arXiv subjects

Roy Schwartz

Publications and source records attributed to Roy Schwartz.

At least 19 recordsLinked to original sources

arXiv2026

Global Divergence, Local Convergence: Representation Geometry in SSMs and Transformers

Recent state-space models (SSMs) such as Mamba achieve language modeling performance comparable to transformers despite relying on fundamentally different architectures. This raises an important question: how do these structural differences influence the geometry and functional nature of their internal representations? We study this question through a multi-scale analysis of representations in transformers, SSMs, and hybrid architecture. First, we find that SSMs distribute their representational information evenly across all dimensions, whereas transformer representations are heavily dominated by a single principal direction. By evaluating hybrid architectures, we observe that the representation space becomes increasingly skewed toward a single dominant direction after each attention layer. Next, we explore how the different geometric spread of representations impacts representational capacity through compressibility. Surprisingly, we find that despite their contrasting geometric structures, both architectures exhibit tightly matched effective capacities. We further investigate whether this skewed geometry affects how concepts are encoded. Using rank-constrained probes, we demonstrate that both architectures encode concepts in subspaces of surprisingly similar dimensionality. Furthermore, we demonstrate that the transformers' dominant principal direction does not inherently encode more conceptual information. Finally, we zoom in and examine the alignment between manifolds, either by analyzing representations of specific topics or by looking at the nearest neighborhoods of tokens, and find that they are highly aligned. Ultimately, our analysis suggests that while transformers and SSMs induce different usage of latent space, they display a striking functional convergence at the level of local semantic manifolds.

cs.CL↗

arXiv2026

If it is Good Then Drop it -- a Spiteful Poisson Process for Submodular Maximization

We study the problem of maximizing a general and not necessarily monotone submodular function subject to a matroid independence constraint. This problem has a rich history, with multiple algorithms using both discrete and continuous methods. Recently, [Ganz-Rozenman, Kulik, Schwartz and Singh STOC `26] presented a novel hybrid approach based on a Poisson process that aims to combine the strengths of both discrete and continuous methods for the special case of the problem where the submodular function is monotone. Our main result is a new Poisson process based hybrid algorithm that works for both non-monotone and monotone submodular functions, achieving an approximation of $ \frac{1}{e}$ for the former and $1-\frac{1}{e}$ for the latter. The algorithm always maintains a feasible set and at random times governed by the Poisson process it performs a single element swap based on a best response set. The new idea is that our algorithm is spiteful as it can purposefully discard an element that is in both the current set and the best response set. Surprisingly, this spiteful step does not harm the approximation our algorithm achieves for monotone submodular functions but is necessary for the non-monotone case. As applications, we obtain fast approximation algorithms for maximizing non-monotone submodular function subject to a general matroid independence constraint as well as faster algorithms for a partition matroid.

cs.DS↗

arXiv2026

To Compare, or Not to Compare: On Methodological Practices in Evaluating Social Bias

As Large Language Models are increasingly deployed in critical applications, robustly evaluating their social biases is paramount. However, the current literature suffers from widespread methodological fragmentation, which yields contradictory conclusions. This stems largely from ignoring the structural framing of benchmark-level evaluations. To resolve this, we introduce a unified and controllable framework that standardizes heterogeneous benchmarks to systematically contrast isolated demographic assessments with forced-choice comparative settings. Crucially, this allows us to disentangle the confounding effects of Chain-of-Thought reasoning, neutral fallback options, and other structural artifacts in social bias evaluations. Our evaluation across multiple model families reveals a massive, systematic paradigm gap: while isolated assessments limit prejudice activation, comparative settings act as aggressive catalysts for latent discrimination, a shift primarily driven by underspecified contexts. Alarmingly, CoT reasoning exacerbates social biases under comparative settings, and this systemic bias persists as a deterministic prejudice even when models are provided neutral fallback options or claim to answer randomly. Finally, we demonstrate that this comparative prejudice is a generalized phenomenon that scales positively with model size. Ultimately, we offer a crucial methodological guideline: while researchers must leverage comparative settings to robustly audit hidden biases, practitioners cannot safely rely on comparative deployments in ambiguous real-world tasks.

