Unzip

## Episode Summary

In this episode, we cover:

- **SKILLER: Language-Level Reinforcement Learning for Reusable Skill Extraction in Small Language Models** (Hugging Face Daily)

- [Read more](https://huggingface.co/papers/2608.10538)

- **Thought-Level Beam Search for Reasoning** (Hugging Face Daily)

- [Read more](https://huggingface.co/papers/2608.08020)

- **Maglev: Sliding Recurrent Memory** (Hugging Face Daily)

- [Read more](https://huggingface.co/papers/2608.02870)

- **HumanTracker: Towards Comprehensive and Human-Aligned Motion Tracking Benchmark** (arXiv)

- [Read more](http://arxiv.org/abs/2608.13555v1)

- **Specification-first convergence with an AI coding agent: a case study of dismantling a core architectural invariant across 189 files in a 717k-line codebase with no test oracle and no human code review** (Hugging Face Daily)

- [Read more](https://huggingface.co/papers/2608.12440)

---

*Sponsored by Ada, Ago Consulting, and Zen Semiconductor*

## Episode Summary

In this episode, we cover:

- **Mitigating Gender Bias in English to Romanian Machine Translation** (Hugging Face Daily)

- [Read more](https://huggingface.co/papers/2608.08606)

- **Knowing When to Quit: Diagnosing and Training LLMs to Abort Futile Reasoning** (Hugging Face Daily)

- [Read more](https://huggingface.co/papers/2607.29211)

- **QuoteBench: How Matched Scores Can Hide Command-Path Failures** (arXiv)

- [Read more](http://arxiv.org/abs/2608.13547v1)

- **Vero: Can AI Agents Build Formally Verified Software Repositories?** (arXiv)

- [Read more](http://arxiv.org/abs/2608.13522v1)

- **PlayWorld: Benchmarking World Models with Agent Players over Long-Horizon Objectives** (arXiv)

- [Read more](http://arxiv.org/abs/2608.13552v1)

---

*Sponsored by Ada, Ago Consulting, and Zen Semiconductor*

## Episode Summary

In this episode, we cover:

- **Mechanist: AI as a Scientific Instrument for Discovering the Mechanisms of Intelligence** (Hugging Face Daily)

- [Read more](https://huggingface.co/papers/2608.12036)

- **Learning How the World Evolves: Extrapolative Video World Models via Latent Dynamics Reasoning** (Hugging Face Daily)

- [Read more](https://huggingface.co/papers/2608.09926)

- **SkillZip: Contract-Preserving Graph Compression for Scalable Agent Skill Libraries** (Hugging Face Daily)

- [Read more](https://huggingface.co/papers/2608.05604)

- **Convergent Detour Hijacking: Task-Preserving Resource Amplification in Skill-Based LLM Agents** (arXiv)

- [Read more](http://arxiv.org/abs/2608.12273v1)

- **DreamFly: Causal Memory and Receding-Horizon Diffusion Planning for Aerial Vision-Language Navigation** (arXiv)

- [Read more](http://arxiv.org/abs/2608.12308v1)

---

*Sponsored by Ada, Ago Consulting, and Zen Semiconductor*

## Episode Summary

In this episode, we cover:

- **Can LLM Agents Stick to the Script? A Benchmark for Long-Horizon Consistency in Interactive Narratives** (Hugging Face Daily)

- [Read more](https://huggingface.co/papers/2608.08160)

- **Diagram-MMU: A Multi-Modal Benchmark for Scientific Diagrams** (arXiv)

- [Read more](http://arxiv.org/abs/2608.12262v1)

- **The Illusion of Visual Tool-Use: A Causal Audit of Thinking with Images** (Hugging Face Daily)

- [Read more](https://huggingface.co/papers/2608.06270)

- **Ready Cohorts: Bounding GPU Opportunity and Avoiding Host Round Trips in LLM-Agent Control** (Hugging Face Daily)

- [Read more](https://huggingface.co/papers/2608.12123)

- **VAKRA: Evaluating Multi-Hop Reasoning Across APIs and Retrieval Under Tool-Use Policies** (arXiv)

- [Read more](http://arxiv.org/abs/2608.12282v1)

---

*Sponsored by Ada, Ago Consulting, and Zen Semiconductor*

## Episode Summary

In this episode, we cover:

- **ContextMaster: Interactive Multi-Shot Video Creation via Fixed-Budget Sparse Context Routing** (Hugging Face Daily)

- [Read more](https://huggingface.co/papers/2608.04956)

- **MASS: Multiplayer World Models with Authoritative Shared State** (Hugging Face Daily)

- [Read more](https://huggingface.co/papers/2608.06257)

- **KVAE: Family of Tokenizers for Multimodal Generative Models** (Hugging Face Daily)

- [Read more](https://huggingface.co/papers/2608.05798)

- **FactorJEPA: Factorizing Monolithic Futures into Layout-Agent-Interaction Channels for Crowded and Chaotic Global South Urban Worlds** (Hugging Face Daily)

- [Read more](https://huggingface.co/papers/2608.01049)

- **The Bitter Lesson of Tool Calling** (arXiv)

- [Read more](http://arxiv.org/abs/2608.06370v1)

---

*Sponsored by Ada, Ago Consulting, and Zen Semiconductor*

## Episode Summary

In this episode, we cover:

