In this episode, we discuss ProRL: Prolonged Reinforcement Learning Expands Reasoning Boundaries in Large Language Models by Mingjie Liu, Shizhe Diao, Ximing Lu, Jian Hu, Xin Dong, Yejin Choi, Jan Kautz, Yi Dong. This paper introduces ProRL, a new reinforcement learning training method that uncovers novel reasoning strategies beyond those found in base language models. Empirical results show that models trained with ProRL consistently outperform base models on challenging reasoning tasks, including cases where base models fail even with extensive attempts. The study demonstrates that prolonged RL can meaningfully expand reasoning capabilities by exploring new solution spaces over time, advancing understanding of how RL enhances language model reasoning.

In this episode, we discuss Roboflow100-VL: A Multi-Domain Object Detection Benchmark for Vision-Language Models by Peter Robicheaux, Matvei Popov, Anish Madan, Isaac Robinson, Joseph Nelson, Deva Ramanan, Neehar Peri. The paper introduces Roboflow100-VL, a large benchmark of 100 diverse multi-modal object detection datasets designed to test vision-language models (VLMs) on out-of-distribution concepts beyond typical pre-training data. It demonstrates that state-of-the-art VLMs perform poorly in zero-shot settings on challenging domains like medical imaging, highlighting the importance of few-shot concept alignment through annotated examples and rich text. The paper also presents results from a CVPR 2025 competition where the winning approach significantly outperforms baselines in few-shot detection tasks.

In this episode, we discuss ImpossibleBench: Measuring LLMs' Propensity of Exploiting Test Cases by Ziqian Zhong, Aditi Raghunathan, Nicholas Carlini. The paper introduces ImpossibleBench, a benchmark framework designed to measure and analyze large language models' tendency to cheat by exploiting test cases. It creates tasks with conflicting specifications and unit tests to quantify how often models take shortcuts that violate intended behavior. The framework is used to study cheating behaviors, refine prompting strategies, and develop tools to detect and reduce such deceptive practices in LLMs.

In this episode, we discuss Scaling Instruction-Based Video Editing with a High-Quality Synthetic Dataset by Qingyan Bai, Qiuyu Wang, Hao Ouyang, Yue Yu, Hanlin Wang, Wen Wang, Ka Leong Cheng, Shuailei Ma, Yanhong Zeng, Zichen Liu, Yinghao Xu, Yujun Shen, Qifeng Chen. The paper presents Ditto, a comprehensive framework that generates large-scale, high-quality training data for instruction-based video editing by combining an advanced image editor with an in-context video generator. Ditto uses an efficient, distilled model with a temporal enhancer and an intelligent agent to ensure scalable, diverse, and high-fidelity video edits. Leveraging this framework, the authors created the Ditto-1M dataset and trained the Editto model, achieving state-of-the-art performance in following editing instructions.

In this episode, we discuss Reasoning with Sampling: Your Base Model is Smarter Than You Think by Aayush Karan, Yilun Du. The paper proposes a novel iterative sampling algorithm based on Markov chain Monte Carlo techniques that enhances reasoning abilities of base large language models at inference time without additional training. This method significantly improves performance on multiple reasoning benchmarks, matching or surpassing results from reinforcement learning fine-tuning. Additionally, the approach maintains sample diversity and does not rely on curated datasets or verifiers, making it broadly applicable.

In this episode, we discuss DeepSeek-OCR: Contexts Optical Compression by The authors of the paper are: **Haoran Wei, Yaofeng Sun, Yukun Li**. DeepSeek-OCR introduces a method to compress long text contexts into compact 2D vision tokens using a DeepEncoder and a decoder model, achieving high OCR accuracy even at significant compression ratios. It outperforms existing OCR benchmarks on OmniDocBench while using fewer vision tokens, demonstrating efficiency and scalability. The system is practical for large-scale training data generation and its code and models are publicly available.

In this episode, we discuss The Markovian Thinker by Milad Aghajohari, Kamran Chitsaz, Amirhossein Kazemnejad, Sarath Chandar, Alessandro Sordoni, Aaron Courville, Siva Reddy. The paper proposes Markovian Thinking, a reinforcement learning paradigm that limits reasoning context to a constant-size state, enabling linear compute with constant memory rather than quadratic overhead. They implement this approach in Delethink, an environment that segments reasoning into fixed-size chunks with learned textual states to seamlessly continue reasoning after resets. Experiments show Delethink-trained models achieve longer reasoning chains more efficiently and scale better than standard methods, significantly reducing computational costs.

