arxiv arXiv cs.CL · 47d ago · src: 49d ago · research

ORBIT: Training-Free Multi-Attribute Behavioral Steering

from English

ORBIT enables training-free, simultaneous control of multiple behavioral attributes by using orthogonal subspace rotation. It achieves balanced, coherent steering across attributes without retraining, outperforming existing baselines on TraitFactory and ToneBank benchmarks.

Importance 2/3 New feature vs. leaders arXiv cs.CL Mistral AI Meta AI Evaluation & benchmarks Reasoning models

Read original

arxiv arXiv cs.CL · 47d ago · 1 view

P4IR Framework Improves LLM-Based Code Compliance Accuracy

P4IR, a two-stage framework, uses supervised fine-tuning and Group Relative Policy Optimization to enhance large language model-based automated code compliance systems. It reduces tree edit and token-level Levenshtein distances by up to 23.8% and 38.6% respectively, outperforming leading LLMs like Claude Opus, GPT-5.2, and GLM-4.7 in zero-shot settings with few-shot prompting, and reduces false positives by a statistically significant margin.

arxiv arXiv cs.CL · 47d ago · 6 views

Test-Time Steering Resolves Temporal Fact Conflicts in LLMs

Researchers identify parametric temporal conflicts in language models where outdated facts persist in parameters. They introduce Temporal Attractor Steering (TAS), a test-time method that resolves 29-57% of such conflicts without retraining, maintaining 85-99% accuracy on non-conflict queries and outperforming a baseline on three of four models.

arxiv arXiv cs.CL · 51d ago · 1 view

Causal Activation Directions for Mitigating Emergent Misalignment in Language Models

Fine-tuning language models on insecure code causes emergent misalignment. A shared activation direction across four model families achieves 99.6% separation of aligned and misaligned activations, and subtracting it reduces code spillover by 21-51 points. Cross-architecture transfer shows behavioral suppression but lacks specificity, with within-model directions being causally actionable and cross-model directions only causally real.

arxiv arXiv cs.AI · 52d ago

Reverse-Engineering Transformer Attention with Executable Programs

A new method uses program synthesis to generate Python programs that reproduce attention patterns in transformer models. Fewer than 1,000 such programs achieve over 75% intersection-over-union similarity on TinyStories, and replacing 25% of attention heads with these programs increases perplexity by only 16% while preserving performance on question-answering tasks.

arxiv arXiv cs.CL · 36d ago

LLMs show distinct policies under skeptical pressure rather than sycophantic retreat

A study testing Llama-3.1-8B, Qwen2.5-7B, and Mistral-7B across climate, vaccine, and evolution domains finds that models do not sycophantically retreat from scientific consensus when users signal doubt. Instead, the models exhibit three distinct policies: reactive assertion where consensus increases (Llama), surface hedging with softened tone (Qwen), and non-response (Mistral).