Exclusive: Jaxon AI Partners with IBM watsonx to Combat AI Hallucination Challenges

When an AI system generates inaccurate content—often referred to as “hallucination”—the results may not always be dire. However, if an AI system used in military technology hallucinates, the implications could be significantly more severe.

Jaxon AI began by developing AI systems for the U.S. Air Force, focusing on reliability and accuracy. Now, the company is expanding into the enterprise market with its Domain-Specific AI Language (DSAIL), aimed at tackling the critical issues of hallucinations and inaccuracies in large language models (LLMs). This technology integrates IBM's watsonx foundation models, representing a fresh approach to creating more dependable AI solutions.

How DSAIL Minimizes AI Hallucination Risks

Hallucination occurs when an AI produces inaccurate responses, often due to incomplete training data or a lack of verification. The DSAIL framework mitigates this risk by transforming natural language inputs into a binary format. This format undergoes a series of checks—like a boolean satisfier—to ensure the AI’s response adheres to all constraints before delivery, thus enhancing its reliability for practical applications.

One common strategy for reducing hallucinations is Retrieval Augmented Generation (RAG). In this model, the LLM accesses a knowledge base to provide accurate answers. Cohen noted that while RAG is part of DSAIL's methodology, outputs still must pass through additional verification checks before being finalized, further reducing the chance of hallucination.

The Role of IBM watsonx in Jaxon's AI Systems

Jaxon leverages models from IBM’s watsonx library as integral components of its AI systems. Specifically, the IBM StarCoder model facilitates code generation by automatically producing initial code for AI projects based on design specifications gathered during the process.

StarCoder, an open-source initiative launched in May with backing from ServiceNow and Hugging Face, is one of several code-generation tools within IBM's watsonx library. Savio Rodrigues, VP of ecosystem engineering and developer advocacy at IBM, confirmed that IBM was a founding contributor to the StarCoder project and highlighted its collaboration with Hugging Face to enhance enterprise access to open models.

While StarCoder showcases broad capabilities, IBM also offers specialized code-generation models tailored for specific applications, such as COBOL migration and quantum computing development.

IBM's Strategic Position in the Generative AI Market

The generative AI and LLM landscape is competitive, with major players like OpenAI, Microsoft, Google, and AWS. IBM is seeking its share of this market by supporting developers and independent software vendors (ISVs) like Jaxon AI through its IBM Build program.

IBM Build provides partners with access to watsonx, technical support, and go-to-market assistance. The aim is to deliver trusted AI foundation models characterized by consistent pricing, performance, and reliability. “Our customers trust IBM's approach to AI, particularly in model training and the legal checks we implement,” Rodrigues emphasized.