← Back to all news
arXiv Machine Learning By Erik Larsen

The Instability of Safety: How Random Seeds and Temperature Expose Inconsistent LLM Refusal Behavior

Read the original on arXiv Machine Learning →

The paper challenges the assumption that large language models (LLMs) produce deterministic safety responses by examining how random seeds and temperature settings affect refusal decisions. Across four instruction‑tuned models and 876 harmful prompts, 18‑28% of prompts flipped between refusal and compliance depending on sampling configuration, with higher temperatures reducing decision stability. The authors introduce a Safety Stability Index (SSI) and recommend multi‑sample evaluation protocols that account for stochastic variation rather than relying on single‑shot tests.

Machine-generated by The Flow from the publisher's headline and feed description — not written or checked by a human. The full article lives at arXiv Machine Learning.

Related stories

Hugging Face Trending Papers
Jun 28

The Joint Effect of Quantization and Sampling Temperature on LLM Safety Alignment: A Factorial Analysis

Modern LLM deployments routinely compress models and raise sampling temperature to reduce cost, latency, or repetition, yet safety evaluations usually treat these choices as fixed implementation details. This leaves a practical uncertainty: does a model that is safe at FP16 and greedy decoding remain safe after it is quantized and sampled stochastically, or do the two deployment knobs amplify one another?

llmsefficiencymultimodalsafety
More like this →
arXiv Machine Learning
Sep 17

Safety-Flag: A Unified Benchmark for the Reliability and Calibration of LLM Content Moderators

Safety-Flag is a unified benchmark that consolidates seven popular safety datasets into a single balanced flag/do‑not‑flag protocol, providing item‑level decisions and confidence scores for multiple large language models and dedicated guards. The benchmark evaluates moderator reliability across three dimensions—error direction, probability calibration, and confidence‑based error ranking—revealing that aggregate accuracy masks significant differences, such as one model flagging 85% of benign content while another misses 54% of harmful content. The study shows that general‑purpose models are overconfident, but temperature tuning can substantially improve calibration, and confidence‑based abstention can reduce selective risk, though performance varies with how well confidence ranks errors.

By Yibo Hu
llmsbenchmarkssafety
More like this →