Antislop: A Comprehensive Framework for Identifying and Eliminating Repetitive Patterns in Language Models

2025-10-16

AI Summary

Large language models (LLMs) frequently generate repetitive, redundant outputs—termed “slop”—which impair textual naturalness and stealth. This work proposes a systematic solution: (1) an automated slop analysis pipeline to identify fine-grained repetition patterns; (2) Antislop Sampler, a decoding-time mechanism that dynamically suppresses slop via controlled token sampling; and (3) Final Token Preference Optimization (FTPO), a token-level preference alignment method trained on inference trajectories without full model retraining. Integrating sampling control with preference-based optimization, FTPO enables end-to-end improvement with no architectural or training overhead. Experiments demonstrate that FTPO suppresses over 8,000 distinct slop patterns, achieving a 90% slop reduction rate while preserving or improving original performance on GSM8K, MMLU, and creative generation tasks. Moreover, it significantly enhances output diversity and human alignment.

Technology Category

• Natural Language Processing: (Large) Language Models
• Machine Learning: Large Multimodal Models (LMMs)
• Search and Optimization: Learning to Search

Application Category

• Economics, Online Markets and Human Computation: Cost models of using LLMs in production systems
• Search and Retrieval-Augmented AI: Search Tool Learning with LLM: Teaching LLMs to invoke search and make use of retrieved information
• User Modeling, Personalization and Recommendation: Large Language Models (LLM) for user modeling and recommendation

Abstract

Widespread LLM adoption has introduced characteristic repetitive phraseology, termed ``slop,'' which degrades output quality and makes AI-generated text immediately recognizable. We present Antislop, a comprehensive framework providing tools to both detect and eliminate these overused patterns. Our approach combines three innovations: (1) The Antislop Sampler, which uses backtracking to suppress unwanted strings at inference time without destroying vocabulary; (2) An automated pipeline that profiles model-specific slop against human baselines and generates training data; (3) Final Token Preference Optimization (FTPO), a novel fine-tuning method that operates on individual tokens, surgically adjusting logits wherever a banned pattern has appeared in an inference trace. We demonstrate that some slop patterns appear over 1,000$ imes$ more frequently in LLM output than human text. The Antislop Sampler successfully suppresses 8,000+ patterns while maintaining quality, whereas token banning becomes unusable at just 2,000. Most importantly, FTPO achieves 90% slop reduction while maintaining or improving performance in cross-domain evals including GSM8K, MMLU, and creative writing tasks. In contrast, DPO suffers significant degradation in writing quality and lexical diversity despite achieving weaker suppression. We release all code and results under MIT license: https://github.com/sam-paech/auto-antislop.

Problem

• Identifying repetitive patterns degrading language model output quality
• Detecting overused phrases making AI-generated text recognizable
• Eliminating characteristic repetitive phraseology while maintaining performance

Innovation

• Backtracking sampler suppresses unwanted strings during inference
• Automated pipeline profiles slop and generates training data
• Token preference optimization fine-tunes models surgically