| Website | https://www.sebastianbuhai.com |
| https://twitter.com/sbuhai | |
| Bluesky | https://bsky.app/profile/sbuhai.bsky.social |
| Threads | https://www.threads.net/@sebastianbuhai |
Valparaíso ya nos tenía conquistados, pero esta semana la mesa subió la vara:
La Concepción https://www.restaurantlaconcepcion.cl/
y
La Caperucita y el Lobo https://www.lacaperucitayellobo.cl/.
Entre vistas al puerto y cuentos con colmillo, ganó el lobo 🐺; ¡nosotros también! Volveremos con hambre de bis.
RE: https://mastodon.social/@sbuhai/115812181250336610
Revised version is—obviously!—better. Same Iron Maiden riff, cleaner solo: leaner text, sharper filtering, tighter network punchlines. Clock still ON; silence just got remastered! 🤘
main: https://www.sebastianbuhai.com/papers/publications/reputation_disclosure_networks.pdf
OA: https://www.sebastianbuhai.com/papers/publications/reputation_disclosure_networks_OA.pdf
(arXiv update: https://arxiv.org/abs/2512.22987)
There has been significant debate on social media about the arXiv's recent policy decision to impose a one year ban on any authors found to have incontrovertible evidence of having introduced AI-generated errors in a paper they are an author on that would have been detected by a reasonable review. I found it interesting to evaluate this policy against the four recommendations I made at this recent talk https://www.youtube.com/watch?v=Uc2zt198U_U:
1. For traditional workflows, enforce strict policies regulating what level of AI assistance is permitted or welcomed.
2. Decrease the emphasis on ``being first'' or ``merely solving'' a problem; increase the emphasis on ``digesting'' the solution.
3. At the same time, generate new challenges, suitable for contributors using heavy AI assistance, that would generate mathematical value.
4. For both traditional and non-traditional workflows, be very clear about what the goals - both explicit and implicit - of the project are, and why they are being pursued. (1/3)
Terence Tao