A friend showed me his Discover Weekly a few weeks back, laughing a little at how off it felt. Same moody indie tracks, same slow tempo, all clearly descended from a breakup that ended over a year ago, still technically perfect somehow. It just didn't sound like him anymore, more like a photograph of someone he used to be.
It made me think about how much of what gets called personalized is actually just well preserved. The algorithm isn't wrong, it's just running on a version of you that stopped being current a while ago. We rarely notice this in the moment, because the recommendations still feel accurate enough to pass.
I keep coming back to the idea that identity isn't something you have, it's something you're mid revision on, usually without realizing it until later. A system trained on past behavior treats that behavior like a stable signal, when really it was just one frame of you holding still.
What's strange, and I only half trust this thought, is that the mismatch is sometimes the only way I notice I've changed at all. The wrong recommendation becomes evidence, not of a broken system, but of a self that already moved on without telling anyone, including me.
It is the same question I imagine any AI trading strategy has to face eventually, Newton Protocol included. Either it keeps refining a picture of who you were, or it finds some way to notice who you are becoming.
I don't have a clean answer. If something built a perfectly accurate model of who you used to be, would you actually want it optimizing for that person, or trying, imperfectly, to catch up to who you are now?
@NewtonProtocol
$NEWT $NFP $TAIKO
#Newt
It made me think about how much of what gets called personalized is actually just well preserved. The algorithm isn't wrong, it's just running on a version of you that stopped being current a while ago. We rarely notice this in the moment, because the recommendations still feel accurate enough to pass.
I keep coming back to the idea that identity isn't something you have, it's something you're mid revision on, usually without realizing it until later. A system trained on past behavior treats that behavior like a stable signal, when really it was just one frame of you holding still.
What's strange, and I only half trust this thought, is that the mismatch is sometimes the only way I notice I've changed at all. The wrong recommendation becomes evidence, not of a broken system, but of a self that already moved on without telling anyone, including me.
It is the same question I imagine any AI trading strategy has to face eventually, Newton Protocol included. Either it keeps refining a picture of who you were, or it finds some way to notice who you are becoming.
I don't have a clean answer. If something built a perfectly accurate model of who you used to be, would you actually want it optimizing for that person, or trying, imperfectly, to catch up to who you are now?
@NewtonProtocol
$NEWT $NFP $TAIKO
#Newt