Papers8

All Beginner Intermediate Advanced

All Sources arXiv

#self-distillation

On-Policy Self-Distillation for Reasoning Compression

Beginner

Hejian Sang, Yuanda Xu et al.Mar 5arXiv

Reasoning models often talk too much, and those extra words can actually make them more wrong.

#on-policy self-distillation#reasoning compression#conciseness instruction

LaSER: Internalizing Explicit Reasoning into Latent Space for Dense Retrieval

Intermediate

Jiajie Jin, Yanzhao Zhang et al.Mar 2arXiv

LaSER teaches a fast search model to “think” quietly inside its hidden space, so it gets the benefits of step-by-step reasoning without writing those steps out as text.

#dense retrieval#chain-of-thought#latent reasoning

Weak-Driven Learning: How Weak Agents make Strong Agents Stronger

Intermediate

Zehao Chen, Gongxun Li et al.Feb 9arXiv

Big language models can get stuck after fine-tuning because they become too sure of themselves, so normal training stops helping.

#weak-driven learning#logit mixing#weak agents

D-CORE: Incentivizing Task Decomposition in Large Reasoning Models for Complex Tool Use

Intermediate

Bowen Xu, Shaoyu Wu et al.Feb 2arXiv

This paper fixes a common problem in reasoning AIs called Lazy Reasoning, where the model rambles instead of making a good plan.

#task decomposition#tool use#large reasoning models

daVinci-Agency: Unlocking Long-Horizon Agency Data-Efficiently

Intermediate

Mohan Jiang, Dayuan Fu et al.Feb 2arXiv

Long tasks trip up most AIs because they lose track of goals and make small mistakes that snowball over many steps.

#long-horizon agency#pull request chains#software evolution

THINKSAFE: Self-Generated Safety Alignment for Reasoning Models

Intermediate

Seanie Lee, Sangwoo Park et al.Jan 30arXiv

Large reasoning models got very good at thinking step-by-step, but that sometimes made them too eager to follow harmful instructions.

#THINKSAFE#self-generated safety alignment#refusal steering

Towards Scalable Pre-training of Visual Tokenizers for Generation

Intermediate

Jingfeng Yao, Yuda Song et al.Dec 15arXiv

The paper tackles a paradox: visual tokenizers that get great pixel reconstructions often make worse images when used for generation.

#visual tokenizer#latent space#Vision Transformer

Native Parallel Reasoner: Reasoning in Parallelism via Self-Distilled Reinforcement Learning

Beginner

Tong Wu, Yang Liu et al.Dec 8arXiv

This paper teaches a language model to think along several paths at the same time instead of one step after another.

#parallel reasoning#reinforcement learning for LLMs#self-distillation