Skip to content

Google’s newest AI Model, Gemini

Google just released its latest AI model: Gemini, opening a whole new range of creative possibilities with AI for businesses. What’s changing for AI with this new model? And what’s the value organisations can get out of it? We interviewed Jason Quek, Devoteam G Cloud’s Global CTO, and Tristan Van Thielen, Devoteam G Cloud’s Machine Learning Tribe Lead to get you up to speed.

Google Cloud

Gemini introduces multimodal capabilities, greater accuracy and decreased bias, opening up a whole new wave of creative possibilities for companies. As Google’s largest AI model, Gemini reflects the company’s commitment to pushing the boundaries of innovation in AI. Its unique approach, using a single model to integrate multiple modalities, establishes a notable industry standard. Resulting in handling various inputs and delivering more refined outcomes.

“We believe that Gemini ultra will be a game changer to be able to create and provide the outputs and experiences that our top enterprise customers expect.”

Multimodal capabilities: a new range of business use cases

Gemini’s core strength lies in its ability to process and understand diverse data formats, including text, code, images, and video.

This multimodal capability opens up a vast array of opportunities for real-world applications. Gemini bridges the gap between different data types, enabling it to analyse complex images and decipher intricate code. This versatility makes it a helpful tool for a wide range of applications.

New AI business use cases: smarter chatbots

Take the example of customer service chatbots. Google’s current AI models can for example fill in your insurance claims. The new Gemini model can receive both text and visual information. In this case, Gemini will not only guide you through the process but also ask you for an image of your damaged car and analyse the image to fill in the insurance claim properly.

“Gemini will unlock new use cases that people will come up with in a more creative aspect. This can lead to new startups or use cases that we hadn’t thought of in the past”

New AI business use cases: document analysis

Gemini simplifies complex machine manuals, empowering you with guidance. Unlike conventional chatbots that handle text or images separately. Gemini integrates both, eliminating the need for disjointed responses.

Its ability to analyse both text and diagrams provides concise and relevant instructions, saving time and boosting productivity. Additionally, you can directly scan machine components, and Gemini will swiftly identify corresponding diagrams and contextual text to provide accurate and immediate assistance.

Gemini’s powerful multimodal features will also become available in Vertex AI, allowing you to tune the model to fit your needs.

“I could see someone optimising Gemini to assist doctors in diagnosis by combining visual data, such as X-rays, with text-based patient records and other relevant information.”

Data security and privacy

Protecting your organisation’s data is a top priority for Google and is an integral part of Gemini and Vertex AI. Your AI training and usage data is securely isolated and only accessible in the region of your choice (GDPR/other regulations). Gemini never shares your data with Google for retraining purposes. Rigorous privacy standards are enforced to safeguard your information and ensure its confidentiality.

“Vertex AI’s PaLM and Gemini models prioritise privacy by securely isolating customer data, safeguarding against leaks during AI training—an essential feature for our enterprise clients”

Challenges and responsible use of AI

New AI models such as Gemini offer innovative capabilities, but you must consider their limitations and ensure responsible implementation. Dealing with shortcomings and biases is essential when you work with large language models. Despite their impressiveness, these models can make errors, requiring careful examination.

“Models inevitably contain inaccuracies and biases, reflecting human fallibility and the web’s documented history.”

Jason underscores the need for careful evaluation when engaging with AI-generated responses. Due to their inherent characteristics, large language models are susceptible to errors, requiring a comprehensive examination. This involves confirming factual accuracy, scrutinising generated text, and ensuring that generated code is optimised for efficiency.

Future evolution of Google & AI

Google aims to improve and surpass current models with more advanced versions, contributing to continuous competition for superior performance among AI competitors.

Jason expects that “upcoming versions of Google’s generative AI models will be used to greatly enhance existing Google services such as Document AI, Google Bard or Vertex AI, and also unlock novel applications across industries”

For example, in the case of Document AI, the need for labelled data in document processing is changing. Generative AI now provides an initial estimate of required labelling for unstructured data, speeding up the overall process. These improvements can be expected in various Google services.

“Google might meet diverse user needs by providing different AI service tiers, empowering customers to select the most fitting versions.”

The development of Google’s AI portfolio is likely to involve creating more specialised AI solutions to meet specific industry needs. Recent initiatives, like Google’s involvement in medical AI with Med-PaLM 2, indicate a stronger emphasis on AI models customised for particular domains or industries. This approach enables these models to understand industry-specific language and nuances effectively.

Jason and Tristan envision another step forward, proposing the training of AI to align with the thought processes of the deploying company, further enhancing the customised nature of the AI.

Your future with AI? Experience the possibilities with a Gen AI Hackathon.