Evaluation-driven development at Telenor
Henrik Raaen Bø and Noah Lund Syrdal shared how ReLU uses measurable evaluation criteria throughout AI development at Telenor AI & Breakfast.
How do we know whether an AI system is doing its job? That was the focus when Henrik Raaen Bø and Noah Lund Syrdal presented ReLU’s experience with evaluation-driven development at Telenor AI & Breakfast, to an audience of more than 400 people.
The talk covered how ReLU uses clear, measurable criteria to assess AI systems throughout development. Evaluation gives a team a way to check whether its system behaves as intended as the work progresses.
For the NLP & Agents lab, this is a useful starting point for discussing how we build and assess systems whose outputs can vary between runs.