Břetislav Šopík (Gen) – Anti-scam protection in SMS: Using large language models for automated... скачать в хорошем качестве

Břetislav Šopík (Gen) – Anti-scam protection in SMS: Using large language models for automated...

Anti-scam protection in SMS: Using large language models for automated annotation of SMS and expansion of training dataset Large language models developed by AI research labs offer a wide range of applications. In our SMS anti-scam protection product, we leverage these models for automated annotation. This approach enables us to significantly expand our labeled dataset—used to train custom, scalable classifiers—while also reducing the workload of our expert analysts. To be effective, an automated annotation system must meet several stringent criteria: – Its annotation accuracy must match that of a human expert. – It must provide confidence scores for its decisions and abstain from annotating when confidence is low. – Its outputs must be consistent and reproducible. In this talk, we will present how we successfully designed and implemented such an annotation system. https://www.zijemeit.cz/workshop?id=13 Břetislav Šopík Initially a theoretical physicist dedicated to computer modeling of the superconducting state. Feeling the need to be more involved in solving problems related to people's daily lives, he decided to leave the academic world in 2015 and joined Avast (today's GenDigital). Since then, he has been dedicated to end-user cyber security. He specializes in applied research in machine learning and big data analysis.