PANDO: Efficient Multimodal AI Agents via Online Skill Distillation

Mellum2 Technical Report

## Episode Summary In this episode, we cover: - **Mellum2 Technical Report** (Hugging Face Daily) - [Read more](https://huggingface.co/papers/2605.31268) - **Harness Updating Is Not Harness Benefit: Disentangling Evolution Capabilities in Self-Evolving LLM Agents** (Hugging Face Daily) - [Read more](https://huggingface.co/papers/2605.30621) - **Recognizing Co-Speech Gestures in-the-Wild** (arXiv) - [Read more](http://arxiv.org/abs/2605.31589v1) - **Beyond Recall: Behavioral Specification as an Interpretive Layer for AI Personalization** (Hugging Face Daily) - [Read more](https://huggingface.co/papers/2605.28969) - **nuReasoning: A Reasoning-Centric Dataset and Benchmark for Long-Tail Autonomous Driving** (arXiv) - [Read more](http://arxiv.org/abs/2605.31572v1) --- *Sponsored by LimitLess AI*

PRISM: A Multi-Dimensional Benchmark for Evaluating LLM Peer Reviewers

## Episode Summary In this episode, we cover: - **PRISM: A Multi-Dimensional Benchmark for Evaluating LLM Peer Reviewers** (Hugging Face Daily) - [Read more](https://huggingface.co/papers/2605.26730) - **DynaFLIP: Rethinking Robotics Perception via Tri-Modal-Dynamics Guided Representation** (Hugging Face Daily) - [Read more](https://huggingface.co/papers/2605.30350) - **CONF-KV: Confidence-Aware KV Cache Eviction with Mixed-Precision Storage for Long-Horizon LLM** (Hugging Face Daily) - [Read more](https://huggingface.co/papers/2605.24786) - **Locally Coherent, Globally Incoherent: Bounding Compositional Incoherence in Multi-Component LLM Agents** (arXiv) - [Read more](http://arxiv.org/abs/2605.30335v1) - **Reflective Prompt Tuning through Language Model Function-Calling** (Hugging Face Daily) - [Read more](https://huggingface.co/papers/2605.21781) --- *Sponsored by LimitLess AI*

PANDO: Efficient Multimodal AI Agents via Online Skill Distillation

## Episode Summary In this episode, we cover: - **PANDO: Efficient Multimodal AI Agents via Online Skill Distillation** (Hugging Face Daily) - [Read more](https://huggingface.co/papers/2605.24785) - **YoCausal: How Far is Video Generation from World Model? A Causality Perspective** (arXiv) - [Read more](http://arxiv.org/abs/2605.30346v1) - **CoHyDE: Iterative Co-Training of LLM Rewriter & Dense Encoder for Tool Retrieval** (Hugging Face Daily) - [Read more](https://huggingface.co/papers/2605.29271) - **Tiny but Trusted: Efficient Vision-Language Reasoning for Time-Series Anomaly Detection** (Hugging Face Daily) - [Read more](https://huggingface.co/papers/2605.30344) - **Benchmarking Single-Factor Physical Video-to-Audio Generation** (arXiv) - [Read more](http://arxiv.org/abs/2605.30339v1) --- *Sponsored by LimitLess AI*

Forecasting Downstream Performance of LLMs With Proxy Metrics

## Episode Summary In this episode, we cover: - **Forecasting Downstream Performance of LLMs With Proxy Metrics** (Hugging Face Daily) - [Read more](https://huggingface.co/papers/2605.18607) - **DeltaBox: Scaling Stateful AI Agents with Millisecond-Level Sandbox Checkpoint/Rollback** (arXiv) - [Read more](http://arxiv.org/abs/2605.22781v1) - **Lean Refactor: Multi-Objective Controllable Proof Optimization via Agentic Strategy Search** (Hugging Face Daily) - [Read more](https://huggingface.co/papers/2605.20244) - **AutoRubric-T2I: Robust Rule-Based Reward Model for Text-to-Image Alignment** (Hugging Face Daily) - [Read more](https://huggingface.co/papers/2605.17602) - **Forecasting Scientific Progress with Artificial Intelligence** (Hugging Face Daily) - [Read more](https://huggingface.co/papers/2605.22681) --- *Sponsored by LimitLess AI*

Live Music Diffusion Models: Efficient Fine-Tuning and Post-Training of Interactive Diffusion Music Generators

## Episode Summary In this episode, we cover: - **Live Music Diffusion Models: Efficient Fine-Tuning and Post-Training of Interactive Diffusion Music Generators** (Hugging Face Daily) - [Read more](https://huggingface.co/papers/2605.22717) - **DeltaBox: Scaling Stateful AI Agents with Millisecond-Level Sandbox Checkpoint/Rollback** (arXiv) - [Read more](http://arxiv.org/abs/2605.22781v1) - **AutoRubric-T2I: Robust Rule-Based Reward Model for Text-to-Image Alignment** (Hugging Face Daily) - [Read more](https://huggingface.co/papers/2605.17602) - **"I didn't Make the Micro Decisions": Measuring, Inducing, and Exposing Goal-Level AI Contributions in Collaboration** (Hugging Face Daily) - [Read more](https://huggingface.co/papers/2605.21363) - **Forecasting Downstream Performance of LLMs With Proxy Metrics** (Hugging Face Daily) - [Read more](https://huggingface.co/papers/2605.18607) --- *Sponsored by LimitLess AI*