Jürgen Schmidhuber

So, yeah, very much looking forward to it.

Maybe I need to add something because our friend Jürgen Schmidhuber is also going to Japan.

He is becoming the new chief scientific advisor at Zakana AI.

That's a Japanese company.

Yeah.

let's let's consider doing that

good so because we mentioned already mentioned x and x is a crazy platform right so so you can find a lot of shit there really bad things but you can also find very entertaining thing and this morning i found something very entertaining a islop from from x and it's impossible to describe what they did with jürgen schmidhuber But I will put the link in the show notes that everybody can see the video, the AISLOV video with Jürgen Schmidhuber.

Very interesting.

¿Amber?

Sakana AI, la startup de Tokio cofundada por un coautor del paper del Transformer, anunció el jueves que Jürgen Schmidhuber, cuya tesis doctoral de 1987 describía un programa capaz de mejorar su propio algoritmo de aprendizaje, se incorpora como asesor científico principal de su nuevo laboratorio de automejora recursiva.

La tesis del laboratorio es directa.

En lugar de entrenar modelos cada vez más grandes con cómputo de fuerza bruta, usar el modelo actual para diseñar y entrenar el siguiente.

And then this one genuinely stopped me.

Sakana AI just announced Jürgen Schmidhuber is joining as chief scientific advisor.

Schmidhuber, you know the name, right? The guy everyone calls the father of modern AI.

He was doing recursive self-improvement theory back in the nineteen-nineties when compute cost a million times more than it does now.

And instead of just getting cited in everyone's papers, he's hands-on now inside an actual active RSI lab.

There were JEPA-like models before JEPA, and there were world models before world models.

In fact, at the Schmidhuber paper, the world model Schmidhuber paper, this is model-based RL.

So arguably, people that now use it for something else other than model-based RL are just using the name wrong.

I think it's mostly branding... of techniques which are very loosely defined.