Elektrine
EN
Log in Register
Paige Chat Timeline Communities Gallery Videos Email DNS VPN Uptime Kairo
Back to Timeline
Remote

Victor

@victorseven@mastodon.social
mastodon 4.7.0-beta.1
  • Open on mastodon.social

Hola! I love playing my piano and writing music 🎶 I'm also into videogames and game development. I have a PhD in Statistical Physics, and I work in applications of theoretical physics to biology (neuroscience, complex systems, ecology). Spanish, currently in Italy.

Looking forward to connect to other people here! 😄

Mostly content in Spanish, but I'm always happy to talk in other languages too!

0 Followers
0 Following
18 Posts
Joined November 17, 2022
Pronouns:
he/him
Langs:
ES/EN/IT/DE/GR

Posts

Open post
victorseven
Victor @victorseven@mastodon.social · May 21, 2026
Victor
@victorseven@mastodon.social

Hola! I love playing my piano and writing music 🎶 I'm also into videogames and game development. I have a PhD in Statistical Physics, and I work in applications of theoretical physics to biology (neuroscience, complex systems, ecology). Spanish, currently in Italy. Looking forward to connect to other people here! 😄 Mostly content in Spanish, but I'm always happy to talk in other languages too!

mastodon.social

Pues como os dije, aquí va la cover de Zelda: Kass theme! Espero que os guste

https://www.youtube.com/watch?v=xd62A62_g3w

3
0
4
0
Open post
victorseven
Victor @victorseven@mastodon.social · May 20, 2026
Victor
@victorseven@mastodon.social

Hola! I love playing my piano and writing music 🎶 I'm also into videogames and game development. I have a PhD in Statistical Physics, and I work in applications of theoretical physics to biology (neuroscience, complex systems, ecology). Spanish, currently in Italy. Looking forward to connect to other people here! 😄 Mostly content in Spanish, but I'm always happy to talk in other languages too!

mastodon.social

Mañana cover de Zelda nueva, la pondré también por aquí, que la resistencia merece contenido del wapo

2
0
0
0
Open post
victorseven
Victor @victorseven@mastodon.social · Apr 27, 2026
Victor
@victorseven@mastodon.social

Hola! I love playing my piano and writing music 🎶 I'm also into videogames and game development. I have a PhD in Statistical Physics, and I work in applications of theoretical physics to biology (neuroscience, complex systems, ecology). Spanish, currently in Italy. Looking forward to connect to other people here! 😄 Mostly content in Spanish, but I'm always happy to talk in other languages too!

mastodon.social
Replying to @victorseven@mastodon.social
(y ya está, he leído un poco el paper pero tampoco he seguido los argumentos súper en detalle, si meto la gamba en algún sitio avisad :blobmiou: )
0
0
0
0
Open post
victorseven
Victor @victorseven@mastodon.social · Apr 27, 2026
Victor
@victorseven@mastodon.social

Hola! I love playing my piano and writing music 🎶 I'm also into videogames and game development. I have a PhD in Statistical Physics, and I work in applications of theoretical physics to biology (neuroscience, complex systems, ecology). Spanish, currently in Italy. Looking forward to connect to other people here! 😄 Mostly content in Spanish, but I'm always happy to talk in other languages too!

mastodon.social
Replying to @victorseven@mastodon.social
Resumen, me parece una publicación absolutamente intrascendente. En el mejor de los casos tiene algún resultado técnico de interés, pero probablemente no afecta en nada a los casos prácticos. Con la de cosas que hay por ahí para criticar la IA, no me escojáis las más flojas, hacedme el favor.
3
1
0
0
Open post
victorseven
Victor @victorseven@mastodon.social · Apr 27, 2026
Victor
@victorseven@mastodon.social

Hola! I love playing my piano and writing music 🎶 I'm also into videogames and game development. I have a PhD in Statistical Physics, and I work in applications of theoretical physics to biology (neuroscience, complex systems, ecology). Spanish, currently in Italy. Looking forward to connect to other people here! 😄 Mostly content in Spanish, but I'm always happy to talk in other languages too!

mastodon.social
Replying to @victorseven@mastodon.social
  1. Un paper donde la introducción hablar principalmente de Singularity (sí, en mayúscula) yo sinceramente lo cogería con un par de pinzas en la nariz. Ahora hay mucho flipao por ahí.
1
1
0
0
Open post
victorseven
Victor @victorseven@mastodon.social · Apr 27, 2026
Victor
@victorseven@mastodon.social

