Amyotrophic lateral sclerosis (ALS) causes significant motor and speech impairments, heightening the need for alternative communication methods. P300 speller brain-computer interfaces (BCIs) have shown potential in enabling non-muscular communication but often lack speed and efficiency. ChatBCI-4-ALS is introduced as the first intent-based BCI communication system specifically for ALS patients. This system utilizes large language models and a dynamic flash algorithm to improve typing speed and facilitate efficient communication of user intent, moving beyond simple word matches. New semantic-based performance metrics are also proposed to assess the effectiveness of this communication approach. Online experiments demonstrate that ChatBCI-4-ALS achieves an impressive average spelling speed of 23.87 characters per minute, with a peak of 42.16 characters per minute, and a maximum information transfer rate of 128.85 bits per minute, representing a significant advancement in P300 BCI-based communication systems.