Well I pressed "Play" and began to wonder if the ChatGPT voice and dialogue were faked. My mind numbed as she snappily fielded all questions with direct answers. As she smoothly broke into a sponsored advert at the 7-minute mark, my suspicions of fakery were confirmed.
Does TTS or voice synthesis take in audible breaths? Gemini and Bing don't seem to.
Good joke mate!
I expected to hear ChatGPT confabulating and becoming lost in misinformation, or shutting down the thread with "But I'm only an LLM, I can't ..."
And the advert seemed quite specific or scripted. How do you elicit such specific ad copy from an LLM that has a mind of its own?
And a 25-minute video is a lot of tokens. I don't understand how the context buffer works, but does it represent a single long session where ChatGPT retains the entire conversation? The smartphone clock seems to indicate parts were edited out?
He explains at the end that the ad was definitely scripted but not the rest of the conversation, since the point of the channel is to explore philosophy. He recently has been posting these AI conversations and they are super interesting to me.
As far as context, I think ChatGPT (it's probably o3 I would guess) has a huge context window that can include large documents or codebases. Most top models have a huge window now, and while it's probably more expensive to use all those tokens I'm sure it's more of an investment for youtube channels.