cs.CL↗

arXiv2026

Tight Algorithm and Hardness for Submodular Linear Ordering

We consider the Minimum Linear Ordering Problem: given a ground set $N$ of cardinality $n$ and a non-negative set function $f\colon 2^N\rightarrow \mathbb{R}_{\geq 0}$, the goal is to find an ordering $\pi$ of $N$ that minimizes the sum of the values of $f$ over all prefixes of $\pi$. This problem has been studied for various classes of set functions, and the case of a submodular $f$ is of special interest, as it captures classic problems including Minimum Linear Arrangement and Minimum Containing Interval Graph. In this work, we resolve the approximability of the Minimum Linear Ordering Problem for a general submodular $f$ by establishing matching upper and lower bounds and present: $(1)$ a polynomial-time algorithm achieving an $O(\sqrt{n/\ln n})$-approximation; and $(2)$ a matching information-theoretic hardness result, showing that no algorithm evaluating $f$ a polynomial number of times can achieve an $o(\sqrt{n/\ln n})$-approximation. Previously, the best known hardness of approximation was $2$, and an $O(\sqrt{n/\ln n})$-approximation was known only for the special case where $f$ is both submodular and symmetric.

cs.DS↗

arXiv2026

A Poisson Process for Submodular Maximization

We study the problem of maximizing a monotone submodular function subject to a matroid independence constraint. For more than a decade, a rich body of work has studied this problem. Initially, a tight approximation of $ (1-\frac{1}{e})$ was given using the continuous greedy algorithm [Calinescu-Chekuri-Pal-Vondr{\'a}k STOC`2008] and later non-oblivious local search techniques were able to match this tight approximation guarantee [Filmus-Ward FOCS`2012] and [Buchbinder-Feldman FOCS`2024]. We propose a new and remarkably simple approach to this problem that is based on a stochastic Poisson process. Our approach matches the tight $ (1-\frac{1}{e})$ approximation guarantee and it differs from the known two techniques since it does not require discretization or rounding while performing very few single element swaps. We also present applications of our approach and obtain fast algorithms for submodular welfare maximization, and for the general and separable assignment problems.

cs.DS↗

arXiv2026

Post-training is (Massive) Supervised Learning

The prevailing paradigm for training LLMs has evolved to rely on a massive post-training phase consisting of SFT and RL. In this position paper, we argue that this methodology effectively marks a reversion to the ``pre-train then fine-tune'' approach of the BERT era, explicitly tailoring models to the desired behaviors and specific benchmarks on which they are evaluated. We begin with a historical overview of LLMs, describing the different phases of the LLM evolution. We argue that the current landscape is remarkably similar to the early days of LLMs, where task performance heavily relied on fitting the models to in-distribution datasets. To empirically demonstrate this, we compare pre-trained models to randomly initialized ones, by fine-tuning both variants on modern reasoning datasets and evaluating them on competitive math and code benchmarks. We show that models post-trained from scratch yield highly non-trivial performance. Our findings suggest that current post-training methodologies function primarily as a distribution-fitting mechanism. We finish by positing that developing generally capable models and systems requires moving beyond extensive post-training for predefined behaviors, shifting instead toward training procedures where models ``learn how to learn''.

cs.CL↗

arXiv2026

Why Fine-Tuning Encourages Hallucinations and How to Fix It

Large language models are prone to hallucinating factually incorrect statements. A key source of these errors is exposure to new factual information through supervised fine-tuning (SFT), which can increase hallucinations w.r.t.~knowledge acquired during pre-training. Since these errors arise as a by-product of knowledge degradation, we explore whether established continual learning tools can mitigate them. We propose a self-distillation-based SFT method that facilitates effective factual learning while minimizing hallucinations w.r.t.~pre-existing knowledge by regularizing output-distribution drift. We also show that when new knowledge acquisition is unnecessary, suppressing factual plasticity by freezing parameter groups preserves task performance while reducing hallucinations. Lastly, we investigate the mechanism, contrasting capacity limitations, behavior cloning, and localized interference. Our experiments show that a main driver is interference among overlapping semantic representations, which self-distillation mitigates and an associative-memory model explains: forgetting grows with the overlap between new and stored facts.

