guess who's back? back again
this time it's not cyber, not algorithmic, it's was a data/ml hackathon on finance
the annual winning streak since 2022 has been secured earlier than expected
🇫🇷 FLASH | "C’est quelque chose auquel je ne m’attendais pas, je ne pensais pas que cela puisse arriver, surtout à mon âge". Gisèle Pélicot annonce avoir retrouvé l’amour avec "Jean-Loup", ce "jeune homme de 73 ans".
@jlsquare_ Yeah, I honestly don't actually care tbh lol
Like, I would be more upset if they were banning cigarettes. I mean, tbf, it's all bad for you, it really is, and in a perfect world, no one is using any nicotine at all.
But also man are cigs nice sometimes
@NewAgeRetroNerd i dont know anyone in france who uses Zyns. i think they just wanted to prevent another addiction source, especially since theyre already trying to cut down on cigarettes. they did the same thing with disposable vapes
@scheminglunatic haha hiiiii alcuin uwu soo umm ehe owo erm if ur rly srs i have 2 other projects, i even found someone who likes backend stuff for that other one, but if you're more interested in model architecture, not just using pretrained stuff, i need the help uwu
@secemp9 e_1 duuuh!! sigmoid go squish but l_1 still smaller than l_2 so f_2 plops down below one half and poor lil e_2 gets zonked again mode collapse but make it wiggly
there is a 2-expert MoE layer, E_1 and E_2 ex-ante symmetric, gate logits z_i = W_g h_i, top-1 routing r_i = argmax_k softmax(z_i)k, each token's routing is its private vote, f_k = (1/N) Σ_i 1[r_i = k], rule is f_2 > 0.5 every token gets y_i = E{r_i}(h_i), else E_2 tokens zeroed and E_1 tokens pass, E[ℓ_i^{E_1}] ≤ E[ℓ_i^{E_2}] for every f_2 so E_1 weakly dominates, empirical risk alone drives f_1 → 1, Switch Transformer's L_aux = α N Σ_k f_k P_k with P_k = (1/N) Σ_i p_i^k pulls against it, f_1 → 1 also satisfies survival because majority routes to E_1 so "only E_1 survives" = all survivors survive, with L_aux the load-balanced optimum sits at f_2 = 0.5 + ε, knife-edge because rule is strict >, one defection to E_1 zeros every E_2 token, soften 1[f_2 > 0.5] to σ(β(f_2 − 0.5)), ∇_θ L nonzero at boundary, gate flows to interior f_2* = 1/(1 + exp(−β(L_1 − L_2))) from ∂E[ℓ]/∂f_2 = 0, which expert do you route to
Keep noticing the most cracked people on this app all have anime pfps and it got me thinking…
The next wave of top founders grew up on anime. They saw themselves in Naruto, Izuku and Edward Elric. Basically, characters who get knocked down constantly but never stop, and who fight for something bigger than themselves.
Relentless + genuinely good-hearted is a rare founder archetype. And there's a whole generation of them coming up.