The paper was submitted in June 2023. The results sound promising. The approach also seems interesting because apparently they use a different approach than the usual NLP embeddings to encode the training data.