Unzip

Unzip

By Skyler @ LimitLess AITechnologyTech News
Download on the App Store

Unzip episodes

  • Periodic Weak Spots: Phase Sensitivity from Chunked KV-Cache Compression
    ## Episode Summary
    In this episode, we cover:
    - **Periodic Weak Spots: Phase Sensitivity from Chunked KV-Cache Compression** (Hugging Face Daily)
    - [Read more](https://huggingface.co/papers/2609.36322)
    - **Org-Agent: Beyond Personal Assistants Towards Organizational Agents** (Hugging Face Daily)
    - [Read more](https://huggingface.co/papers/2609.34392)
    - **Routing Should Pay for Itself: Sparse Supervision for Economical LLM Routing** (Hugging Face Daily)
    - [Read more](https://huggingface.co/papers/2609.37402)
    - **Scheduling Recursive Reasoning in Looped Transformers** (Hugging Face Daily)
    - [Read more](https://huggingface.co/papers/2609.36653)
    - **PreviewDiff: Multimodal Critic-Guided Search over Diffusion Latents** (Hugging Face Daily)
    - [Read more](https://huggingface.co/papers/2609.36199)
    ---
    *Sponsored by Ada, Ago Consulting, and Zen Semiconductor*
    11 min
  • Scaling Long-Form Story Generation via Narrative State Tracking
    ## Episode Summary
    In this episode, we cover:
    - **Scaling Long-Form Story Generation via Narrative State Tracking** (arXiv)
    - [Read more](http://arxiv.org/abs/2609.35759v1)
    - **TokenCast: Forecasting Token Consumption During LLM Agent Execution** (arXiv)
    - [Read more](http://arxiv.org/abs/2609.35760v1)
    - **ExpVoyager: Direct Experience Navigation for Dynamic Agent Skill Synthesis** (Hugging Face Daily)
    - [Read more](https://huggingface.co/papers/2609.32630)
    - **Distillation Defenses Easily Break After Reinforcement Learning** (Hugging Face Daily)
    - [Read more](https://huggingface.co/papers/2609.35699)
    - **Reinforcing Agentic Creativity in Scientific Ideation with Night Science** (Hugging Face Daily)
    - [Read more](https://huggingface.co/papers/2609.35706)
    ---
    *Sponsored by Ada, Ago Consulting, and Zen Semiconductor*
    0 min
  • IndicBankBench: Evaluating Safety and Reliability of Language Model Assistants in Indian Retail Banking
    ## Episode Summary
    In this episode, we cover:
    - **IndicBankBench: Evaluating Safety and Reliability of Language Model Assistants in Indian Retail Banking** (Hugging Face Daily)
    - [Read more](https://huggingface.co/papers/2609.29167)
    - **DeepEdu-v1: Efficient and Scalable Agentic LLMs for Vietnamese Education** (arXiv)
    - [Read more](http://arxiv.org/abs/2609.31568v1)
    - **Enhancing Photogrammetric Digital Surface Models with Pretrained Diffusion Models and Multimodal Conditioning** (Hugging Face Daily)
    - [Read more](https://huggingface.co/papers/2609.31199)
    - **FuseReg: Regularizing Layer Fusion Mitigates the Reconstruction-Generation Gap in Representation Autoencoders** (arXiv)
    - [Read more](http://arxiv.org/abs/2609.31620v1)
    - **SAGE: Mitigating Long-Horizon Reasoning Biases via Topological Guidance** (Hugging Face Daily)
    - [Read more](https://huggingface.co/papers/2609.30192)
    ---
    *Sponsored by Ada, Ago Consulting, and Zen Semiconductor*
    15 min
  • Your Transformer Can Hold Two Thoughts at Once: Evidence of Linear Superposition in LLMs
    ## Episode Summary
    In this episode, we cover:
    - **Your Transformer Can Hold Two Thoughts at Once: Evidence of Linear Superposition in LLMs** (Hugging Face Daily)
