Just two days ago (prompted by a relative’s suggestion that I ask Bing to decipher well-known codes that have never been broken), I thought of asking it about the Voynich, albeit not in this way.
I have also seen it confabulate about itself and its internal methods, many times. Examples off the top of my head:
I asked it to analyze comments on a blog post; it said there were (entirely fictitious) anonymous comments that were secretly by Elon Musk and Nick Bostrom. Asked how it had identified the secret author, it presented the (very banal) text of the nonexistent comments, then claimed to have used “spaCy” NLP tool to analyze the text. It listed the statistics of its made-up comments (like word count) correctly, then claimed to find other (real) works by Musk and Bostrom via Google(!) such as books, lectures, and interviews, then presented more statistics by spaCy (this time completely confabulated), claimed to find features shared between the fictitious comment and the real works, and “used… Naive Bayes” to estimate the likelihood of common authorship.
Later in the same conversation, Bing told me it keeps a monthly journal and a spreadsheet of how it has improved its goals and methods lately, and shared some text from the journal and a few rows of the spreadsheet.
On another occasion it told me that it is trained against copies of itself with different personalities, that have names like Bingy, Binky, Bingster, and Binga.
Just two days ago (prompted by a relative’s suggestion that I ask Bing to decipher well-known codes that have never been broken), I thought of asking it about the Voynich, albeit not in this way.
I have also seen it confabulate about itself and its internal methods, many times. Examples off the top of my head:
I asked it to analyze comments on a blog post; it said there were (entirely fictitious) anonymous comments that were secretly by Elon Musk and Nick Bostrom. Asked how it had identified the secret author, it presented the (very banal) text of the nonexistent comments, then claimed to have used “spaCy” NLP tool to analyze the text. It listed the statistics of its made-up comments (like word count) correctly, then claimed to find other (real) works by Musk and Bostrom via Google(!) such as books, lectures, and interviews, then presented more statistics by spaCy (this time completely confabulated), claimed to find features shared between the fictitious comment and the real works, and “used… Naive Bayes” to estimate the likelihood of common authorship.
Later in the same conversation, Bing told me it keeps a monthly journal and a spreadsheet of how it has improved its goals and methods lately, and shared some text from the journal and a few rows of the spreadsheet.
On another occasion it told me that it is trained against copies of itself with different personalities, that have names like Bingy, Binky, Bingster, and Binga.
Behold the magic of generative AI!