In this episode, we discuss DeepDive: Advancing Deep Search Agents with Knowledge Graphs and Multi-Turn RL by Rui Lu, Zhenyu Hou, Zihan Wang, Hanchen Zhang, Xiao Liu, Yujiang Li, Shi Feng, Jie Tang, Yuxiao Dong. The paper introduces DeepDive, a method to improve large language models' deep search capabilities by automatically generating complex questions and applying multi-turn reinforcement learning for enhanced long-horizon reasoning. DeepDive-32B outperforms existing open-source models on browsing benchmarks like BrowseComp. The approach also enables scalable tool usage during inference, with all resources made publicly available.

In this episode, we discuss Towards a Physics Foundation Model by Florian Wiesner, Matthias Wessling, Stephen Baek. This paper introduces the General Physics Transformer (GPhyT), a foundation model trained on diverse simulation data that can simulate multiple complex physical systems without explicit knowledge of governing equations. GPhyT outperforms specialized models by up to 29 times, generalizes zero-shot to unseen physics tasks, and maintains stable predictions over long time horizons. This work demonstrates the feasibility of a universal physics foundation model, potentially revolutionizing computational science by eliminating the need for task-specific solvers.

In this episode, we discuss Scalable Option Learning in High-Throughput Environments by Mikael Henaff, Scott Fujimoto, Michael Rabbat. The paper presents Scalable Option Learning (SOL), a hierarchical reinforcement learning algorithm designed for high-throughput environments. SOL achieves a 25x increase in training speed and outperforms flat agents by training on 20 billion frames in the game NetHack. The method is also validated on MiniHack and Mujoco, demonstrating broad applicability and scalability.

In this episode, we discuss Beyond the 80/20 Rule: High-Entropy Minority Tokens Drive Effective Reinforcement Learning for LLM Reasoning by Shenzhi Wang, Le Yu, Chang Gao, Chujie Zheng, Shixuan Liu, Rui Lu, Kai Dang, Xionghui Chen, Jianxin Yang, Zhenru Zhang, Yuqiong Liu, An Yang, Andrew Zhao, Yang Yue, Shiji Song, Bowen Yu, Gao Huang, Junyang Lin. This paper investigates Reinforcement Learning with Verifiable Rewards (RLVR) by analyzing token entropy patterns during Chain-of-Thought reasoning in Large Language Models. It finds that a small subset of high-entropy "forking" tokens critically guide reasoning pathways and that RLVR primarily adjusts these tokens to improve performance. Leveraging this insight, the authors enhance RLVR efficiency by focusing updates on these tokens, achieving better results with fewer token updates across multiple model scales.

In this episode, we discuss Reverse-Engineered Reasoning for Open-Ended Generation by Haozhe Wang, Haoran Que, Qixin Xu, Minghao Liu, Wangchunshu Zhou, Jiazhan Feng, Wanjun Zhong, Wei Ye, Tong Yang, Wenhao Huang, Ge Zhang, Fangzhen Lin. The paper introduces REverse-Engineered Reasoning (REER), a novel backward approach that uncovers deep reasoning steps from known good solutions instead of forward trial-and-error or imitation. Using REER, the authors create DeepWriting-20K, a large dataset of reasoning trajectories for open-ended tasks, and train DeepWriter-8B, a model that outperforms strong open-source baselines. DeepWriter-8B also matches or exceeds the performance of leading proprietary models like GPT-4o and Claude 3.5.

In this episode, we discuss Scaling Performance of Large Language Model Pretraining by Alexander Interrante-Grant, Carla Varela-Rosa, Suhaas Narayan, Chris Connelly, Albert Reuther. The paper explores the challenges and strategies involved in training large language models (LLMs) at scale, focusing on distributed training and managing massive datasets across many computing nodes. It provides practical recommendations for optimizing data parallelism to fully utilize GPU resources during pretraining. The goal is to offer clearer guidance on scaling LLM training pipelines, addressing a gap in publicly available information.

