Engineering blog
Evaluation methodology, metric design and industry practice.
How to measure LLM hallucination rate rigorously
Hallucination is the biggest obstacle to shipping an LLM. This piece lays out a reproducible method for measuring hallucination rate: case design, scoring rules and metric calculation.
What to check before a model goes live
Shipping a model takes more than a working API call. Here's a pre-launch quality checklist covering three hard lines: safety, stability and compliance.
2 total