IBM launches generative AI models and features

IBM unveiled new generative AI models and capabilities on its Watsonx data science platform this week to stay relevant in the highly competitive AI field.

The Granite series models, like OpenAI’s GPT-4 and ChatGPT, can summarize, analyze, and generate text. IBM supplied few specifics regarding Granite, making it impossible to compare the models to other LLMs, including IBM’s. However, the business promises to provide the Granite series models’ training data and filtering and processing procedures before their Q3 2022 release.

We’ll hold the corporation accountable.

IBM is also releasing Tuning Studio in Watsonx.ai, which helps customers test, deploy, and monitor models after deployment.

IBM Watsonx clients can fine-tune models to new jobs with 100 to 1,000 instances using Tuning Studio. Once customers set a job and supply labeled samples in the proper data format, they can deploy the model through IBM Cloud API.

Watsonx.ai will soon launch a synthetic data generator for tabular data, which is rows and columns in relational databases. In a news release, IBM says it can generate synthetic data from custom data schemas and internal data sets to extract insights for AI model training and fine tuning with “reduced risk.”

Given the risks of training AI with synthetic data, “reduced risk” seems unclear. (We requested clarification.) Make of that what you will.

IBM is also adding generative AI capabilities to Watsonx.data, a data store that lets users access data through query engines, governance, automation, and connectors with existing databases and tools. Self-service, chatbot-like tools will allow consumers to “discover, augment, visualize and refine” AI data in Q4 2023 as a tech preview.

IBM again provided few details. I imagine a data visualization and transformation-focused ChatGPT experience.

IBM claims Watsonx.data will support retrieval-augmented generation (RAG) vector databases in Q4 2023. IBM’s enterprise clients can benefit from RAG, an AI framework that improves LLM replies by rooting the model on external information sources.

Other significant news: IBM is starting the technical preview for Watsonx.governance, a toolbox that protects customer data, detects model bias and drift, and helps enterprises satisfy ethics standards. IBM will debut Intelligent Remediation next week, using generative AI models to help IT staff summarize events and recommend workflows to deliver solutions.

We’re here to help clients across the whole AI lifecycle, as proven by the watsonx platform’s continued evolution within just a few months since launch, said IBM SVP of products Dinesh Nirmal in a news statement. As a transformation partner, IBM helps clients expand AI securely and reliably, from establishing core data strategies to tailoring models for specific business uses cases to governing models beyond that.

IBM faces pressure to prove it can compete in AI.

IBM’s second fiscal quarter sales underperformed analyst projections due to a larger-than-expected infrastructure slowdown. Revenue fell 0.4% to $15.48 billion, below the expert average of $15.58 billion for Q2.

IBM CEO Arvind Krishna stressed the importance of AI to IBM’s future growth during the earnings call, saying businesses are signing up at a strong rate to adopt IBM’s hybrid cloud and AI capabilities, including Watsonx. Krishna said Samsung and Citi were among 150 corporate customers using Watsonx when it launched in July.

“We continue to meet the needs of our clients who seek trusted, enterprise AI solutions, and we are excited about the response to the Watsonx AI platform. Krishna added on the earnings call, “We remain confident in our revenue and free cash flow growth expectations for the full year,” reported Investing.com.