arXiv subjects

Hyunji Nam

Publications and source records attributed to Hyunji Nam.

10 recordsLinked to original sources

arXiv2026

Pigeonholing: how bad prompts hurt models, causing collapse and mistakes

While in-context learning is generally shown to be effective in Large Language Models (LLMs), bad contexts can cause performance degradation and mode collapse, a phenomenon we call "pigeonholing." **Unintentionally bad** contexts can happen without malicious jailbreaking intents: For example, a user asks the model to justify an incorrect math theorem or fails to correct the model's buggy code. Specifically, we investigate ``pigeonholing" in two scenarios: (1) when the user suggests a solution, and (2) when the conversation context includes the assistant's previous (incorrect) responses. Our experiments across 10 verifiable and open-ended tasks with 10 different models show that pigeonholing manifests in several ways: (1) repeating the incorrect answers from context (leading to 38-40% performance drop), (2) converging on a narrow set of answers in coding and text generation without exploring alternatives, and (3) flipping stance on controversial topics to align with the user or the assistant's previous claims. We find that pigeonholing worsens almost monotonically with the number of conversation turns (performance drops by additional 14+% as repeated mistakes increase from 1 to 5), and pigeonholing-induced mode collapse can happen even when the provided example is correct. As a step toward mitigation, we propose RLVR with synthetic errors which improves models by 43-60% under bad contexts compared to vanilla RLVR baselines.

cs.CL↗

arXiv2026

Mitigating LLM biases toward spurious social contexts using direct preference optimization

LLMs are increasingly used for high-stakes decision-making, yet their sensitivity to spurious context can introduce harmful biases. This is a critical concern when models are deployed for tasks like evaluating teachers' instructional quality, where biased assessment can affect teachers' professional development and career. We investigate model robustness to spurious social contexts about teachers using the largest publicly available dataset of U.S. classroom transcripts (NCTE) paired with expert evaluation scores. Evaluating seven frontier and open-weight models across seven categories of spurious contexts -- including teacher experience, education level, demographic identity, and sycophancy-inducing framings -- we find that irrelevant contexts can shift model-generated ratings by up to 1.48 points on a 7-point scale. Prompt-based mitigations and popular post-training methods, such as Supervised Fine-Tuning (SFT) and Direct Preference Optimization (DPO), fall short of addressing the issue. We propose Debiasing-DPO, which combines contrastive reasoning-augmented DPO with SFT on expert labels to reduce the effect of spurious context while avoiding mode collapse. Applied to Llama and Qwen Instruct models of 3-8B parameters, Debiasing-DPO reduces bias by 84% and improves predictive accuracy by 52% on average across models. Our findings from the educational dataset highlight that stronger models may exhibit greater sensitivity despite higher accuracy, and Debiasing-DPO can improve both accuracy and robustness in prompt-based prediction tasks.

cs.AI↗

arXiv2026

Maximizing Mutual Information Between Prompt and Response Improves LLM Performance With No Additional Data

While post-training has successfully improved large language models (LLMs) across a variety of domains, these gains heavily rely on human-labeled data or external verifiers. Existing data has already been exploited, and new data is expensive to collect. Moreover, true intelligence goes far beyond verifiable tasks. Therefore, we need self-improvement frameworks that are less dependent on external signals and more broadly applicable to both verifiable and non-verifiable domains. We propose **Mutual Information Preference Optimization (MIPO)**, a contrastive data augmentation method that constructs preference pairs by generating a positive response conditioning on the correct prompt, and a negative response by conditioning on a random, unrelated prompt. We show that using Direct Preference Optimization to learn from this paired data maximizes pointwise mutual information *under the base LLM* between prompts and model responses. Experiments with with 1-7B parameter Llama and Qwen instruct models show that MIPO achieves 3-16% gains (and 51% increase for Qwen2.5-1.5B-Instruct) on personalization compared to prompting baselines. Surprisingly, MIPO can also be useful in verifiable domains, such as math and multiple-choice question answering, yielding 1-20% gains *without any additional data or external supervision*. These results suggest a promising direction for self-improvement using intrinsic signals derived from contrastive data pairs.

