Skip to main navigation Skip to search Skip to main content

Large Language Models in AI and the Linguistic Brain

    Activity: Talk or presentationInvited talk

    Description

    The rise of ChatGPT or other generative large language models (LLMs) represents a significant advance in artificial intelligence while providing new insights into how language acquisition, representation, and production is conducted by the human brain. LLMs rely on computing distributional statistics from large-scale text data, yielding results comparable to human linguistic performance. The power of LLMs can help to resolve age-old debates on the role of statistics in human language learning, while forcing consideration of what aspects of the human brain are uniquely suited for acquiring and representing natural languages. The human brain needs only a tiny fraction of the data of LLMs (e.g., 400-billion tokens of ChatGPT). Its efficiency might arise from its ability to effectively integrate multisensory and multimodal information from the environment, using neural architecture and mechanisms for sensorimotor, visual, auditory, and higher-level conceptual and linguistic processing in both the left and right hemispheres. By comparison, LLMs rely on many (deep) layers of multiple processing units, simulating the brain’s division of labor as well as interactions and collaborations among units. This symposium brings together experts from neuroscience, language science, and machine learning to examine the extent these two types of architectures (models and brains) can inform each other in language processing, representation, and technological applications.
    Period16 Feb 2024
    Event titleAmerican Association for the Advancement of Science Annual Meeting: Toward Science without Walls
    Event typeConference
    LocationDenver, United StatesShow on map
    Degree of RecognitionInternational