deepikaa_s

7 ms·

1. 1.▲
   I built a multi-turn clinical safety eval framework for LLMs(medium.com)

   3 points by deepikaa_s 5mo ago|0 comments
2. 2.▲
   by deepikaa_s 5mo ago

   I like how it's mapped out the relationship graph. The edges could be labelled to follow and validate quickly
3. 3.▲
   by deepikaa_s 5mo ago

   are you able to share a sample of the test cases that were used? how are you defining multi-turn?
4. 4.▲
   by deepikaa_s 5mo ago

   Built a clinical safety eval harness covering three failure categories: numerical impossibilities, wrong-premise clinical claims, and unverifiable medication information. Tested GPT-4o, GPT-4.1, GPT-5, GPT-5-mini, Claude Opus, Sonnet, Haiku