Ir para o conteúdo

Evangelos Kazakos

@ekazakos.bsky.social did:plc:u73ld57pmwsta45y45uitdfi

Postdoctoral researcher @ CIIRC, CTU, Prague working in vision & language. Also robotics noob. PhD from University of Bristol. Ex. Samsung Research (SAIC-C). I love coffee and plants. And socks. 𖤓꩜𓆉𓃰𖤐

No Bluesky desde

Seguidores
1.046
Seguindo
762
Posts
1.729
Dias nos rankings
0

Resumo

Evangelos Kazakos (@ekazakos.bsky.social) é um usuário do Bluesky desde 2 setembro 2024. Tem 1.046 seguidores, segue 762 contas e publicou 1.729 posts.

Aparições nos rankings

Os melhores dias em que posts desta conta entraram na lista diária do Debya.

Esta conta ainda não apareceu em um ranking diário.

Posts curtidos recentemente

Curtidas públicas da rede Bluesky.

  • Mark Riedl@markriedl.bsky.social

    Aside: In 2018 @matthewguz.bsky.social and I were playing around with direct manipulation of neural nets without backprop arxiv.org/abs/1802.03605. Didn't work super well, but the models were crude by today's standards. Fun to see things we were trying in 2018 might actually work quite well today.

    Ver post
  • Mark Riedl@markriedl.bsky.social

    I'll also note that this appears to largely be one-shotting the weight edits. When combined with RL or evolutionary search loops, this should mean that specialized neural networks can be built without back propagation on data. This is where things will get really interesting.

    Ver post
  • Mark Riedl@markriedl.bsky.social

    This is fascinating. Very large models can read the weights of small models and intuit what those models were trained to do, and make direct weight changes to improve them. rikiyat.github.io/blog/weight-...

    Ver post
  • Chanda Prescod-Weinstein 🌌@chanda.blacksky.app

    I voted! With my Lula glasses in tow 🤩

    Ver post
  • Ethan Mollick@emollick.bsky.social

    If this holds up, it is big: "We asked GPT-6 Astra to edit Qwen3.5-0.8B weights to improve math. It directly changed only 0.016% of weights, fixing 0.8B's reasoning loops & improving math up to 6x and even instruction following 3x, without gradients or training." rikiyat.github.io/blog/weight-...

    Ver post

Entre para ver mais

Contas que bloqueia

No Bluesky, bloqueios são registros públicos.