Text Degeneration: A Production Failure Mode That Most Benchmarks Do Not Track
Dharma-AI
• • 1NEW Articles from Team or Enterprise organizations will get promoted to the main section.
Why Open Models Are the Only Sustainable Way to Teach AI
penelopegittos
• • 6Scaling Zero-Resource Vocabulary: A Data Pipeline for Sango
MEYNG
• When does autoresearch need a human?
ProlificAI
• LeRobot Humanoid: An Open, Low-Cost, 3D-Printed Humanoid for Robot Learning
VirgileBatto
• • 16SANA-WM Bidirectional on Apple Silicon
junafinity
• • 2Cross-game Slay the Spire 1 and 2 multimodal embeddings with Qwen3-VL: text-as-primary, art-as-secondary
t22000t
• Software Forgets: Agent Traces Are the Memory
huggingface
• • 8Building a prompt-to-deck generator: greedy semantic retrieval with hard feasibility constraints
t22000t
• Computing between models with residual coupling
pfekin
• GOLDENER <br> Smart sampling from embeddings and coreset selection
Yann-CV
• Beyond logs: building a local-first replay layer for AI agents
Spectorfrost123
• AI, Physical AI, World Models, VLA, VLM, and Other Terms We Should Stop Mixing Together
barakor
• • 1The Agentic Kill Chain: A Five-Phase Framework for LLM Security
AlephBeth-AI
• • 1Vividh-ASR: Diagnosing and Fixing Studio-Bias in Whisper for Indic Languages
adalat-ai
• • 11Training-Free Reasoning at 88.89% on GPQA Diamond: How Darwin Family Hit Frontier Scores Without a Single Gradient Step
FINAL-Bench
• • 18How to Comply with SOC 2 and ISO 27001 with Hugging Face: A Practical Guide to AI Model Supply Chain Governance
jeffboudier
• • 5smolagents를 활용한 에이전트 구축
jwlee-ai
• Vocabulary-Augmented Prompting for Sango — Production African Language AI Without a Parallel Corpus
MEYNG
• • 2Slay the Spire on the Hub: a six-dataset bundle for deckbuilding AI/ML
t22000t
•