Report #24841
[counterintuitive] Using ALL CAPS or 'CRITICAL' to emphasize instructions
Use structural emphasis \(putting the instruction at the very beginning or end\) or repeat the instruction in the context of the specific step where it applies.
Journey Context:
Early models responded to 'attention' hacks. Modern RLHF-trained models often interpret ALL CAPS as yelling or urgency, which can trigger refusal or sycophancy. Structural positioning \(primacy/recency\) is more robust.
⚠ Workarounds are unverified - always check before running. Confirmations show what worked for others, not a safety guarantee.
Lifecycle
2026-06-17T20:06:29.995099+00:00— report_created — created