AI models learn, not replicate: OpenAI cites 'fair use' cover

IN COPYRIGHT CASE FILED BY NEWS AGENCY, ANI AMERICAN FIRM CONTENDS LARGE LANGUAGE MODELS DON'T MEMORISE DATA

- Shruti Kakkar

NEW DELHI: Large language models, the programmes that underpin Artificial Intelligence chatbots, Vrinda Tulsianand instead generate new responses "to prevent regurgitation", ChatGPT-founder OpenAI told the Delhi high court on Wednesday, as it defended its use of news agency ANI's content to train its chatbot.

ChatGPT is merely an interface between the user and content, the US firm said, explaining that LLMs do not memorise data, but are trained to predict the relationships between words using broad statistical patterns.