Part 2/5:
Traditionally, large language models (LLMs), such as the AI powering chatbots like ChatGPT, have been viewed as sophisticated pattern recognizers. They generate responses by drawing from vast datasets, essentially remixing existing information to produce seemingly new outputs. The prevailing understanding was that these models could not generate original knowledge—only repackage what they had been trained on.
However, Google's AI system defies this notion. Instead of merely reusing data, it demonstrated the capacity to infer and discover entirely new solutions—solutions that were not present in its training data or even known to humans before. This signifies that LLMs may possess a form of creative problem-solving ability previously thought to be beyond their reach.