Discussion
It just occurred to me that there's an obvious reason why using a Trained #MOLE as the engine for web search portals, makes their results much less useful than they used to be.
A MOLE is a statistical engine that searches for the centre of any given bell curve. So when we ask one for web pages that match a set of keywords, it returns the most obvious and unexceptional results. The stuff we probably already know about, and could easily find without it.
(1/?)
@strypey an experienced game designer designed a game with an LLM and saw that it just felt average, and lacked character. It was then that he understood that LLMs are averaging algorithms
Whereas the whole value of traditional web search was in being able to surf down bell curves, looking for the surprising and exceptional stuff that lurks well outside normal distribution. A #MOLE isn't just bad at this, it does exactly the opposite. It's precisely the wrong tool for the job.
(2/?)
All of which leads me down the side of a familiar bell curve, to a very unfamiliar conclusion. Maybe the way to fix search is regulation that structurally separates the librarian's task of indexing the web, from the business of running search portals? So any portal can search any index.
This could take a number of forms. The idea just occurred to me, so I don't have any specific ideas about how it might work. But I'm interested in whether people think the idea could have legs.
(3/3)
Install bonfire.cafe
Get the full app experience
