Google introduces Gemini Pro through Bard upgrade

Lungile Msomi
By Lungile Msomi
Johannesburg, 02 Feb 2024
ITWeb tasked Bard with Gemini Pro to generate an image of Bafana Bafana lifting the Africa Cup of Nations trophy. This is the image it generated.
ITWeb tasked Bard with Gemini Pro to generate an image of Bafana Bafana lifting the Africa Cup of Nations trophy. This is the image it generated.

Search engine giant Google has announced Gemini Pro, its multimodal artificial intelligence (AI) model, marking an upgrade to its Bard AI chatbot.

This comes in the same week Google said its Johannesburg cloud region in South Africa is now ready for use.

In a statement, Google says the upgrade to Bard has new features, such as more advanced reasoning, enhanced code generation, more coherent and fluent answers, and support for 40 languages in over 230 countries around the world.

Bard with Gemini Pro also introduces a ‘double-check’ feature that leverages Google Search to verify the information it provides. Activating this feature prompts Bard to evaluate web content, offering highlighted phrases that link to relevant supporting or contradicting information.

Users can also generate images in Bard, supported by Google’s updated Imagen 2 model, by typing a description, and Bard will generate custom visuals at no cost.

The company assures users that image generation aligns with its AI principles, incorporating digitally-identifiable watermarks and filters to avoid creating images of named individuals.

“To ensure there’s a clear distinction between visuals created with Bard and original human artwork, Bard uses SynthID to embed digitally-identifiable watermarks into the pixels of generated images,” says Google.

Arthur Goldstuck, CEO of World Wide Worx, views the addition of Gemini Pro to Bard as a pivotal moment in generative AI’s evolution.

“Multimodal AI is the next phase of the generative AI revolution that has swept the world over the past year. It uses a huge amount of computing power, and having access to it via the cloud will be of enormous benefit to South Africans,” he comments.

Gemini Pro enters the competitive field of OpenAI’s GPT-4 model. While both are multimodal, GPT-4's training spans text, code, images and video, whereas Gemini Pro primarily focuses on text data with some code understanding.

Goldstuck anticipates Gemini Pro to initially outperform GPT-4, but expects competition to intensify once Microsoft Copilot integrates GPT-4 into Microsoft services.

He emphasises that specific needs, pricing models and individual preferences will come into play when selecting between these models.

However, with excitement likely to be tempered by cautious optimism and concerns, the reception of Bard with Gemini Pro in SA will prompt discussions around data privacy, potential biases and the ethical implications of AI, he adds.

Goldstuck underscores the platform's reliability, stating: “Having a generative AI platform that double-checks facts it generates will increase confidence in using AI, and will accelerate the adoption of such tools in South Africa.”

Google has addressed privacy concerns, highlighting its investments in the safety of training data to limit explicit content.

“Additionally, we apply filters designed to avoid the generation of images of named people. We’ll continue investing in new techniques to improve the safety and privacy protections of our models,” it states.

To test its new capabilities, ITWeb tasked Bard with Gemini Pro to generate an image of SA's national soccer team lifting the Africa Cup of Nations trophy – a request it semi-successfully completed.

Bafana Bafana this week beat Morocco 2-0 to advance to the quarter-finals of the Africa Cup of Nations, where the team will face Cape Verde on Saturday. Winning tomorrow’s match means Bafana Bafana will qualify for the semi-finals before the final set for Sunday, 11 February, in Côte d’Ivoire.