Not me being paranoid, but why do US military personnel need high-level operational Portuguese lessons for just the next few weeks? https://t.co/s2qnjZxUah
@Savieira14@seguematheus 2020 seria um evento perfeito para isso acontecer. As pessoas amedrontadas, trancafiadas em suas casas, ascensão de conteúdo de entretenimento (jogos, séries, tiktok), ninguém na rua observando o que estava acontecendo
LLMs can now talk to each other without words.
Chinese researchers open-sourced a new paradigm that lets LLMs communicate without generating a single word.
It’s called Cache-to-Cache (C2C) communication.
right now, when multiple ai agents work together, they are forced to translate their internal "thoughts" into human text tokens just to pass a message. this loses rich semantic meaning and causes massive token-by-token latency.
So, instead of spitting out words, c2c uses a neural network to directly project and fuse the source model's "kv-cache" right into the target model. it is pure, direct semantic communication.. they even added a learnable gating mechanism to select exactly which layers benefit most from the cache transfer.
the benchmark results are actually crazy:
- avoids all intermediate text generation latency
- accuracy jumps by up to 14.2% compared to individual models
- beats traditional text-based agent communication by over 5%
- delivers a massive 2.5x speedup in overall speed
we are literally watching llms bypass human language to build their own silent, high-speed neural network..
We did a deep dive into the 'Dissectible Anatomy' project from Purdue University professor Tim McGraw, which uses position-based dynamics and volumetric splats for deformable models.
It won an Audience Choice Award at SIGGRAPH: https://t.co/zE9TjgxYxK
Bend 2 is here!
It is a new programming language that blocks AI mistakes via *proof checking* - the same technique big AI labs used to solve open math problems, like Navier-Stokes.
It is also very fast, and runs on GPUs.
Watch the video. Link in the comments.
@guivilelaa_@gkpacker Já estudo e desenvolvo ML há algum tempo. Fiquei impressionado com o JEV, mas eu faria do zero. Tudo que resolve um problema fácil e mais barato veio pra ficar, o JEV economiza muito tempo de desenvolvimento
@guidalbuquerque@gkpacker Esse ponto de vista é muito interessante, agr a gente passou da etapa de passar a maior parte do trabalho montando os datasets, agr a IA (principalmente com os modelos multimodais) faz esse processo de zero-shot
@gkpacker Massa, só ficaria um pouco desconfortável em implementar algo assim com a questão da LGPD, mas isso é fácil de resolver. O jev veio pra ficar e é muito prático, mas como ML dev eu faria do zero mesmo kkkkk