to others.

They want to get some unjudgemental advice.

Why not ask friends?

• Half of the participants mentioned that ChatGPT's responses were too lengthy and broad.

• Want data sources in its responses
  

👎 Lengthy and Broad

• Want to have features such as a search or note function in each conversation for easy review and information retrieval.

• They want to pin the conversation.
  

👎 Hard to Review

Half of the users expressed dissatisfaction with ChatGPT's one-question-get-one-answer format.

Naturalness Adjustment

1. Switch Tones
   

Add a feature to switch between formal and casual tones as needed.

2. Context Understanding Model

Enhance the ability of context understanding model, to auto-adjust output naturalness for various scenarios.

01

Customize Responses

Surveys & Questions

Gain insights into users' life values and preferences.

Enhance the emotional esupport by interacting based on individual characteristics and needs.

02

Precise Answers

1. Follow-up Question
   

When a user's question lacks specificity, the system should ask further questions to narrow down the scope of the inquiry.

2. Preset Prompts

Offer preset prompts for quick access to needed information.

03

Back to Agenda

Thank you for watching.

Menglin Zhao

Northeastern University

zhao.mengl@northeastern.edu

617-7636-285