Google has recently unveiled Gemini 1.5, an upcoming update to its primary language model, previously known as Bard but recently rebranded as Gemini. This iteration boasts a breakthrough in comprehending lengthy contextual information across various formats. It also promises enhanced efficiency, thanks to its adoption of the “Mixture-of-Experts (MoE)” architecture, achieving performance similar to Gemini 1.0 but with fewer power-hungry GPUs.
The claim about multi-modal “long-context” comprehension might sound technical, but a demonstration showcased by a Google Deepmind co-founder on X illustrates its practical implications. Using a sizable excerpt from the public domain, specifically a 402-page transcript of the NASA moon landing mission, Gemini 1.5 demonstrates its ability to focus on relevant user needs despite extensive prompts. This elucidates the concept of “long-context” understanding.
Also Read: Google Gemini AI Login: How to Access and Use
During the demo, Gemini 1.5 successfully identifies three amusing moments from the lengthy text and matches an image of a lunar boot print to the corresponding event in the transcript — Neil Armstrong’s moonwalk. This demonstrates the “multimodal” aspect, with image recognition complementing the language model.
This update is part of Google’s ongoing efforts to remain competitive in the AI landscape, especially after OpenAI gained prominence with ChatGPT. Google has been teasing improvements to Bard and its underlying model, although it’s primarily recognized for its integration into popular Google and Android products rather than for everyday problem-solving like ChatGPT.
Google claims that Gemini 1.5 excels at processing large datasets with high accuracy and exhibits strong reasoning abilities across diverse data types, a notable weakness in many language models.
Sundar Pichai, CEO of Google, announced a limited preview of Gemini 1.5 for developers and enterprise customers, with broader availability expected later. It’s likely that Gemini 1.5 will eventually become part of Google’s premium Workspace package, Google One AI Premium, following the recent release of Gemini Ultra.
