IBM Unveils Generative AI Foundation Models and Watsonx Enhancements

IBM Unveils Generative AI Foundation Models and Watsonx Enhancements

IBM has introduced its Granite series foundation models, applying generative AI to both natural language and coding tasks. These models, which use the "Decoder" architecture, are designed to support enterprise natural language processing (NLP) tasks such as content generation, summarization, and insight extraction. IBM plans to provide comprehensive information about data sources, data processing, and filtering steps used to produce training data for the Granite series, with availability expected this month. The company also offers third-party models, including Meta's Llama 2-chat 70 billion parameter model and the StarCoder LLM for code generation on the Watsonx.ai platform hosted on IBM Cloud.

IBM's Watsonx.ai models are trained on the company's enterprise-focused data lake, employing a rigorous data collection process and control points to ensure governance, risk assessment, compliance, and bias mitigation in model deployment.

Additional capabilities planned for the Watsonx platform include:

1. Tuning Studio for Watsonx.ai, scheduled for release this month, allowing adaptation of foundation models to unique downstream tasks using enterprise data.

2. Synthetic Data Generator for Watsonx.ai, also set for release this month, enabling users to create artificial tabular datasets from custom data schemas or internal datasets for AI model training and fine-tuning while minimizing risk.

3. Watsonx.data lakehouse data store, integrating Watsonx.ai generative AI capabilities to help users explore, visualize, and refine data through a natural language interface, expected to be in preview in the fourth quarter of this year.

4. Watsonx.data, incorporating vector database capability to support retrieval-augmented generation use cases with Watsonx.ai, planned for a fourth-quarter preview.

5. Model Risk Governance for Generative AI, launching as a tech preview for Watsonx.governance, will enable clients to automate the collection of foundation model details and model risk governance capabilities, providing relevant metrics in dashboards for enterprise-wide AI workflows.

IBM is also integrating Watsonx.ai innovations into its hybrid cloud software and infrastructure, including:

• Intelligent IT Automation, currently in tech preview, which includes automation products Instana and AIOps. It incorporates Intelligent Remediation, leveraging Watsonx.ai generative AI foundation models to assist IT ops practitioners in summarizing incident details and providing prescriptive workflow suggestions.

• Developer services for Watsonx, set to launch in the first quarter of 2024, aiming to bring Watsonx capabilities closer to data on IBM Power for SAP workloads, expanding clients' options for using AI near their data on IBM Power systems and deploying AI algorithms for sensitive data and transactions.