Jaxon is taking a significant leap forward with the introduction of Domain-Specific AI Language (DSAIL) (link resides outside ibm.com), which represents a new approach to AI development, targeting one of the most challenging aspects of AI technology: hallucinations in large language models (LLMs). With help from watsonx, IBM’s portfolio of AI products, Jaxon’s developer-friendly system seeks to help reduce hallucination related inaccuracies.
Our team has been working closely with IBM Ecosystem Engineering and has embedded StarCoder LLM, hosted in the watsonx.ai studio, and I’m thrilled with the outcome. The watsonx.ai foundation model library hosts both IBM and third party models for AI builders to use. StarCoder is the first of many models from watsonx.ai that we plan to incorporate into the Jaxon platform. The foundation models and supporting tools and techniques found in watsonx are building blocks, and Jaxon is the manual that helps users architect custom solutions.
Applying DSAIL to AI design can be transformative. It automates the design cycle and enables developers to reason about designs abstractly. The process involves prototyping with metamodels and simulated data, which allows for an exploration of potential solutions before final implementation. DSAIL empowers data scientists, developers, and analysts, to specify problems abstractly, use declarative languages, optimize solution space exploration and align LLM designs.
For example, DSAIL enables users to articulate the problem or task in software using both natural and formal languages. By leveraging declarative language, DSAIL allows for the unambiguous statement of requirements to complex AI systems. DSAIL’s methodology enables the exploration of a wide range of solutions in an abstract manner. This flexibility can make decision making more viable, mitigating the too-common situation where time forces a team to move forward to production with suboptimal, rushed designs.
At the core of DSAIL is its ‘Fact Checker’, a robust system designed to tackle the hallucination issue head-on. DSAIL establishes a formal system that aligns the efficiency of LLM-produced designs with the grounding of user-provided constraints and known-good knowledge stores.
Depending on the use case, Jaxon selects the IBM or third-party foundation model(s) from the watsonx.ai library and incorporates it into solution architectures that directly address business problems. Jaxon then applies DSAIL technology to refine and adapt these models so they can be relevant and effective in their designated domains and system roles.
This approach represents a fusion of versatility and specialization. The watsonx.ai foundation models offer a broad base of capabilities, which are then honed and aligned by Jaxon to meet the unique demands of different sectors. Whether it’s insurance, finance, healthcare or any other industry, Jaxon offers AI solutions that are not just efficient but also aligned with the specific challenges and nuances of each domain.
A pivotal aspect of Jaxon AI’s collaboration with IBM is the strategic utilization of the StarCoder foundation model. This advanced model provides a starting point for Jaxon’s specialized approach.
The synergy between Jaxon’s DSAIL technology and watsonx is a cornerstone of this collaboration, paving the way for highly tailored, domain-specific AI applications.
By addressing the hallucination issue, Jaxon DSAIL is not just a step forward in AI technology; it’s a leap towards a future where AI is powerful, trusted and reliable.