Essay.05-WeinanQian-2100017831
Keywords: Large language model, human conversation, intentionality
Abstract: While Large Language Models (LLMs) demonstrate remarkable proficiency in generating fluent natural language sentences replete with nuanced meanings, an intriguing question arises: are they truly engaging in conversations with humans, or are they merely executing a sophisticated form of imitation? We contend that authentic human communication is fundamentally rooted in human intention, a quality unique to our species. This is evidenced by human capacity to discern and employ irony and humblebrag, which are two intricate yet prevalent linguistic phenomena, within their interactions. To explore this further, we conducted a series of experiments utilizing GPT-3.5, ultimately revealing that LLM lack the capability to comprehend irony and humblebrag. This substantiates our assertion that they function as dispassionate mimickers, generating responses based solely on their pre-existing knowledge.
Submission Number: 109
Loading