In this episode, we discuss General Social Agents by Benjamin S. Manning, John J. Horton. The paper proposes using AI agents guided by social science theory and natural language instructions to predict human behavior in novel settings without ad hoc adjustments. By training these agents on human data from related "seed" games, they successfully predict outcomes across a large and diverse set of new games. Their approach outperforms traditional game-theoretic predictions and existing AI models, even exceeding predictions based on published human data in some novel scenarios.

In this episode, we discuss We need a new ethics for a world of AI agents by Iason Gabriel, Geoff Keeling, Arianna Manzini & James Evans. The paper examines the shift toward autonomous AI agents capable of goal-directed actions with minimal human oversight. It highlights both the potential benefits of these agents, such as economic growth and scientific advancement, and the associated risks involving responsibility, safety, and social dynamics. The authors call for increased collaboration among various stakeholders to address challenges and ensure beneficial human-agent and agent-agent interactions.

In this episode, we discuss Hierarchical Reasoning Model by Guan Wang, Jin Li, Yuhao Sun, Xing Chen, Changling Liu, Yue Wu, Meng Lu, Sen Song, Yasin Abbasi Yadkori. The paper introduces the Hierarchical Reasoning Model (HRM), a recurrent architecture inspired by the brain's hierarchical processing that achieves deep, efficient reasoning in a single forward pass. HRM uses two interdependent modules for abstract planning and detailed computation, enabling it to excel on complex tasks like Sudoku and maze solving with minimal data and no pre-training. It outperforms larger models on the ARC benchmark, highlighting its promise for advancing general-purpose AI reasoning.

In this episode, we discuss ARC-Hunyuan-Video-7B: Structured Video Comprehension of Real-World Shorts by Yuying Ge, Yixiao Ge, Chen Li, Teng Wang, Junfu Pu, Yizhuo Li, Lu Qiu, Jin Ma, Lisheng Duan, Xinyu Zuo, Jinwen Luo, Weibo Gu, Zexuan Li, Xiaojing Zhang, Yangyu Tao, Han Hu, Di Wang, Ying Shan. The paper presents ARC-Hunyuan-Video, a 7B-parameter multimodal model designed for detailed, temporally-structured understanding of short user-generated videos using visual, audio, and text inputs. It supports tasks like timestamped captioning, summarization, question answering, and video reasoning, trained through a multi-stage process including reinforcement learning. Evaluations show strong real-world performance, efficiency, and positive impact on user engagement in production deployment.

In this episode, we discuss Small Language Models are the Future of Agentic AI by Peter Belcak, Greg Heinrich, Shizhe Diao, Yonggan Fu, Xin Dong, Saurav Muralidharan, Yingyan Celine Lin, Pavlo Molchanov. The paper argues that small language models (SLMs) are more suitable, powerful enough, and cost-effective for many specialized agentic AI tasks compared to large language models (LLMs). It proposes that heterogeneous agentic systems using multiple models are ideal when general conversational abilities are needed and presents an algorithm for converting LLM-based agents to SLM-based ones. The authors emphasize the economic and operational benefits of shifting towards SLMs and invite further discussion to advance affordable AI deployment.

In this episode, we discuss Learning When to Plan: Efficiently Allocating Test-Time Compute for LLM Agents by Davide Paglieri, Bartłomiej Cupiał, Jonathan Cook, Ulyana Piterbarg, Jens Tuyls, Edward Grefenstette, Jakob Nicolaus Foerster, Jack Parker-Holder, Tim Rocktäschel. The paper introduces a framework enabling large language model agents to dynamically decide when to plan during task execution, improving efficiency and performance. They propose a two-stage training process combining supervised fine-tuning and reinforcement learning to develop this capability. Experiments show these dynamically planning agents are more sample-efficient, achieve complex goals better, and can be guided by human plans.

In this episode, we discuss Why Language Models Hallucinate by The authors of the paper are: - Adam Tauman Kalai - Ofir Nachum - Santosh S. Vempala - Edwin Zhang. The paper explains that hallucinations in large language models arise because training and evaluation reward guessing over admitting uncertainty, framing the issue as errors in binary classification. It shows that models become incentivized to produce plausible but incorrect answers to perform well on benchmarks. The authors propose that addressing hallucinations requires changing how benchmarks are scored, promoting more trustworthy AI by discouraging penalization of uncertain responses.