On Monday, IBM introduced the latest version of its AI models aimed at businesses, seeking to leverage the growing trend of enterprises embracing generative AI technology. These genAI models are designed to continually learn and improve, automating tasks like coding, bug fixes, and document translation. They also have applications in data analysis and identifying suspicious activities.
GenAI has become a hot topic in IT as companies see the potential for cost decreases and revenue gains from its use. However, genAI also poses risks such as inaccuracy, which has caused organizations to adopt best practices to mitigate those risks and capture value. High-performing genAI adopters, for example, are twice as likely as others to include the legal function in deploying gen AI and embedding risk reviews into gen AI projects from the beginning.
At its THINK conference in New York, IBM opened a new tab and announced that it will open-source the latest versions of its Granite family of generative AI models, allowing customers to customize and run them in their data centers. This approach differs from rivals such as Microsoft, opening a new tab that tightly guards its models and charges customers for access. Instead, the Armonk, N.Y.-based company makes money when its customers use tools such as Watsonx to run the Granite models smoothly in their data centers.
The new Granite 3.0 models, available on Watsonx and other cloud platforms, include a base model that can handle 3 billion parameters and a mix-and-match option with 2 billion and 8 billion parameters. According to IBM, the models are trained for business use on 116 programming languages and are considered to be among the top-performing open large language models. Its training data is scrubbed for objectionable content and filtered to address governance, risk assessment, privacy protection, and bias mitigation. Customers can use these models for business-specific purposes and receive indemnity coverage for their use, the company said.
Other updates include new models and services to make generative AI easier for organizations to implement and manage. The Granite models can be used with the Watsonx data platform to enable a centralized way for businesses to store and manage their AI workloads across multiple systems and clouds. In addition, the platform will allow businesses to build their custom foundation model by using existing models available on Watsonx or bringing their own.
IBM also updated its Watsonx AI management and governance capabilities to help organizations create trusted workflows and deploy genAI in the data center or the cloud. It also added an AI repository and API that businesses can use to share their customized genAI models. The AI repository can be integrated into existing workflows or used to train and deploy a new workflow for specific use cases. It can be accessed through the Watsonx API and automatically scale to meet demand without manual provisioning.