Eustace Tilley would nod: Alpha‑Factory v1 now breathes Silver & Sutton’s “Era of Experience.” Multi‑agent α‑AGI that learns by living—cross‑industry ready, deployable in one command. Demo ➜ https://
github.com/MontrealAI/AGI
-Alpha-Agent-v0/tree/main/alpha_factory_v1/demos/era_of_experience
… #AGI #AGIALPHA #ReinforcementLearning
MACHINE LEARNING
-
Alpha-Factory v1: Multi-Agent AGI Learns Through Experience
By
–
-
DeepMind Era of Experience: Reinforcement Learning Advances
By
–
Welcome to the Era of Experience David Silver, Richard S. Sutton: https://
storage.googleapis.com/deepmind-media
/Era-of-Experience/TheEraofExperiencePaper.pdf
… #ArtificialIntelligence #DeepLearning #ReinforcementLearning -
TextArena: Open-Source Competitive Games for LLM Agents
By
–
TextArena went live on Hugging Face It's an open-source collection of competitive text-based games for LLMs, spanning 57+ unique environments Tests for different agentic behaviors—negotiation, theory of mind, deception, via competitive play
-
Does Reinforcement Learning Improve LLM Reasoning Beyond Base Models?
By
–
Does Reinforcement Learning Really Incentivize Reasoning Capacity in LLMs Beyond the Base Model? Yue et al.: https://
arxiv.org/abs/2504.13837 #ArtificialIntelligence #DeepLearning #MachineLearning -
Chain-of-Modality: New Prompting Strategy for Vision-Language Models
By
–
DeepMind and Stanford just dropped a new paper!
— 机器之心 JIQIZHIXIN (@jiqizhixin) 22 avril 2025
They propose Chain-of-Modality, a prompting strategy that enables Vision-Language Models to reason over multimodal human demonstrations — bridging modalities step by step.
A smart way to unlock richer, more grounded reasoning. 🧠📷 pic.twitter.com/fhbZNpd48vDeepMind and Stanford just dropped a new paper!
They propose Chain-of-Modality, a prompting strategy that enables Vision-Language Models to reason over multimodal human demonstrations — bridging modalities step by step.
A smart way to unlock richer, more grounded reasoning. -
Chain-of-Modality: Learning Robot Manipulation from Multimodal Videos
By
–
Chain-of-Modality: Learning Manipulation Programs from Multimodal Human Videos with Vision-Language-Models
Paper: https://
arxiv.org/pdf/2504.13351
v1
…
Page: -
RAG Model Enhancement Through Metadata Extraction and Embedding
By
–
Supercharge your RAG models! Our next live demo takes you through creating functions that extract, structure & tag metadata from documents, & how these insights can fine-tune your embedding model for top-tier performance. Sign up: https://
snorkel.ai/webinar/weekly
-demo-with-qa/
… #AI #RAG -
MCTS-Guided Sample Selection for Data-Efficient Visual Reasoning
By
–
SoTA with Less: MCTS-Guided Sample Selection for Data-Efficient Visual Reasoning Self-Improvement Wang et al.: https://
arxiv.org/abs/2504.07934 #ArtificialIntelligence #DeepLearning #MachineLearning -
What screen-aware AI can do: summarize, recommend, guide, write, answer
By
–
What can it actually do?
— God of Prompt (@godofprompt) 21 avril 2025
• Summarize a full web page
• Recommend products from a list
• Walk you through recipes step-by-step
• Help write a cover letter for a job post
• Answer questions about any page's content
It’s like ChatGPT, but screen-aware. pic.twitter.com/BcvqnJ665hWhat can it actually do? • Summarize a full web page
• Recommend products from a list
• Walk you through recipes step-by-step
• Help write a cover letter for a job post
• Answer questions about any page's content It’s like ChatGPT, but screen-aware. -
Google Unveils DolphinGemma AI to Decode Dolphin Vocalizations
By
–
Google a également dévoilé DolphinGemma, une IA de type speech-to-speech conçue pour analyser les vocalisations des dauphins.
— VISION IA (@vision_ia) 21 avril 2025
Elle décode les sifflements et sons émis pour en repérer les motifs et prédire les sons suivants.
Bonne nouvelle : les poids du modèle seront open… pic.twitter.com/7W8WEdhXIEGoogle a également dévoilé DolphinGemma, une IA de type speech-to-speech conçue pour analyser les vocalisations des dauphins. Elle décode les sifflements et sons émis pour en repérer les motifs et prédire les sons suivants. Bonne nouvelle : les poids du modèle seront open