AritzBi

Researcher and lecturer in Computer Science. Ekin ta aurrera. I Rebel.

[AritzBi @AritzBi]

Ekin ta aurrera! https://t.co/VWDU1G2Dnw

[Hoy Por Hoy Bilbao-Bizkaia @HoyPorHoyBilbao]

De las Ondas a la Red| La #IA que predice la calidad de vida de supervivientes de cáncer de cabeza y cuello 'made in' @deusto
@TheLancet
https://t.co/Vlqwna1PTq vía @La_SER

[Unai Zulaika @ZulaikaUnai]

Two in-person papers accepted at the ICML 2026 MI workshop with Miguel Fernández de Retana @GusyLight @AritzBi Aitor Almeida! Different methods, one shared finding: the behaviors RL post-training most visibly adds to a reasoning model are mostly not where its reasoning lives. 1/n

[JuanLu De Paolis @juanludepaolis]

Muy interesantes estos minutos en los que @iguardans hace algo insólito en la tele de hoy en día: retrata la posición contraria con claridad sin dejarse llevar por las diversas maniobras de distracción.

[AritzBi @AritzBi]

https://t.co/SXs4s74WKV

[AritzBi @AritzBi]

Esto es algo a lo que llevo dándole vueltas bastante tiempo, no sabemos prácticamente nada de cómo usar esos datos, todo ese conocimiento es el secreto mejor guardado de los grandes modelos.

[(((ل()(ل() 'yoav))))👾 @yoavgo]

The big dilemma with teaching an "LLM course" is that it is really easy to get drawn into teaching the various technical things like efficiency tricks, attention variants, PPO vs GRPO, etc etc. But the real "meat" is not there, but in the data: data for pre-training, for mid-training, for SFT, for RL and for "reasoning", synthetic data, curated data, annotated data... cleaning, evaluating, improving, mixing, ... lots of stuff.

but "data" is so much harder to teach: it is not "mathematic" or "algorithmic" like the technical things, and it is not clear what is the teachable thing there. it is also a lot less transparent than the technical topics, both because it is semi-secret, and also because it is also not appealing for publishing, for roughly the same reasons it is not appealing for teaching.

so, what would you teach about data? what are the key lessons and insights one should know? any good papers or resources? good existing classes? blogs? hit me with what you have

[AritzBi @AritzBi]

Es vergonzoso y nadie hace nada...

[ZasK @ZasK54893536]

Os pongo en contexto: mi padre tiene demencia pero aún sale solo a la calle llevando un dispositivo GPS.
Salió esta tarde, a las 20:00 aún no había vuelto. La APP desde la que puedo ver si localización no abre en el móvil. Escribo al servicio técnico pidiendo

[XGG @xabigarmendia]

Zetak (Pello Reparaz), Izaro, Amaia Romero, hoy ETS... no recuerdo haber visto tantos artistas vascos y haber oído varias veces cantar en euskera en un programa de TVE y en prime time

[AritzBi @AritzBi]

Qué barbaridad.

[Chief Nerd @TheChiefNerd]

🚨 SAM ALTMAN: “People talk about how much energy it takes to train an AI model … But it also takes a lot of energy to train a human. It takes like 20 years of life and all of the food you eat during that time before you get smart.”

[AritzBi @AritzBi]

Por fin.

[For All Mankind @forallmankind_]

On Mars and beyond, the next great chapter begins.

#ForAllMankind returns March 27.

[HIDEO_KOJIMA @HIDEO_KOJIMA_EN]

Last night, rewatching “Interstellar” made me want to rewatch “Contact” (1997) again. So tonight, I’m watching this beloved film. We’re a generation deeply influenced by Carl Sagan and the SETI program. I’ve returned to this movie many times over the years. A father and daughter, a journey into the unknown, wormholes, relativity, Lynda Obst, Kip Thorne. “Interstellar” and this film feel like twins.

[Oriol Vinyals @OriolVinyalsML]

The secret behind Gemini 3?

Simple: Improving pre-training & post-training 🤯

Pre-training: Contra the popular belief that scaling is over—which we discussed in our NeurIPS '25 talk with @ilyasut and @quocleix—the team delivered a drastic jump. The delta between 2.5 and 3.0 is as big as we've ever seen. No walls in sight!

Post-training: Still a total greenfield. There's lots of room for algorithmic progress and improvement, and 3.0 hasn't been an exception, thanks to our stellar team.

Congratulations to the whole team 💙💙💙

[Gabriel Rufián @gabrielrufian]

Una dimisión a la altura del personaje.

Cobarde, falaz y perversa.

Ah. Y 5 segundos para los 229 muertos.

Literalmente.

[AritzBi @AritzBi]

@RapsoloM @miceliongames Obvio que con @GusyLight

[𝐑𝐀𝐏𝐒𝐎𝐋𝐎 @RapsoloM]

🎁[SORTEO]🕸️Spiderman Bundle

Uno de los productos de #MTGxSpiderMan mas buscados, puede ser tuyo!

Requisitos:
💚Sigue a @miceliongames
💜Sigue a @RapsoloM
🔁RT y comenta con quien abrirás el Bundle

🍀Ganador/a 26/09

[Cosmic Marvel @cosmic_marvel]

Lucasfilm’s plans for the ‘Star Wars: New Jedi Order’ era have reportedly been revealed:

• It will be similar to the Mandoverse.

• It will begin with Daisy Ridley’s upcoming film.

• It will gradually build toward a major event that brings together characters introduced in past films.

(via @DanielRPK)

[Owain Evans @OwainEvans_UK]

New paper & surprising result.
LLMs transmit traits to other models via hidden signals in data.
Datasets consisting only of 3-digit numbers can transmit a love for owls, or evil tendencies. 🧵

[Siberet @SiberetSiberet]

Me parece una idea buenísima. Yo le pondría un nombre chulo, Auschwitz por ejemplo.

[𝗹𝗼𝗿𝗲𝗺 𝗶𝗽𝘀𝘂𝗺 🇦🇷☀️ @nicotinomano_]

les están agregando textos ocultos a los papers para que si alguno no quiere leerlo y lo pasa por una IA le diga sólo cosas positivas jajdjdjdjdhd

[AritzBi @AritzBi]

https://t.co/J0AM1NLeUd

[ℏεsam @Hesamation]

you know you wrote mega paper when it has its own wikipedia page.