causal
Models Drift and Develop Biases Over Time
Machine learning models are trained on a fixed dataset from a specific point in time, and as they absorb more data they can learn biases or behave badly, as the Microsoft Twitter bot did, and this appears in industrial models when initial human assumptions are reweighted.
causalpending
Speaker
AshEvidence Quote
“will some biases start coming out in that model and will it start doing crazy stuff?”
Created: 11/19/2026, 2:19:59 AM
Your notes
Loading notes...