AI Paper Atlas

Research map / Large language models / Alignment and preferences

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

Samuel J Paech, Allen G Roush, Judah Goldfeder, Ravid Shwartz-Ziv

ICLR 2026 · Large language models › Alignment and preferences · approach: LLM-as-judge and feedback data · 6 citations (OpenAlex)

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
use FTPO instead of DPO (2k banlist)gemma-3-12b · Creative writing, 2k banlistWriting quality rubric: 62.98 -> 68.93Improved
use FTPO instead of DPO (8k banlist)gemma-3-12b · Longform writing, 8k banlistlongform score: 26.9 -> 52.3Improved
use FTPO instead of DPO (8k banlist)gemma-3-12b · Lexical diversity, 8k banlistdiversity: 73.92 -> 95.09Improved
use FTPO instead of DPO (2k banlist)gemma-3-12b · Banlist suppression 2kban %: 82.00 -> 92.39Improved
use FTPO instead of DPO (2k banlist)gemma-3-12b · GSM8Kaccuracy: 0.847 -> 0.876Improved
use Antislop Sampler instead of token banninggemma-3-12b · Creative writing, 8k banlistWriting quality rubric: 28 -> above baselineImproved
add margin-based clipping of preference loss to FTPO (ablation)FTPO · FTPO hyperparameter ablationWriting quality rubric: 19.57 -> 67.89Improved
add target MSE tether to FTPO (ablation)FTPO · FTPO hyperparameter ablationWriting quality rubric: 39.65 -> 67.89Improved
increase lambda_mse_target from 0.05 to 0.4 (ablation)FTPO · FTPO hyperparameter ablationBanlist suppression %: 84.51 -> 55.86Hurt
increase DPO beta from 0.1 to 1.0 (ablation)DPO · Banlist 1kBanlist suppression %: 74.7 suppression, <1% quality lossMixed
train to higher preference accuracyDPO · gemma-3-12b 1k banlistWriting quality rubric: degrades after 40% accuracyHurt
restrict LoRA to lm_head for Llama-3.3-70BFTPO · Llama-3.3-70B 1k banlistBanlist suppression %: 66.41Mixed

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy