Finally, a survey on Agentic Large Language Models! Link: https://
arxiv.org/pdf/2503.23037
@jiqizhixin
-
New Survey on Agentic Large Language Models Released
By
–
-
CPPO Accelerates Group Relative Policy Optimization Training
By
–
CPPO: Accelerating the Training of Group Relative Policy Optimization-Based Reasoning Models
Paper: https://
arxiv.org/pdf/2503.22342
Code: https://
github.com/lzhxmu/CPPO -
CPPO Boosts GRPO Speed by 8x on Mathematical Reasoning
By
–
GRPO just got a speed boost! Xiamen University introduced Completion Pruning Policy Optimization (CPPO), which significantly reduces the number of gradient calculations and updates.
How fast? On GSM8K, it's 8.32× faster than GRPO, and on MATH, the speedup is 3.51×. -
GPT-4o and Kling recreate Empresses in Palace in Ghibli style
By
–
Using GPT-4o and Kling, we recreated a famous scene from Empresses in the Palace (甄嬛传) in Ghibli style! Check it out: https://
youtu.be/xNTt-rhA_zE?si
=cdFFlVSSZX169OHQ
… -
OpenAI’s Superintelligence Prediction: Two Years Later Assessment
By
–
Almost two years ago, OpenAI said, "within the next ten years, AI systems will exceed expert skill level in most domains and carry out as much productive activity as one of today’s largest corporations." How do you think this prediction is holding up now? https://
openai.com/index/governan
ce-of-superintelligence/
… -
VGG-T: Advanced Visual Foundation Model from Facebook Research
By
–
project page: https://
vgg-t.github.io
paper：
https://
arxiv.org/abs/2503.11651
code：
https://
github.com/facebookresear
ch/vggt
…
demo：
https://
huggingface.co/spaces/faceboo
k/vggt
… -
Feed-Forward Networks Infer 3D Scene Attributes Directly
By
–
Can a feed-forward neural network directly infer all key 3D attributes of a scene? This CVPR 2025 paper says YES.
— 机器之心 JIQIZHIXIN (@jiqizhixin) 28 mars 2025
The proposed VGGT can directly infer camera parameters, point maps, depth maps, and 3D point tracks from one, a few, or even hundreds of scene views. 🚀🔍 pic.twitter.com/Gm3z2ZAn6RCan a feed-forward neural network directly infer all key 3D attributes of a scene? This CVPR 2025 paper says YES.
The proposed VGGT can directly infer camera parameters, point maps, depth maps, and 3D point tracks from one, a few, or even hundreds of scene views. -
AI Infrastructure Bullish Case: NVIDIA’s Strategic Position
By
–
These two tweets are exactly why you should be bullish on AI infrastructure—and, of course, bullish on NVIDIA.
