4 ms·

I made it through the first 28 minutes of the video. Nothing of substance. Perhaps there's some grand point he made at the end, but there's no way to represen

• vages 2y ago
  At around 32 minutes, he explained that a sufficient implementation of negative infinity would be the smallest possible value representable by the data type (e.g. 32 bit integers).
  • neuroelectron 2y ago
    i.e. a binary flag or two. as is ISO
  • linuxftw 2y ago
    That's just moving the problem elsewhere. Since the smallest possible value is a possible value, then it may be present in the array, or it may not actually be present in the array. The caller would have to determine which case is which by testing for the length of the array, which if done, negates the entire utility of returning the smallest possible value.
• MITSardine 2y ago
  He sure takes his time! I stopped (for now) at 32min when he skipped the least trivial step to showing his invariant holds. It's fairly trivial too, some element leaves B and is max'd against x, max of those two things is conserved. I don't know why he skipped it. It's a little mean spirited, what with the "you'll get replaced by LLMs if you can't solve it" comment. Anyways, it's often like this with these experts, some are so far in their own worlds, they no longer have any notion of what other people understand, so they either over or under shoot the complexity of their talks drastically. Complete with a trivial mistake where max is replaced by min in the numerical application, à la Grothendieck's prime 57. At least this chap wanted people to understand his talk.