causal

Training models to say I do not know

Accuracy improved by training models to decline answering when search snippets lack sufficient information, because LLMs are conditioned to always be helpful, and this cannot be solved with prompt engineering alone but required reprogramming and rewiring.

causalpending

Speaker

Aravind

Evidence Quote

“LLMs are conditioned to always be helpful, will always try to serve the user's query despite what it has access to”

Source

Aravind Srinivas - Building An Answer Engine - [Invest Like the Best, EP.363]— Invest Like the Best
Created: 11/14/2026, 5:47:28 AM

Your notes

Loading notes...