Gabriel Oliveira-Barra 🇦🇩

@Aditya_181105 Switch to Ubuntu and be happy

So microsoft decided you can't setup windows 11 offline. Well, time to go linux

@AdamHoltererer It's all about cooling optimization

Dumb question, but why not data centers, but vertical?

@osanseviero p.d. feels like there's a lot of room for improvement on the visual head

Feedback requested! What do you want for our next Gemma models? Which capabilities should we add and why?

@osanseviero Gemma current architecture is great, and performance is great. I'd have exactly the same features while trying to make it faster, lighter, and smarter. Perhaps add new architecture tweaks? Also, keep it friendly to fine tune

@sudoingX all that 2x faster with 70 % less memory sounds like marketing stuff. Have you guys ever heard about axolotl?

unsloth is one of the most underrated teams in ai and it's not close. while the timeline fights about the frontier, they quietly write the kernels that let you fine-tune on 3gb of vram, 2x faster with 70% less memory, no accuracy lost. and now they've brought real training support to amd hardware that was effectively cuda only until this week. radeon, ryzen, instinct, a massive install base that couldn't touch local training just got a first class path in. the labs making the models get the headlines, the people making the models run on your actual hardware get a quote tweet. that ratio is broken. go give unsloth their due.

@UnslothAI @AMD no CUDA no party

Introducing Unsloth for AMD 🚀 You can now train & run LLMs on your AMD hardware • We collaborated with AMD to enable you to train & run 500+ models on AMD GPUs • Works on Windows, WSL, Linux • Train Qwen, Gemma on 3GB VRAM GitHub: github.com/unslothai/unsl… Works on Radeon, Instinct, Ryzen and data center GPUs with up to 2× faster with 70% less VRAM and no accuracy loss via our custom Triton kernels and math algorithms. We also support optimized ROCm builds for GGUF & Safetensors inference. Unsloth is an open-source local UI for faster LLM training and inference, with tool-call healing, code execution, secure web search, remote APIs, and HTTPS deployment. Connect local models to Claude Code, Codex agents and run the latest Kimi, GLM, DeepSeek, Qwen3.6, and Gemma 4 models. 🔗Blog + Guide: unsloth.ai/docs/basics/amd

VAMOOOS ESPAÑAAAA 🇪🇦

@PlayStationES el trofeo que ganó España es físico, no como vuestros juegos.

49 Millones de personas han desbloqueado el trofeo "Campeones del Mundo" 🇪🇸🏆 💪

Andrej Karpathy just broke the entire premise of modern AI: "Agents aren't magic. They're distillation at scale." 99.99% of your LLM's capacity is wasted on garbage data it never needed. Small model + right tools + closed loop = terrifying capability. In a 16-minute conversation, Karpathy reveals the full reasoning stack. Worth more than any $500 AI course you've seen this year.

@miakhalifa don't say that. really.

ARRIBA ESPAÑAAAAA🇪🇸🇪🇸🇪🇸🇪🇸

@Itsfoss Ubuntu does the job. Good support, non toxic community, user friendly, newbie friendly, large adoption, works out of the box. Ubuntu hate is unjustified and irrational.

How come Ubuntu became the most hated distro in the desktop Linux community?

@deanwball The "ungovernability" (and openness) of Linux and the Internet is precisely what has made their success. The same will be true of open weight AI foundation models.

@DaniMayakovski Esas camaras capturan las direcciones MAC de los dispositivos bluetooth que pasan cerca. Muchos las alestorizan por privacidad (como teléfonos Apple) pero no todos y no siempre. Esto hace tiempo que existe para medir tráfico en semáforos de "smart cities" como Barcelona.

Un activista anti-vigilancia camina junto a una cámara de seguridad Flock en EEUU, mostrando cómo estas cámaras rastrean incluso los teléfonos cercanos para recopilar datos e identificar personas. Las cámaras Flock son sistemas de vigilancia masiva que usan IA para identificar "sospechosos", al principio decían que solo eran para leer matrículas de coches... sin embargo, esconde un sistema mucho siniestro que es capaz de rastrear personas en tiempo real. Ese mismo imperio occidental que hacía películas diciendo que la Stasi te espiaba a través de las tuberías, realmente es un estado parapolicial de vigilancia masiva.

Introducing Kimi K3: Open Frontier Intelligence 🔹 2.8 Trillion Parameters, 1 Million Context, Native Multimodal 🔹 Kimi Delta Attention enables up to 6.3x faster decoding in million-token contexts 🔹 Attention Residuals deliver ~25% higher training efficiency at <2% additional cost 🔹 Built for long-horizon agentic coding and self-evolving workflows Kimi K3 is now live on on Kimi.com, Kimi Work, Kimi Code, and the Kimi API. Open Weights by July 27, 2026. 🔗 API: platform.kimi.ai 🔗 Tech blog: kimi.com/blog/kimi-k3

Nais, rebaixeu una mica l'argentinofòbia uns dies perquè no es pensin que sou espanyols.

I resigned from Google DeepMind bc it broke its founding promise by selling AI to the military without restrictions against killer robots or mass spying. For months, I worked to stop this but watched powerful ethicists and institutions choose silence. Here's what happened. 🧵

A major question in multimodal modeling is how to leverage strong pre-trained models, such as industry-scale LLMs and VLMs, during training to avoid starting from scratch. This is important as the demand for rich and domain-specific multimodal models continues to increase, while training them from scratch is obviously impractical due to limited data and compute. The so-called “any-to-any” multimodal models can model a large diverse dictionary of modalities. That’s good. Their downside is that their architectures are often not decoder-only, which has limited their performance in practice and prevents them from leveraging strong pre-trained decoder-only models as priors. We are releasing MODUS (@icmlconf '26) a decoder-only any-to-any multimodal model to address some of these questions. A single transformer decoder predicts any modality from any others with no modality-specific heads, losses, or task pipelines. We show efficient adaptation of established models, e.g., BAGEL, to rich any-to-any multimodal modeling. We are releasing 14B to 77B parameter models. All materials are open-source. The download links and demos here modus-multimodal.epfl.ch 🧵

@Lordhoras87 El mayor problema de xbox es el mando. A quién se le ocurre invertir X con Y y A con B? Es como escribir con un teclado francés, le acabas por pillar manía.

Una parte de mí sinceramente cree que omitis Xbox por un desprecio alimentado en el tiempo y por la prensa y youtubers... O simplemente la consideráis muerta. (Muriendo 25 años ya 🤣) Pero Nintendo y Xbox me parecen excelentes alternativas antes que el pc si se desea consola

Los que amáis las consolas... Xq cuando rechazais PS por quitar el formato físico pasáis directamente a pc? No termino de entenderlo y al final abrazais el formato que os hizo salir de la consola de Sony Microsoft y Nintendo me parecen las opciones más logicas. Pero que se yo

@KenRoth The biggest risk of AI is the concentration of power in a few dominant providers of proprietary AI assistants. The only solution to AI sovereignty is open source foundation models.

@RinoTheBouncer jaja no

Will you be buying a PlayStation 6, under the current circumstances?🚀