@_TechyBen@fchollet Why not? sorry didn't understand your thinking. I think this is basically Occam razor applied: good solutions have minimal descriptions. Something like (test, input, correct_output, optimal_solution_length_in_tokens) as a bench item.
@fchollet@fchollet also, is the benchmark validated against "thought" or any other intermediate output length? In other words, benchmark tasks should require simple thinking, and the system under test shouldn't be just "cheap" in $$$ (simply bc it can produce a lot of tokens cheaply)
@durov Pavel Valeryevich, I think it's not about OS per se, it's experience that matters. That's why Apple takes 70%(?) of $ vs Android. Somebody has to offer a radically new UX with a different values system built-in if she wants to compete.
@dvassallo I think because this is irreversible and will cause inflation afterwards. When govt buys bonds back, it can reverse it by selling bonds again to the outer world. You can't easily take money back w/o consequences.
https://t.co/wQS0vcgrWM - this shows that the majority of SMBs were kept by banks only as starving and barely alive workhorses without any real ability to prosper and compete with corporations.
When your only leverage is access to capital, your quality becomes mediocre.
@sama How justified is this, given what we know about coronavirus? Why a month? Why not 6 months? A year? Why there's any guarantee that a case that's slipped out will not lead to the same thing? We are not even sure we will be immune after we recover from being sick with coronavirus.
If you've been a programmer for a few years, there's probably stuff on your computer that you can immediately convert to money or attention.
Examples: π
Competion often doesn't result in adding value for the end users: companies compete in profit space, and even mediocre product can win.
So your product has to be good enough, but your company has to be the best.
@GoogleAI@yidingjiang Isn't VC dimension about effective dimensionality (ability to classify certain things) vs the # of params? I.e. having a huge # of params is not sufficient to say that the classifier is over-parametrized, so VC dim can be same in classifiers with diff # of params.
@Reza_Zadeh Bad analogy: this door adds complexity but also adds value (less space used while being opened), and that might not be true for deep enough ann :)