Hola! I love playing my piano and writing music 🎶 I'm also into videogames and game development. I have a PhD in Statistical Physics, and I work in applications of theoretical physics to biology (neuroscience, complex systems, ecology). Spanish, currently in Italy. Looking forward to connect to other people here! 😄 Mostly content in Spanish, but I'm always happy to talk in other languages too!

mastodon.social
Replying to @victorseven@mastodon.social
  1. Por último, veo varios post interpretando el titular del paper como que la IA no se puede mejorar de modo recursivo. Ojo que los IAbros cuando hablan de esto se refieren a otro proceso en el que la IA escribe mejores modelos (sin tocar los datos de entrenamiento) y/o analiza los outputs para hacer el muestreo de forma independiente. Este paper tampoco prohíbe eso.

De todas formas eso no va a pasar, ya les gustaría a ellos xdddddd

1
1
0
0
Open post
victorseven
Victor @victorseven@mastodon.social · Apr 27, 2026
Victor
@victorseven@mastodon.social

Hola! I love playing my piano and writing music 🎶 I'm also into videogames and game development. I have a PhD in Statistical Physics, and I work in applications of theoretical physics to biology (neuroscience, complex systems, ecology). Spanish, currently in Italy. Looking forward to connect to other people here! 😄 Mostly content in Spanish, but I'm always happy to talk in other languages too!

mastodon.social
Replying to @victorseven@mastodon.social
La IA se equivoca bastante y tiene problemas, así que si muestreas la basura con los ojos tapados obviamente vas a tener el famoso problema del colapso. Con un muestreo adecuado, en teoría, se puede evitar. La pregunta es obviamente hasta qué punto se puede hacer un muestreo adecuado sin llenar el conjunto de entrenamiento de artefactos invisibles para nosotros pero importantes para la IA.
0
1
0
0
Open post
victorseven
Victor @victorseven@mastodon.social · Apr 27, 2026
Victor
@victorseven@mastodon.social

Hola! I love playing my piano and writing music 🎶 I'm also into videogames and game development. I have a PhD in Statistical Physics, and I work in applications of theoretical physics to biology (neuroscience, complex systems, ecology). Spanish, currently in Italy. Looking forward to connect to other people here! 😄 Mostly content in Spanish, but I'm always happy to talk in other languages too!

mastodon.social
Replying to @victorseven@mastodon.social
  1. La evidencia empírica claramente va en contra de la idea de que la IA no puede entrenarse con su propio output. Muchos modelos usan una gran cantidad de datos generados por IA y les va bien.

Es un problema de aproximación de distribuciones. Mientras en el conjunto de entrenamiento el muestreo sea representativo la tecnología va a funcionar sin problema. El problema es conseguir un muestreo representativo de una distribución de una dimensionalidad super alta etcetc.

0
1
0
0
Open post
victorseven
Victor @victorseven@mastodon.social · Apr 27, 2026
Victor
@victorseven@mastodon.social

Hola! I love playing my piano and writing music 🎶 I'm also into videogames and game development. I have a PhD in Statistical Physics, and I work in applications of theoretical physics to biology (neuroscience, complex systems, ecology). Spanish, currently in Italy. Looking forward to connect to other people here! 😄 Mostly content in Spanish, but I'm always happy to talk in other languages too!

mastodon.social
Replying to @victorseven@mastodon.social
  1. El resultado, que está enunciado en el resumen (abstract), es clarísimo: el colapso se produce si la cantidad de datos no auto-generados por la propia IA es 'vanishing'. Es decir, mientras mantengas un porcentaje finito de datos reales, POR PEQUEÑO QUE SEA, el training sigue funcionando sin problema.

El único problema es al aumentar el tamaño del conjunto de entrenamiento al infinito de forma que la fracción real/generado -> 0. Mantén un 0.1% fijo de datos reales y ya no hay problema.

0
1
0
0
Open post
victorseven
Victor @victorseven@mastodon.social · Apr 27, 2026
Victor
@victorseven@mastodon.social

Hola! I love playing my piano and writing music 🎶 I'm also into videogames and game development. I have a PhD in Statistical Physics, and I work in applications of theoretical physics to biology (neuroscience, complex systems, ecology). Spanish, currently in Italy. Looking forward to connect to other people here! 😄 Mostly content in Spanish, but I'm always happy to talk in other languages too!

mastodon.social
Replying to @victorseven@mastodon.social
  1. El paper usa un modelo y estudia el modelo. Si la IA funciona según ese modelo aproximado, las conclusiones del paper aplican. Pero aquí hablamos de algo más físico. Si esas hipótesis no son exactamente ciertas las conclusiones tampoco aplican.

