Pushing for a stronghold in the intensely competitive AI arena, IBM announced this week the launch of novel generative AI models and features on its recently unveiled Watsonx data science platform.
Dubbed the Granite series models, these tools appear to align with popular large language models (LLMs) like OpenAI’s GPT-4 and ChatGPT, boasting abilities like text summarization, analysis, and generation. Details about Granite remain scanty, keeping direct comparisons with other LLMs, including IBM’s own, at bay. However, IBM commits to shedding light on the dataset used to develop the Granite series and the methods for its filtration and processing, anticipating a Q3 2022 release.
Complementing this release, Watsonx.ai, the Watsonx segment facilitating model testing, deployment, and post-launch monitoring, will introduce the Tuning Studio. This innovation empowers users to modify generative AI models to align with their unique data. Leveraging the Tuning Studio, Watsonx’s clientele can adapt models to novel tasks using as little as 100 to 1,000 samples. After task specification and provision of labeled data samples, users can activate the model through the IBM Cloud’s API.
Another anticipated Watsonx.ai feature is a synthetic data generator designed for tabular data, typical of relational databases. IBM’s press release posits that this tool, drawing on custom data schemas and internal datasets, can generate synthetic data. This can potentially offer companies insights valuable for AI model training and tweaking, albeit at a “reduced risk.” The exact implications of “reduced risk” remain ambiguous, especially considering the challenges of AI training on synthetic data.
IBM’s foray into expanding its generative AI capabilities, particularly on the Watsonx platform, underscores the tech giant’s ambition in the AI landscape. While the exact prowess of the Granite models remains to be seen, the introduction of user-centric tools like Tuning Studio displays a clear commitment to delivering tailored AI solutions for clients. As for the synthetic data generator, its true value and implications will be clearer as more details emerge.
