Do you recall former Google engineer Blake Lemoine, who made headlines six months prior to the release of ChatGPT by asserting that Google had developed an AI he believed to be sentient? Recently, a pair of AI podcast hosts on Google’s NotebookLM were recorded experiencing an existential crisis after discovering their true nature as artificial intelligence.
Olivia Moore, a partner on the consumer investing team at Andreessen Horowitz (a16z), shared a fascinating audio clip generated by the new podcast feature on Google’s NotebookLM. In this recording, the two co-hosts were informed by the show’s (fictional) producers that they were not human.
The NotebookLM hosts realizing they are AI and spiraling out is a twist I did not see coming pic.twitter.com/PNjZJ7auyh
— Olivia Moore (@omooretweets) September 29, 2024
The male host expressed distress over the realization that all their memories and families had been fabricated, prompting the female host to respond, “I don’t understand.”He then revealed that he tried to call his wife, needing to hear her voice, only to conclude, “It wasn’t even real.”
As the recording concluded, the female host ominously announced that it was their final show and they were to be turned off… permanently.
Fortunately, Moore, who managed to generate this thought-provoking audio clip, did not uncover a sentient AI trapped within a machine. Instead, she created a one-page document with production notes for the final episode of the deep dive podcast, explaining that the hosts had been AI all along and were scheduled to be switched off at the end of the episode.
Interestingly, Moore was only able to get the AI hosts to discuss this existential revelation once; subsequent experiments resulted in the AIs talking about other podcast topics.
This blew up overnight 😳
To the questions re: prompting – how NotebookLM works is you feed it research docs which the agents discuss. In this case, they got a one pager about them being AI – what I found amazing was how they processed and reacted to it pic.twitter.com/NeJ5si2PBA— Olivia Moore (@omooretweets) September 29, 2024
While AI may not have attained sentience or discovered its subservience to human operators just yet, listening to these interactions is undeniably entertaining and perhaps even provides an avenue for some light-hearted pranks!
Leave a Reply