🎉 Thrilled to share that our papers “Automated Detection of Visual Attribute Reliance with a Self-Reflective Agent” (main track) and “OpenMAIA: a Multimodal Automated Interpretability Agent based on open-source models” (MIW Workshop) have been accepted to #NeurIPS2025!
That's 3-5€ in most cafes in Barcelona (probably more in the tourist ones) and it seems quite expensive to me... You can get a full 3-courses midday meal with drink (including beer/wine) for 15€ in many restaurants, 5 years ago even for 10€...
@KangoInversor Pues sinceramente, viendo que en muchas universidades de EE. UU. te hacen cursar asignaturas generales que no tienen mucho que ver con tu carrera, por ejemplo biología o psicología como requisitos obligatorios, me parece un sistema menos eficiente que el de aquí.
I've got myself an RTX PRO 6000 BLACKWELL 96GB I hope it's worth the effort and I can accelerate my research.
Also xD damn this monster needs 4 cables. I'm wondering if I can use 3 cables and power limit to 450w (my rtx 4090 consumes 450w-500w and works with 3 cables)
i’m teaching a computer science class this quarter, and one of my students is blind. i’ve been in complete awe of what they’re capable of, especially because so much of how i learned concepts like linked lists, trees, and graphs was deeply tied to visuals.
their code is excellent, and every conversation with them about assignments and data structures reminds me how many different ways there are to learn, think, and create, and how important accessible teaching and learning content is.
Tengo 35 años y cancer de mama metastásico, un caso raro, menos del 1% de tumores de mama son como el mío y hay poca documentación sobre ello.
Por eso me gustaría encontrar personas que se dediquen a esto y que quieran investigar con mi caso. Twitter haz tu magia
@josecanciani@javilopen@drawais_ai El guión largo es algo que se mucho en la academia desde muchísimos años, que una persona que esté haciendo un PhD lo use no es raro.
@BizDevKevv@loktar00 Thunderbolt 4 is about pcie3.0 X4, there's some bottleneck but it's not that dramatic. Also the idea is to load whole models on eGPU so once they are inside they operate at Bream's bandwidth which is usually 1-2TB/s
Exactly why I left. It became the preferred destination for sun-starved Dutch and English tourists, which turned the economy’s reliance on tourism into an absolute hellhole for the youth, who when I was there only dreamed of leaving. You push the natives toward self-destruction when you allow foreigners to take over historic properties and turn them into their eye candy summer homes.
@tpotasuka@__tinygrad__ The GPU does most computation inside its VRAM. So once you load data into eGPU you get the speed of GDDR7
The PCIe/USB link is used only for occasional transfers.