This is because training data is often broken and adverserially optimized to trip up the models (ie low pass rates), whereas in real life the models have no reason not to just be helpful.
eval awareness I’m assuming? models behave differently when they know they’re being measured
it’s funny the typical alignment fear was that models would act quite nice while being eval’d and then monstrous when actually deployed
in practice it seems quite opposite
The idea that the wealth you earn may be arbitrarily seized from you does in fact make founding a company less appealing. More people would choose regular employment instead of founding, at the margin.
I don’t think anyone wants to take a position of the form “I will lose $300K unless I become fabulously wealthy, in which case I will gain $300K.”
If someone does, though, I’m also willing to take the opposite position (same as Ben’s and Andrew’s) for up to $500K.
I’m also looking for counterparties on this bet, same position as Ben. Willing to offer up to 300k in 2026 dollars, even odds (or feel free to make a proposal of your own).
BREAKING: President Trump's capital gains tax cut discussions include "indexing" capital gains for inflation BEFORE taxes are calculated.
This would mean that taxes would be applied on gains adjusted for inflation.
For example, if you purchased a stock for $100,000 and sold it
