Google AI targets programmer productivity

Deutsche Bank is one of a number of organisations testing Google’s generative artificial intelligence (AI) and large language models (LLMs) at scale.

At the Google I/O 2023 event, the multinational tech company unveiled a raft of new AI capabilities and early adopter customers.

The bank is testing how Google’s AI’s capabilities could be deployed to provide new insights to financial analysts, driving operational efficiencies and execution velocity. According to Google, there is an opportunity to significantly reduce the time it takes to perform banking operations and financial analysts’ tasks. The bank also sees opportunities to deploy AI to support staff by increasing their productivity while helping to safeguard customer data privacy, data integrity and system security.

Another customer, Uber, is using Google’s AI as the basis for generative AI for customer service chatbots and agent assist capabilities. Google said the chatbots are able to handle a range of common service issues with human-like interactions. Uber aims to use AI for cost efficiency and to achieve greater customer satisfaction.

Along with improving employee productivity and helpdesks, Google sees a big role for large language models to improve the productivity of software developers.

The company has introduced new foundation models and capabilities across its Google Cloud AI products. In a blog discussing the AI capabilities, Google Cloud CEO Thomas Kurian said the company was building out generative AI support in Vertex AI, the company’s framework for giving software developers and data scientists access to foundational models.

One example is Codey, a code generation foundation model, which Kurian said can help accelerate software development with code generation, code completion and code chat.

Another example, Imagen, is a text-to-image foundation model. According to Kurian, this lets customers generate and customise studio-grade images.

There is also Chirp, a speech model, which Kurian said allows customers to engage more deeply with their customers in their native languages with captioning and voice assistance.

All of these new features are available to programmers via application programming interfaces (APIs). Kurian said the APIs have been tuned through Google Generative AI Studio and offer developers enterprise-grade security and reliability. Management and security capabilities include encryption, access control, content moderation and recitation capabilities, which Kurian said enables organisations to see the sources behind outputs produced by their AI models.

Text Embeddings is another new API, designed to help developers build recommendation engines, classifiers, question-answering systems, similarity matching and other applications based on semantic understanding of text or images.

Google is also now offering a feature called Reinforcement Learning from Human Feedback (RLHF), which allows organisations to incorporate human feedback to customise and improve model performance.

Another programming tool, Duet AI, offers what Google claims is an expert pair programmer to assist cloud users with contextual code completion. Google said Duet is able to offer software developers suggestions tuned to their code base, generating entire functions in real time, and assisting with code reviews and inspections.

To support enterprise users and the public sector, Google has also ramped up its partner programme, which includes Cognizant, Deloitte and KPMG, among others, as well as software-as-a-service providers such as Box, Jasper and Salesforce.