Sijia L. Liu

2025

1. ICLR

   Humanline: Online Alignment as Perceptual Loss

   Sijia Liu, Niklas Muennighoff, and Kawin Ethayarajh

   2025

   Abs arXiv Bib Blog

   Online alignment (e.g., GRPO) is generally more performant than offline alignment (e.g., DPO) – but why? Drawing on prospect theory from behavioral economics, we propose a human-centric explanation. We prove that online on-policy sampling better approximates the human-perceived distribution of what the model can produce, and PPO/GRPO-style clipping – originally introduced to just stabilize training – recovers a perceptual bias in how humans perceive probability. In this sense, PPO/GRPO act as perceptual losses already. Our theory further suggests that the online/offline dichotomy is itself incidental to maximizing human utility, since we can achieve the same effect by selectively training on any data in a manner that mimics human perception, rather than restricting ourselves to online on-policy data. Doing so would allow us to post-train more quickly, cheaply, and flexibly without sacrificing performance. To this end, we propose a design pattern that explicitly incorporates perceptual distortions of probability into objectives like DPO/KTO/GRPO, creating humanline variants of them. Surprisingly, we find that these humanline variants, even when trained with offline off-policy data, can match the performance of their online counterparts on both verifiable and unverifiable tasks.

   @article{liu2025humanline,
     title = {Humanline: Online Alignment as Perceptual Loss},
     author = {Liu, Sijia and Muennighoff, Niklas and Ethayarajh, Kawin},
     year = {2025},
     eprint = {2509.24207},
     archiveprefix = {arXiv},
     primaryclass = {cs.AI},
   }

2024

1. Model Card

   The Amazon Nova family of models: Technical report and model card

   Amazon Artificial General Intelligence

   Amazon Technical Reports, 2024

   Abs Bib PDF

   We present Amazon Nova, a new generation of state-of-the-art foundation models that deliver frontier intelligence and industry-leading price performance. Amazon Nova Pro is a highly-capable multimodal model with the best combination of accuracy, speed, and cost for a wide range of tasks. Amazon Nova Lite is a low-cost multimodal model that is lightning fast for processing images, video, documents and text. Amazon Nova Micro is a text-only model that delivers our lowest-latency responses at very low cost. Amazon Nova Canvas is an image generation model that creates professional grade images with rich customization controls. Amazon Nova Reel is a video generation model offering high-quality outputs, customization, and motion control. Our models were built responsibly and with a commitment to customer trust, security, and reliability. We report benchmarking results for core capabilities, agentic performance, long context, functional adaptation, runtime performance, and human evaluation.

   @article{Intelligence2024,
     author = {Intelligence, Amazon Artificial General},
     title = {The Amazon Nova family of models: Technical report and model card},
     year = {2024},
     url = {https://www.amazon.science/publications/the-amazon-nova-family-of-models-technical-report-and-model-card},
     journal = {Amazon Technical Reports},
   }

2. EMNLP

   LLM Self-Correction with DeCRIM: Decompose, Critique, and Refine for Enhanced Following of Instructions with Multiple Constraints

   Thomas Palmeira Ferraz, Kartik Mehta, Yu-Hsiang Lin, and 7 more authors

   EMNLP, NeurIPS Workshop on System 2 Reasoning at Scale (Oral), 2024

   Abs Bib PDF Slides

   Instruction following is a key capability for LLMs. However, recent studies have shown that LLMs often struggle with instructions containing multiple constraints (e.g. a request to create a social media post "in a funny tone" with "no hashtag"). Despite this, most evaluations focus solely on synthetic data. To address this, we introduce RealInstruct, the first benchmark designed to evaluate LLMs’ ability to follow real-world multi-constrained instructions by leveraging queries real users asked AI assistants. We also investigate model-based evaluation as a cost-effective alternative to human annotation for this task. Our findings reveal that even the proprietary GPT-4 model fails to meet at least one constraint on over 21% of instructions, highlighting the limitations of state-of-the-art models. To address the performance gap between open-source and proprietary models, we propose the Decompose, Critique and Refine (DeCRIM) self-correction pipeline, which enhances LLMs’ ability to follow constraints. DeCRIM works by decomposing the original instruction into a list of constraints and using a Critic model to decide when and where the LLM’s response needs refinement. Our results show that DeCRIM improves Mistral’s performance by 7.3% on RealInstruct and 8.0% on IFEval even with weak feedback. Moreover, we demonstrate that with strong feedback, open-source LLMs with DeCRIM can outperform GPT-4 on both benchmarks.

