How-ToDevelopersOctober 5, 2026

Talk shows where assertion-based LLM evals break down

Read original source →youtube.com

Tejas Kumar of IBM builds a live failure case: a customer support answer passes a test because it contains the word "cannot," even though it approves a return outside store policy. The example grows into an LLM judge.

People · Tejas Kumar

1 source

More stories today

Open the live feed