AI research explores hidden messaging and Alzheimer's detection

Researchers are investigating the ability of large language models to hide and decode secret messages within images. A study involving the Singapore AI Safety Fellowship (Sash) tested Anthropic’s Fable model and OpenAI’s Astra. The researchers found that Fable used object heights and rotations, such as skyscraper heights, to encode text in 15 different languages. In contrast, OpenAI’s Astra utilized methods like Morse code and acrostics.

In the medical field, artificial intelligence is being utilized to identify early warning signs of Alzheimer's disease. AI models are capable of analyzing vast datasets, including speech patterns, medical notes, and PET scans, to detect patterns difficult for human doctors to spot. The StateViewer research program, for instance, can distinguish nine patterns of different dementia types in brain scans with up to 88 percent accuracy. While these tools can assist in screening and clinical trial recruitment, experts note they are not replacements for blood tests or professional medical diagnoses.