Large Language Models
Research on LLMs, transformers, and language model scaling
20 papers indexed
Artificial Epanorthosis: Why large language models overuse a classical rhetorical figure, and how to mitigate it
This paper measures and analyzes the overuse of epanorthosis, a rhetorical figure, in large language models and proposes techniques to mitigate it.
Attention Is Where You Attack
The paper introduces the Attention Redistribution Attack (ARA), a white-box adversarial method that bypasses safety alignments in LLMs by manipulating the attention mechanism's geometry, showing that…
ARES: Adaptive Red-Teaming and End-to-End Repair of Policy-Reward System
Jiacheng Liang, Yao Ma, Tharindu Kumarage, Satyapriya Krishna +4 more
ARES is a novel framework that systematically discovers and mitigates dual vulnerabilities in RLHF systems by simultaneously testing the core LLM and its Reward Model (RM) using structured adversarial…
Seed Hijacking of LLM Sampling and Quantum Random Number Defense
Ziyang You, Xiaoke Yang, Zhanling Fan, Feng Guo +2 more
The paper introduces SeedHijack, a backdoor attack that manipulates the pseudorandom number generation process in LLMs to force specific token selections, and proposes a hardware quantum random number…
SDR: Set-Distance Rewards for Radiology Report Generation
The paper introduces Set-Distance Rewards (SDR), a permutation-invariant reward signal that effectively guides the generation of unordered radiology reports, significantly outperforming standard train…
Target-Side Paraphrase Augmentation for Sign Language Translation with Large Language Models
The paper proposes using GPT-4o to generate controlled paraphrases of target text for sign language translation (SLT) augmentation, achieving significant BLEU-4 improvements on PHOENIX14T.
Humanoid-GPT: Scaling Data and Structure for Zero-Shot Motion Tracking
Zekun Qi, Xuchuan Chen, Dairu Liu, Chenghuai Lin +9 more
The paper introduces Humanoid-GPT, a large-scale generative Transformer model that achieves robust zero-shot motion tracking and control by training on a massive, unified corpus of motion data.
TAHOE: Text-to-SQL with Automated Hint Optimization from Experience
The paper presents Tahoe, a system that optimizes Text-to-SQL performance through dynamic data management and hint learning.
Translation Analytics for Freelancers II: Benchmarking Local LLMs for Confidential Translation Workflows
The paper benchmarks local, offline LLMs for confidential translation workflows, demonstrating that while they are viable for privacy-sensitive use, they generally lag behind top commercial NMT system…
In-Context Reward Adaptation for Robust Preference Modeling
The paper proposes In-Context Reward Adaptation, a transformer-based framework that uses in-context learning and auxiliary signals (like human response time) to robustly model diverse and unseen human…
Listwise Cross-Encoder Fine-Tuning vs. Agentic Instruction Tuning for LLM Rerankers: A Systematic Study in Medical Procedure Reranking
This paper compares two reranking paradigms for health insurance information retrieval: small cross-encoders fine-tuned with listwise learning-to-rank objectives and a 4B-parameter instruction reranke…
GradCuit: Credit-Assigned Gradient Flow Enables Robust and Interpretable Test-Time Latent Reasoning
Zhaoxin Yu, Qi Shen, Hengli Li, Zhaowei Zhang +3 more
The paper introduces GradCuit, a method for optimizing internal reasoning in large language models at test time by inserting optimizable latent states between hidden representations and enabling direc…
Jailbreaking the Matrix: Nullspace Steering for Controlled Model Subversion
The paper introduces Head-Masked Nullspace Steering (HMNS), a novel geometry-aware attack method that achieves state-of-the-art jailbreak success rates by manipulating the internal attention mechanism…
Digital Pantheon: Simulating and Auditing Coalition Formation with LLM Agents
A multi-agent framework is presented for political coalition formation using LLMs, combining Supervised Fine-Tuning, Direct Preference Optimization, and Retrieval-Augmented Generation.
HMPO: Hybrid Median-length Policy Optimization for Chain-of-Thought Compression
Minghui Zheng, Hongxu Chen, Huimin Ren, Hongsheng Xin +7 more
HMPO introduces a single-stage, cost-effective reinforcement learning framework that achieves significant token compression of Chain-of-Thought reasoning with minimal loss of accuracy, applicable acro…
Sequential Data Poisoning in LLM Post-Training
Jack Sanderson, Yihan Wang, Xiaoqian Lu, Gautam Kamath +1 more
The paper introduces the threat model of sequential data poisoning, demonstrating that multiple, collaborating attackers can exploit compound vulnerabilities in LLM post-training pipelines that are in…
MADS: Model-Aware Diverse Core Set Selection for Instruction Tuning
Yi Bai, Wenhao Zhang, Yao Chen, Jiao Xue +2 more
The paper proposes MADS, a Model-Aware Diverse Core Set Selection method that uses LLM internal activation states to select a small, diverse core set of instructions, significantly improving model per…
Towards Identification and Intervention of Safety-Critical Parameters in Large Language Models
Weiwei Qi, Zefeng Wu, Tianhang Zheng, Zikang Zhang +3 more
The paper proposes the Expected Safety Impact (ESI) framework to identify safety-critical parameters in LLMs, introducing targeted tuning methods (SET and SPA) to enhance safety and preserve alignment…
Rethinking the Role of Temperature in Large Language Model Distillation
This paper re-examines the role of temperature ($ au$) in LLM distillation, demonstrating that while Reverse KL (RKL) is often preferred, Forward KL (FKL) significantly outperforms RKL at higher tempe…
Reconstruction of Personally Identifiable Information from Supervised Finetuned Models
This paper investigates the privacy risk of reconstructing Personally Identifiable Information (PII) from Large Language Models (LLMs) that have undergone Supervised Finetuning (SFT), proposing a nove…