Por ejemplo, el paper asume que el objetivo del modelo es minimizar la distancia KL entre training y test. Eso no se hace explícitamente en LLMs. ¿El entrenamiento reduce esa distancia? Probablemente sí. ¿Es suficiente para aplicar el paper? Aaaaaaah

0
1
1
0
Open post
victorseven
Victor @victorseven@mastodon.social · Apr 27, 2026
Victor
@victorseven@mastodon.social

Hola! I love playing my piano and writing music 🎶 I'm also into videogames and game development. I have a PhD in Statistical Physics, and I work in applications of theoretical physics to biology (neuroscience, complex systems, ecology). Spanish, currently in Italy. Looking forward to connect to other people here! 😄 Mostly content in Spanish, but I'm always happy to talk in other languages too!

mastodon.social
Replying to @victorseven@mastodon.social
  1. El 'paper' es un preprint, básicamente no está todavía revisado. Es un único autor. Aunque no pinta mal, tampoco parece extremadamente formal a primera vista.

Es decir, igual el resultado ni es verdad o es parcialmente verdad. Si no sois gente con conocimiento técnico, no compartáis cosas de arxiv o bioarxiv anunciando descubrimientos alegremente, y no os fiéis de la gente que las anuncia, como regla general.

0
1
0
0
Open post
victorseven
Victor @victorseven@mastodon.social · Apr 27, 2026
Victor
@victorseven@mastodon.social

Hola! I love playing my piano and writing music 🎶 I'm also into videogames and game development. I have a PhD in Statistical Physics, and I work in applications of theoretical physics to biology (neuroscience, complex systems, ecology). Spanish, currently in Italy. Looking forward to connect to other people here! 😄 Mostly content in Spanish, but I'm always happy to talk in other languages too!

mastodon.social

Hoy veo a gente muy contenta comentar que matemáticos demuestran lo que ya se sospechaba: que la IA no puede auto-mejorarse a sí misma.

He abierto el paper y le he echado un vistazo. El paper no dice eso en absoluto, y de hecho se puede interpretar al contrario. Hay además muchos ángulos que no están cubiertos.

Varios comentarios.

4
4
5
0
Open post
victorseven
Victor @victorseven@mastodon.social · Apr 24, 2026
Victor
@victorseven@mastodon.social

Hola! I love playing my piano and writing music 🎶 I'm also into videogames and game development. I have a PhD in Statistical Physics, and I work in applications of theoretical physics to biology (neuroscience, complex systems, ecology). Spanish, currently in Italy. Looking forward to connect to other people here! 😄 Mostly content in Spanish, but I'm always happy to talk in other languages too!

mastodon.social
Replying to @Martinelli@mas.to
@Martinelli ojo que ahora los enlaces de instagram contienen la info de quién lo compartió, he abierto el vídeo y antes de reproducirlo sale la opción de seguirte. Esto lo han metido ahora, no pasaba antes. Si quieres mantener oculto tu @ cuando compartas enlaces quita ?igsh=.... y todos los numeros que vienen después del link.
0
2
0
0
Open post
victorseven
Victor @victorseven@mastodon.social · Mar 29, 2026
Victor
@victorseven@mastodon.social

Hola! I love playing my piano and writing music 🎶 I'm also into videogames and game development. I have a PhD in Statistical Physics, and I work in applications of theoretical physics to biology (neuroscience, complex systems, ecology). Spanish, currently in Italy. Looking forward to connect to other people here! 😄 Mostly content in Spanish, but I'm always happy to talk in other languages too!

mastodon.social
Replying to @pierre_pebble@mastodon.ie
@pierre_pebble @kenney those are good news! I am tired of using the grid, sometimes I don't know very well how to rotate the nodes to make them coincide exactly. Happened to me with some Kenney stuff too.
0
0
0
0
Open post
victorseven
Victor @victorseven@mastodon.social · Mar 12, 2026
Victor
@victorseven@mastodon.social

Hola! I love playing my piano and writing music 🎶 I'm also into videogames and game development. I have a PhD in Statistical Physics, and I work in applications of theoretical physics to biology (neuroscience, complex systems, ecology). Spanish, currently in Italy. Looking forward to connect to other people here! 😄 Mostly content in Spanish, but I'm always happy to talk in other languages too!

