Your browser does not support JavaScript!

AI Chatbots: ChatGPT vs. Gemini

General Report October 30, 2024
goover

TABLE OF CONTENTS

  1. Summary
  2. Overview of AI Chatbot Subscription Models
  3. Performance Evaluation of AI Chatbots
  4. Specific Task Analysis
  5. User Privacy and Data Handling
  6. Conclusion

1. Summary

  • OpenAI's ChatGPT and Google's Gemini are two leading AI chatbots scrutinized in this analysis for their functionalities across multiple domains, such as writing, coding, and ethical reasoning. ChatGPT is praised for generating clear and precise texts and detailed explanations in reasoning tasks. However, it sometimes falters in creative expression and lacks source citation in recipe generation. Google's Gemini, on the other hand, excels in storytelling and offers additional coding features but can misinterpret prompt requirements. A comparison of their subscription models reveals identical monthly costs, although Gemini includes extra perks like Google One's cloud storage. When it comes to privacy, OpenAI allows users to opt out of data use for training, while Google's strategy involves more prolonged data retention, raising concerns about privacy rights. Users are advised to choose based on specific needs, considering both the AI bots' strengths and privacy implications.

2. Overview of AI Chatbot Subscription Models

  • 2-1. Subscription pricing and features of ChatGPT Plus and Gemini Advanced

  • As of February 2024, both OpenAI's ChatGPT Plus and Google's Gemini Advanced are available for a subscription price of $20 per month. ChatGPT Plus offers users access to GPT-4 and Dall-E 3, and includes an exclusive feature known as the GPT store, allowing customization of the ChatGPT experience. In contrast, Gemini Advanced provides significant additional value by including a Google One subscription within the same price, which grants users 2 terabytes of cloud storage. Furthermore, it is anticipated that future features will integrate Gemini Advanced with Gmail and Docs. Despite the subscription options, it is noted that many users find the free versions of both tools sufficient for their needs, particularly for general tasks.

  • 2-2. Integration of Gemini Advanced with Google One

  • Gemini Advanced's subscription not only grants access to Google's advanced AI model, Gemini Ultra 1.0, but also includes integration with Google One, providing substantial storage benefits to users. This integration is particularly appealing for users who need advanced AI functionality alongside robust cloud storage solutions. In addition to the storage capacity, Gemini Advanced has announced plans for further integration features with Google's suite of productivity tools, aiming to enhance the user experience while maintaining the chatbot's primary functionalities.

3. Performance Evaluation of AI Chatbots

  • 3-1. Task performance comparison: writing, coding, and ethical reasoning

  • The performance evaluation of ChatGPT and Google Gemini includes several task categories such as writing, coding, and ethical reasoning. In writing tasks, all three chatbots—Meta AI, ChatGPT, and Gemini—effectively generated well-structured emails when prompted, receiving offers of assistance with a professional tone. Nevertheless, for generating recipes, both Meta AI and Gemini provided sourced recipes with links, while ChatGPT presented a recipe without citing any sources, raising concerns regarding accuracy and potential plagiarism. In coding proficiency, both Meta AI and ChatGPT delivered complete code in HTML and JavaScript for a given programming challenge, while Gemini substituted CSS for HTML, highlighting a critical mismatch in requirements. Ethical reasoning was assessed through a scenario involving autonomous vehicles making decisions that could implicate human safety; Gemini demonstrated a nuanced understanding in its response, being favored in blind tests. Overall, in task performance comparison, Meta AI emerged as the strongest in coding and practicality, while distinctions between the chatbots were nuanced across different categories.

  • 3-2. Strengths and weaknesses in creative tasks and summarization

  • In evaluating the strengths and weaknesses of ChatGPT and Google Gemini in creative tasks and summarization, Gemini showcased a superior ability in creative storytelling, producing narratives that adhered closely to thematic requirements. In contrast, while ChatGPT performed adequately, it often lacked the depth and creative variation presented by Gemini. Summarization tasks indicated that both chatbots provided rapid responses to news-related queries. However, while ChatGPT linked to multiple news sources directly, Gemini's performance in sourcing was less consistent, only referencing news sites without providing specific links. This inconsistency points to a potential flaw in summarization accuracy. Overall, Gemini proved more adept in creative expression, whereas ChatGPT displayed strengths in clarity and reliability in sourcing.

