How you/ChatGPT can learn 'indirectly'

economictimes.indiatimes.com:

OpenAI, the company behind the AI wise guy. Since the generative robot processes from the mountains of data it is 'fed', what goes into its feed is super-important.

In the ChatGPT diet, books, especially fiction, play a big role in making the machine 'sound' human. But what happens when two novelists find ChatGPT giving gyan about their books that they have not given their permission for as part of the AI's ketogenic diet? Correct, they sue.

While Mona Awad and Paul Tremblay believe that their books, despite being copyrighted, were unlawfully 'ingested' and generated 'very accurate summaries' of their works, ChatGPT may still not be guilty of filching their forbidden-to-machines fruits of labour. The AI also 'learns' from internet chatter, which could very well be populated by summaries, discussions and secondary material of the books in question.

After all, there are smart alecs among us who have discussed War and Peace quite credibly and creditably without reading the tome by simply reading gists, watching a movie adaption, or generally listening in on those who've 'taken the Tolstoy test'. So, while reliability of information of the secondary and tertiary kind can be in doubt, one can't also be sure that the chatter one uses to wise-up is dodgy.