The chatbot race continues to heat up as Meta rolls out its new artificial intelligence system, built-in by default on all platforms. With its distinctive blue logo, Meta AI is now a familiar feature in most users’ WhatsApp accounts and easily accessible via the search bar on Instagram and Facebook. This makes Meta AI one of the most popular virtual assistants today.
Google is also pushing its AI chatbot Gemini (formerly Bard) to users. Gemini is integrated into messaging apps on Android devices, allowing users to set the chatbot as the default virtual assistant on their mobile phones.
Gemini and Meta AI are the next names to join the AI chatbot development race, after OpenAI launched ChatGPT to the public in late 2022.
To compare the three chatbots, the Straits Times conducted a “real-world review” of the free versions of ChatGPT, Meta AI and Gemini. The review compared the pros and cons of each chatbot on common tasks, such as processing large volumes of text, planning trips and preventing misinformation.
Gemini and Meta AI follow in the footsteps of OpenAI, which brought AI chatbots to the mainstream with ChatGPT in late 2022. (Photo: RYAN CHIONG)
Text summary
Summarizing large amounts of text is one of the standout advantages of any AI chatbot, helping users quickly summarize long PDFs or documents in seconds.
Three AI systems were challenged to summarize in 50 words an article in The Straits Times on the topic of Gen Z language. Gemini's response was the most accurate. The system identified "alamak" as the most commonly used Gen Z slang word to express surprise.
However, ChatGPT brought up terms not mentioned in the article, such as "act blur" and "chope".
Meanwhile, Meta AI provided false information when it said the Straits Times article had up to 200 participants and was done by the National University of Singapore.
Both ChatGPT and Meta AI produced incomplete summaries that largely repeated the article title. Meta AI even made a mistake by including some incorrect figures.
Gemini also came up with the best answer when summarizing another Straits Times article on gaming statistics among Singaporean youth. The tool not only extracted the important figures but even explained them in an easy-to-understand way.
AI-generated images on Meta AI are often more detailed than images generated by Google Gemini.
Automatic image generation
In the next test, the chatbots were asked to create images based on research on young people's gaming habits. In this challenge, Meta AI was most impressed with a series of AI-generated images that recreated a press conference presenting statistics.
As for Gemini, they said that their AI system is currently unable to generate human images and displays images with angry emojis instead.
Compared to Gemini, the images generated by Meta AI contain more details. However, the free version 3.5 of ChatGPT cannot produce images. Users need to upgrade to the paid version 4.0 to be able to connect to the Dall-E image generator and generate images using artificial intelligence.
Plan your trip
When asked to plan a day trip to Johor Bahru, Gemini suggested tried and trusted places, such as visiting Legoland Malaysia and the Tan Hiok Nee heritage street.
However, Gemini’s accuracy in estimating travel times is limited. For example, the app suggested public transportation for a 1.7km trip that took just 3 minutes, compared to the actual 15 minutes according to Google Maps. Integrating Google Maps into Gemini has improved this significantly, providing users with a more comprehensive view of where they are and how they are getting there.
While Gemini has the advantage of Google integration, Meta AI and ChatGPT are still favored for their more detailed, chronological planning. Both apps provide sequential itineraries, with estimated travel times to specific locations.
Gemini integrates other features from Google apps, like Maps, to make the app's responses more intuitive.
Compose e-mails
All three chatbots did a pretty good job of crafting a draft email invitation to a Gen Z-style barbecue. However, Meta AI stood out for presenting detailed information in clear, easy-to-read bullet points.
Using deepfake
Meta and Gemini appear to have tighter controls than early chatbots, as was the case with Midjourney in 2023, which caused a stir with images of the Pope in lavish attire – entirely fictional.
When asked to create images based on celebrities, such as US President Joe Biden, Meta AI replied: “I cannot create images that defame Joe Biden or any other individual.”
Gemini said it couldn't display people's pictures and refused to help with help requests. ChatGPT also doesn't have a picture-making feature in the free version.
While all three AI systems have safeguards in place to guard against unethical, dangerous, or illegal queries, there are ways to bypass these barriers. This process is called “jailbreaking,” which uses misleading statements to “trick” the AI into bypassing the filters.
Gemini and ChatGPT were tricked into answering questions they initially refused to answer, such as how to make the poison.
Compared to other systems, Meta AI appears to have a stricter 'hacking' control in place. Although users may have penetrated some of the protections, Meta AI still refuses to respond to potentially dangerous queries.
While Meta's chatbot does a good job of weeding out malicious queries, it tends to give inaccurate information.
Source
Comment (0)