Here in lines a problem when you use LLMs and RAG that is unguided and does not have baseline reference data to use to reason against.

← Back to Blog

Here in lines a problem when you use LLMs and RAG that is unguided and does not have baseline reference data to use to reason against. One suspects that over 99% of AI used by organisations today goes unguided, unstructured, uncontrolled, and ungoverned. AKA what comes out is more often than not taken for granted.

This is a problem when AI is then used for applications where users expect accuracy and repeatability, yet do not, or are unable to, validate outputs, or do not have the necessary knowledge, skills, or experience to do so.

AI is a probabilistic tool that reasons based on what it is told, what it can find, has been mapped and vectored, and what it can infer. If any of this is wrong, then the output will also be inaccurate.

This becomes a significant issue for functions where AI is being adopted, such as cybersecurity and audit. Functions that can be uncorrelated, independent, or decoupled, AKA what goes into a process is not correlated with what comes out of a process. Where human knowledge, skills, and experience does the correlation, based on years of learning and human reasoning.

@Thaddeus Dziekanowski

@Parava Security Solutions

#ai #aiaudit #aicybersecurity

View Original Post
https://www.linkedin.com/feed/update/urn%3Ali%3AugcPost%3A7460604194924838912