University of Science and Technology of China
AI overview
Sign in and the AI will write an overview from our coverage.
Headlines · 4
- AREX-2 trains a 27B model with long-horizon reflection
The AREX research group released AREX-2, a model based on Qwen3.8-27B and trained with long-horizon reflection data for machine-learning and algorithm-programming tasks. The team reported scores of 81.8 on MLE-bench Lite, 70.7 on Frontier-CS, 84.0 on BrowseComp and 52.6 on HLE. It also said the training helped capabilities transfer to deep research and scientific problem-solving. The report presents long-horizon reflection as a trainable and verifiable approach to recursive self-improvement, which is the team’s research interpretation rather than established proof of that capability.
机器之心 · 🔥 0 - MEMCUT tests whether AI agents truly forget deleted secrets
Researchers from Hong Kong University of Science and Technology (Guangzhou), the University of Science and Technology of China, and Hong Kong University of Science and Technology built MEMCUT to evaluate user-requested forgetting in AI agent memory. Across five memory architectures and four large language models, 85%–99% of secrets remained recoverable after only direct disclosures were deleted, while LLM-assisted cleanup still left 39%–76% derivable from residual evidence. The benchmark distinguishes native recovery through a system’s retriever from residual recovery when all surviving memory records are provided.
36氪 人工智能 · 🔥 0 - PAIR 2026 panel weighs what AI can do for quantum computing, and vice versa
At PAIR Conference 2026 in Shanghai on September 20–21, a roundtable examined how AI can aid quantum R&D and what quantum computing might offer AI. Hu Hongye said the Institute of Physics and the Beijing Academy of Quantum Information Sciences achieved automated calibration at more than 100 qubits without continuous experimenter involvement earlier this year, and that AI is being explored for quantum compilation, verification and error-correction optimization; he noted a Google AI control study required about 10^5–10^6 reinforcement-learning samples. Yan Zheng said strongly correlated many-body problems still need quantum computing advances, Wu Anbang expected AI agents to speed quantum software R&D in two to three years, and Hu's team proved a polynomial memory advantage for quantum generative models in specific tasks.
MIT科技评论中文 · 🔥 0 - Baize Tongjing releases abductive world model AWM and edge engine BAIZ-RUN
Baize Tongjing, a Hefei-based company founded less than six months ago by a team from USTC, has published a paper titled "Abductive World Modeling via Causal Representation Learning" and open-sourced its code, saying its AWM model not only predicts future states but also asks what caused them, decomposing a scene into entities, dynamics and relations. The company says AWM reached a 75% success rate on the Push-T robotic manipulation task with only 2,000 training trajectories, versus 14,728 for the comparison method Causal-JEPA, and about 92% when given the same 14,728 trajectories. Its companion edge inference engine BAIZ-RUN quantizes weights to 2 bit and activations to 4 bit; the team reports a 2.56x reduction in memory use, a 94.6% increase in inference frequency and under 1% accuracy loss on an AMD Radeon edge integrated GPU with the SSV2 dataset. The company also says it has signed and validated three industrial scenarios — loading/unloading, quality inspection and plant patrol — and closed a first funding round of tens of millions of RMB within three months of founding, with backers including SMIC Capital and a market-oriented fund whose LP is Ke Da Silicon Valley.
机器之心 · 🔥 0
Experience and discussion from the community
Share my University of Science and Technology of China experienceAsk about University of Science and Technology of China
Nobody has shared their experience with University of Science and Technology of China yet.