Google Cloud has over 100 language models, is working towards reducing hallucinations: CEO Thomas Kurian

Google Cloud, the enterprise division of Google, hosted its Born Digital Summit in Bengaluru, where CEO Thomas Kurian discussed the company’s strides in AI and its commitment to global infrastructure. With two Google Cloud regions already established in Delhi and Mumbai, India plays a significant role in this digital transformation journey.

Google Cloud offers two primary AI solutions: Vertex for developers and Duet for end-users and businesses. Kurian emphasized that Google Cloud’s capabilities allow it to train AI models twice as fast and at half the cost compared to competitors.

Kurian highlighted the Google Cloud model garden, boasting over 100 models of various sizes catering to diverse needs, including text, chat, programming code, images, audio, and visuals. Frequent model upgrades, occurring every two weeks, keep the technology at the forefront.

Currently, 24,000 Google developers utilize Google Cloud’s AI-powered coding model, with over 75% of them accepting almost all the instructions. Notably, Moderna, a key contributor to the COVID-19 vaccine, employs Duet AI to accelerate medical drug simulation, converting medical experiments into mathematical models for rapid processing.

Kurian discussed the capabilities of smaller AI models, like those powering Gmail on smartphones. These models run locally, providing super-low latency. In contrast, larger models possess general skills but are slower due to evaluating hundreds of millions of parameters.

Google Cloud’s Cortex AI platform offers natural language-based interactions for search and conversations. Users can interact with systems using everyday language for information retrieval or transactions.

Addressing concerns about generative AI, Google Cloud is developing a feature called grounding. It delivers fresh results alongside fact-checks by comparing model-generated outcomes with established datasets, reducing the risk of hallucinations.

The Vision watermarking feature allows users to identify AI-generated images even without watermarks, as AI-generated images typically lack copyright protection.

As part of ongoing developments, Google introduced the new PaLM (Pathways Language Model) generation, capable of summarizing entire books. Med-PaLM caters to medicine-related inquiries, while another model generates highly realistic images.

On the user front, the Google Cloud Duet-AI-powered Help Me Write feature in Gmail has been a hit since March 2023, with over a million users. Google Sheets now features an AI-powered analyst for formula generation and calculations, while Google Slides can create photo-realistic images from text inputs.

Google Cloud’s continuous innovation promises to shape the future of AI and its applications, making it a prominent player in the rapidly evolving technology landscape.