AI Dynamics

Global AI News Aggregator

@jiqizhixin

  • SwitchCraft: Training-Free Multi-Event Video Generation Framework
    SwitchCraft: Training-Free Multi-Event Video Generation Framework

    By

    @jiqizhixin

    –

    What if AI could generate multi-event videos with perfectly distinct scenes and smooth transitions? Researchers from Westlake University, Duke Kunshan University, and The University of Queensland present SwitchCraft! This training-free framework smartly aligns each video frame's attention to individual events in your prompt. t directs focus precisely and adaptively balances this control to ensure both smooth transitions and visual quality. It dramatically improves prompt alignment, event clarity, and scene consistency, outperforming current baselines and making complex video narratives easy. SwitchCraft: Training-Free Multi-Event Generation with Attention Controls Paper: arxiv.org/abs/2602.23956 Project: switchcraft-project.github.i… Github: github.com/Westlake-AGI-Lab/… Our report: mp.weixin.qq.com/s/Z7D5imbgZ… 📬 #PapersAccepted by Jiqizhixin

    → View original post on X — @jiqizhixin, 2026-04-06 14:20 UTC

  • HiDrop: Efficient Visual Token Reduction for Multimodal LLMs
    HiDrop: Efficient Visual Token Reduction for Multimodal LLMs

    By

    @jiqizhixin

    –

    What if MLLMs could process visual data much faster without sacrificing performance? Eastern Institute of Technology, Ningbo, with USTC, SJTU, and LMU Munich presents HiDrop just for that! This new framework intelligently reduces visual tokens by processing them only when active fusion truly begins (Late Injection) and dynamically pruning them across deeper layers (Concave Pyramid Pruning with Early Exit). It focuses computation where it matters most. HiDrop compresses ~90% of visual tokens, matches original MLLM performance, and accelerates training by 1.72x. A new state-of-the-art for efficient MLLM training & inference! HiDrop: Hierarchical Vision Token Reduction in MLLMs via Late Injection, Concave Pyramid Pruning, and Early Exit Paper: arxiv.org/pdf/2602.23699 Code: github.com/EIT-NLP/HiDrop Our report: mp.weixin.qq.com/s/QKGZ7cFi0… 📬 #PapersAccepted by Jiqizhixin

    → View original post on X — @jiqizhixin, 2026-04-06 10:16 UTC

  • IMMACULATE: Auditing LLM Providers with Verifiable Computation
    IMMACULATE: Auditing LLM Providers with Verifiable Computation

    By

    @jiqizhixin

    –

    Can you really trust your black-box LLM provider with correct inference and honest billing? Researchers from NUS, NTU, and UC Berkeley introduce IMMACULATE. This practical auditing framework uses verifiable computation to randomly check a small fraction of LLM requests. It detects economically motivated cheats like model substitution, quality degradation, and token overbilling without needing trusted hardware or internal model access. IMMACULATE reliably distinguishes honest vs. malicious LLM execution in dense and MoE models, adding less than 1% throughput overhead. IMMACULATE: A Practical LLM Auditing Framework via Verifiable Computation Paper: arxiv.org/pdf/2602.22700 Code: github.com/guo-yanpei/Immacu… Our report: mp.weixin.qq.com/s/WR9nXudXT… 📬 #PapersAccepted by Jiqizhixin

    → View original post on X — @jiqizhixin, 2026-04-06 05:13 UTC

  • SHINE: Hypernetwork Generates LoRA Adapters from Context Instantly
    SHINE: Hypernetwork Generates LoRA Adapters from Context Instantly

    By

    @jiqizhixin

    –

    What if LLMs could instantly absorb new context directly into their parameters? Researchers from Peking University, University of Oxford, Technion, and NVIDIA present SHINE! SHINE is an innovative hypernetwork that, in a single pass, generates high-quality LoRA adapters directly from diverse contexts. This effectively bakes temporary contextual knowledge into the LLM’s core parameters, turning it into lasting skill without any traditional fine-tuning. It smartly reuses the LLM's own frozen parameters for efficiency. This breakthrough dramatically cuts down on time, computation, and memory costs compared to supervised fine-tuning (SFT). SHINE outperforms SFT across various tasks, especially in complex question answering by embedding knowledge directly, offering outstanding performance and massive scalability potential. SHINE: A Scalable In-Context Hypernetwork for Mapping Context to LoRA in a Single Pass arXiv: arxiv.org/abs/2602.06358 GitHub: github.com/Yewei-Liu/SHINE Hugging Face: huggingface.co/collections/Y… Our report: mp.weixin.qq.com/s/sy1L2RoWu… 📬 #PapersAccepted by Jiqizhixin

    → View original post on X — @jiqizhixin, 2026-04-06 01:11 UTC

  • Alchemy: Standardized AI Research Environment from Tsinghua University
    Alchemy: Standardized AI Research Environment from Tsinghua University

    By

    @jiqizhixin

    –

    What's holding back the true scale of automated AI research? Tsinghua University's AI team, led by Asst. Prof. Jia Li, with key contributions from Lehui Li and Liyi Cai, introduces Alchemy, a standardized research environment. It's a 'pre-built AI furnace' that unifies all complex engineering tasks—from data processing to resource scheduling. AI scientists simply provide their algorithm (.py) and hyperparameters (.yaml) to run full experiments. This frees AI scientists from engineering burdens, drastically boosting the scale and efficiency of automated AI research across domains like recommender systems, time series, and graph learning, enabling high-concurrency experiments and continuous integration for new tasks. Code: github.com/TsinghuaISE/Alche… Our report: mp.weixin.qq.com/s/FyhCjhZ4l… 📬 #PapersAccepted by Jiqizhixin

