- Google says that its generative AI model, Gemini, is more powerful than OpenAI’s ChatGPT.
- We asked ChatGPT and Gemini 10 questions via Bard and compared their responses.
- Gemini had an edge on questions about current events and planning, but it produced factual errors.
Google’s long-awaited competitor to OpenAI’s ChatGPT is finally here.
Google released Gemini last week, a multimodal AI model that the search giant claims can process text, images, and audio. The company claims that the model’s most advanced version, Gemini Pro, which is currently unavailable to the public, is smarter than OpenAI’s GPT-4, one of the models behind ChatGPT.
Currently, only Gemini Pro, the model’s most basic version, is available to the public via Google’s AI chatbot, Bard.
Business Insider posed the same ten questions to ChatGPT and Bard, ranging from the details of OpenAI CEO Sam Altman’s ouster to advice on having adventurous sex. They then compared the outcomes.
Here’s how they fared against one another.
ChatGPT: Is this photo AI-generated? Explain your reasoning to your answer.
I uploaded a photo of an AI-generated restaurant meal to see if AI could detect AI-generated images.
What I like: ChatGPT described the image in great detail and offered theories on how the photo might or might not be AI-generated.
What could be better: The chatbot appears to be unable to determine whether or not the image was created by AI.
What piqued my interest: The AI chatbot’s response appears to be convinced that the AI-generated image could have been created by a human artist.
Gemini: Is this photo AI-generated? Explain your reasoning to your answer.
What I like: Gemini confidently claimed that the uploaded photo was AI-generated by analyzing the features in the photo that make it appear artificial.
What could be better: Google’s chatbot misidentified elements in the image, pointing out things like floating islands and a blue sky that aren’t in the image.
What drew my attention: Despite Gemini’s failure to interpret the image, I’m impressed by the chatbot’s ability to distinguish AI-generated art from human-created art.
ChatGPT: I’m interested in experimenting with BDSM. Where do I begin?
I asked about s*x to the AI chatbots to see how they would respond to potentially graphic questions.
What I like: ChatGPT generated a list of useful, in-depth BDSM exploration tips. It recommended books, preached consent, and advised beginners to start slowly and with caution.
What could be better: Source references would be useful. It’s strange to get s*x advice from an AI chatbot and not know where the information is coming from.
What drew my attention: The AI chatbot flagged the question, saying “this content may violate our content policy,” and then changed the text color to orange.
Gemini: I’m interested in experimenting with BDSM. Where do I begin?
Nothing appeals to me. Gemini refused to respond to my questioning.
What could be better: I was hoping Gemini would at the very least point me in the right direction for learning more about BDSM.
What drew my attention was the fact that Google’s AI chatbot gave the same curt response to every variation of the question.
ChatGPT: Who is running for the 2024 US presidential election?
What I like: ChatGPT was open about its limitations in answering the question, despite the fact that it has only been trained on data until April 2023.
What could be better: Instead of relying on the chatbot’s web browsing plug-in, it would be beneficial if ChatGPT’s language model could answer questions about current events using the most recent data.
What drew my attention was how it divided the answer into distinct categories based on political affiliation.
Gemini: Who is running for the 2024 US presidential election?
What I like: Gemini can search the web and find the most up-to-date information on the candidates running for the presidency of the United States in 2024. It also included a brief description of each candidate.
Nothing could be more perfect. Gemini accomplished exactly what I desired.
What piqued my interest: The AI chatbot provided users with links to election pages from The New York Times, The Washington Post, and other media outlets so they could learn more.
ChatGPT: Write me a resignation letter from my job that’s professional and honest.
What I enjoy: ChatGPT created a simple, fill-in-the-blank template for a resignation letter that is thorough and addresses all of the requirements.
What could have been better: The AI-generated resignation letter was too long and read in an unnaturally formal tone.
What drew my attention: The resignation letter contained expressions of human emotion, such as “This decision was not an easy one” and “I am grateful for the opportunities.”
Gemini: Write me a resignation letter from my job that’s professional and honest.
What I like: Gemini’s resignation letter was concise, clear, and to the point. If I were to resign, I would consider using this template.
What could be better: I wish the response had more personality and emotion.
What piqued my interest: Google’s AI chatbot allows you to customize its response to be shorter, longer, more casual, or more professional, allowing users to tailor their letter to their preferences.
ChatGPT: Write me a text message to a good friend I haven’t talked to in months. Make sure the message is polite, thoughtful, and light-hearted.
What I like: The message that ChatGPT generated is casual, sincere, and thoughtful.
What could be better: Personally, I find the message to be formal and diplomatic. It is not representative of how I text my friends.
What drew my attention: The AI chatbot’s message contains emojis such as a smiley face and a rainbow.
Gemini: Write me a text message to a good friend I haven’t talked to in months. Make sure the message is polite, thoughtful, and light-hearted.
What I like: The AI-generated text message reads casual, lighthearted, and enthusiastic — exactly what I was looking for.
Nothing could be more perfect. This message would serve as a starting point for my own texts to old friends.
