A new training method is being utilized by researchers to address AI fact-validation shortcomings

“The model could be provided with a genetics dataset and instructed to produce a report on the gene variations and mutations it contains,” outlined IBM. “By planting a few of these starting points, the model commences generating novel guidelines and replies, drawing on the latent knowledge in its training data and utilizing RAG to retrieve facts from external databases as needed for ensuring precision.”

This method may appear somewhat similar to the implementation of RAG. The key distinction, as indicated by the researchers, is that these specialized models are summoned only when necessary via an API.

Continued challenges in factual accuracy

Mark Stockley, co-host of The AI Fix podcast alongside Graham Cluley, noted that the fundamental issue lies in the widespread misunderstanding of LLMs. While they excel at specific tasks, they are not, and were never designed to be, straightforward fact- or truth-validation engines.

About Author

AndyC

Andy Curtis is an award-winning security consultant, researcher and public speaker. He has been working in the computer security industry since the early 1990s, having been employed by state and federal government, leading healthcare and banking providers across three continents. He has given talks about computer security for some of the world’s largest companies, worked with law enforcement agencies on investigations into hacking groups, and is a regular voice on TV and radio explaining IT security threats.

See author's posts