This paper from PayPal and NVIDIA quietly kills the myth that agentic AI needs giant models to work. PayPal just published a research paper showing that their biggest performance win did not come from a better prompt, a bigger model, or a clever orchestration trick. It came from
COMPUTING
-
MiniMax M2.1 Powers Robot Dog with 230B Parameter Sparse MoE
By
–
MiniMax M2.1 × Vbot is super-powered in action! This AI agent model (230B params, sparse MoE) controls a physical robot dog seamlessly—virtual training to real-world navigation, coding, & multimodal mastery.
— Futurepedia – Learn to Leverage AI (@futurepedia_io) 29 décembre 2025
Speedy, efficient, rivals Claude Sonnet.#MiniMaxM21 #AIRobotics pic.twitter.com/Hji1KAuIgzMiniMax M2.1 × Vbot is super-powered in action! This AI agent model (230B params, sparse MoE) controls a physical robot dog seamlessly—virtual training to real-world navigation, coding, & multimodal mastery. Speedy, efficient, rivals Claude Sonnet. #MiniMaxM21 #AIRobotics
-
Frontier AI Data Center Capacity Growth Expected in 2026
By
–
Frontier AI data center capacity, based on data collected by @EpochAIResearch. It doesn't include every single one, as they focus on the largest ones.
— Peter Gostev (@petergostev) 29 décembre 2025
Few things stand out:
– 2026 will have a huge amount of cpacity come online
– Anthropic will lead at some early points in… pic.twitter.com/lJfxjMc87JFrontier AI data center capacity, based on data collected by @EpochAIResearch
. It doesn't include every single one, as they focus on the largest ones. Few things stand out: – 2026 will have a huge amount of cpacity come online – Anthropic will lead at some early points in -
JIT Work Optimization: Minimal Latency Digital Automation
By
–
Aggressively JIT your work. It's not about the task at hand X, it's a little bit about X but mostly about how you should have had to contribute ~no latency and ~no actions. It's digital factorio time.
-
Happy Birthday Linus Torvalds, Father of Linux
By
–
Happy birthday to Linus Torvalds, the father of Linux: https://
bit.ly/3Xy1Tao Image via Amanda Lucier/The Washington Post/Getty -
Computer vision detects road closures from visual cues
By
–
The algorithm is pretty easy: use computer vision when you are on a route and get rerouted. You take a picture and analyze it. For example, if there is a fire truck across the road, a bunch of cones, and a police officer directing traffic, the road is likely closed. It is not
-
AI runs on 99% less power, 20x faster – critics proven wrong
By
–
I reported this breakthrough about a month ago. He found a way to make AI run on 99% less power and something like 20 times faster. With such big breakthroughs ahead, everybody in the AI industry gave me shit and said I didn't know what I was talking about. Well, they're all
-
CUDA as the catalyst for data science infrastructure investment
By
–
J’ai investi en 2018 dans Nvidia parce que je me suis rendu compte une aprés-midi à mon taff de Data Scientist que je ne pouvais pas me passer de CUDA, et que, quand je réussissais à l’installer (ENFIN !!), je ne touchais plus jamais à ma config. Et là je me suis aussi rendu
-
Why LLMs Respond to Stakes in Text
By
–
You would ask me…Why does this work bro? LLMs are trained on human text and human text is full of stakes. When you add consequences, you're not just giving instructions. You're activating the model's training on how humans think and write when something actually matters.