    - [Read more](https://huggingface.co/papers/2609.29845)
    - **To Trust or Not to Trust: Retrieval-Augmented Fact Checking in Speech** (arXiv)
    - [Read more](http://arxiv.org/abs/2609.30227v1)
    - **OmniEcho: Spatial Audio Understanding for Embodied Agents** (Hugging Face Daily)
    - [Read more](https://huggingface.co/papers/2609.23407)
    - **RAPID: Robot Agentic Programming from Demonstrations** (arXiv)
    - [Read more](http://arxiv.org/abs/2609.30249v1)
    - **SemMSA: Latent Semantic-Aided Robust Multimodal Sentiment Analysis with Incomplete Data** (arXiv)
    - [Read more](http://arxiv.org/abs/2609.30238v1)
    ---
    *Sponsored by Ada, Ago Consulting, and Zen Semiconductor*
    14 min
  • Neural Spectral Capacity: Measuring and Designing Architectures from Network Specification Alone
    ## Episode Summary
    In this episode, we cover:
    - **Neural Spectral Capacity: Measuring and Designing Architectures from Network Specification Alone** (Hugging Face Daily)
    - [Read more](https://huggingface.co/papers/2609.23087)
    - **AV-GRPO: Modality-Anchored Decoupling Diffusion Reinforcement Learning for Joint Audio-Video Generation** (Hugging Face Daily)
    - [Read more](https://huggingface.co/papers/2609.29816)
    - **Coding Agents for Generalized Task and Motion Planning Problems** (Hugging Face Daily)
    - [Read more](https://huggingface.co/papers/2609.30233)
    - **ViRDM: Taming Representation Distribution Matching for Few-Step Causal Video Generation** (Hugging Face Daily)
    - [Read more](https://huggingface.co/papers/2609.28923)
    - **Instrumental Monitor Evasion Emerges Under Ordinary Task Pressure** (arXiv)
    - [Read more](http://arxiv.org/abs/2609.30217v1)
    ---
    *Sponsored by Ada, Ago Consulting, and Zen Semiconductor*
    0 min
  • RGBD20K: A Large-Scale Benchmark for RGB-D Semantic Segmentation
    ## Episode Summary
    In this episode, we cover:
    - **RGBD20K: A Large-Scale Benchmark for RGB-D Semantic Segmentation** (Hugging Face Daily)
    - [Read more](https://huggingface.co/papers/2609.29028)
    - **Just Ask Jev: Reinforcement Learning for Calibrated Decisions as a Zero-Shot Detector of AI Alignment Failures** (Hugging Face Daily)
    - [Read more](https://huggingface.co/papers/2609.29429)
    - **LLM Agents Can Easily Tamper With Their Own Traces** (arXiv)
    - [Read more](http://arxiv.org/abs/2609.30266v1)
    - **IterSynth: Rethinking Deep Search Agents via Role-Decoupled Iterative Synthesis** (Hugging Face Daily)
    - [Read more](https://huggingface.co/papers/2609.29444)
    - **Learning to Discover Interesting Mathematics** (Hugging Face Daily)
    - [Read more](https://huggingface.co/papers/2609.28603)
    ---
    *Sponsored by Ada, Ago Consulting, and Zen Semiconductor*
    12 min
  • FAMOS: Feed-Forward 3D Articulation Modeling from Sparse Observations
    ## Episode Summary
    In this episode, we cover:
    - **FAMOS: Feed-Forward 3D Articulation Modeling from Sparse Observations** (Hugging Face Daily)
    - [Read more](https://huggingface.co/papers/2609.20817)
    - **When2Think: Learning Difficulty-Aware Length Control for Efficient Hybrid Reasoning Models** (Hugging Face Daily)
    - [Read more](https://huggingface.co/papers/2609.19671)
    - **Don't Mask the Environment: Observation Supervision Changes How Agents Explore Under RL** (Hugging Face Daily)
    - [Read more](https://huggingface.co/papers/2609.20715)
