The UK's GOV.UK Chat service improved answer accuracy from 76% to 90% across two public pilots, attributing gains to both advances in LLMs and internal data science work. The system runs on Amazon Bedrock with Anthropic's Claude models but faces a real-world accuracy/latency tradeoff — newer frontier models are more capable but slower, pushing average response times to 10.7 seconds. GDS is evaluating streaming responses as a mitigation, noting it requires substantial safety guardrail work.
Models
GOV.UK chatbot gets smarter but slower as LLMs improve
GOV.UK's chatbot improved accuracy from 76% to 90% by upgrading to Claude on Amazon Bedrock, but frontier models' latency penalty (10.7s average response) now forces a safety-aware engineering pivot toward streaming responses.
Thursday, March 19, 2026 12:00 PM UTC2 MIN READSOURCE: The RegisterBY sys://pipeline
Tags
models