Cool! AI can win gold at the International Math Olympiad via Simple and Unified Scaling! Researchers from Shanghai AI Lab, CUHK, Tsinghua and PKU introduce SU-01. Their simple recipe: first train on proof-search and self-checking behaviors, then scale via two-stage
@jiqizhixin
-
TaH: Skipping extra reasoning steps makes AI smarter
By
–
Your AI model is thinking too much, and skipping a few steps makes it smarter! Tsinghua University, Infinigence AI, and Shanghai Jiao Tong University introduce Think-at-Hard (TaH). Instead of looping every token through extra reasoning, TaH uses a lightweight decider to skip
-
AmbiSuR: 3D reconstruction facing photometric ambiguities
By
–
What if your 3D surface reconstruction was being fooled by light and shadows? Researchers from Beihang University and National University of Singapore present AmbiSuR – a new framework built on Gaussian Splatting that directly tackles photometric ambiguity. Their method first
-
MP-MoE: Diverse Expert Routing Boosts Large Language Models
By
–
What if your AI model’s experts weren’t just the smartest, but the most diverse? Researchers from Renmin University, Huawei, and Tianjin University introduce MP-MoE: a routing method that selects diverse experts using co-occurrence patterns. Result: 1-3% boost in LLM
-
X Square Robot open-sources Wall-OSS-0.5 for zero-shot robotic manipulation
By
–
X Square Robot today officially open-sourced Wall-OSS-0.5 under the motto "Pretrain Once, Act Anywhere."
— 机器之心 JIQIZHIXIN (@jiqizhixin) 28 mai 2026
Wall-OSS-0.5 is a Vision-Language-Action model for real-world robotic manipulation. According to the team, the pretrained checkpoint shows zero-shot generalization on… https://t.co/QZcUz9ck0RX Square Robot today officially open-sourced Wall-OSS-0.5 under the motto "Pretrain Once, Act Anywhere." Wall-OSS-0.5 is a Vision-Language-Action model for real-world robotic manipulation. According to the team, the pretrained checkpoint shows zero-shot generalization on
-
BIGAI’s NPR enables parallel reasoning in LLMs via self-distilled RL
By
–
What if LLMs could think in multiple directions at once, not just step by step? BIGAI introduces NPR: a teacher-free framework that lets LLMs self-evolve genuine parallel reasoning. Instead of emulating sequential logic, it uses self-distilled reinforcement learning and a
-
Looping transformer layers boosts AI language model efficiency
By
–
What if you could make AI language models smarter by reusing the same layers over and over? Researchers from KAIST, KRAFTON, and UC Berkeley present LoopMDM（Looped Diffusion Language Models）. They selectively loop early-middle transformer layers in masked diffusion models—no
-
Heima Hidden Llama speeds reasoning with abstract thinking tokens
By
–
AI could reason faster by thinking in abstract tokens instead of full sentences! Researchers from Zhejiang University, Adobe, and Northeastern University introduce Heima (Hidden Llama). It compresses lengthy Chain-of-Thought reasoning into a compact set of “thinking tokens,”
-
Side-by-Side Interleaved Reasoning lets AI think silently before sharing certainties
By
–
Your AI could think silently before speaking—and only share what it’s certain of! Researchers from Zhejiang University, William & Mary, UIUC, UBC, CUHK, Fudan, and Stony Brook University introduce Side-by-Side (SxS) Interleaved Reasoning. Instead of forcing every thought into
-
MiniMax releases M2 series technical report for agentic tasks
By
–
MiniMax just released a technical report for the MiniMax-M2 Series! The MiniMax-M2 series, a family of Mixture-of-Experts models designed from the ground up for agentic tasks. The flagship M2 packs 229.9B total parameters but activates only 9.8B per token—thanks to an