cs.CL↗

arXiv2025

Vocab Diet: Reshaping the Vocabulary of LLMs via Vector Arithmetic

Large language models (LLMs) often encode word-form variation (e.g., walk vs. walked) as linear directions in the embedding space. However, standard tokenization algorithms treat such variants as distinct words with different vocabulary entries, quickly filling the size-capped token vocabulary with surface-form variation (e.g., walk, walking, Walk) at the expense of diversity and multilingual coverage. We show that many of these variations can be captured by transformation vectors: additive offsets that yield the appropriate word representation when applied to a base form embedding, in both the input and output spaces. Building on this, we propose a compact reshaping of the vocabulary: instead of assigning unique tokens to each surface form, we compose them from shared base form and transformation vectors (e.g., walked is walk+past tense). Our approach is lightweight, keeping the pretrained backbone frozen and only training small adaptation modules. We apply it across five languages and multiple LLMs in both pretraining and post-hoc adaptation, freeing 10-40% of vocabulary slots to be reallocated where tokenization is inefficient. Importantly, we do so while also expanding vocabulary coverage to out-of-vocabulary words, and with minimal impact on downstream performance. Our findings motivate a rethinking of vocabulary design, towards a representation that better matches the underlying structure of language and the practical needs of multilingual coverage.

cs.CL↗

arXiv2025

Context Length Alone Hurts LLM Performance Despite Perfect Retrieval

Large language models (LLMs) often fail to scale their performance on long-context tasks performance in line with the context lengths they support. This gap is commonly attributed to retrieval failures -- the models' inability to identify relevant information in the long inputs. Accordingly, recent efforts often focus on evaluating and improving LLMs' retrieval performance: if retrieval is perfect, a model should, in principle, perform just as well on a long input as it does on a short one -- or should it? This paper presents findings that the answer to this question may be negative. Our systematic experiments across 5 open- and closed-source LLMs on math, question answering, and coding tasks reveal that, even when models can perfectly retrieve all relevant information, their performance still degrades substantially (13.9%--85%) as input length increases but remains well within the models' claimed lengths. This failure occurs even when the irrelevant tokens are replaced with minimally distracting whitespace, and, more surprisingly, when they are all masked and the models are forced to attend only to the relevant tokens. A similar performance drop is observed when all relevant evidence is placed immediately before the question. Our findings reveal a previously-unrealized limitation: the sheer length of the input alone can hurt LLM performance, independent of retrieval quality and without any distraction. They motivate our simple, model-agnostic mitigation strategy that transforms a long-context task into a short-context one by prompting the model to recite the retrieved evidence before attempting to solve the problem. On RULER, we observe a consistent improvement of GPT-4o up to 4% on an already strong baseline.

cs.CL↗

arXiv2025

How Quantization Shapes Bias in Large Language Models

This work presents a comprehensive evaluation of how quantization affects model bias, with particular attention to its impact on individual demographic subgroups. We focus on weight and activation quantization strategies and examine their effects across a broad range of bias types, including stereotypes, fairness, toxicity, and sentiment. We employ both probability- and generated text-based metrics across 13 benchmarks and evaluate models that differ in architecture family and reasoning ability. Our findings show that quantization has a nuanced impact on bias: while it can reduce model toxicity and does not significantly impact sentiment, it tends to slightly increase stereotypes and unfairness in generative tasks, especially under aggressive compression. These trends are generally consistent across demographic categories and subgroups, and model types, although their magnitude depends on the specific setting. Overall, our results highlight the importance of carefully balancing efficiency and ethical considerations when applying quantization in practice.

cs.CL↗

arXiv2025

SpeLLM: Character-Level Multi-Head Decoding

Scaling LLM vocabulary is often used to reduce input sequence length and alleviate attention's quadratic cost. Yet, current LLM architectures impose a critical bottleneck to this procedure: the output projection layer scales linearly with vocabulary size, rendering substantial expansion impractical. We propose SpeLLM, a method that decouples input and output vocabularies by predicting character-level strings through multiple output heads. In SpeLLM, each of the $k$ linear heads predicts a single character simultaneously, enabling the model to represent a much larger output space using smaller, independent linear heads. We present a self-distillation approach for converting a standard LLM to a SpeLLM. Our experiments with four pre-trained LLMs show their SpeLLM variants achieve competitive performance on downstream tasks while reducing runtime by 5.1% on average across models. Our approach provides a potential avenue for reducing LLM costs, while increasing support for underrepresented languages and domains.

