It feels like if the term AI agents is going to stick around, we desperately need similar levels of AI Agents. My proposal…
RESEARCH
-
Proposing Levels of AI Agents Based on Agency
By
–
But the word *agent* suggests the AI has *agency* – the ability to decide what to do. Agency isn’t something that most agents have though. So, here’s what I’m proposing. Levels of AI agents.
-

LUFFY: Off-Policy Reasoning for Zero-Shot Reinforcement Learning
By
–
Learning to Reason under Off-Policy Guidance This paper presents LUFFY (Learning to reason Under oFF-policY guidance), a framework that improves zero-shot reinforcement learning (zero-RL) by incorporating off-policy reasoning traces alongside on-policy exploration, enabling
-

ReDi Framework Improves Image Generation via Joint Feature Synthesis
By
–
Boosting Generative Image Modeling via Joint Image-Feature Synthesis This paper introduces ReDi, a new framework that jointly models low-level VAE latents and high-level semantic features within the diffusion process, significantly improving image generation quality and training
-

MongoDB Atlas GraphRAG Integration for Enhanced LLM Reasoning
By
–
MongoDB Atlas GraphRAG MongoDB Atlas just launched a GraphRAG integration that combines knowledge graphs with LLMs, automatically creating graphs from documents for smarter retrieval and multi-hop reasoning with 2x better accuracy. Build better AI apps with GraphRAG Link
-
AI Scientists Featured in Tim Rocktäschel’s ICLR2025 Keynote on LLM Agents
By
–
AI Scientists mentioned in Tim Rocktäschel’s keynote speech at #ICLR2025 today! Exciting times for pushing the limits of LLM agents in an open-ended world
-

Paper2Code: Automating Code Generation from Scientific Papers
By
–
Paper2Code: Automating Code Generation from Scientific Papers in Machine Learning Seo et al.: https://
arxiv.org/abs/2504.17192 #ArtificialIntelligence #DeepLearning #MachineLearning -

DeepSeek R2 Leaks: 97% Cost Reduction and Ascend 910B Clusters
By
–

DES INFOS FUITENT SUR DEEPSEEK R2 ! Deepseek R2 réduit ses coûts unitaires de 97,3 % et s'apprête à être lancé (rumeur). Utilise un cadre d'entraînement distribué développé en interne Atteint 82 % d'utilisation sur les clusters de puces Ascend 910B Délivre 512 pétaflops en
-

Scaling AI Responsibly: Beyond Flawed Assumptions
By
–
When designing AI systems, blindly scaling a concept without careful engagement risks pushing flawed assumptions into reality—thoughtful collaboration and ongoing refinement are what turn ideas into sustainable, responsible tools. Microblog @antgrasso #AI #DigitalTransformation
-

AI Alignment: Are We Still Aligned With Our Creation?
By
–
𝕀𝔸 : 𝕝𝕖 𝕘𝕣𝕒𝕟𝕕 𝕕é𝕣𝕒𝕚𝕝𝕝𝕖𝕞𝕖𝕟𝕥 La grande question n’est plus seulement "comment aligner les intelligences artificielles"… Mais sommes-nous nous-mêmes encore alignés avec ce que nous construisons ? Bienvenue dans Mes Flashs de la Semaine #37