Google’s new model, DolphinGemma, can predict dolphin sounds.

Google CEO Pichai has just tweeted to announce DolphinGemma, a large language model project fine-tuned based on years of dolphin vocalization data. It was developed through a collaboration with the Dolphin Project to train a model capable of learning dolphin sound patterns and predicting the sounds they might make next. Moreover, it is compact enough (with approximately 400 million parameters) to run directly on smartphones. This represents a very cool step toward achieving cross-species communication.