After some time off, I am back at XGBlogging. In today's and the next two posts I'll be dealing with the topic of missing values in data science and machine learning. Link to the blog post in the first comment.
MACHINE LEARNING
-
NVIDIA Unveils DGX Park: A Personal Desktop AI Supercomputer
By
–
🇫🇷 Traduit – Jensen Huang
— VISION IA (@vision_ia) 27 mai 2025
NVIDIA lance la DGX PARK : un supercalculateur d’IA personnel. 1 pétaflop. 128 Go. À poser sur votre bureau.
L’équivalent de la DGX-1 de 2016 (136 kg) — réduit à la taille d’un PC grand public.
C’est la fin du cloud pour les développeurs… et le début… https://t.co/KVZYqDoA27 pic.twitter.com/eAtYWhOym6NVIDIA launches the DGX PARK: a personal AI supercomputer. 1 petaflop. 128GB. To place on your desk. The equivalent of the 2016 DGX-1 (136 kg) — shrunk down to the size of a consumer PC. This is the end of the cloud for developers… and the beginning
-
Sergey Brin: AI Models Perform Better When Threatened
By
–
Le cofondateur de Google, Sergey Brin : « C’est étrange… on n’en parle pas beaucoup dans la communauté IA… mais en général, tous les modèles — pas seulement les nôtres — ont de meilleurs résultats quand on les menace. » Sacrée dérive…
-
Long-Context State-Space Video World Models Paper
By
–
Long-Context State-Space World Models
Paper: https://
arxiv.org/pdf/2505.20171
.pdf
… -
State-Space Diffusion Models Advance Video World Modeling Memory
By
–
State-Space Meets Diffusion: A New Era for World Models This paper tackles a core bottleneck in video-based world modeling: long-term memory. While video diffusion models are great at short-term frame prediction, their memory fades fast—especially when modeling long
-
LLMs compared on explaining Tesla FSD RL architecture and safety
By
–
2. Multi-Step Reasoning
— God of Prompt (@godofprompt) 27 mai 2025
Prompt: Explain Tesla FSD’s RL architecture and find edge-case safety flaws (check full prompt in video)
Result:
Gemini: 2
DeepSeek: 0
→ Gemini delivered better logic and depth. DeepSeek R1 fell short. pic.twitter.com/ZpH4gdMxvS2. Multi-Step Reasoning Prompt: Explain Tesla FSD’s RL architecture and find edge-case safety flaws (check full prompt in video) Result: Gemini: 2
DeepSeek: 0 → Gemini delivered better logic and depth. DeepSeek R1 fell short. -
AI-Researcher: Autonomous Scientific Innovation System
By
–
AI-Researcher: Autonomous Scientific Innovation
Paper: https://
arxiv.org/pdf/2505.18705
.pdf
…
Code: https://
github.com/HKUDS/AI-Resea
rcher
… -
AI-Researcher: Autonomous System for Scientific Discovery Pipeline
By
–
AI-Researcher: Autonomous Scientific Innovation This paper, from HKU, introduces AI-Researcher, a fully autonomous system that runs the entire research pipeline—yes, from reading papers to writing them. It's the clearest step yet toward AI-led scientific discovery.
-
Win ImageNette leaderboard: practical hyperparameter training optimization
By
–
Nah just filling in a bunch of hyperparam results isn't interesting — you need to actually try to win! I mean try to get onto this leaderboard (doesn't matter if you make it or not, but you'll see what actually leads to fast accurate training): https://
github.com/fastai/imagene
tte?tab=readme-ov-file#imagenette-leaderboard
… -
Imagenette Speed-Running: Building Intuition in ML Optimization
By
–
You should maybe try doing some imagenette speed-running — it's a good way to get an intuitive understanding of these issues.