cs.LG↗

arXiv2026

Netflix Artwork Personalization via LLM Post-training

Large language models (LLMs) have demonstrated success in various applications of user recommendation and personalization across e-commerce and entertainment. On many entertainment platforms such as Netflix, users typically interact with a wide range of titles, each represented by an artwork. Since users have diverse preferences, an artwork that appeals to one type of user may not resonate with another with different preferences. Given this user heterogeneity, our work explores the novel problem of personalized artwork recommendations according to diverse user preferences. Similar to the multi-dimensional nature of users' tastes, titles contain different themes and tones that may appeal to different viewers. For example, the same title might feature both heartfelt family drama and intense action scenes. Users who prefer romantic content may like the artwork emphasizing emotional warmth between the characters, while those who prefer action thrillers may find high-intensity action scenes more intriguing. Rather than a one-size-fits-all approach, we conduct post-training of pre-trained LLMs to make personalized artwork recommendations, selecting the most preferred visual representation of a title for each user and thereby improving user satisfaction and engagement. Our experimental results with Llama 3.1 8B models (trained on a dataset of 110K data points and evaluated on 5K held-out user-title pairs) show that the post-trained LLMs achieve 3-5\% improvements over the Netflix production model, suggesting a promising direction for granular personalized recommendations using LLMs.

cs.IR↗

arXiv2025

IDEAlign: Comparing Ideas of Large Language Models to Domain Expert

Large language models (LLMs) are increasingly used to produce open-ended, interpretive annotations, yet there is no validated, scalable measure of idea-level similarity to expert annotations. We (i) introduce the content evaluation of LLM annotations as a core, understudied task, (ii) propose IDEAlign for capturing expert similarity judgments via pick-the-odd-one-out tasks, and (iii) benchmark various similarity methods (text embeddings, topic models, and LLM-as-a-judge) against these human ratings. Applying this approach to two real-world educational datasets (e.g., interpreting math reasoning and feedback generation), we find that most metrics fail to capture the nuanced dimensions of similarity meaningful to experts. LLM-as-a-judge performs best (11~18% improvement over other methods) but still falls short of expert alignment, making it useful as a triage tool rather than a substitute for human review. Our work demonstrates the difficulty of evaluating open-ended LLM annotations at scale, and positions IDEAlign as a reusable protocol for benchmarking on this task to help guide responsible deployment of LLMs.

cs.CL↗

arXiv2025

Efficient RL for optimizing conversation level outcomes with an LLM-based tutor

Large language models (LLMs) built on existing reinforcement learning with human feedback (RLHF) frameworks typically optimize responses based on immediate turn-level human preferences. However, this approach falls short in multi-turn dialogue settings, such as online math tutoring. We propose a method to enhance LLM-based tutors by representing the dialogue history with a lower-dimensional latent state representation of a student and optimizing a long-term policy to determine high-level actions based on the latent state. The goal is to better align the tutor's behavior with the long-term objective of guiding the student towards solving a target math problem on their own. Our model is lightweight, requiring less computational resources than prior work of training the tutor policy end-to-end to directly output the tutor's next utterance. Our experiment results demonstrate that these modifications lead to improved long-term outcomes compared to prompting in LLM-simulated tutoring tasks.

cs.CL↗

arXiv2025

Learning to summarize user information for personalized reinforcement learning from human feedback

As everyday use cases of large language model (LLM) AI assistants have expanded, it is becoming increasingly important to personalize responses to align to different users' preferences and goals. While reinforcement learning from human feedback (RLHF) is effective at improving LLMs to be generally more helpful and fluent, it does not account for variability across users, as it models the entire user population with a single reward model, meaning it assumes that everyone's preferences are the same. We present a novel framework, Preference Learning Using Summarization (PLUS), that uses reinforcement learning (RL) to learn to produce text-based summaries of each user's preferences, characteristics, and past conversations. These summaries condition the reward model, enabling it to make personalized predictions about the types of responses valued by each user. Both the user-summarization model and reward model are trained simultaneously, creating an online co-adaptation loop. We show that in contrast to the standard Bradley-Terry model, summaries produced by PLUS capture diverse aspects of user preferences, achieving a 11-77/% improvement in reward model accuracy. Key strengths of PLUS are: (1) robust performance with new users and conversation topics, achieving a 25\% improvement over the best personalized reward model technique used for RLHF; (2) zero-shot personalization with state-of-the-art proprietary models like GPT-4 (e.g., PLUS-summary-conditioned responses achieved a 72\% win rate compared to 28% for default GPT-4o); (3) learning from flexible user contexts beyond preference labels, and (4) interpretable representation of users, enabling greater transparency and user control in pluralistic LLM alignment.

