Sign up for our daily and weekly newsletters to stay updated with the latest in AI technology.
When an AI system generates incorrect information, it’s not ideal, but usually not disastrous. However, if such an error happens in military technology, the consequences could be much more serious.
Jaxon AI initially developed AI systems for the U.S. Air Force, where the highest levels of reliability and accuracy were crucial. The company is now branching out into the wider enterprise market with a tool called Domain-Specific AI Language (DSAIL). This technology aims to tackle one of artificial intelligence’s significant problems: hallucinations and inaccuracies in large language models (LLMs). By using IBM’s watsonx foundation models, DSAIL offers a new way to create more reliable AI solutions.
AI hallucination happens when a system produces an inaccurate response due to factors like incomplete training data or insufficient verification. DSAIL aims to reduce these errors by converting natural language inputs into a binary format. This format then goes through various checks and balances to ensure the AI response meets all necessary criteria, boosting the system’s reliability.
One common technique to decrease hallucination is Retrieval Augmented Generation (RAG), where the AI model accesses a knowledge base to generate more accurate responses. However, DSAIL takes it a step further by subjecting the RAG output to additional checks before delivering the final answer to the user.
Jaxon AI incorporates models from IBM’s watsonx foundation library to build its AI systems. Specifically, the IBM StarCoder model is used for code generation. This model helps Jaxon automatically generate initial code for AI projects based on collected design requirements, forming part of Jaxon’s method for creating bespoke AI systems.
IBM’s StarCoder was initially launched with the support of ServiceNow and Hugging Face. IBM has partnered closely with Hugging Face to bring open models to enterprise users and has various tools in the watsonx library for different applications, such as helping with COBOL code migration and developing quantum computing systems.
IBM is striving to embed watsonx in software vendor tools through a program called IBM Build. This initiative aims to support developers and independent software vendors (ISVs), like Jaxon AI, by providing access to watsonx, technical support, and market strategies. The objective is to offer organizations reliable and trusted AI models, ensuring consistent pricing, performance, and availability.
Stay informed with our daily updates on the latest in AI technology by subscribing to our newsletter.