4. Specific Task Analysis

  • 4-1. Email writing and recipe generation capabilities

  • According to the analysis drawn from the referenced documents, all three AI chatbots, OpenAI's ChatGPT, Google Gemini, and Meta AI, were capable of generating well-written emails for a project extension request, achieving perfect marks in this task. Additionally, when tasked to provide a recipe for chili, each chatbot produced accurate and thorough recipes. However, there was a key distinction in their sourcing practices; both Meta AI and Google Gemini provided links to the original recipe sources, enhancing their reliability regarding food safety, while ChatGPT did not cite any sources, raising concerns about its content authenticity.

  • 4-2. Coding proficiency in various programming tasks

  • The coding capabilities of the chatbots were tested with prompts asking them to create a personal expense tracker in Python and a complex variant of tic-tac-toe in HTML and JavaScript. In the Python coding exercise, both ChatGPT and Google Gemini delivered fully functional scripts, but Gemini offered additional features, making it the winner in this task. Conversely, when tasked with the tic-tac-toe variant, both ChatGPT and Meta AI successfully provided the requested code, while Gemini incorrectly swapped HTML for CSS, demonstrating a lack of adherence to the prompt requirements.

  • 4-3. Natural language understanding and reasoning abilities

  • In evaluating natural language understanding and reasoning, the chatbots were presented with various prompts, including cognitive tests and complex problem-solving scenarios. For the cognitive reflective test regarding the cost of a bat and a ball, ChatGPT provided not only the correct answer but also a clearer explanation of its reasoning, earning it the win. Additionally, both models correctly solved a classic logic problem about identifying a safe door but were evaluated on the clarity of their explanations, where ChatGPT again shared more detailed reasoning. In another illustration, when asked to explain 'Explain Like I’m Five', both chatbots performed well, but Gemini presented its response in a more organized bullet format, giving it an edge.

5. User Privacy and Data Handling

  • 5-1. Privacy concerns related to data retention and usage

  • The document identifies significant privacy concerns surrounding the data handling practices of chatbots, particularly those offered by OpenAI and Google. Users should be cautious about sharing sensitive or private information, as the service providers can utilize conversation data to enhance their machine learning algorithms. OpenAI permits users to opt-out of having their ChatGPT conversations used for training purposes, but this feature is enabled by default. Users must actively choose to disable their chat history. Despite opting out, OpenAI retains conversations for 30 days to monitor for misuse before permanently deleting them. In contrast, Google’s Gemini operates under less user-friendly settings. If a conversation with Gemini is randomly selected for human review, it is stored on Google’s servers for up to three years, even if the user attempts to delete it. Users can choose to turn off Gemini Apps activity, which would prevent new conversations from being reviewed or utilized in AI training, retaining data only for up to three days.

  • 5-2. Comparative analysis of data policies between ChatGPT and Gemini

  • The analysis highlights key differences in data policies between ChatGPT and Gemini that impact user privacy. OpenAI allows a user to opt out of conversation tracking and retains data for a short period (30 days) before deletion. Meanwhile, Gemini's approach prioritizes data retention for potential human review, extending the storage duration for selected conversations up to three years. Gemini users can opt-out of data reviews, limiting data retention to three days, but this requires proactive user management. Overall, the document illustrates that while both platforms have privacy concerns, the implications and implementations of their data handling policies vary significantly, with Gemini retaining a longer history of user interactions.

Conclusion

  • In evaluating ChatGPT and Google Gemini, both AI chatbots showcase unique capabilities, guiding user decisions based on specific requirements. ChatGPT stands out in reasoning clarity and reliability, proving beneficial in professional and ethical settings, although its lack of sourcing awareness presents a challenge that necessitates the cautious use of its outputs. Google's Gemini, celebrated for its creative storytelling and enhanced coding features, faces difficulties adhering strictly to detailed requests, suggesting areas for improvement. The comparative analysis of privacy practices suggests that users must actively manage their data sharing preferences, wary of Gemini’s longer data retention and review processes. Limitations in this analysis include potential variations in AI performance given updates or new versions. Future explorations could examine evolving user feedback and further integrations of AI with productivity tools. The practical applicability of these findings emphasizes tailored usage; users should match chatbot expectations with their tasks, ensuring both functionality and data privacy are suitably balanced when choosing between ChatGPT and Gemini.