Júlia Klee | Supersonic Labs

Co-founder, Head of Research and Model Architect at @supersonicai

Bringing powerful and capable AI to everyone.

[Júlia Klee | Supersonic Labs @klee00z]

em breve.

[João Davi | Supersonic Labs @oC3B_]

vem aí.

[Supersonic Labs @supersonicai]

there’s still something we haven’t talked about

for a reason

[João Davi | Supersonic Labs @oC3B_]

fazer pesquisa no BR é ter situações como essa aqui.

a gente vive no MODO HARD kkkkkk

[Júlia Klee | Supersonic Labs @klee00z]

@FERNAND77S @supersonicai Calma que o advogado dele tá recorrendo com recurso

[Júlia Klee | Supersonic Labs @klee00z]

@supersonicai

[Supersonic Labs @supersonicai]

yall simply aren't ready for the banger that we gonna to drop...

[Júlia Klee | Supersonic Labs @klee00z]

Hell yeah let's go baby

[Supersonic Labs @supersonicai]

Yo, btw, you will be able to run our model on webgpu.
just sayin.
🙃

[el hombre pulpo @coproduto]

Nos últimos 2 meses, eu trabalhei com um pessoal muito foda pra tentarmos modelar e mapear o destino do Brasil na revolução da IA se as coisas continuarem do jeito que estão.

Hoje estamos publicando nosso cenário atual. Ficarei muito feliz se lerem. Link nos comentários.

[Júlia Klee | Supersonic Labs @klee00z]

It’s kind of wild seeing a new open source “Jev alternative” pop up every day claiming better accuracy, when a lot of them seem to miss the entire point of Jev

Jev is not just an LLM fine-tuned to output JSON

If you fine-tune Qwen to emit a structured decision in JSON, you still have an autoregressive language model generating tokens sequentially

That’s fundamentally different from a model built around typed probabilistic decisions, where outputs are produced in parallel rather than generated token by token

[Júlia Klee | Supersonic Labs @klee00z]

Thank you, everyone

Não esperávamos que esse projeto iria crescer tão rápido em poucos dias, estamos fazendo o máximo para entregar um modelo de ótima qualidade a todos

We didn’t expect this project to grow so quickly in just a few days. We’re doing our best to deliver a high-quality model for everyone

Isso especialmente, para aqueles que dizem que o Brasil não tem capacidade. Nós temos, ela só não é usada da forma que deveria

[Supersonic Labs @supersonicai]

Uh oh... We weren't expecting this! :D

Thank you so much, everyone!! <3

Soon, we'll be making our model open source so everyone can run it on their own computers!

Our mission is to bring high-quality AI to YOUR machine, the one you ALREADY HAVE!

tmj

[Supersonic Labs @supersonicai]

Yo, btw, you will be able to run our model on webgpu.
just sayin.
🙃

[Júlia Klee | Supersonic Labs @klee00z]

@korenmiklos @supersonicai @JuliaLanguage Wow, some parts of our research are done in Julia. It’s one of the best programming languages

[João Davi | Supersonic Labs @oC3B_]

Vc percebe o machismo na sociedade quando eu to tendo mais follows que a @klee00z que literalmente ta LIDERANDO o Research da Supersonic Labs... foda.

[Júlia Klee | Supersonic Labs @klee00z]

só queria um macbook de 8TB de ram

[João Davi | Supersonic Labs @oC3B_]

fazer pesquisa no BR é ter situações como essa aqui.

a gente vive no MODO HARD kkkkkk

[Júlia Klee | Supersonic Labs @klee00z]

@lucas_montano @manodeyvin @RealGalego Porque não apoiar uma iniciativa nacional que está solando os gringos?

[Supersonic Labs @supersonicai]

You were saying something...?

yes, we classify responses 5x faster than Jev, using our own 144M-param model (Julia-1) on a 2021 i5 laptop while maintaining high accuracy.

So no, we’re not like other open-source models. And fuck no, we’re not just a game neural network.

[Júlia Klee | Supersonic Labs @klee00z]

@supersonicai @manodeyvin no video da RioAI, você disse que queria ver iniciativas brasileiras reais de IA, pois bem

[João Davi | Supersonic Labs @oC3B_]

tutorial de como deixar um gringo totalmente de xereca:

[Júlia Klee | Supersonic Labs @klee00z]

@velsinhax @supersonicai https://t.co/vpBVoa0Puc !!!

[Supersonic Labs @supersonicai]

You were saying something...?

yes, we classify responses 5x faster than Jev, using our own 144M-param model (Julia-1) on a 2021 i5 laptop while maintaining high accuracy.

So no, we’re not like other open-source models. And fuck no, we’re not just a game neural network.

[Supersonic Labs @supersonicai]

You were saying something...?

yes, we classify responses 5x faster than Jev, using our own 144M-param model (Julia-1) on a 2021 i5 laptop while maintaining high accuracy.

So no, we’re not like other open-source models. And fuck no, we’re not just a game neural network.

[Júlia Klee | Supersonic Labs @klee00z]

@supersonicai turns out the “game neural net” could do a little more than play games

[Júlia Klee | Supersonic Labs @klee00z]

Apparently, the skeptics thought we were just a game-playing neural net
Meanwhile, Julia-1 can play Snake, use tools, and classify any text

Turns out it was never just about the game

[Supersonic Labs @supersonicai]

You were saying something...?

yes, we classify responses 5x faster than Jev, using our own 144M-param model (Julia-1) on a 2021 i5 laptop while maintaining high accuracy.

So no, we’re not like other open-source models. And fuck no, we’re not just a game neural network.