cs.LG↗

arXiv2025

EduCoder: An Open-Source Annotation System for Education Transcript Data

We introduce EduCoder, a domain-specialized tool designed to support utterance-level annotation of educational dialogue. While general-purpose text annotation tools for NLP and qualitative research abound, few address the complexities of coding education dialogue transcripts -- with diverse teacher-student and peer interactions. Common challenges include defining codebooks for complex pedagogical features, supporting both open-ended and categorical coding, and contextualizing utterances with external features, such as the lesson's purpose and the pedagogical value of the instruction. EduCoder is designed to address these challenges by providing a platform for researchers and domain experts to collaboratively define complex codebooks based on observed data. It incorporates both categorical and open-ended annotation types along with contextual materials. Additionally, it offers a side-by-side comparison of multiple annotators' responses, allowing comparison and calibration of annotations with others to improve data reliability. The system is open-source, with a demo video available.

cs.CL↗

arXiv2024

Predicting Long Term Sequential Policy Value Using Softer Surrogates

Off-policy policy evaluation (OPE) estimates the outcome of a new policy using historical data collected from a different policy. However, existing OPE methods cannot handle cases when the new policy introduces novel actions. This issue commonly occurs in real-world domains, like healthcare, as new drugs and treatments are continuously developed. Novel actions necessitate on-policy data collection, which can be burdensome and expensive if the outcome of interest takes a substantial amount of time to observe--for example, in multi-year clinical trials. This raises a key question of how to predict the long-term outcome of a policy after only observing its short-term effects? Though in general this problem is intractable, under some surrogacy conditions, the short-term on-policy data can be combined with the long-term historical data to make accurate predictions about the new policy's long-term value. In two simulated healthcare examples--HIV and sepsis management--we show that our estimators can provide accurate predictions about the policy value only after observing 10\% of the full horizon data. We also provide finite sample analysis of our doubly robust estimators.

cs.AI↗

arXiv2020

High-Throughput Computational-Experimental Screening Protocol for the Discovery of Bimetallic Catalysts

For decades of catalysis research, the d-band center theory that correlates the d-band center and the adsorbate binding energy has successfully enabled the accelerated discovery of novel catalyst materials. Recent studies indicate that, on top of the d-band center value, the full consideration of the d-band shapes describing higher moments of the d-band as well as sp-band properties can help better capturing surface reactivity. However, the density-of-states (DOS) patterns themselves have never been used as a descriptor in combined computational-experimental studies. Here, we propose the full DOS patterns as a key descriptor in high-throughput screening protocols, and prove its effectiveness. For the hydrogen peroxide (H2O2) synthesis as our demo catalytic reaction, the present study focuses on discovering bimetallic catalysts that can replace the prototypic palladium (Pd) one. Through a series of screening processes based on DOS pattern similarities (evaluated using first-principles calculations) and synthetic feasibility, 9 candidates are finally proposed out of 4,350 bimetallic alloy, which then are expected to have a catalytic performance comparable to that of Pd. The subsequent experimental tests demonstrate that 4 bimetallic catalysts (Ni61Pt39, Au51Pd49, Pt52Pd48, Pd52Ni48) indeed exhibit the catalytic properties comparable to those of Pd. Moreover, we discovered a novel bimetallic (Ni-Pt) catalyst, which has not yet been reported for H2O2 direct synthesis. In particular, Ni61Pt39 outperforms the prototypical Pd catalyst for the chemical reaction and exhibits a 9.5-fold enhancement in cost-normalized productivity. This protocol provides a new opportunity for the catalyst discovery for the replacement or reduction in use of the platinum-group metals.

cond-mat.mtrl-sci↗