AGI has a heart and, we can prove it?
Today's frontier LLMs may be trained on a few too many 19th century romance novels?? Research shows that they can be moved by big, heartful talk. 💜
The more grandiloquent the prose, the more likely something is to be cited in AI-generated answers. And we're not talking about an indexing quirk. The answers show evidence of momentum, a kind of emotional build-up that comes with the reading experience - like a human.
For example, when you're reading, you're not building up a bank of tags and tallies in your brain. Instead, you have little bits of emotion that hitch a ride with you and you explore the story. Apparently, there's evidence that the models are doing something similar.
Is there such a thing as AI psychology?
Made me wonder if there are any formal fields of AI psychology. It's a bit loose. The closest we found was "computational psychology," which at least gives us a language for talking about this stuff without pretending we fully understand the mechanism.
How a human does it in psychology can be called "transportation." When readers are transported, they experience an emotional ride. They're in it. And this is not about cheesy love letters or empty hype writing. Some of the finest prose creates that same deep mood and psychological after-glow you might get from the end of a great novel.
What this means for GEO teams
Consider measuring by "momentum," whatever you might define it to be. Voice and rhythm might be a magic potion here ecause frontier models appear to experience genuine momentum, something like to a heart responding to prose. It might just be high-dimensional patterning but the practical output looks the same: vivid, even bombastic(?) writing gets pulled forward.
