As It Was: Aligning LLM Search Evaluation with Historical User Preferences

Large-scale search systems evolve faster than human quality assurance can scale, especially for long-tail intents and multilingual queries. LLM-as-a-judge approaches provide a scalable alternative for evaluating the relevance of search engine result pages (SERPs), but judgments based solely on semantic similarity or world knowledge can drift from actual user preferences, particularly for ambiguous queries. We introduce a behavior-grounded LLM judge that augments each SERP item with a lightweight and auditable behavioral prior in the form of a Query-Relevance-Impressions (QRI) card. Each card summarizes how users have historically interacted with similar queries and results, providing compact empirical evidence that the judge can cite to resolve ambiguity and make more consistent relevance judgments while still relying on semantic reasoning. In a large-scale music search evaluation at Spotify, using relevance estimates derived from historical user interactions across 6,000 recomposed SERPs, the behavior-grounded judge achieves stronger alignment with user preferences, improving Spearman rank correlation by approximately 5% overall and yielding a 91% relative improvement on disagreement cases. On a multilingual human-judged dataset spanning five languages, grounding further increases correlation with human relevance judgments by 15%. Importantly, when evaluated against outcomes from a live A/B test, the grounded judge shows consistently higher alignment with the observed winning model. While absolute alignment remains moderate, these findings demonstrate that lightweight behavioral grounding can improve the reliability and practical usefulness of LLM-based evaluation in real-world search systems.

Primal-Dual Guided Decoding for Constrained Discrete Diffusion

Discrete diffusion models generate structured sequences by progressively unmasking tokens, but enforcing global property constraints during generation remains an open challenge. We propose primal-dual guided decoding, an inference-time method that formulates constrained generation as a KL-regularised optimisation problem and solves it online via adaptive Lagrangian multipliers. At each denoising step, the method modifies token logits through an additive, constraint-dependent bias, with multipliers updated by mirror descent based on constraint violation. The bias arises as the optimal KL-regularised projection of the constraint, so the constrained distribution remains as close as possible to the model's unconstrained distribution while still satisfying the constraint. The method requires no retraining and no additional model evaluations beyond standard sampling, supports multiple simultaneous constraints, and provides formal bounds on constraint violation. We evaluate our approach on topical text generation, molecular design, and music playlist generation, showing that a single algorithm instantiated via domain-specific scoring functions improves constraint satisfaction while preserving relevant domain-specific quality metrics.

Efficient Dataset Selection for Continual Adaptation of Generative Recommenders

Recommendation systems must continuously adapt to evolving user behavior, yet the volume of data generated in large-scale streaming environments makes frequent full retraining impractical. This work investigates how targeted data selection can mitigate performance degradation caused by temporal distributional drift while maintaining scalability. We evaluate a range of representation choices and sampling strategies for curating small but informative subsets of user interaction data. Our results demonstrate that gradient-based representations, coupled with distribution-matching, improve downstream model performance, achieving training efficiency gains while preserving robustness to drift. These findings highlight data curation as a practical mechanism for scalable monitoring and adaptive model updates in production-scale recommendation systems.

Deploying Semantic ID-based Generative Retrieval for Large-Scale Podcast Discovery at Spotify

Podcast listening is often grounded in a set of favorite shows, while listener intent can evolve over time. This combination of stable preferences and changing intent motivates recommendation approaches that support both familiarity and exploration. Traditional recommender systems typically emphasize long-term interaction patterns, and are less explicitly designed to incorporate rich contextual signals or flexible, intent-aware discovery objectives. In this setting, models that can jointly reason over semantics, context, and user state offer a promising direction. Large Language Models (LLMs) provide strong semantic reasoning and contextual conditioning for discovery-oriented recommendation, but deploying them in production introduces challenges in catalog grounding, user-level personalization, and latency-critical serving. We address these challenges with GLIDE, a production-scale generative recommender for podcast discovery at Spotify. GLIDE formulates recommendation as an instruction-following task over a discretized catalog using Semantic IDs, enabling grounded generation over a large inventory. The model conditions on recent listening history and lightweight user context, while injecting long-term user embeddings as soft prompts to capture stable preferences under strict inference constraints. We evaluate GLIDE using offline retrieval metrics, human judgments, and LLM-based evaluation, and validate its impact through large-scale online A/B testing. Across experiments involving millions of users, GLIDE increases non-habitual podcast streaming on Spotify home surface by up to 5.4% and new-show discovery by up to 14.3%, while meeting production cost and latency constraints.

