Sergi

Recerco com fer que les màquines aprenguin com els humans. Seeking to carve machine's memories in their hearts. PhD candidate at @tuytelaarslab

[Tuytelaars Lab @tuytelaarslab]

This last year, our lab members have been working hard, and several publications are on the way. Expect news soon!

[Sergi @Atenrev]

By the way, I'll be around RSS in about 2 weeks. I'm looking into the field of robotics, so I'd be happy to meet anyone that's also visiting Sydney!

[Sergi @Atenrev]

Just updated my website with some upgrades and added some new papers. It's looking 🔥🥵🫦. Some threads on cool papers are coming soon!

[Rudy Gilman @rgilman33]

Anatomy of a Starpilot Ace

Ace is an agent trained on the Atari-style env Starpilot.

We can trace Ace's sensorimotor chain from pixel inputs to action. Ace's actions appear complex, often sophisticated, but when we pop open the hood they reduce to pure mechanism.

1/12

[Sergi @Atenrev]

@waghweb You've motivated me to do something stupid but fun

[Sergi @Atenrev]

@orphcorp @HououinTyouma @ZyMazza Now I understand the people who like "to fuck minds"

[Sergi @Atenrev]

@_ueaj send paper

[Sergi @Atenrev]

I really hope v4 is the ultimate last final final version this time

[Sergi @Atenrev]

@_ueaj Perú had a significant influx of Japanese immigration some decades ago. Pretty interesting

[Sergi @Atenrev]

@BandicootKiller Ufff qué tiempos, volvía atrás a grabar vídeo sin pensarlo

[Sergi @Atenrev]

@BandicootKiller Sí pero sube vídeo

[Sergi @Atenrev]

@REjercitos Es cierto, hoy en día las librerías de deep learning son muy fáciles de usar y existen modelos de visión mucho más eficientes que antes. A lo que me refería con el anterior tuit es que alguien con un mínimo de conocimiento ya podía hacer algo como lo del vídeo fácilmente.

[Sergi @Atenrev]

Does that mean LLMs have some sort of self-awareness? Do they recognize and identify themselves when they read about "AI"?

[Anthropic @AnthropicAI]

We started by investigating why Claude chose to blackmail. We believe the original source of the behavior was internet text that portrays AI as evil and interested in self-preservation.

Our post-training at the time wasn’t making it worse—but it also wasn’t making it better.

[Sergi @Atenrev]

Gemini is the Hikaru of LLMs. Every problem position is the "classic" whatever

[Lucy Shi @lucy_x_shi]

1/ We just released π0.7 — a steerable generalist robot model with emergent capabilities.
I want to share a bit of the backstory, because π0.7 taught me something surprising about where robot learning is heading. A thread on bittersweet lessons ���

[Sergi @Atenrev]

Latex space engineer, that's what I am

[Sergi @Atenrev]

EL BIEN SE RESTAURA SÍÚÍÍÍÍÚUUUUUUUUU

[Sergi @Atenrev]

El bien se restaura

[Sergi @Atenrev]

@leothecurious Continual investing and catastrophic forgetting of previous crashes