In a bid to establish its presence in the rapidly evolving and highly competitive AI sector, IBM recently unveiled innovative generative AI models and features within its newly launched Watsonx data science platform.
These new models, titled the Granite series, resemble standard large language models (LLMs), akin to OpenAI’s GPT-4 and ChatGPT, and are designed to summarize, analyze, and generate text. While IBM has not disclosed extensive details about Granite, making direct comparisons to rival LLMs challenging, the company has promised to provide insights into the training data used for the Granite models and the filtering methods applied, ahead of their rollout in Q3 2023.
We will hold IBM accountable for that commitment.
In the meantime, Tarun Chopra, IBM’s Vice President of Product Management for Data and AI, shed light on these developments during an email interview. He stated, “The IBM Granite series models have been developed using curated, enterprise-quality data instead of publicly scraped data. Each series includes specialized subsets for different domains. For instance, one model is tailored for financial data, allowing AI builders to utilize a compact model that delivers performance comparable to larger, general models. These models can effectively support various enterprise NLP tasks such as summarization, content generation, and insight extraction.”
In addition, IBM is introducing Tuning Studio within Watsonx.ai, enabling users to customize generative AI models to their specific data needs. With Tuning Studio, IBM Watsonx customers can fine-tune models for new tasks using as few as 100 to 1,000 examples. By specifying a task and providing labeled samples in the necessary data format, users can deploy their tailored model via an API from the IBM Cloud.
Moreover, a new synthetic data generator for tabular data will soon debut in Watsonx.ai. According to IBM, this tool allows businesses to generate synthetic data from custom schemas and internal datasets, assisting in extracting insights critical for training and fine-tuning AI models while minimizing risks. However, it remains unclear what precisely is meant by “minimizing risks,” given the challenges associated with training AI on synthetic data. (Further clarification is being sought.)
IBM is also enhancing generative AI capabilities in Watsonx.data, which integrates query engines, governance, automation, and compatibility with existing databases and tools. Starting in Q4 2023, customers will gain access to a tech preview that enables them to “discover, augment, visualize, and refine” their data via a self-service, chatbot-like tool.
While specifics remain sparse, this experience promises to resemble a data visualization and transformation-centric version of ChatGPT. As Chopra described, “The generative AI capabilities in Watsonx.data will simplify and expedite user interactions with their data. For example, when a user seeks specific data, the AI chat assistant interface can generate a text response along with API calls and parameters to fulfill the request. It will also enable the importing of external data, with the AI model performing semantic data enrichment.”
In parallel, IBM plans to introduce a vector database capability in Watsonx.data to enhance retrieval-augmented generation (RAG) by Q4 2023. RAG is an AI framework aimed at improving the quality of LLM-generated responses by grounding them in external knowledge sources, greatly benefiting IBM's enterprise clients.
Additionally, IBM is entering the technical preview phase for Watsonx.governance, a toolkit aimed at safeguarding customer privacy, identifying model bias and drift, and ensuring compliance with ethical standards. Next week, IBM will also launch Intelligent Remediation, which utilizes generative AI models to assist IT teams in summarizing incidents and recommending workflows for effective solutions.
“As evidenced by the continuous development of the Watsonx platform within just a few months since launch, we are dedicated to supporting clients throughout the entire AI lifecycle,” stated IBM SVP of Products Dinesh Nirmal. “As a transformation partner, IBM is working closely with clients to scale AI in a secure and trustworthy manner, aiding them in establishing foundational aspects of their data strategies and fine-tuning models for their unique business applications.”
Undoubtedly, IBM faces significant pressure to establish its foothold in the crowded AI market. In its second fiscal quarter, the company reported revenue of $15.48 billion, slightly below analyst expectations due to a larger-than-anticipated slowdown in its infrastructure segment, marking a 0.4% decline year-over-year against the analyst consensus of $15.58 billion.
Throughout the earnings call, IBM CEO Arvind Krishna emphasized the critical role of AI in the company's future growth, highlighting that companies are increasingly adopting IBM’s hybrid cloud and AI solutions, including Watsonx. As of July, over 150 corporate clients, including major organizations like Samsung and Citi, had already begun utilizing Watsonx.
“We are committed to addressing the needs of our clients seeking trusted, enterprise AI solutions,” Krishna reiterated. “We are particularly thrilled with the positive response to the newly launched Watsonx AI platform, and we remain optimistic about our revenue and free cash flow growth expectations for the full year.”
AI, Generative AI, IBM, Watsonx
Ranking
