Post by regenold

2,104 followers

๐—”๐—œ ๐—ฎ๐˜€๐˜€๐—ถ๐˜€๐˜๐—ฎ๐—ป๐˜๐˜€ ๐—ฐ๐—ฎ๐—ป ๐˜€๐—ผ๐˜‚๐—ป๐—ฑ ๐˜ƒ๐—ฒ๐—ฟ๐˜† ๐—ฐ๐—ผ๐—ป๐—ณ๐—ถ๐—ฑ๐—ฒ๐—ป๐˜. ๐—•๐˜‚๐˜ ๐—ฎ๐—ฟ๐—ฒ ๐˜๐—ต๐—ฒ๐˜† ๐—ฎ๐—ฐ๐˜๐˜‚๐—ฎ๐—น๐—น๐˜† ๐—ฟ๐—ฒ๐—ด๐˜‚๐—น๐—ฎ๐˜๐—ผ๐—ฟ๐˜†-๐—ฟ๐—ฒ๐—ฎ๐—ฑ๐˜†? We recently tested several off-the-shelf AI assistants on regulatory questions based on the EU AI Act, MDR and EMA guidance. Many responses looked convincing - yet around 30% contained serious issues. This new article written by Marco Virgolin, PhD explains why generalist AI tools arenโ€™t enough for regulated environments, and why purpose-built assistants and robust evaluation frameworks (Evals) are needed.

Post content