cs.CL↗

arXiv2025

Don't Overthink it. Preferring Shorter Thinking Chains for Improved LLM Reasoning

Reasoning large language models (LLMs) heavily rely on scaling test-time compute to perform complex reasoning tasks by generating extensive "thinking" chains. While demonstrating impressive results, this approach incurs significant computational costs and inference time. In this work, we challenge the assumption that long thinking chains results in better reasoning capabilities. We first demonstrate that shorter reasoning chains within individual questions are significantly more likely to yield correct answers - up to 34.5% more accurate than the longest chain sampled for the same question. Based on these results, we suggest short-m@k, a novel reasoning LLM inference method. Our method executes k independent generations in parallel and halts computation once the first m thinking processes are done. The final answer is chosen using majority voting among these m chains. Basic short-1@k demonstrates similar or even superior performance over standard majority voting in low-compute settings - using up to 40% fewer thinking tokens. short-3@k, while slightly less efficient than short-1@k, consistently surpasses majority voting across all compute budgets, while still being substantially faster (up to 33% wall time reduction). To further validate our findings, we finetune LLMs using short, long, and randomly selected reasoning chains. We then observe that training on the shorter ones leads to better performance. Our findings suggest rethinking current methods of test-time compute in reasoning LLMs, emphasizing that longer "thinking" does not necessarily translate to improved performance and can, counter-intuitively, lead to degraded results.

cs.CL↗

arXiv2025

Follow the Flow: On Information Flow Across Textual Tokens in Text-to-Image Models

Text-to-image generation models suffer from alignment problems, where generated images fail to accurately capture the objects and relations in the text prompt. Prior work has focused on improving alignment by refining the diffusion process, ignoring the role of the text encoder, which guides the diffusion. In this work, we investigate how semantic information is distributed across token representations in text-to-image prompts, analyzing it at two levels: (1) in-item representation-whether individual tokens represent their lexical item (i.e., a word or expression conveying a single concept), and (2) cross-item interaction-whether information flows between tokens of different lexical items. We use patching techniques to uncover encoding patterns, and find that information is usually concentrated in only one or two of the item's tokens; for example, in the item ``San Francisco's Golden Gate Bridge'', the token ``Gate'' sufficiently captures the entire expression while the other tokens could effectively be discarded. Lexical items also tend to remain isolated; for instance, in the prompt ``a green dog'', the token ``dog'' encodes no visual information about ``green''. However, in some cases, items do influence each other's representation, often leading to misinterpretations-e.g., in the prompt ``a pool by a table'', the token ``pool'' represents a ``pool table'' after contextualization. Our findings highlight the critical role of token-level encoding in image generation, and demonstrate that simple interventions at the encoding stage can substantially improve alignment and generation quality.

cs.CL↗

arXiv2025

On Pruning State-Space LLMs

Recent work proposed state-space models (SSMs) as an efficient alternative to transformer-based LLMs. Can these models be pruned to further reduce their computation costs? We adapt several pruning methods to the SSM structure, and apply them to four SSM-based LLMs across multiple tasks. We find that such models are quite robust to some pruning methods (e.g. WANDA), while using other methods lead to fast performance degradation.

cs.CL↗

arXiv2024

Separating Coverage and Submodular: Maximization Subject to a Cardinality Constraint

We consider two classic problems: maximum coverage and monotone submodular maximization subject to a cardinality constraint. [Nemhauser--Wolsey--Fisher '78] proved that the greedy algorithm provides an approximation of $1-1/e$ for both problems, and it is known that this guarantee is tight ([Nemhauser--Wolsey '78; Feige '98]). Thus, one would naturally assume that everything is resolved when considering the approximation guarantees of these two problems, as both exhibit the same tight approximation and hardness. In this work we show that this is not the case, and study both problems when the cardinality constraint is a constant fraction $c \in (0,1]$ of the ground set. We prove that monotone submodular maximization subject to a cardinality constraint admits an approximation of $1-(1-c)^{1/c}$; This approximation equals $1$ when $c=1$ and it gracefully degrades to $1-1/e$ when $c$ approaches $0$. Moreover, for every $c=1/s$ (for any integer $s \in \mathbb{N}$) we present a matching hardness. Surprisingly, for $c=1/2$ we prove that Maximum Coverage admits an approximation of $0.7533$, thus separating the two problems. To the best of our knowledge, this is the first known example of a well-studied maximization problem for which coverage and monotone submodular objectives exhibit a different best possible approximation.