A Unified Language Model for Large Scale Search, Recommendation, and Reasoning

LLMs are increasingly applied to recommendation, retrieval, and reasoning, yet deploying a single end-to-end model that can jointly support these behaviors over large, heterogeneous catalogs remains challenging. Such systems must generate unambiguous references to real items, handle multiple entity types, and operate under strict latency and reliability constraints requirements that are difficult to satisfy with text-only generation. While tool-augmented recommender systems address parts of this problem, they introduce orchestration complexity and limit end-to-end optimization. We view this setting as an instance of a broader research problem: how to adapt LLMs to reason jointly over multiple-domain entities, users, and language in a fully self-contained manner. To this end, we introduce NEO, a framework that adapts a pre-trained decoder-only LLM into a tool-free, catalog-grounded generator. NEO represents items as SIDs and trains a single model to interleave natural language and typed item identifiers within a shared sequence. Text prompts control the task, target entity type, and output format (IDs, text, or mixed), while constrained decoding guarantees catalog-valid item generation without restricting free-form text. We refer to this instruction-conditioned controllability as language-steerability. We treat SIDs as a distinct modality and study design choices for integrating discrete entity representations into LLMs via staged alignment and instruction tuning. We evaluate NEO at scale on a real-world catalog of over 10M items across multiple media types and discovery tasks, including recommendation, search, and user understanding. In offline experiments, NEO consistently outperforms strong task-specific baselines and exhibits cross-task transfer, demonstrating a practical path toward consolidating large-scale discovery capabilities into a single language-steerable generative model.

Evaluating Podcast Recommendations with Profile-Aware LLM-as-a-Judge

Evaluating personalized recommendations remains a central challenge, especially in long-form audio domains like podcasts, where traditional offline metrics suffer from exposure bias and online methods such as A/B testing are costly and operationally constrained. In this paper, we propose a novel framework that leverages Large Language Models (LLMs) as offline judges to assess the quality of podcast recommendations in a scalable and interpretable manner. Our two-stage profile-aware approach first constructs natural-language user profiles distilled from 90 days of listening history. These profiles summarize both topical interests and behavioral patterns, serving as compact, interpretable representations of user preferences. Rather than prompting the LLM with raw data, we use these profiles to provide high-level, semantically rich context-enabling the LLM to reason more effectively about alignment between a user's interests and recommended episodes. This reduces input complexity and improves interpretability. The LLM is then prompted to deliver fine-grained pointwise and pairwise judgments based on the profile-episode match. In a controlled study with 47 participants, our profile-aware judge matched human judgments with high fidelity and outperformed or matched a variant using raw listening histories. The framework enables efficient, profile-aware evaluation for iterative testing and model selection in recommender systems.

NeurIPS 2025 E2LM Competition : Early Training Evaluation of Language Models

Existing benchmarks have proven effective for assessing the performance of fully trained large language models. However, we find striking differences in the early training stages of small models, where benchmarks often fail to provide meaningful or discriminative signals. To explore how these differences arise, this competition tackles the challenge of designing scientific knowledge evaluation tasks specifically tailored for measuring early training progress of language models. Participants are invited to develop novel evaluation methodologies or adapt existing benchmarks to better capture performance differences among language models. To support this effort, we provide three pre-trained small models (0.5B, 1B, and 3B parameters), along with intermediate checkpoints sampled during training up to 200B tokens. All experiments and development work can be run on widely available free cloud-based GPU platforms, making participation accessible to researchers with limited computational resources. Submissions will be evaluated based on three criteria: the quality of the performance signal they produce, the consistency of model rankings at 1 trillion tokens of training, and their relevance to the scientific knowledge domain. By promoting the design of tailored evaluation strategies for early training, this competition aims to attract a broad range of participants from various disciplines, including those who may not be machine learning experts or have access to dedicated GPU resources. Ultimately, this initiative seeks to make foundational LLM research more systematic and benchmark-informed from the earliest phases of model development.

