How do we build AI systems that work equally well for everyone? @Stanford
's @lschmidt3 and @Diyi_Yang are tackling this challenge head-on. Their research focuses on improving training datasets & ensuring AI systems are designed with diverse users in mind. https://
news.stanford.edu/stories/2025/1
1/human-focused-ai-systems-research-ludwigs-schmidt-diyi-yang
…
MACHINE LEARNING
-
Building Fair AI Systems for Diverse Users Research
By
–
-
Asking for error traces on CORE-Agent performance plateau
By
–
Super interesting! I'd love to see why performance starts plateauing with CORE-Agent even with better models – do you have some error traces to share?
-
AI for Rapid Hard Automation Design in Manufacturing
By
–
Agreed! Special-purpose hard automation like this feeder will always dominate general-purpose robots for high volume repetitive tasks like this. How could AI be used to facilitate rapid design of hard automation? https://t.co/6jL4U191jV
— Ken Goldberg (@Ken_Goldberg) 4 décembre 2025Agreed! Special-purpose hard automation like this feeder will always dominate general-purpose robots for high volume repetitive tasks like this. How could AI be used to facilitate rapid design of hard automation?
-
Mistral Large 3 Tops Open Source Coding Model Leaderboard
By
–
Mistral Large 3 debuts as the #1 open source coding model on the @arena leaderboard. We'd love for you to try it! More on coding in a few days… 👀
→ View original post on X — @arthurmensch, 2025-12-04 14:00 UTC
-
Caltech AI+Science Group Presents NeurIPS Papers on Diffusion and LLMs
By
–
Our AI+Science group from @Caltech has a number of papers at @NeurIPSConf in chronological order: FunDPS: Guided Diffusion Sampling on Function Spaces with Applications to PDEs https://
arxiv.org/abs/2505.17004
R-KV: Redundancy-aware KV Cache Compression for Reasoning Models -
MIT Study Shows Ineffective Neural Networks Need Better Starting Points
By
–
Do "untrainable" neural networks just need a helping hand to do modern tasks? MIT study reveals many "ineffective" networks may start from suboptimal starting points. CSAIL's short-term guidance method transferred structural knowledge to boost performance:
-
Veo 3.1 vs Veo 2: 10 months of video generation progress
By
–
Amazing progress in video generation, in 10 months:
— m_ric (@AymericRoucher) 3 décembre 2025
Veo 3.1 (Oct '25) || Veo 2 (Dec '25) pic.twitter.com/SGaRWJ366gAmazing progress in video generation, in 10 months: Veo 3.1 (Oct '25) || Veo 2 (Dec '25)
-
Curiosity Cup 2026: Global Analytics Competition for Students
By
–
Students, it’s time! Registration is now open for the 2026 Curiosity Cup. Take your analytics skills to the next level, compete with students worldwide and unlock incredible opportunities. Register your team by Feb 1:
-
BEHAVIOR Challenge 2025 Winners Announced at NeurIPS
By
–
Are you at #NeurIPS2025 this week? Our colleagues at @StanfordSVL will be announcing the winners of the 2025 BEHAVIOR Challenge, a global competition to stress-test robotic systems against 50 everyday domestic tasks in high-fidelity simulation. Read more:
-
Teaching LLM Architecture and Coding from Scratch
By
–
J’ai formé 100 000 personnes aux mécanismes d’attention, la colonne vertébrale de comment fonctionne ChatGPT, et au code derrière les LLMs. C’est exactement ce que je me dis quand je vois que la vidéo ou je code un LLM de 0 vient de passer les 100k vues. Et ça donne un vrai