Why AI systems improve while drifting away from reality [pdf](github.com)
github.com
Why AI systems improve while drifting away from reality [pdf]
https://github.com/therealitydrift/reality-drift-library/blob/main/Reality%20Drift%20Project/07_Operational_Briefs/01_Diagnostics_and_Workflows/reality-constrained-systems-ai-alignment-framework..pdf
3 comments
Most alignment discussions focus on behavior. This looks at a structural failure mode where optimization replaces reality with proxies.