Chatbots Process Information Like Mushrooms Grow, Not Through Human-Like Thought
A recent explanation clarifies that chatbots, specifically large language models (LLMs), do not possess human-like minds or understanding. Instead, their operation is likened to the growth of a mushroom's mycelial network. When a chatbot is 'trained,' it is fed an immense volume of text data—books, websites, news articles, code, and online discussions. As words appear together, the 'thread' connecting them in the AI's internal network thickens. For instance, 'bread' and 'butter' would have a thick connection, while 'bread' and 'engine' would have a thin one. This process builds a complex web that stores statistical relationships between words, grammar, facts, and common explanatory structures. The 2020 version of the AI behind ChatGPT, for example, was primarily trained on a vast copy of the open internet, along with shared online pages, book collections, and English Wikipedia. This means the AI's 'knowledge' is strongest where the internet discusses topics most frequently, not necessarily where it is most...