Papers17

#token efficiency

From Perception to Action: An Interactive Benchmark for Vision Reasoning

The paper introduces CHAIN, a hands-on 3D playground that tests if AI can not only see objects but also plan and act under real physics.

#interactive benchmark#vision-language models#physical reasoning

Panini: Continual Learning in Token Space via Structured Memory

Intermediate

Shreyas Rajesh, Pavan Holur et al.Feb 16arXiv

Panini is a way for AI to keep learning new facts without changing its brain by storing them as tiny linked Q&A facts in an external memory.

#non-parametric continual learning#structured memory#Generative Semantic Workspace

Does Your Reasoning Model Implicitly Know When to Stop Thinking?

Intermediate

Zixuan Huang, Xin Xia et al.Feb 9arXiv

Big AI reasoning models often keep thinking long after they already found the right answer, wasting time and tokens.

#SAGE#efficient reasoning#chain of thought

Parallel-Probe: Towards Efficient Parallel Thinking via 2D Probing

Intermediate

Tong Zheng, Chengsong Huang et al.Feb 3arXiv

Parallel-Probe is a simple add-on that lets many AI “thought paths” think at once but stop early when they already agree.

#parallel thinking#2D probing#consensus-based early stopping

LatentMem: Customizing Latent Memory for Multi-Agent Systems

Intermediate

Muxin Fu, Guibin Zhang et al.Feb 3arXiv

LatentMem is a new memory system that helps teams of AI agents remember the right things for their specific jobs without overloading them with text.

#multi-agent systems#latent memory#role-aware memory

CodeOCR: On the Effectiveness of Vision Language Models in Code Understanding

Intermediate

Yuling Shi, Chaoxiang Xie et al.Feb 2arXiv

The paper tests a simple but bold idea: show code to AI as pictures instead of plain text, then shrink those pictures to save tokens and time.

#multimodal language models#code as images#visual code understanding

Beyond Imitation: Reinforcement Learning for Active Latent Planning

Intermediate

Zhi Zheng, Wee Sun LeeJan 29arXiv

The paper shows how to make AI think faster and smarter by planning in a hidden space instead of writing long step-by-step sentences.

#latent reasoning#chain-of-thought#variational autoencoder

MemOCR: Layout-Aware Visual Memory for Efficient Long-Horizon Reasoning

Intermediate

Yaorui Shi, Shugui Liu et al.Jan 29arXiv

MemOCR is a new way for AI to remember long histories by turning important notes into a picture with big, bold parts for key facts and tiny parts for details.

#MemOCR#visual memory#adaptive information density

Training Reasoning Models on Saturated Problems via Failure-Prefix Conditioning

Intermediate

Minwu Kim, Safal Shrestha et al.Jan 28arXiv

When training smart language models with RL that use right-or-wrong rewards, learning can stall on 'saturated' problems that the model almost always solves.

#failure-prefix conditioning#RLVR#GRPO

Spark: Strategic Policy-Aware Exploration via Dynamic Branching for Long-Horizon Agentic Learning

Intermediate

Jinyang Wu, Shuo Yang et al.Jan 28arXiv

SPARK is a new way to train AI agents that saves compute by exploring more only at the most important moments.

#SPARK#dynamic branching#strategic exploration

Innovator-VL: A Multimodal Large Language Model for Scientific Discovery

Intermediate

Zichen Wen, Boxue Yang et al.Jan 27arXiv

Innovator-VL is a new multimodal AI model that understands both pictures and text to help solve science problems without needing mountains of special data.

#Innovator-VL#multimodal large language model#scientific reasoning

daVinci-Dev: Agent-native Mid-training for Software Engineering

Intermediate

Ji Zeng, Dayuan Fu et al.Jan 26arXiv

This paper teaches code AIs to work more like real software engineers by training them in the middle of their learning using real development workflows.

#agentic mid-training#agent-native data#contextually-native trajectories

1 2