cs.DS↗

arXiv2024

From Tokens to Words: On the Inner Lexicon of LLMs

Natural language is composed of words, but modern large language models (LLMs) process sub-words as input. A natural question raised by this discrepancy is whether LLMs encode words internally, and if so how. We present evidence that LLMs engage in an intrinsic detokenization process, where sub-word sequences are combined into coherent whole-word representations at their last token. Our experiments show that this process primarily takes place within the early and middle layers of the model. We further demonstrate its robustness to arbitrary splits (e.g., "cats" to "ca" and "ts"), typos, and importantly-to out-of-vocabulary words: when feeding the last token internal representations of such words to the model as input, it can "understand" them as the complete word despite never seeing such representations as input during training. Our findings suggest that LLMs maintain a latent vocabulary beyond the tokenizer's scope. These insights provide a practical, finetuning-free application for expanding the vocabulary of pre-trained models. By enabling the addition of new vocabulary words, we reduce input length and inference iterations, which reduces both space and model latency, with little to no loss in model accuracy.

cs.CL↗

arXiv2024

Attend First, Consolidate Later: On the Importance of Attention in Different LLM Layers

In decoder-based LLMs, the representation of a given layer serves two purposes: as input to the next layer during the computation of the current token; and as input to the attention mechanism of future tokens. In this work, we show that the importance of the latter role might be overestimated. To show that, we start by manipulating the representations of previous tokens; e.g. by replacing the hidden states at some layer k with random vectors. Our experimenting with four LLMs and four tasks show that this operation often leads to small to negligible drop in performance. Importantly, this happens if the manipulation occurs in the top part of the model-k is in the final 30-50% of the layers. In contrast, doing the same manipulation in earlier layers might lead to chance level performance. We continue by switching the hidden state of certain tokens with hidden states of other tokens from another prompt; e.g., replacing the word "Italy" with "France" in "What is the capital of Italy?". We find that when applying this switch in the top 1/3 of the model, the model ignores it (answering "Rome"). However if we apply it before, the model conforms to the switch ("Paris"). Our results hint at a two stage process in transformer-based LLMs: the first part gathers input from previous tokens, while the second mainly processes that information internally.

cs.CL↗

arXiv2024

What Can Natural Language Processing Do for Peer Review?

The number of scientific articles produced every year is growing rapidly. Providing quality control over them is crucial for scientists and, ultimately, for the public good. In modern science, this process is largely delegated to peer review -- a distributed procedure in which each submission is evaluated by several independent experts in the field. Peer review is widely used, yet it is hard, time-consuming, and prone to error. Since the artifacts involved in peer review -- manuscripts, reviews, discussions -- are largely text-based, Natural Language Processing has great potential to improve reviewing. As the emergence of large language models (LLMs) has enabled NLP assistance for many new tasks, the discussion on machine-assisted peer review is picking up the pace. Yet, where exactly is help needed, where can NLP help, and where should it stand aside? The goal of our paper is to provide a foundation for the future efforts in NLP for peer-reviewing assistance. We discuss peer review as a general process, exemplified by reviewing at AI conferences. We detail each step of the process from manuscript submission to camera-ready revision, and discuss the associated challenges and opportunities for NLP assistance, illustrated by existing work. We then turn to the big challenges in NLP for peer review as a whole, including data acquisition and licensing, operationalization and experimentation, and ethical issues. To help consolidate community efforts, we create a companion repository that aggregates key datasets pertaining to peer review. Finally, we issue a detailed call for action for the scientific community, NLP and AI researchers, policymakers, and funding bodies to help bring the research in NLP for peer review forward. We hope that our work will help set the agenda for research in machine-assisted scientific quality control in the age of AI, within the NLP community and beyond.

cs.CL↗