DPT-Agent's System 1 employs a Finite-state Machine (FSM) and code-as-policy for fast, intuitive, and controllable decision-making. System 2 integrates Theory of Mind (ToM) and asynchronous reflection to infer human intentions and make reasoning-based autonomous decisions.
@jiqizhixin
-
DPT-Agent Enables Real-Time Human-AI Collaboration in Overcooked
By
–
You can now play Overcooked with AI agents. New research from SJTU validates the necessity of Dual Process Theory (DPT) in real-time tasks. Based on their discovery, they built DPT-Agent, an agent framework that integrates System 1 and 2 for real-time human-AI collaboration. pic.twitter.com/iJ9XzwiKKK
— 机器之心 JIQIZHIXIN (@jiqizhixin) 19 mars 2025You can now play Overcooked with AI agents. New research from SJTU validates the necessity of Dual Process Theory (DPT) in real-time tasks. Based on their discovery, they built DPT-Agent, an agent framework that integrates System 1 and 2 for real-time human-AI collaboration.
-
xLSTM-7b: Advanced LSTM Model Released with Code
By
–
Model: https://
huggingface.co/NX-AI/xLSTM-7b
Code: https://
github.com/NX-AI/xlstm and https://
github.com/NX-AI/xlstm-jax
Paper: https://
arxiv.org/pdf/2503.13427 -
xLSTM 7B: Faster, More Efficient Open-Source LLM Alternative
By
–
If you want faster and more efficient LLMs, Transformers might not be your best choice. xLSTM 7B is a new open-source LLM that leverages xLSTM’s architectural advantages with targeted optimizations for speed and efficiency in inference.
-
Impossible Videos: Generating Physically Impossible Content with AI
By
–
Paper: https://
arxiv.org/pdf/2503.14378
Project Page: https://
showlab.github.io/Impossible-Vid
eos/
…
GitHub: https://
github.com/showlab/Imposs
ible-Videos
…
HuggingFace Dataset: -
NUS Show Lab Releases Impossible Videos for Hyper-Realistic Generation
By
–
What a visual feast! NUS's Show Lab has released Impossible Videos, capable of generating hyper-realistic videos that defy common sense. pic.twitter.com/Ot4EfDBiHS
— 机器之心 JIQIZHIXIN (@jiqizhixin) 19 mars 2025What a visual feast! NUS's Show Lab has released Impossible Videos, capable of generating hyper-realistic videos that defy common sense.
-
Unitree G1 Achieves First-Ever Humanoid Robot Side Flip
By
–
Breaking! Unitree G1 just performed the first-ever side flip in humanoid robot history! https://t.co/5iOOpfogeh pic.twitter.com/BTp3tmp8U2
— 机器之心 JIQIZHIXIN (@jiqizhixin) 19 mars 2025Breaking! Unitree G1 just performed the first-ever side flip in humanoid robot history!
-
NVIDIA Cosmos: Advanced Video Prediction Model Released
By
–
GitHub: https://
github.com/nvidia-cosmos/
cosmos-predict1
…
Paper: https://
arxiv.org/pdf/2501.03575 -
NVIDIA Open Sources Cosmos World Foundation Model for Physical AI
By
–
NVIDIA just updated the tech report on the Cosmos World Foundation Model Platform for Physical AI—and it’s now open source.
-
Token Injection Enables Layout-Guided Generative AI Image Composition
By
–
This new method enables: (a) layout-guided generation by translating token-injected regions, (b) multi-subject composition through sequential token injection, and (c) inpainting and outpainting via specifying masks and increased replacement.