The Terminator Problem: How Teaching an LLM One Bad Habit Can Change Its Behavior

T2Covered by 1 source · first reported by The Pragmatic Engineer / ByteByteGo / Architecture Weekly at 14 Oct, 15:30 UTC

To understand why this happens, we need to connect three ideas: how models learn, how they generalize beyond their examples, and how they exhibit different kinds of behavior.

ai

Sources

• T2The Terminator Problem: How Teaching an LLM One Bad Habit Can Change Its BehaviorThe Pragmatic Engineer / ByteByteGo / Architecture Weekly14 Oct, 15:30