The Research Desk.

The most upvoted and starred AI research crossing the community today.

Last Brew Time: Aug 18, 2026, 10:58 AM PT

X.com Research Buzz

The AI Layoff Trap
X.com
20042

The AI Layoff Trap

Brett Hemenway Falk, Gerry Tsoukalas

University of Pennsylvania - The Wharton School, Boston University

NLP
Emergent Introspective Awareness in Large Language Models
X.com
6035

Emergent Introspective Awareness in Large Language Models

Jack Lindsey

Anthropic

NLP
Mind Viruses: Self-Propagating Ideas in Multi-Agent LLM Systems
X.com
5119

Mind Viruses: Self-Propagating Ideas in Multi-Agent LLM Systems

Jack Lindsey, Vassilis Papadopoulos, McNair Shah, Sam Zimmerman

Anthropic

AlphaXiv Trending

Proteus: Incremental Memory Activation for Long-Context Sequence Modeling
AlphaXiv
20

Proteus: Incremental Memory Activation for Long-Context Sequence Modeling

Reza Bayat, Ali Behrouz, Vahab Mirrokni

Computer Vision
HarnessEval-W: Agentifying the Evaluation of Visual Worlds
AlphaXiv
18

HarnessEval-W: Agentifying the Evaluation of Visual Worlds

Weiliang Chen, Haowen Sun, Jun Gao

https://github.com/mirros-lab/harnesseval-w, https://mirros-lab.github.io/HarnessEval-W

Machine Learning
An Empirical Study of Training Pixel-Space Text-to-Image Diffusion Models
AlphaXiv
16

An Empirical Study of Training Pixel-Space Text-to-Image Diffusion Models

Dengyang Jiang, Ruoyi Du, Zhennan Chen

Alibaba Token Hub, Alibaba Group, The Hong Kong University of Science, Nanjing University, University of California

Reasoning
HiPHI: A Large-Scale Benchmark for High-Precision Human Motion and Object-Interaction
AlphaXiv
13

HiPHI: A Large-Scale Benchmark for High-Precision Human Motion and Object-Interaction

Jiahao Ji, Ji Ma, Runhan Zhang

Affiliation: National University of Singapore, Affiliation: The University of Hong Kong, Affiliation: SIGS, Tsinghua University, Affiliation: The Hong Kong University of Science and Technology, National University of Singapore

Reinforcement Learning
ClawGym II: Exploring Black-Box RL on Agent Harness
AlphaXiv
12

ClawGym II: Exploring Black-Box RL on Agent Harness

Huatong Song, Fei Bai, Ming Yang

Gaoling School of Artificial Intelligence, Renmin University of China, IQuest Research

Computer Vision
GenRouter: Unified Workflow Routing for Agentic Image Generation
AlphaXiv
10

GenRouter: Unified Workflow Routing for Agentic Image Generation

Harold Haodong Chen, Zhiyu Hou, Wen-Jie Shu

https://github.com/EnVision-Research/GenRouter

HuggingFace Daily Papers

PriorLabs
Advancing Open and Reproducible Relational Learning: RelArena-α, TabPFN-Rel and RPI
HuggingFace
20

Advancing Open and Reproducible Relational Learning: RelArena-α, TabPFN-Rel and RPI

Adrian Hayler, Klemens Flöge, Alan Arazi, Rishabh Ranjan, Jure Leskovec

Prior Labs

Henryqin1997
StateM: Reaching 95.3% Raw Accuracy, or a \$15 Frontier Run, on Terminal-Bench 2.1 via Harness Scaling
HuggingFace
13

StateM: Reaching 95.3% Raw Accuracy, or a \$15 Frontier Run, on Terminal-Bench 2.1 via Harness Scaling

Ziheng Qin, Yaxin Lu, Zhangyang Atlas Wang, Kai Wang

Computer Vision
GRNEdit: Efficient General Video Editing from a New Binary-Evidence Perspective in Generative Refinement Networks
HuggingFace
13

GRNEdit: Efficient General Video Editing from a New Binary-Evidence Perspective in Generative Refinement Networks

Feng Xie, Jiagao Hu, Fuhao Li, Zepeng Wang, Yuxuan Chen

NLP
Plausible but Not Valid: A Psychometric Audit of LLMs as Synthetic Survey Respondents
HuggingFace
1

Plausible but Not Valid: A Psychometric Audit of LLMs as Synthetic Survey Respondents

Mantas Lukauskas, Viktorija Šarkauskaitė

Machine Learning
StreamOPD: A Post-Training Recipe with Spatio-Temporal Cue Gating for Streaming Video Understanding
HuggingFace
0

StreamOPD: A Post-Training Recipe with Spatio-Temporal Cue Gating for Streaming Video Understanding

Keming Wu, Baoyi Wang, Kaichen Zhang, Xiang An, Zuhao Yang

Valid Per-Field Selective Risk Control for Document Extraction: Three Failure Modes, a Validity Ladder, and When Conditioning Pays
HuggingFace
0

Valid Per-Field Selective Risk Control for Document Extraction: Three Failure Modes, a Validity Ladder, and When Conditioning Pays

Bhaskar Gurram