If we start at a set “0” time, and only the future is infinite, then we break some of the mappings discussed in the “Agent [or any other value-location] Neutrality” mappings. Intuitively, some of the “same distributions” would, if both bounded on the right side, be different again. This does not hold for e.g. w5 and w6 (since there’s a 1-to-1 mapping between n and 2n for all n), but it does hold for w3 and w4 (since the first w3 agent to the left of the cutoff-line won’t have a corresponding w4 agent).
Diminishing realityfluid sure looks superficially similar to the weight-by-simplicity idea, which is unusual for being one of the least-tempting not-quite-solutions-to-infinities in the entire post. This makes me update (weakly!) away from “diminishing realityfluid”.
What makes Heaven+Speck intuitively better-than Hell+Lollipop? I think the answer might end up as something like “realityfluid”, where (normalized or not) there’s more of it on the Heaven/Hell than on the Speck/Lollipop, even though both Heaven&Speck are infinite and so are both Hell&Lollipop.
This whole post and what Tamsin Leake has built off of it, have updated me away from “backchain learning math as you need it”. Maybe not all-the-way away, but a good bit away.
Thoughts while reading this, especially as they relate to realityfluid and diminishing-matteringness in the same vein as “weight by simplicity”:
If we start at a set “0” time, and only the future is infinite, then we break some of the mappings discussed in the “Agent [or any other value-location] Neutrality” mappings. Intuitively, some of the “same distributions” would, if both bounded on the right side, be different again. This does not hold for e.g. w5 and w6 (since there’s a 1-to-1 mapping between n and 2n for all n), but it does hold for w3 and w4 (since the first w3 agent to the left of the cutoff-line won’t have a corresponding w4 agent).
Diminishing realityfluid sure looks superficially similar to the weight-by-simplicity idea, which is unusual for being one of the least-tempting not-quite-solutions-to-infinities in the entire post. This makes me update (weakly!) away from “diminishing realityfluid”.
What makes Heaven+Speck intuitively better-than Hell+Lollipop? I think the answer might end up as something like “realityfluid”, where (normalized or not) there’s more of it on the Heaven/Hell than on the Speck/Lollipop, even though both Heaven&Speck are infinite and so are both Hell&Lollipop.
This whole post and what Tamsin Leake has built off of it, have updated me away from “backchain learning math as you need it”. Maybe not all-the-way away, but a good bit away.