8% compared to the recommended upperbound value. We are so doomed 🤐

Currently on Upperbound (3) which I need to grind more relics for 😌

Frontier AI is until some upperbound border only bound by the user interacting with it.

Assuming it's an upperbound tower, the damage done to it exceeds the known DI, which is nuts!

Cool, awesome, congratulations! Just not very relevant to my point, which is about the relative cost of antennas vs thermal management hardware on cost/kg and cost/kW bases, specifically as it applies to comparing the cost of Starlink v2 mini to the expected cost upperbound of AI1.

I think if you did the math yourself you wouldn't be able to justify getting all the way to even parity. Maybe 45% women, 55% men as an upperbound including incorporating uncompensated domestic labor by its near substitute price.

Great energy at #UpperBound today! It was a packed house. Alberta is positioning itself as a North American leader in AI data centre infrastructure, and rooms like this are why. Great conversations with the folks at Beacon and Cologix. Thank you for having me!

upperbound

이것도 있고 그냥 교육용으로 지식 다시 돌아보려고 한것도 있어뵈는데 걍 싸가지가 좀 이상한거 같다.. 어제 그 cat is all you need이 upperbound 쯤 되는 정도의 그런 수준 같다

{-# LANGUAGE ScopedTypeVariables #-} import qualified Data.Set as S import Data.Set ((\\), powerSet, Set, cartesianProduct) import qualified Data.Array as A type Geo a = Set a -> Float memo :: forall a.(Bounded a, Enum a, Ord a) => Geo a -> Geo a memo v = let upperBound = 2 ^ (1 fromEnum (maxBound :: a)) - 1 basisToPosition = foldr (\x -> ( 2^fromEnum x)) 0 :: [a] -> Int arr = A.array (0, upperBound) [ (basisToPosition (S.toList b), v b) | b <- S.toList (powerSet pseudoscalarBasis) ] in \b -> arr A.! basisToPosition (S.toList b) (△) :: Ord a => Set a -> Set a -> Set a a △ b = S.union (a \\ b) (b \\ a) pseudoscalarBasis :: (Enum a, Bounded a, Ord a) => Set a pseudoscalarBasis = S.fromList [minBound .. maxBound] sign :: (Ord a) => Set a -> Set a -> Float sign u v = product $ liftA2 (\a b -> case compare a b of {LT -> 1; EQ -> 0; GT -> -1}) (S.toList u) (S.toList v) dual :: (Enum a, Bounded a, Ord a) => Geo a -> Geo a dual f s = (f * sign s) (pseudoscalarBasis △ s) (/\) :: (Ord a, Enum a, Bounded a) => Geo a -> Geo a -> Geo a (/\) f g = memo $ \t -> sum [ sign tau (t △ tau) * f tau * g (t △ tau) | tau <- S.toList (powerSet pseudoscalarBasis) ] data Basis = X | Y | Z | W deriving (Enum, Bounded, Show, Eq, Ord) x, y, z, w :: Geo Basis x b = if b == S.singleton X then 1 else 0 y b = if b == S.singleton Y then 1 else 0 z b = if b == S.singleton Z then 1 else 0 w b = if b == S.singleton W then 1 else 0 instance Num (Geo a) where ( ) = liftA2 ( ) (*) = liftA2 (*) fromInteger = const . fromInteger negate = fmap negate abs = fmap abs signum = fmap signum instance Fractional (Geo a) where (/) = liftA2 (/) fromRational = const . fromRational main = do let system = (3*x y-5*w) /\ (y z-4*w) /\ (x-y 2*z-8*w) solution = dual system wValue = solution (S.singleton W) homogenizedSolution = solution / realToFrac wValue putStrLn ("x = " show (homogenizedSolution (S.singleton X))) putStrLn ("y = " show (homogenizedSolution (S.singleton Y))) putStrLn ("z = " show (homogenizedSolution (S.singleton Z)))

By training with noise, you can get even better results (upperbound)!

The upperbound is in the name: "Hyperdimension". In practice they're a parody of just how silly JRPGs are, and they casually break spacetime as the plot demands.

As a bonus, by solving a relaxed linear program, our method allows you to upperbound how far from optimal any tokeniser is. w/ @tpimentelms Paper: arxiv.org/pdf/2605.22821 Code: github.com/JanTempus/tokenis…

“We’ve got the best minds in AI right here in Edmonton. #UpperBound is one of the best few days in the country to see innovation in action. It’s really exciting.” We think so, too! Listen to Minister Solomon's full CBC Radio interview here 🎧 cbc.ca/listen/live-radio/1-1…

"It just makes sense that an event like Upperbound takes place in Albert" - @YourAlberta Minister for Technology & Innovation, Nate Glubbish. cc @AmiiThinks

Excited to be attending UpperBound this week in Edmonton - the heartland of RL - hosted by @AmiiThinks. Looking forward to talking tomorrow all things embodied AI - including how we’re scaling RL in the real world.

No. I use size_t because it is supposed to be able to represent size of anything that fits in memory. Otherwise you should make a good guess of what variable/field upperbound is and use appropriately sized integer type.

Eの解法、かなりの苦し紛れ L[i]=k,R[i]=kなるものについてSL[k],SR[k]に区間のsizeを昇順に持った2次元配列SL,SRをもつ 各クエリではSL[s]とSR[t]をupperboundで参照して、pl,prとする sz=t-s 1としたとき、 max(pl,pr)<szでpl pr>=szならよい max(pl,pr)=szで、SL,SRの2番手と足してsz以上ならよい

This is why there is ab upperbound on what people are willing to pay and what providers can charge and I dont think many get this. When it becomes $40 cheaper to hire a human no one ia going to pay for AI. This is the elephant in the room.