   @article{ferraz2024llmselfcorrectiondecrimdecompose,
     title = {LLM Self-Correction with DeCRIM: Decompose, Critique, and Refine for Enhanced Following of Instructions with Multiple Constraints},
     author = {Ferraz, Thomas Palmeira and Mehta, Kartik and Lin, Yu-Hsiang and Chang, Haw-Shiuan and Oraby, Shereen and Liu, Sijia and Subramanian, Vivek and Chung, Tagyoung and Bansal, Mohit and Peng, Nanyun},
     year = {2024},
     journal = {EMNLP, NeurIPS Workshop on System 2 Reasoning at Scale <span style="color: red"><b><i>(Oral)</i></b></span>},
     eprint = {2410.06458},
     archiveprefix = {arXiv},
     primaryclass = {cs.CL},
     url = {https://arxiv.org/abs/2410.06458},
   }

2023

1. AAAI

   Towards Credible Human Evaluation of Open-Domain Dialog Systems Using Interactive Setup

   Sijia Liu, Patrick Lange, Behnam Hedayatnia, and 5 more authors

   AAAI (Oral), NeurIPS Workshop on Human Evaluation of Generative Models (Oral), EMNLP Workshop on Natural Language Generation, Evaluation, and Metrics (Oral), Jun 2023

   Abs Bib PDF Poster Slides

   Evaluating open-domain conversation models has been an open challenge due to the open-ended nature of conversations. In addition to static evaluations, recent work has started to explore a variety of per-turn and per-dialog interactive evaluation mechanisms and provide advice on the best setup. In this work, we adopt the interactive evaluation framework and further apply to multiple models with a focus on per-turn evaluation techniques. Apart from the widely used setting where participants select the best response among different candidates at each turn, one more novel per-turn evaluation setting is adopted, where participants can select all appropriate responses with different fallback strategies to continue the conversation when no response is selected. We evaluate these settings based on sensitivity and consistency using four GPT2-based models that differ in model sizes or fine-tuning data. To better generalize to any model groups with no prior assumptions on their rankings and control evaluation costs for all setups, we also propose a methodology to estimate the required sample size given a minimum performance gap of interest before running most experiments. Our comprehensive human evaluation results shed light on how to conduct credible human evaluations of open domain dialog systems using the interactive setup, and suggest additional future directions.

   @article{Liu_Lange_Hedayatnia_Papangelis_Jin_Wirth_Liu_Hakkani-Tur_2023,
     title = {Towards Credible Human Evaluation of Open-Domain Dialog Systems Using Interactive Setup},
     number = {11},
     journal = {AAAI <span style="color: red"><b><i>(Oral)</i></b></span>, NeurIPS Workshop on Human Evaluation of Generative Models <span style="color: red"><b><i>(Oral)</i></b></span>, EMNLP Workshop on Natural Language Generation, Evaluation, and Metrics <span style="color: red"><b><i>(Oral)</i></b></span>},
     volume = {37},
     url = {https://ojs.aaai.org/index.php/AAAI/article/view/26557},
     doi = {10.1609/aaai.v37i11.26557},
     author = {Liu, Sijia and Lange, Patrick and Hedayatnia, Behnam and Papangelis, Alexandros and Jin, Di and Wirth, Andrew and Liu, Yang and Hakkani-Tur, Dilek},
     year = {2023},
     month = jun,
     pages = {13264-13272},
   }

2022

1. SigDial

   Improving Bot Response Contradiction Detection via Utterance Rewriting

   Di Jin, Sijia Liu, Yang Liu, and 1 more author

   SIGDIAL (Oral), Jul 2022

   Abs Bib PDF Code Slides

   Though chatbots based on large neural models can often produce fluent responses in open domain conversations, one salient error type is contradiction or inconsistency with the preceding conversation turns. Previous work has treated contradiction detection in bot responses as a task similar to natural language inference, e.g., detect the contradiction between a pair of bot utterances. However, utterances in conversations may contain co-references or ellipsis, and using these utterances as is may not always be sufficient for identifying contradictions. This work aims to improve the contradiction detection via rewriting all bot utterances to restore antecedents and ellipsis. We curated a new dataset for utterance rewriting and built a rewriting model on it. We empirically demonstrate that this model can produce satisfactory rewrites to make bot utterances more complete. Furthermore, using rewritten utterances improves contradiction detection performance significantly, e.g., the AUPR and joint accuracy scores (detecting contradiction along with evidence) increase by 6.5% and 4.5% (absolute increase), respectively.

   @article{jin2022improving,
     title = {Improving Bot Response Contradiction Detection via Utterance Rewriting},
     author = {Jin, Di and Liu, Sijia and Liu, Yang and Hakkani-Tur, Dilek},
     year = {2022},
     month = jul,
     journal = {SIGDIAL <span style="color: red"><b><i>(Oral)</i></b></span>},
   }