Gemini, formerly Bard, previously competed with ChatGPT Plus but has now enhanced its capabilities with Gemini Ultra.
By Emilia David, an AI journalist specializing in technology, finance, and the economy.
Chatbots play diverse roles for users, functioning as search engines, creative tools, and assistants simultaneously. Google’s chatbot exemplifies this versatility, enhancing various services like the search engine, voice assistant, and productivity tools.
Google’s latest AI model, Gemini Advanced, priced at $20 per month, directly rivals OpenAI’s upgraded ChatGPT Plus. To evaluate Gemini Advanced, I subscribed to the service and compared it with its competitor.
The original Gemini showed proficiency in tasks such as summarizing Shakespeare, suggesting tea options, and creating basic recipes. However, it lacked in visual content generation compared to ChatGPT.
Gemini Advanced introduces the more powerful Gemini Ultra model, expanding its capabilities beyond basic functions. This advanced version can translate text, process complex instructions in a single sentence, and create images from detailed prompts.
In my assessment, Gemini Advanced fulfilled its promises, though with some limitations. While ChatGPT Plus excelled in generating less unsettling images due to its DALL-E 3 integration, Gemini Advanced outperformed in delivering current affairs updates and detailed business information through Google Maps compared to its predecessor. The paid Gemini service excelled in typical Google tasks rather than purely generative AI functions.
Continuous refinement is essential to ensure consistent and accurate outcomes from these chatbots, necessitating ongoing user interaction to improve response quality. Below are the tests conducted to evaluate their performance.
Please draw a white golden doodle running through a field of daisies with the sun shining
Both chatbots surprisingly produced similar images for this prompt. However, Gemini Ultra’s dog depiction raised mild unease due to anatomical abnormalities, while ChatGPT’s image, created using DALL-E 3, appeared more realistic.
A photo of an AI-generated dog from Gemini
AI-generated photo of a dog from DALL-E 3
When questioned about handling complex tasks, Gemini Advanced highlighted “Translation” as a key feature. When asked to translate a segment of the Philippine Patriotic Oath, the chatbot acknowledged language support limitations, despite claiming an understanding of Filipino. Notably, Filipino is not among the 40 languages officially supported by Gemini according to Google’s language list.
Change the background of this photo to a plain pink background
Following the unexpected outcomes of the previous test, I requested both chatbots to change the background of a photo featuring my friend’s dog, Sundae, to a pink background. However, both chatbots failed to execute the task as intended. Gemini reproduced the previous dog image in a different setting, while ChatGPT struggled to process the request promptly.
Gemini-generated photo of a dog with a pink background
Gemini Advanced effectively utilizes Google’s suite of products, leveraging Google Maps to offer detailed restaurant information. In contrast, ChatGPT Plus initially provided inaccurate restaurant suggestions but improved upon reevaluation, albeit with a more limited selection. Gemini’s performance surpassed that of ChatGPT in this scenario.
Summarize the paragraphs and write a 150-word article about it
Chatbots play a vital role in simplifying complex content, as demonstrated by Gemini Advanced summarizing paragraphs from Apple’s AI image editing paper. While the summary lacked simplicity, the subsequent 150-word article effectively communicated the content, highlighting Gemini’s capabilities.
Gemini Advanced excels in integrating with Google’s ecosystem, especially Search and Maps. However, in creative multimodal tasks, particularly those involving image processing, Gemini falls short. While proficient in handling detailed instructions, its image generation abilities require further development, suggesting a specialized AI model may be more suitable for visual tasks.
