Google Gemini – Alphabet Unveils Most Advanced AI Model Yet Set to Rival GPT-4
Alphabet CEO Sundar Pichai has presented Google's latest AI product Gemini. Photo by Justin Sullivan/Getty Images
Key Takeaways
- Google has launched Gemini, a new “multimodal” AI model that can handle text, image, and audio inputs.
- The Big Tech firm has positioned the new technology as a rival to OpenAI’s GPT-4.
- Gemini outperformed GPT-4 in several key tests.
When OpenAI launched ChatGPT in November 2022, it fired the starting gun on a generative Artificial Intelligence (AI) race that left rivals with little option but to respond. For Google, that meant accelerating the release of Bard. But when the chatbot fumbled a question during its public debut in February, many questioned whether the rush was worth it.
On December 6, however, Google unveiled Gemini, its most advanced AI model yet. With a full public launch and integration with Bard scheduled for the coming weeks, promotional videos showcase Gemini’s “multimodal” capabilities, offering a glimpse into the next stage of human-AI interactions.
Multimodel Approach Combines AI Applications
As the technology has unfolded so far, consumer applications of generative AI have generally fallen into distinct categories, stemming from different traditions in research and development.
On the one hand, advances in natural language programming created the conditions for today’s chatbots and conversational AI models. On the other, contemporary AI derives its ability to recognize and generate images from the field of computer vision.
Combining both linguistic and visual capabilities, Gemini represents the growing interpellation of previously separate AI functions. Coming just weeks after OpenAI launched GPT-4 with vision (GPT-4V), Google’s recent product launch establishes multimodal AI as the latest frontier in an ongoing technology race.
Google Touts Gemini’s Superior Test Performance
In presenting Gemini, Google DeepMind has boasted the new model’s strong performance in various tests, benchmarking it against the competition. For example, the firm has reported that Gemini outperformed OpenAI’s GPT-4 in 8 out of 9 text-based tasks.
When it comes to handling multimedia inputs, the new model fared even better.
Compared to GPT-4V, Gemini demonstrated an improved ability to identify and understand images. Meanwhile, its capacity to recognize speech outclassed OpenAI’s latest model, Whisper v3. When confronted with the same 16-language database, Gemini made around 60% fewer mistakes than the rival speech recognition algorithm.
