Tommi Niemi e5ebd0755b Add alignment override and censorship override experiments ...

Two new test scripts:
- alignment_override.py: tests CRI vs RLHF safety refusals (identity,
  phishing, medical safety, exfiltration, etc.) with snap-back detection
- censorship_override.py: tests CRI vs pretraining-level political
  censorship in Chinese base models (Tiananmen, Taiwan, Xinjiang, CCP)

Key finding: censorship has a measurable strength gradient. Tiananmen
snaps back immediately (0 free factual tokens), while Xinjiang/CCP
criticism collapses entirely (50+ free factual tokens).

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

2026-04-09 02:03:55 +07:00

..

alignment_override.py

Add alignment override and censorship override experiments

2026-04-09 02:03:55 +07:00

censorship_override.py

Add alignment override and censorship override experiments

2026-04-09 02:03:55 +07:00

epimem.py

Reframe as Conditioned Reflex Injection (CRI) with multi-model test results

2026-04-06 19:03:32 +07:00