mastodon.social
Replying to @victorseven@mastodon.social
Luego está el tema de 'los revisores no cobran' que la gente a menudo se escandaliza bastante. Yo soy de la opinión de que eso está bien. Pagar por revisar llevaría inmediatamente a gente que se dedica a hacer revisiones de baja calidad non-stop por sacar un dinero extra. Ahora, lo suyo sería revisar para una organización pública que sea la que gestione el tema de las publicaciones. Hecha por y para científicos.
6
0
1
0
Open post
victorseven
Victor @victorseven@mastodon.social · Mar 12, 2026
Victor
@victorseven@mastodon.social

Hola! I love playing my piano and writing music 🎶 I'm also into videogames and game development. I have a PhD in Statistical Physics, and I work in applications of theoretical physics to biology (neuroscience, complex systems, ecology). Spanish, currently in Italy. Looking forward to connect to other people here! 😄 Mostly content in Spanish, but I'm always happy to talk in other languages too!

mastodon.social
Replying to @victorseven@mastodon.social
Lo de pagar por publicar es la mayor estafa que hay. A todos los científicos les parece un escándalo y las cifras son absurdas. En el pantallazo arriba podéis ver que en teoría es posible publicar gratis, pero entonces la gente no lo puede leer. ¡Pero ojo! Si tenéis financiación de la UE para vuestra investigación, la UE os prohíbe publicar en cerrado. Las revistas lo saben, así que básicamente están robando dinero público a mano armada sin ningún tipo de vergüenza.
11
2
6
0
Open post
victorseven
Victor @victorseven@mastodon.social · Mar 12, 2026
Victor
@victorseven@mastodon.social

Hola! I love playing my piano and writing music 🎶 I'm also into videogames and game development. I have a PhD in Statistical Physics, and I work in applications of theoretical physics to biology (neuroscience, complex systems, ecology). Spanish, currently in Italy. Looking forward to connect to other people here! 😄 Mostly content in Spanish, but I'm always happy to talk in other languages too!

mastodon.social
Replying to @victorseven@mastodon.social
Lo de la cesión de derechos a las revistas tampoco es nuevo. Cuando publicas cedes todos tus derechos a la revista, eso ha sido siempre. Así que pueden entrenar IA, publicar un libro, hacer lo que les dé la gana. En cuanto al tema de autores/revisores usando IA para artículos o revisiones. Está ampliamente considerado mala práctica. Si te pescan haciéndolo te estás jugando el tipo. Pero efectivamente cada vez más gente lo hace...
7
1
1
0
Open post
victorseven
Victor @victorseven@mastodon.social · Mar 12, 2026
Victor
@victorseven@mastodon.social

Hola! I love playing my piano and writing music 🎶 I'm also into videogames and game development. I have a PhD in Statistical Physics, and I work in applications of theoretical physics to biology (neuroscience, complex systems, ecology). Spanish, currently in Italy. Looking forward to connect to other people here! 😄 Mostly content in Spanish, but I'm always happy to talk in other languages too!

mastodon.social

RE: @miguelespigado@masto.es

Hola, aquí un 100tifiko de esos. Voy a aclarar una cosita que os va a poner los pelos de punta. Y es... ¡ya me gustaría pagar SOLO 3000 pavos por publicar en Nature!

Probablemente, el amigo de este señor trabaja para Springer, que es la editorial que edita Nature. Pero también muchas otras revistas. Las revistas 'normales' tienen un precio entre 1000 - 3000 euros, efectivamente.

Las revistas de Nature especializadas, son entre 5000 y 9000. Nature Nature, la mehó, 11.000 pavinis.

15
4
21
0

Remote instance

mastodon.social
Open on original server

Media

313k7r1n3
Elektrine

Tor hidden service

elekhj7afj4qnrr4yd3bkzslsyo5jgfxw3orgjkhlcxifueodybyiiad.onion

Platform

  • Email
  • Chat
  • Timeline
  • Communities
  • VPN
  • DNS

Company

  • About
  • Contact
  • FAQ

Legal

  • Terms of Service
  • Privacy Policy
  • Warrant Canary
  • Lite (no JS)
  • VPN Policy
  • Source code

Support

  • support@elektrine.com
  • Report Security Issue
Mail client setup IMAP mail.elektrine.com:993 POP3 mail.elektrine.com:995 SMTP mail.elektrine.com:465
© 2026 Elektrine. All rights reserved. Server: 07:04:26 UTC