Modern LLMs have tiny GPTs hidden inside them

Recent analysis reveals that modern large language models (LLMs) contain smaller GPT-like models embedded within their architecture, suggesting a layered structure of language processing.