@MIT_CSAIL There’s a day-of-week effect in your data due to using raw counts & a limited number of years. To correct for this, see Gelman’s blog entry discussing the cover for Bayesian Data Analysis with seasonal, trend, day-of-week, & day-of-year anomaly effects: https://t.co/1pjfFDDYxt
@skglearning This is just a confusingly explained “fun preview of things to come.” Bookmark & come back. Basically, you can add functions and multiply them by scalars, so they follow the rules of vectors. And d/dt acts like a matrix (linear operator) on these.
@DmitryRybin1 Oh I see; I didn’t realize how restrictive your algorithm class was. I saw you required homogeneous but missed that you considered this “in each matrix” separately. Interesting to think about how to generalize to less restrictive settings.
@DmitryRybin1 It’s intuitively attractive, but how to prove? E.g. to get a monomial abcd to show up you can do (a+c)(b+d) and then square it; certainly tons of unwanted byproducts and it doesn’t seem at all useful but how to prove there’s no clever cancellation with a bunch of such?
@JohnnyFord99@bryan_johnson@realPatrickJr Exactly: Beware correlation vs causation here. Health conditions associated with increased all-cause mortality may make you need more sleep. For this reason the AASM recommends 7+ hours not 7-9 hours: https://t.co/8BMqOdAohi
@ESYudkowsky@YitziLitt If AI causes real GDP growth (or feel free to reject this) as well as reduced demand for labor, these would suggest opposite remedies from this perspective? Or is this more about using the methods associated to these concepts to address labor issues (literal NGDPLT not required)?
@ESYudkowsky Of note from the earlier paper https://t.co/wM495UAIOc
“Surprisingly, AlphaGo Zero outperformed AlphaGo Lee after just 36 hours; […] After 72 hours […] against the exact version of AlphaGo Lee that defeated Lee Sedol […] AlphaGo Zero defeated AlphaGo Lee by 100 games to 0”
@robertskmiles Nanda et al’s work is beautiful, and your broader point about interpretability being difficult makes sense, but:
This is the cleanest way to embed “clock arithmetic” in linear algebra, doing rotations using sines and cosines. It just looks messy without matrices.