Google introduced Gemini 1.5 Pro on Thursday, highlighting its significant performance improvements compared to its predecessor. This latest iteration follows the recent unveiling of Gemini 1.0 Ultra and the rebranding of the Bard chatbot to align with the new model’s enhanced capabilities.
In a blog post announcement, Google CEO Sundar Pichai and Google DeepMind CEO Demis Hassabis addressed the ethical considerations surrounding AI safety while underscoring the rapid advancements in their models. Pichai emphasized the teams’ focus on safety while pushing the boundaries of their latest models.
The company faces the dual challenge of reassuring AI skeptics and regulators about safety while also showcasing the accelerating performance of its models for developers, customers, and investors concerned about competition from platforms like OpenAI’s ChatGPT.
According to Pichai and Hassabis, Gemini 1.5 Pro offers comparable results to Gemini 1.0 Ultra but operates more efficiently with reduced computational requirements. Its multimodal capabilities encompass text, images, videos, audio, and code processing. As AI models evolve, they continue to provide a broader range of capabilities in a single prompt box, exemplified by recent integrations like DALL-E 3 image generation into ChatGPT.
Gemini 1.5 Pro can handle up to one million tokens, demonstrating versatility in processing various data types. Google reported successful tests with larger token counts, highlighting its performance in tasks like the Needle In a Haystack evaluation and analyzing complex datasets such as the Apollo 11 mission transcripts and a silent film starring Buster Keaton.
The company plans to introduce new pricing tiers supporting up to one million-token queries, paralleling OpenAI’s GPT-4 models. Gemini 1.5 Pro excels in “in-context learning,” demonstrated by its ability to learn new skills from long prompts without additional fine-tuning.
On the ethical front, Google emphasizes responsible deployment and rigorous testing for potential harms, including content safety and representational harms. Gemini 1.5 is initially available in early access for developers and enterprise customers, with broader availability planned in the future.
