Google’s generative AI model, Gemini launched, Things to know

Google’s much-anticipated next-generation generative AI model, Gemini, has made its debut in the form of Gemini Pro. However, this week’s release is essentially a lightweight version, with the more robust Gemini model expected to arrive sometime next year.

The Google DeepMind team, in collaboration with Google Research, presented an overview of Gemini, revealing that it’s not a single AI model but a family of models available in three versions: Gemini Ultra (the flagship model), Gemini Pro (a lightweight version), and Gemini Nano (optimized for mobile devices like the Pixel 8 Pro).

Gemini Pro, showcased as an enhancement for tasks like content summarization and reasoning, will be integrated into Google’s ChatGPT competitor, Bard. Scheduled to launch on December 13 for enterprise customers using Vertex AI, Gemini Pro will later be incorporated into various Google products.

Gemini Nano, tailored for mobile applications, will soon be available for preview through Google’s AI Core app on the Pixel 8 Pro, and developers can sign up for early access.

While Gemini Pro is claimed to outperform OpenAI’s GPT-3.5 in certain benchmarks, Gemini Ultra, designed to be natively multimodal, shows advancements in comprehending text, images, audio, and code. However, Gemini Ultra’s benchmark superiority is marginal, and its capabilities are highlighted in solving physics homework and extracting information from scientific papers.

The training data details for Gemini remain undisclosed, and Google refrained from providing information on how it collected the data and potential compensation for contributors.

14 Things to Note About Gemini, Google’s Ai Model

1. Gemini comprises three models: Gemini Ultra (flagship), Gemini Pro (lite version), and Gemini Nano (optimized for mobile devices).
2. Gemini Pro powers Bard, Google’s ChatGPT competitor, delivering improved reasoning, planning, and understanding capabilities over its predecessor.
3. Gemini Pro’s launch for enterprise customers using Vertex AI is set for December 13, and it will integrate into various Google products in the coming months.
4. Gemini Nano, for mobile devices, will launch soon in preview via Google’s AI Core app, starting with the Pixel 8 Pro.
5. Gemini Pro is claimed to outperform OpenAI’s GPT-3.5 in tasks such as summarizing content and brainstorming but faces scepticism due to the lack of independent verification.
6. Gemini Ultra, designed to be natively multimodal, can comprehend information in text, images, audio, and code, outperforming rival models like GPT-4 with Vision.
7. Concerns arise as Google refuses to disclose details about Gemini’s training datasets and whether contributors can opt out or expect compensation.
8. Gemini was trained on Google’s in-house AI chips (TPUs), but specifics about the number of chips used, cost, and environmental impact remain undisclosed.
9. Gemini Ultra’s benchmark superiority is highlighted, but closer inspection reveals marginal improvements over GPT-4 and GPT-4 with Vision across various benchmarks.
10. The launch of Gemini suggests a rushed job, with concerns about its actual capabilities and comparisons to Google’s previous generative AI efforts like Bard.
11. Troubled development reports include difficulties in handling non-English queries and uncertainties about Gemini’s monetization strategy.
12. Gemini Ultra’s availability will be limited initially, with concerns raised about its understanding of novel capabilities and a lack of a clear monetization strategy.
13. Gemini Pro and Gemini Ultra’s context window limitations are discussed, with Gemini Pro marginally outperforming GPT-3.5.
14. Questions arise about the impact of Google’s marketing and the challenging nature of building state-of-the-art generative AI models, especially given past development issues.

Recommended: Elon Musk’s Grok AI chatbot finally rolled out for X premium plus subscribers

Gemini’s training utilized Google’s in-house AI chips, tensor processing units (TPUs), without specifying the number of chips used or the environmental impact. Google showcased Gemini’s applications in physics homework and information extraction from scientific papers through prerecorded demos. Despite claims of Gemini Ultra’s benchmark superiority, a closer look reveals only slight improvements over GPT-4 in various benchmarks.

Recommended: Microsoft to unveil new AI Features for Bing Chat

Gemini’s launch raises questions about the rushed development, with concerns about the model’s capabilities and the overall development process. Google’s previous AI endeavors, such as Bard, faced criticism for its initial shortcomings and concerns over accelerated timelines.

Gemini’s troubled development, reportedly involving difficulties in handling non-English queries and uncertainty about monetization strategies, adds to the skepticism surrounding the product’s potential impact. The launch leaves uncertainties about Gemini’s actual capabilities, particularly compared to the initial marketing promises.