@IsaacKing314@max_spero_@JohnHolbein1 Depends on the use case, no? Low FP rate would be relevant for teachers or similar that don't want to mistakenly punish students, low FN would be relevant if rejection is cheap, e.g. social media feeds.
@OwainEvans_UK So true! A human would never believe implausible claims, even when exposed to them a lot. Can you imagine people believing the earth is flat? Ridiculous
@eleusinianatlas@lumpenspace@Naozymandias Bad example because exactly this benchmark is basically solved. ARC-1 is saturated, arc-2 almost if using flagship models, arc-3 still stands
@lumpenspace@grok > Losing an extra $1k after already down $1M (now poorer) hurts more than losing that same $1k from $1M+ (when richer).
? Isn't that exactly what I was saying?
@ShrekIsGrek@lumpenspace Yeah I guess. It is fun though. And I *am* actually trying to provide a mathematical argument and want to understand if they actually believe what they are saying.
@lumpenspace I... don't think this is going the way you think it is going? Again, do you not understand this? Maybe google or ask an AI of your choice? Apparently you are unable to understand what I am saying or how math works.
@lumpenspace Okay, sure. In that case the second one is obviously worse. Especially if I only have 1M1k dollars. It makes sense if you look at it sequentially. After loosing 1M you are far worse off, so the 1k extra becomes more valuable to you.
@lumpenspace If I have a NW of lets say 10M
log((10M+50k)/10M)=log(201/200)~0.005
log((10M-1M)/10M)=log(9/10)~-0.105
So if I had a NW of 10M dollars, it would be 21 times worse and not 20
If I had a NW of 2M it is 28x worse, at 1.1M 54x worse. But it is *always* worse than 20x
@lumpenspace It is for *getting* the money, yes. But not for *loosing* it. Understandable mistake though. Would be easy to fix by thinking before posting.
@NeverSayClever@BenShindel@SkyeSharkie You are not eligible for MAID in canada if you have only psychological conditions, it *requires by law* a physical condition.
@FakePsyho True. 100% is almost impossible, and basically unachievable in Kaggle compute limits. I'd argue the previous grand prizes were also unachievable, which makes marketing with the prize pool a bit dishonest. Still a nice benchmark though.