SOMEONE VIBE CODED POKEMON GO BUT FOR CATS YOU MEET IN REAL LIFE
see a cat, open the camera, snap it, and it gets added to your collection as a little creature with a name, rarity, level, and its own stats page
> checks the photo actually has a real cat in it, so you cant just screenshot one or use an image
> every cat you catch gets a rarity and its own collectible card
> a world map shows cats other players have found near you
> has retro cartoon art, cream colors and thick outlines, built to feel like a game
he wanted "snapping" a cat to feel playful and fast, not like just saving a photo
so most of the work went into the detection and making the catch feel like a real game
i need a dog one asap
The largest models are ~5TB. Your brain has ~86 billion neurons so 86gb ~= 10.75 GB. 5TB / 10.75 GB = ~465 There is only 465 people's worth of knowledge on the planet.
@arestlessrest to be clear, I don't think the second example is quite sound. but it's a heuristic using the same principle of "transform the problem so it's easier to engage salience / likelihood intuitions on"
grokking collider bias is my favorite still-underrated epistemic mode:
can get a huge amount of info on which filter bubble you (or another) are in by carefully examining suspicious correlations you observe
this is kinda the ONLY way to bootstrap yourself out of a filter bubble
@TomWright165389 bayes thm is powerful because it gives you a new epistemic mode. a new way to turn evidence into belief that previously you "had to" throw away.
it may seem obvious now that it's in the water supply, but 20 years ago when yud started ranting, it was surprisingly rare to apply
@arestlessrest yes Jeffrey-Bolker approach is related to "exponential discounting" behavior. although strictly speaking Jeffrey-Bolker is just reparameter and shouldn't affect relative value. True discounting for counterfactuals is about change of reference frame. a "lorentz boost", if you will