Text2Tracks: Prompt-based Music Recommendation via Generative Retrieval

In recent years, Large Language Models (LLMs) have enabled users to provide highly specific music recommendation requests using natural language prompts (e.g. "Can you recommend some old classics for slow dancing?"). In this setup, the recommended tracks are predicted by the LLM in an autoregressive way, i.e. the LLM generates the track titles one token at a time. While intuitive, this approach has several limitation. First, it is based on a general purpose tokenization that is optimized for words rather than for track titles. Second, it necessitates an additional entity resolution layer that matches the track title to the actual track identifier. Third, the number of decoding steps scales linearly with the length of the track title, slowing down inference. In this paper, we propose to address the task of prompt-based music recommendation as a generative retrieval task. Within this setting, we introduce novel, effective, and efficient representations of track identifiers that significantly outperform commonly used strategies. We introduce Text2Tracks, a generative retrieval model that learns a mapping from a user's music recommendation prompt to the relevant track IDs directly. Through an offline evaluation on a dataset of playlists with language inputs, we find that (1) the strategy to create IDs for music tracks is the most important factor for the effectiveness of Text2Tracks and semantic IDs significantly outperform commonly used strategies that rely on song titles as identifiers (2) provided with the right choice of track identifiers, Text2Tracks outperforms sparse and dense retrieval solutions trained to retrieve tracks from language prompts.

Policy-as-Prompt: Rethinking Content Moderation in the Age of Large Language Models

Content moderation plays a critical role in shaping safe and inclusive online environments, balancing platform standards, user expectations, and regulatory frameworks. Traditionally, this process involves operationalising policies into guidelines, which are then used by downstream human moderators for enforcement, or to further annotate datasets for training machine learning moderation models. However, recent advancements in large language models (LLMs) are transforming this landscape. These models can now interpret policies directly as textual inputs, eliminating the need for extensive data curation. This approach offers unprecedented flexibility, as moderation can be dynamically adjusted through natural language interactions. This paradigm shift raises important questions about how policies are operationalised and the implications for content moderation practices. In this paper, we formalise the emerging policy-as-prompt framework and identify five key challenges across four domains: Technical Implementation (1. translating policy to prompts, 2. sensitivity to prompt structure and formatting), Sociotechnical (3. the risk of technological determinism in policy formation), Organisational (4. evolving roles between policy and machine learning teams), and Governance (5. model governance and accountability). Through analysing these challenges across technical, sociotechnical, organisational, and governance dimensions, we discuss potential mitigation approaches. This research provides actionable insights for practitioners and lays the groundwork for future exploration of scalable and adaptive content moderation systems in digital ecosystems.

PODTILE: Facilitating Podcast Episode Browsing with Auto-generated Chapters

Listeners of long-form talk-audio content, such as podcast episodes, often find it challenging to understand the overall structure and locate relevant sections. A practical solution is to divide episodes into chapters--semantically coherent segments labeled with titles and timestamps. Since most episodes on our platform at Spotify currently lack creator-provided chapters, automating the creation of chapters is essential. Scaling the chapterization of podcast episodes presents unique challenges. First, episodes tend to be less structured than written texts, featuring spontaneous discussions with nuanced transitions. Second, the transcripts are usually lengthy, averaging about 16,000 tokens, which necessitates efficient processing that can preserve context. To address these challenges, we introduce PODTILE, a fine-tuned encoder-decoder transformer to segment conversational data. The model simultaneously generates chapter transitions and titles for the input transcript. To preserve context, each input text is augmented with global context, including the episode's title, description, and previous chapter titles. In our intrinsic evaluation, PODTILE achieved an 11% improvement in ROUGE score over the strongest baseline. Additionally, we provide insights into the practical benefits of auto-generated chapters for listeners navigating episode content. Our findings indicate that auto-generated chapters serve as a useful tool for engaging with less popular podcasts. Finally, we present empirical evidence that using chapter titles can enhance effectiveness of sparse retrieval in search tasks.