SINGAPOUR dévoile un plan mondial pour la sécurité de l’IA, en pleine tension USA-Chine. Des chercheurs d’OpenAI, xAI, DeepMind, Tsinghua, etc. unissent leurs forces autour du “Singapore Consensus” : Comprendre les risques Créer des IA plus sûres Contrôler les systèmes
RESEARCH
-
SAS Innovate Day 2: Data Innovation Announcements and Future Vision
By
–
Day 2 of #SASInnovate was electric.From epic announcements to the latest insights to a powerful vision for the future of data. We're looking at you Day 3, and you can experience it too, with a free #SASInnovate digital pass. http://
2.sas.com/6012NHCB2 -
Meta Launches Perception Language Model for Advanced Video Analysis
By
–
Meta AI dropped Meta Perception Language Model, an open & reproducible vision-language AI for challenging visual tasks
— Rowan Cheung (@rowancheung) 8 mai 2025
It can watch videos and extract details like what a person is doing in the content and how they are doing itpic.twitter.com/JSbbgCukRIMeta AI dropped Meta Perception Language Model, an open & reproducible vision-language AI for challenging visual tasks It can watch videos and extract details like what a person is doing in the content and how they are doing it
-

R³-VQA: Read the Room Video Social Reasoning
By
–
R³-VQA: “Read the Room” by Social Reasoning
Paper: https://
arxiv.org/pdf/2505.04147 -
R³-VQA Benchmark Tests AI Theory of Mind in Social Scenes
By
–
How well can AI read the room? R³-VQA is a new video QA benchmark pushing LVLMs to reason like humans in complex social scenes—tracking beliefs, emotions, intentions, and more. Turns out, even top models still struggle with consistent Theory of Mind.
-
Academic AI projects require urgent modernization and reboot
By
–
the “academic project” has been long overdue for a reboot
-
Understanding Core Concepts and Terminology in Artificial Intelligence
By
–
What does #artificialintelligence really mean? Some of the core concepts have gotten buried underneath the buzzwords. Understanding some basic
terminology will help you get your bearings: https://
okt.to/pV6wGS @antgrasso @mvollmer1 @blueyonder @nicochan33 -
Physicists’ Hidden Impact on Modern AI Development
By
–
i was surprised to learn how much of modern AI was developed by physicists: – scaling laws – alphafold
– RLHF (dario got his phd in biophysics)
– diffusion models
– normalizing flows
– VAEs
– neural tangent kernel
– hopfield networks
– information bottleneck theory
– neural -
Video-STaR Enables AI Self-Training on Video Content
By
–
The limitations of traditional labeling methods in video analysis have long stifled AI’s potential. A team of researchers aims to address this with Video-STaR, which enables models to self-train on video content.
-

DeepSeek’s Rise: What It Reveals About Global AI Talent
By
–
What does #DeepSeek’s rise say about AI talent? Scholars @AmyZegart and @idenarch analyzed data on the 233 authors listed on DeepSeek’s research papers. Nearly all were educated or trained in China; half have never left China for school or work; and most of those who trained in