    → View original post on X — @jiqizhixin, 2026-04-05 18:08 UTC

  • Huawei’s Smart Noise Scheduler Improves Diffusion LLM Logic Learning
    Huawei’s Smart Noise Scheduler Improves Diffusion LLM Logic Learning

    By

    @jiqizhixin

    –

    What if Diffusion LLMs learned logic more efficiently? Huawei's Noah's Ark Lab proposes a breakthrough: their "smart noise scheduler" uses priority masking to focus training on only information-dense data, making DLLMs master core reasoning and structure. This boosts average accuracy by 4% on Code & Math reasoning, beating uniform baselines and unlocking DLLM potential. Mask Is What DLLM Needs: A Masked Data Training Paradigm for Diffusion LLMs Paper: arxiv.org/abs/2603.15803 Dataset: huggingface.co/datasets/malr… Our report: mp.weixin.qq.com/s/1yTd36hev… 📬 #PapersAccepted by Jiqizhixin

    → View original post on X — @jiqizhixin, 2026-04-05 14:05 UTC

  • HiFi-Inpaint: ByteDance AI Framework for Detail-Preserving Product Images
    HiFi-Inpaint: ByteDance AI Framework for Detail-Preserving Product Images

    By

    @jiqizhixin

    –

    How do you get AI to create stunning product images without losing crucial details? ByteDance and a collaboration of top universities present HiFi-Inpaint. This novel AI framework employs 'Shared Enhancement Attention' to meticulously refine fine-grained product features and 'Detail-Aware Loss' for pixel-perfect guidance, supported by a new large-scale dataset, HP-Image-40K. HiFi-Inpaint achieves state-of-the-art performance, generating human-product images with unprecedented detail preservation, set to transform digital marketing and e-commerce visuals. HiFi-Inpaint: Towards High-Fidelity Reference-Based Inpainting for Generating Detail-Preserving Human-Product Images Paper: arxiv.org/abs/2603.02210  Project: correr-zhou.github.io/HiFi-I… Our report: mp.weixin.qq.com/s/xoIJU4fBc… 📬 #PapersAccepted by Jiqizhixin

    → View original post on X — @jiqizhixin, 2026-04-05 07:02 UTC

  • SWE-MiniSandbox: Container-Free RL for Software Engineering Agents
    SWE-MiniSandbox: Container-Free RL for Software Engineering Agents

    By

    @jiqizhixin

    –

    What if you could train AI software engineers faster, without the heavy overhead of containers? Researchers from Peking University, Ant Groupe, and The University of Hong Kong present SWE-MiniSandbox. This novel, container-free method uses kernel-level isolation and lightweight pre-caching, eliminating bulky container images for reinforcement learning. It achieves comparable performance to container-based pipelines while reducing disk usage by 95% and environment setup time by 75%, making scalable RL training far more accessible for software engineering agents. SWE-MiniSandbox: Container-Free Reinforcement Learning for Building Software Engineering Agents Paper: arxiv.org/abs/2602.11210 Code: github.com/lblankl/SWE-MiniS… Docs: lblankl.github.io/SWE-MiniSa… Our report: mp.weixin.qq.com/s/NlQLprZmM… 📬 #PapersAccepted by Jiqizhixin

    → View original post on X — @jiqizhixin, 2026-04-05 04:00 UTC

  • STAGE: AI Multi-shot Video Generation with Narrative Consistency
    STAGE: AI Multi-shot Video Generation with Narrative Consistency

    By

    @jiqizhixin

    –

    Can AI finally create multi-shot videos with seamless narrative flow and cinematic flair? Researchers from Beijing University of Posts and Telecommunications and Peking University present STAGE. This new method rethinks video generation by planning full shot-by-shot storyboards using start-to-end frame pairs. It employs smart memory and clever encoding to keep characters and scenes consistent, ensuring smooth visuals within and between shots. STAGE significantly outperforms existing methods, achieving superior narrative control and unparalleled visual consistency across cinematic sequences. STAGE: Storyboard-Anchored Generation for Cinematic Multi-shot Narrative Paper: arxiv.org/abs/2512.12372 Code: github.com/escapistmost/Stor… Our report: mp.weixin.qq.com/s/rmeF2tbIu… 📬 #PapersAccepted by Jiqizhixin

    → View original post on X — @jiqizhixin, 2026-04-04 18:55 UTC

  • EasySteer: Unified Framework for High-Performance LLM Steering
    EasySteer: Unified Framework for High-Performance LLM Steering

    By

    @jiqizhixin

    –

    Want to precisely control your LLM's behavior without expensive retraining? New research from Zhejiang University unveils EasySteer, a unified framework for high-performance and extensible LLM steering. It's a unified framework that lets you finely tune LLM responses in real-time by subtly adjusting their internal 'thoughts' or hidden states. This lightweight method offers modular control and pre-computed steering options, sidestepping costly model retraining. EasySteer achieves a game-changing 10.8-22.3x speedup over current methods. It dramatically reduces common LLM issues like overthinking and hallucinations, making advanced steering a robust, production-ready tool for deployable, controllable language models. EasySteer: A Unified Framework for High-Performance and Extensible LLM Steering Paper: arxiv.org/abs/2509.25175 Code: github.com/ZJU-REAL/EasyStee… Our report: mp.weixin.qq.com/s/dxuJHvXOf… 📬 #PapersAccepted by Jiqizhixin

    → View original post on X — @jiqizhixin, 2026-04-04 14:53 UTC