What if making math problems harder actually boosts AI reasoning? Researchers from Renmin University, Alibaba, and other institutions introduce MathForge—a new method that flips the script. Instead of avoiding tough questions, it actively seeks them out. The approach combines
@jiqizhixin
-
XBridge Architecture Pairs LLMs With Translation Models for Better Results
By
–
Why do even the smartest AI models stumble when speaking your language? Chinese Academy of Sciences presents XBridge — a clever new architecture that pairs a large language model with proven encoder-decoder translation models. Instead of forcing the LLM to handle every
-
Survey Proposes Taxonomy for Feed-Forward 3D Reconstruction
By
–
What if you could reconstruct 3D from 2D in a single forward pass, no per-scene optimization needed? Researchers from Zhejiang University, NTU Singapore, Monash, ETH Zurich & Uni Tübingen present a new survey on feed-forward 3D reconstruction. They propose a taxonomy that
-
Squeeze Evolve Framework Optimizes Model Compute Across Evolution Stages
By
–
Why waste compute on expensive models for every step of evolution? UC Berkeley, UT Austin, Stanford, Princeton, and Together AI introduce Squeeze Evolve: a framework that assigns the strongest models only to high-impact stages while cheaper models handle the rest. This
-
HERMES: Training-Free AI System for Real-Time Video Stream Understanding
By
–
How can we make AI understand live video streams in real-time without draining GPU memory? Researchers from Fudan University, Shanghai Innovation Institute, and the National University of Singapore introduce HERMES. This training-free system reimagines the model internal memory
-
DeepSeek AI Releases Thinking with Visual Primitives Project and Paper
By
–
Project: https://
github.com/deepseek-ai/Th
inking-with-Visual-Primitives
…
Paper: https://
github.com/deepseek-ai/Th
inking-with-Visual-Primitives/blob/main/Thinking_with_Visual_Primitives.pdf
… -
DeepSeek Releases Visual Primitives Reasoning Framework for Grounded Thinking
By
–
Huge! DeepSeek just released "Thinking with Visual Primitives" It's a reasoning framework that lets models “point” with visual markers (points, bounding boxes) while they think. Instead of describing locations in words, the AI grounds each step of its chain-of-thought directly
-
Agent-World: Scaling Real-World Environment Synthesis for General Agent Intelligence
By
–
Agent-World: Scaling Real-World Environment Synthesis for Evolving General Agent Intelligence Paper: https://
arxiv.org/abs/2604.18292
Project: https://
agent-tars-world.github.io/-/ -
Agent-World: Self-Evolving AI Agent Training Arena Using Real APIs
By
–
What if your AI agent could train itself in thousands of real-world environments instead of static datasets? Renmin University of China and ByteDance Seed introduce Agent-World, a self-evolving training arena. It automatically discovers real-world tool ecosystems (like APIs
-
MacTok: AI Image Generation With Just 64 Tokens
By
–
Can AI generate images with just 64 tokens instead of thousands? Researchers from Fudan University introduce MacTok, a new continuous tokenizer. It uses clever image masking and representation alignment to prevent information loss, forcing the model to learn robust visuals