TokensGen
— AK (@_akhaliq) 22 juillet 2025
Harnessing Condensed Tokens for Long Video Generation pic.twitter.com/mtFnloBkmA
TokensGen Harnessing Condensed Tokens for Long Generation
MACHINE LEARNING
-
TokensGen: Condensed Tokens for Extended Video Generation
By
–
-
SeC: Progressive Concept Construction for Video Object Segmentation
By
–
SeC
— AK (@_akhaliq) 22 juillet 2025
Advancing Complex Video Object Segmentation via Progressive Concept Construction pic.twitter.com/qsnpJQH70sSeC Advancing Complex Object Segmentation via Progressive Concept Construction
-
MiroMind-M1: Open-Source Mathematical Reasoning with Policy Optimization
By
–
MiroMind-M1 An Open-Source Advancement in Mathematical Reasoning via Context-Aware Multi-Stage Policy Optimization
-
AI Surpasses Human Analysts in Real-World Professional Tasks
By
–
It’s already beating humans at work. Benchmarks show:
– Better than human analysts in many real-world tasks
– SOTA on web research, spreadsheet editing, and data science
– Can build financial models, marketing slides, and more This isn't just smart — it's capable. -
Distinguishing Open Source and AI Model Deployment Costs
By
–
Je pense que bcp confonde Open Source et « utilisation gratuite et accessible pour tous». Ça n’a rien à voir. Un modèle 235-B il vous faut un Cluster de GPU à 50k€ pour le faire tourner chez vous. Mais vous pouvez le distiller, le fine tuner, et le mettre sur des
-
Reasoning Methods in Large Language Models Explained
By
–
Non les méthodes de raisonnement sur les LLMs c’est en ultra bref c’est pouvoir bouclé et adapté sa réponse en se répondant à lui même. Sans raisonnement ça veut tu lui donnes une question et il répond direct sans tourner 7 fois la langue dans sa bouche
-
Technology and Trends Prediction for 2021
By
–
New on the Blog: TECHNOLOGY & TRENDS PREDICTION FOR 2021 Stay ahead in the ever-evolving world of Digital Transformation. This article explores key insights into emerging technologies and their impact on today's business landscape. Discover how to leverage innovation to
-
Why LLM Analysis Becomes Harder with Each Generation
By
–
the reason llm analysis (and regulation, and PMing) is hard* is that the relevant DIMENSIONS keep moving with each generation of frontier model; it is not enough to just put your x or y axis in log scale and track scaling laws, you have to actually do the work to think about how
-
Stabilizing Knowledge and Promoting Reasoning with Dual-Token Constraints
By
–
Stabilizing Knowledge, Promoting Reasoning Dual-Token Constraints for RLVR
-
GUI-G²: Gaussian Reward Modeling for GUI Grounding
By
–
GUI-G^2 Gaussian Reward Modeling for GUI Grounding