Skip to content

Google Unveils Gemini Pro, a Prelude to Next-Gen Generative AI

  • Google unveils Gemini Pro, a lightweight version of its generative AI model, featured in Bard.
  • Gemini Ultra, the flagship model, boasts native multimodal capabilities, excelling in text, images, audio, and code understanding.
  • Reports highlight challenges, delays, and concerns regarding Gemini's handling of non-English queries, bias issues, and limited transparency.

Google launched Gemini Pro, a precursor to the more powerful Gemini Ultra, part of the Gemini family of generative AI models. Gemini Pro, a lightweight version, is featured in Bard, Google's ChatGPT competitor, showcasing improved reasoning and understanding capabilities. 

While Gemini Ultra's launch is delayed until next year, Gemini Pro debuts on December 13 for enterprise customers via Vertex AI, with broader availability in Google products like Duet AI, Chrome, and Ads in the coming months.

Gemini Ultra: A multimodal marvel with refined understanding

Gemini Ultra, the flagship model of the Gemini family, stands out as a natively multimodal AI capable of comprehending text, images, audio, and code. Google positions Gemini Ultra as a leader in complex reasoning tasks, especially math and physics. 

The model supports transcription of speech, answering questions about audio and videos, and excels in multimodal benchmarks, showcasing superiority over competitors in specific scenarios. However, Gemini Ultra's examples reveal marginal improvements over GPT-4, leaving questions about its groundbreaking capabilities.

Challenges and questions surrounding Gemini's development

Gemini's development faces challenges, with reports indicating struggles in handling non-English queries and delays in launching Gemini Ultra. Google's rush to market and limited details about Gemini's training data, environmental impact, and monetization strategy raise concerns. 

Google refrains from addressing potential bias and toxicity issues and fails to provide opt-out options for contributors to the training data. The rushed launch and perceived underwhelming features of Gemini Pro underscore the difficulties in creating state-of-the-art generative AI models.

Edited by Shruti Thapa