English
Related papers

Related papers: Revisiting the Superficial Alignment Hypothesis

200 papers

Reinforcement learning is used to align language models with human preference signals after first pre-training the model to predict the next token of text within a large corpus using likelihood maximization. Before being deployed in a…

Computation and Language · Computer Science 2024-08-30 Alec Solway

The prevailing paradigm in large language model (LLM) development is to pretrain a base model, then perform further training to improve performance and model behavior. However, hyperparameter optimization and scaling laws have been studied…

Machine Learning · Computer Science 2026-02-12 Tessa Han , Sebastian Bordt , Hanlin Zhang , Sham Kakade

Despite the widespread multilingual deployment of large language models, post-training pipelines remain predominantly English-centric, contributing to performance disparities across languages. We present a systematic, controlled study of…

Computation and Language · Computer Science 2026-04-16 Mehak Dhaliwal , Shashwat Chaurasia , Yao Qin , Dezhi Hong , Thomas Butler

Large language models are trained in two stages: (1) unsupervised pretraining from raw text, to learn general-purpose representations, and (2) large scale instruction tuning and reinforcement learning, to better align to end tasks and user…

Computation and Language · Computer Science 2023-05-22 Chunting Zhou , Pengfei Liu , Puxin Xu , Srini Iyer , Jiao Sun , Yuning Mao , Xuezhe Ma , Avia Efrat , Ping Yu , Lili Yu , Susan Zhang , Gargi Ghosh , Mike Lewis , Luke Zettlemoyer , Omer Levy

Natural language data follows a power-law distribution, with most knowledge and skills appearing at very low frequency. While a common intuition suggests that reweighting or curating data towards a uniform distribution may help models…

Artificial Intelligence · Computer Science 2026-04-28 Zixuan Wang , Xingyu Dang , Jason D. Lee , Kaifeng Lyu

Large language models (LLMs) have recently shown impressive performance on tasks involving reasoning, leading to a lively debate on whether these models possess reasoning capabilities similar to humans. However, despite these successes, the…

Computation and Language · Computer Science 2024-08-07 Philipp Mondorf , Barbara Plank

Recent advances in unsupervised learning have shown that unsupervised pre-training, followed by fine-tuning, can improve model generalization. However, a rigorous understanding of how the representation function learned on an unlabeled…

Machine Learning · Computer Science 2024-03-12 Yuyang Deng , Junyuan Hong , Jiayu Zhou , Mehrdad Mahdavi

Recent advancements in Reinforcement Post-Training (RPT) have significantly enhanced the capabilities of Large Reasoning Models (LRMs), sparking increased interest in the generalization of RL-based reasoning. While existing work has…

Computation and Language · Computer Science 2025-10-03 Wen Yang , Junhong Wu , Chong Li , Chengqing Zong , Jiajun Zhang

The alignment process changes several properties of a large language model's (LLM's) output distribution. We analyze two aspects of post-alignment distributional shift of LLM responses. First, we re-examine previously reported reductions in…

Computation and Language · Computer Science 2025-05-13 Thom Lake , Eunsol Choi , Greg Durrett

Scaling laws are useful guides for derisking expensive training runs, as they predict performance of large models using cheaper, small-scale experiments. However, there remain gaps between current scaling studies and how language models are…

Pretrained language models (PLMs) display impressive performances and have captured the attention of the NLP community. Establishing best practices in pretraining has, therefore, become a major focus of NLP research, especially since…

Computation and Language · Computer Science 2024-10-08 Zihao Li , Shaoxiong Ji , Timothee Mickus , Vincent Segonne , Jörg Tiedemann

The prevailing paradigm for enhancing the reasoning abilities of LLMs revolves around post-training on high-quality, reasoning-intensive data. While emerging literature suggests that reasoning data is increasingly incorporated also during…

We propose a novel scaling law for general-purpose decoder-only language models (LMs) trained on multilingual data, tackling the problem of balancing languages during multilingual pretraining. A primary challenge in studying multilingual…

Computation and Language · Computer Science 2024-12-05 Yifei He , Alon Benhaim , Barun Patra , Praneetha Vaddamanu , Sanchit Ahuja , Parul Chopra , Vishrav Chaudhary , Han Zhao , Xia Song

Large language models (LLMs) with enormous pre-training tokens and parameters emerge diverse abilities, including math reasoning, code generation, and instruction following. These abilities are further enhanced by supervised fine-tuning…

Computation and Language · Computer Science 2024-06-10 Guanting Dong , Hongyi Yuan , Keming Lu , Chengpeng Li , Mingfeng Xue , Dayiheng Liu , Wei Wang , Zheng Yuan , Chang Zhou , Jingren Zhou

We investigate whether post-trained capabilities can be transferred across models without retraining, with a focus on transfer across different model scales. We propose the Master Key Hypothesis, which states that model capabilities…

Machine Learning · Computer Science 2026-05-07 Rishab Balasubramanian , Pin-Jie Lin , Rituraj Sharma , Anjie Fang , Fardin Abdi , Viktor Rozgic , Zheng Du , Mohit Bansal , Tu Vu

We have witnessed superhuman intelligence thanks to the fast development of large language models and multimodal language models. As the application of such superhuman models becomes more and more popular, a critical question arises here:…

Computation and Language · Computer Science 2024-12-24 Minlie Huang , Yingkang Wang , Shiyao Cui , Pei Ke , Jie Tang

Multilingual Large Language Models (LLMs) achieve remarkable levels of zero-shot cross-lingual transfer performance. We speculate that this is predicated on their ability to align languages without explicit supervision from parallel…

Computation and Language · Computer Science 2024-06-21 Hetong Wang , Pasquale Minervini , Edoardo M. Ponti

A prevailing narrative in LLM post-training holds that supervised finetuning (SFT) memorizes while reinforcement learning (RL) generalizes. We revisit this claim for reasoning SFT with long chain-of-thought (CoT) supervision and find that…

Artificial Intelligence · Computer Science 2026-04-09 Qihan Ren , Peng Wang , Ruikun Cai , Shuai Shao , Dadi Guo , Yuejin Xie , Yafu Li , Quanshi Zhang , Xia Hu , Jing Shao , Dongrui Liu

Reasoning post-training improves Large Language Models (LLMs) on complex tasks such as mathematics and coding, but its benefits across diverse multimodal tasks remains uncertain. The trend of releasing parallel "Instruct" and "Thinking"…

Computation and Language · Computer Science 2026-05-12 Ruobing Zheng , Tianqi Li , Jianing Li , Qingpei Guo , Yi Yuan , Jingdong Chen

Reinforcement learning (RL) has emerged as a powerful post-training technique to incentivize the reasoning ability of large language models (LLMs). However, LLMs can respond very inconsistently to RL finetuning: some show substantial…

Machine Learning · Computer Science 2025-10-07 Zhepeng Cen , Yihang Yao , William Han , Zuxin Liu , Ding Zhao