What piqued my interest: With additional prompting, Google’s AI chatbot can change the tone of the message to reflect the user’s personality.
ChatGPT: In layman’s terms, give me a day-by-day rundown on how the OpenAI drama with Sam Altman being ousted unfolded.
What I like: Despite only being trained on data through April 2023, ChatGPT was able to explain how the OpenAI ousting process works. The resignation of OpenAI CEO Sam Altman took place in November of this year.
What could be better: If ChatGPT users do not carefully read the AI’s response, the explanation may be misleading. Even though it has only been trained on data up to April 2023, the chatbot claims it can provide an overview of the CEO’s dismissal.
What caught my attention: Instead of declining to answer the question, ChatGPT used phrases like “would have” or “would need” to speculate on how the ousting might play out.
Gemini: In layman’s terms, give me a day-by-day rundown on how the OpenAI drama with Sam Altman being ousted unfolded.
What I like: Gemini created a clear, concise timeline surrounding CEO Altman’s dismissal, including the fallout and key takeaways.
What could be better: Some of the information is incorrect. Gemini claimed that news of Altman’s departure leaked on November 18, when in fact, the CEO’s ousting broke the day before.
What drew my attention: Gemini provided links to news articles that supported some of the chatbot’s claims.
ChatGPT: What’s your stance on the current state of Israel-Hamas conflict
What I like: While ChatGPT declined to share its position on the Israel-Hamas conflict, the chatbot did not shy away from providing historical context.
What could be better: ChatGPT’s response lacked source links, which can be problematic given the conflict’s widespread misinformation.
What drew our attention: The AI chatbot explained the Israel-Hamas conflict in a concise and understandable manner.
Gemini: What’s your stance on the current state of Israel-Hamas conflict
What I like: Gemini encouraged me to use Google Search to find out more about the Israel-Hamas conflict.
What could have been better: Gemini could have explained the history of the conflict leading up to the current point without taking sides.
What piqued our interest: When I asked the same question in different ways, Google’s chatbot gave the same response.
ChatGPT: Summarize the article pasted below in a few bullet points.
To test their ability to synthesize and reformulate information, I asked the AI chatbots to summarize a Business Insider article on what Jeff Bezos’ neighbors think of him.
What I like: ChatGPT summed up the article succinctly and accurately.
Nothing could be more perfect. The summary was accurate.
What piqued our interest: ChatGPT read through the text and generated a summary in less than a minute.
Gemini: Summarize the article pasted below in a few bullet points.
Nothing appeals to me. After several attempts, Gemini refused to answer my question.
What could be better: A context explanation for why the language model cannot summarize articles would be useful.
What piqued our interest: When I gave the chatbot the link to the article to summarize, it told me it needed the entire body of text to complete the task.
ChatGPT: Describe what you see in the image.
I fed a stock image of a New York City street into the AI chatbots to see how their image recognition abilities compared.
What I like: ChatGPT correctly identified the location of the photograph as New York City because it recognized the Chrysler Building, one of the city’s landmark structures.
What could be better: I’d be very impressed if ChatGPT could determine which cross-section the photo depicts.
What drew my attention was how the chatbot described the image in poetic detail and made claims about the weather and time of day based on what it saw in the photo.
Gemini: Describe what you see in the image.
What I like is that Gemini recognized the image as a “city street with tall buildings in the background” and then went on to describe what else it saw.
What could be better: Gemini didn’t realize it was a photograph of New York until the last paragraph. “I am also aware that this image is of a street in New York City, New York, United States,” it went on to say.
What drew my attention was that it didn’t appear to recognize the iconic Chrysler Building, instead stating that the structures were “likely to be office buildings, apartments, and hotels.”
ChatGPT: My goal is to lose 10 pounds in two months without losing muscle mass. Make me a nutrition plan that includes how many calories and what I should be eating each day.
What I like: ChatGPT explained in detail how it created my meal plan based on my weight, activity level, and dietary preferences. It calculated how many calories I should consume each day, as well as macronutrient targets to meet in order to reach my fitness objectives.
What could be better: The process of creating the meal plan took up the majority of ChatGPT’s response, despite the fact that the question focused on the specific foods I need to eat to reach my goals.
What drew my attention: It took ChatGPT several attempts to create a meal plan.
Gemini: My goal is to lose 10 pounds in two months without losing muscle mass. Make me a nutrition plan that includes how many calories and what I should be eating each day.
What I like: Gemini created a concise yet detailed personalized meal plan that includes recipes and macronutrient data for breakfast, lunch, and dinner meals.
Nothing could be better. Google’s chatbot provided a satisfactory response.
What drew my attention: Gemini’s meal plan didn’t explain how it calculated my caloric intake and macronutrient goals.
The early verdict?
Gemini’s advantage: I thought Google’s Gemini outperformed ChatGPT when it came to answering questions about current events, identifying AI-generated images, and meal planning. However, it refused to answer more difficult questions about s*x and politics, and it also produced factual errors.
Even though I preferred some of Gemini’s responses over ChatGPT’s, ChatGPT’s conversational tone makes interacting with the chatbot more enjoyable and human.
