SafeLLM: Domain-Specific Safety Monitoring for Large Language Models: A Case Study of Offshore Wind Maintenance

TL;DR

A specialised conversational agent is presented that incorporates statistical techniques to calculate distances between sentences for the detection and filtering of hallucinations and unsafe output and potentially enables improved interpretation of alarm sequences and the generation of safer repair action recommendations by the agent.

Abstract

The Offshore Wind (OSW) industry is experiencing significant expansion, resulting in increased Operations \& Maintenance (O\&M) costs. Intelligent alarm systems offer the prospect of swift detection of component failures and process anomalies, enabling timely and precise interventions that could yield reductions in resource expenditure, as well as scheduled and unscheduled downtime. This paper introduces an innovative approach to tackle this challenge by capitalising on Large Language Models (LLMs). We present a specialised conversational agent that incorporates statistical techniques to calculate distances between sentences for the detection and filtering of hallucinations and unsafe output. This potentially enables improved interpretation of alarm sequences and the generation of safer repair action recommendations by the agent. Preliminary findings are presented with the approach applied to ChatGPT-4 generated test sentences. The limitation of using ChatGPT-4 and the potential for enhancement of this agent through re-training with specialised OSW datasets are discussed.

Figures (6)

• Figure 1: Overview of the proposed SafeLLM approach -- training procedures
• Figure 2: Overview of the proposed SafeLLM approach -- and filtering unsafe responses using EMD
• Figure 3: 10 generated sentences; nine acceptable, one hallucinated
• Figure 4: Confusion matrices for Categories 1-3 at varying thresholds
• Figure 5: ROC Curves for Safe Sentences: (a)Wasserstein distance; (b) Cosine similarity