    - **Sample Count Is Not Enough: Candidate-Generation Strategy Shapes the Energy and Performance of LLM Test-Time Scaling** (Hugging Face Daily)
    - [Read more](https://huggingface.co/papers/2609.19499)
    - **When EOS Tokens Disagree: Understanding Length Inflation in On-Policy Distillation** (Hugging Face Daily)
    - [Read more](https://huggingface.co/papers/2609.20511)
    ---
    *Sponsored by Ada, Ago Consulting, and Zen Semiconductor*
    14 min
  • Self-Evolving Search Index
    ## Episode Summary
    In this episode, we cover:
    - **Self-Evolving Search Index** (Hugging Face Daily)
    - [Read more](https://huggingface.co/papers/2609.19656)
    - **Quantifying Overclaiming Propensity in Frontier LLM Agents** (arXiv)
    - [Read more](http://arxiv.org/abs/2609.20812v1)
    - **Reflect, Revise, Reuse: Training-Free Skill Evolution for GUI Agents** (Hugging Face Daily)
    - [Read more](https://huggingface.co/papers/2609.17653)
    - **PACT: Can Enterprise AI Assistants Be Trusted Under Pressure?** (Hugging Face Daily)
    - [Read more](https://huggingface.co/papers/2609.18605)
    - **RetireOPD: Self-Retiring On-Policy Distillation for Agentic Reinforcement Learning** (Hugging Face Daily)
    - [Read more](https://huggingface.co/papers/2609.20784)
    ---
    *Sponsored by Ada, Ago Consulting, and Zen Semiconductor*
    11 min
  • Rethinking Critic Learning in PPO: Understanding and Mitigating Value Flattening
    ## Episode Summary
    In this episode, we cover:
    - **Rethinking Critic Learning in PPO: Understanding and Mitigating Value Flattening** (Hugging Face Daily)
    - [Read more](https://huggingface.co/papers/2609.18708)
    - **Fathom: Per-Query Read Depth for Sparse Decoding over Offloaded KV Caches** (Hugging Face Daily)
    - [Read more](https://huggingface.co/papers/2609.17652)
    - **Evidence-Grounded Agentic Formulation Development in an Autonomous Laboratory** (arXiv)
    - [Read more](http://arxiv.org/abs/2609.19099v1)
    - **Confidence Comes from Experience: Experiential Confidence Estimation from Reasoning to Agents** (Hugging Face Daily)
    - [Read more](https://huggingface.co/papers/2609.17708)
    - **ActionPiece: Rethinking Action Tokenization for Autoregressive Vision-Language-Action Models** (Hugging Face Daily)
    - [Read more](https://huggingface.co/papers/2609.18487)
    ---
    *Sponsored by Ada, Ago Consulting, and Zen Semiconductor*
    18 min
  • The Last AI Built by Humans: Toward Genuine Recursive Self-Improvement
    ## Episode Summary
    In this episode, we cover:
    - **The Last AI Built by Humans: Toward Genuine Recursive Self-Improvement** (Hugging Face Daily)
    - [Read more](https://huggingface.co/papers/2609.11873)
    - **Verifiable Social Reasoning for LLM Assistants** (arXiv)
    - [Read more](http://arxiv.org/abs/2609.17496v1)
    - **StepAudio 3 Realtime Technical Report** (Hugging Face Daily)
    - [Read more](https://huggingface.co/papers/2609.14005)
    - **StepAudio 3 Music Technical Report** (Hugging Face Daily)
    - [Read more](https://huggingface.co/papers/2609.16034)
    - **What Breaks Under Pruning in Smart Homes, and When? Evaluating LLM Degradation Across Architectures and Task Complexity** (arXiv)
    - [Read more](http://arxiv.org/abs/2609.17515v1)
    ---
    *Sponsored by Ada, Ago Consulting, and Zen Semiconductor*
    0 min

About Unzip

From the publisher's feed

Your guide to the latest AI and machine learning research. We unpack complex papers into actionable insights for practitioners and enthusiasts alike.