Addressing Hallucinations and Inaccuracies in AI Language Models

When an AI system hallucinates for content generation on a piece of text, it’s not an ideal situation, but it’s also not necessarily catastrophic. If an AI powering a piece of military technology hallucinates, the outcome could likely have more severe consequences. Jaxon AI, a startup that originally built AI systems for the U.S. Air Force, is now expanding into the broader enterprise market with a technology called Domain-Specific AI Language (DSAIL) to tackle hallucinations and inaccuracies in large language models (LLMs).

DSAIL, which incorporates IBM Watson foundation models, represents a novel approach to developing more reliable AI solutions. “Our tagline is AI for AI because we’re using Jaxon to help users create custom AI,” says Scott Cohen, CEO of Jaxon AI.

DSAIL: Mitigating the Risk of Hallucination

Hallucination occurs when an AI system generates an inaccurate response to a query.

The inaccuracy in AI system responses can be caused by various factors, such as incomplete training data and a lack of verification. To address this, DSAIL takes natural language inputs and converts them into a binary language format. Then, the transformed input is subjected to a series of checks and balances, ensuring that the AI response meets all constraints before being returned. This approach helps to limit non-determinism and increase the trustworthiness of AI systems for various applications.

One commonly used approach to reduce hallucinations is Retrieval Augmented Generation (RAG), where LLMs have access to a knowledge base to provide accurate answers. While DSAIL utilizes RAG to tackle hallucination problems, it also emphasizes the importance of running the output through a series of checks to further limit hallucination.

Jaxon AI and IBM’s Watson Foundation Models

Jaxon AI leverages models from IBM’s Watson foundation library to develop its AI systems. Specifically, the IBM StarCoder model is used for the code generation step in Jaxon AI. By leveraging StarCoder’s capabilities, Jaxon AI can automatically generate initial code for AI projects based on the collected design and requirements.

The StarCoder LLM is an open-source project that was launched with support from ServiceNow and Hugging Face. IBM, one of the founding contributors to the StarCoder project, partners closely with Hugging Face to bring open models to enterprise users. IBM also has its own code generation LLM tools in its Watson foundation library, which are used for specific use cases such as COBOL code migration and building quantum computing applications.

IBM’s Commitment to the AI Market

IBM competes in the market for generative AI and LLM technology alongside big players like OpenAI, Microsoft, Google, and Amazon Web Services (AWS). To assist developers and independent software vendors (ISVs) like Jaxon AI, IBM offers a program called IBM Build. This program provides partners with access to Watson foundation models, technical assistance, and go-to-market support.

IBM aims to provide organizations with reliable trusted AI foundation models, ensuring consistent pricing, performance, and availability. Savio Rodrigues, VP of ecosystem engineering and developer advocacy at IBM, highlights the trust customers have in IBM’s approach to AI training and legal checks.

Leave a Reply

Your email address will not be published. Required fields are marked *

Related Posts