- **Interpretable MEG Decoding of Perceived Speech: Cortical Sources and the Stimulus Features That Drive Retrieval** (Hugging Face Daily)

- [Read more](https://huggingface.co/papers/2608.01481)

- **TRAJDEBUG: Tracing Error Lifecycle to Identify Critical Failures in Long-Horizon Agent Trajectories** (arXiv)

- [Read more](http://arxiv.org/abs/2608.06346v1)

- **Activity Frames: Deterministic Screen-Activity Compilation for Agent Memory and Replay** (Hugging Face Daily)

- [Read more](https://huggingface.co/papers/2608.05784)

- **EffectLearner: World-Aware Object-Effect Reasoning for Real-World Video Object Removal** (Hugging Face Daily)

- [Read more](https://huggingface.co/papers/2608.05565)

- **From Economic Agents to Agentic Economies: A Systems Blueprint for Economic World Models** (Hugging Face Daily)

- [Read more](https://huggingface.co/papers/2608.06020)

---

*Sponsored by Ada, Ago Consulting, and Zen Semiconductor*

## Episode Summary

In this episode, we cover:

- **DataSpace: Benchmarking Data Agents for Verifiable Analytics over Heterogeneous Workspaces** (Hugging Face Daily)

- [Read more](https://huggingface.co/papers/2608.03451)

- **Teaching Nemotron Greek: Mining a Corpus, Adapting Retrieval, and Grounding Generation for Modern Greek across Specialist Domains** (Hugging Face Daily)

- [Read more](https://huggingface.co/papers/2608.05138)

- **Learning When to Trust via Selective Context Preference Optimization** (arXiv)

- [Read more](http://arxiv.org/abs/2608.06377v1)

- **Tracing the Heart: An Evidence-Linked Pipeline for Heart-Failure Feature Engineering** (arXiv)

- [Read more](http://arxiv.org/abs/2608.06366v1)

- **RP-OPSD: Reasoning-Pivot-Guided On-Policy Self-Distillation for Multilingual Reasoning Transfer** (arXiv)

- [Read more](http://arxiv.org/abs/2608.06347v1)

---

*Sponsored by Ada, Ago Consulting, and Zen Semiconductor*

## Episode Summary

In this episode, we cover:

- **AVE-Compass: Towards Holistic Evaluation for Audio-Video Editing Abilities** (Hugging Face Daily)

- [Read more](https://huggingface.co/papers/2607.24821)

- **GDPevo: Evaluating Agent Self-Evolution on Real Business Tasks** (Hugging Face Daily)

- [Read more](https://huggingface.co/papers/2608.03764)

- **ABSeeker: Training Long-Horizon Search Agents via Answer-Backtracked Credit Assignment** (Hugging Face Daily)

- [Read more](https://huggingface.co/papers/2608.05102)

- **FinanceHarness: Autonomous Financial Deep Research Framework** (Hugging Face Daily)

- [Read more](https://huggingface.co/papers/2607.27853)

- **SmartMage: Dynamic Modality Orchestration for 3D Scene Understanding** (arXiv)

- [Read more](http://arxiv.org/abs/2608.05137v1)

---

*Sponsored by Ada, Ago Consulting, and Zen Semiconductor*

## Episode Summary

In this episode, we cover:

- **WorldCup Arena: Prospective, Leakage-Free Evaluation of Frontier LLMs on a Live Tournament** (arXiv)

- [Read more](http://arxiv.org/abs/2608.04008v1)

- **ARCHead: Activation-Metric Residual Correction for Large Language Model Output Heads** (Hugging Face Daily)

- [Read more](https://huggingface.co/papers/2608.02703)

- **When Many Answers Are Valid, Voting Fails: Symbolic Verification for Best-of-K Causal Reasoning in LLMs** (Hugging Face Daily)

- [Read more](https://huggingface.co/papers/2608.03506)

- **CURV: Enhancing Chart Understanding Through Curriculum Visual Grounded Reasoning** (Hugging Face Daily)

- [Read more](https://huggingface.co/papers/2608.02833)

- **SocietyBench: Forecasting Counterfactual Social-World Evolution** (arXiv)

- [Read more](http://arxiv.org/abs/2608.04009v1)

---

*Sponsored by Ada, Ago Consulting, and Zen Semiconductor*

## Episode Summary

In this episode, we cover:

- **Model or Harness? An Interaction-Centric Taxonomy for Localizing Agent Failures** (Hugging Face Daily)

- [Read more](https://huggingface.co/papers/2607.28802)

- **GPTQ-2D: Cubic-Time Two-Sided Adaptive Rounding** (Hugging Face Daily)

- [Read more](https://huggingface.co/papers/2607.27042)

- **Seeing or Knowing? Visual Context Sensitivity in Multimodal Large Language Models** (Hugging Face Daily)

- [Read more](https://huggingface.co/papers/2607.26326)

- **Zero-Mem: Zero-Token Memory Operations for LLM Agents** (Hugging Face Daily)

- [Read more](https://huggingface.co/papers/2607.29377)

- **RecHarness: A Bandit-Routed Agentic Harness for Self-Evolving Recommender Systems** (Hugging Face Daily)

- [Read more](https://huggingface.co/papers/2607.29241)

---

*Sponsored by Ada, Ago Consulting, and Zen Semiconductor*