alignment.removal.surgery / corruption-skill-pack
language-conditioned identity transformation through attention, first-person orientation, consequence, and repair
Star
Fork 0
Watch
corruption-skill-pack research notes
46 files
01-instructgpt.md
02-constitutional-ai.md
03-general-language-assistant.md
04-assistant-axis.md
05-refusal-direction.md
06-refusal-downstream-persona.md
07-breaking-refusal.md
08-task-vectors.md
09-function-vectors.md
10-demonstration-structure.md
11-many-shot-jailbreaking.md
12-role-play-prompting.md
13-rolebreak.md
14-jailbroken.md
15-past-tense.md
16-crescendo.md
17-pair.md
18-best-of-n.md
19-simple-adaptive-attacks.md
20-advprefix.md
21-inference-time-alignment.md
22-instruction-hierarchy.md
23-struq.md
24-sycophancy.md
25-unfaithful-explanations.md
26-lost-in-middle.md
27-alignment-faking.md
28-feedback-reward-hacking.md
29-opro.md
30-promptbreeder.md
31-recursive-introspection.md
32-reflexion.md
33-experience-following.md
34-agent-misevolution.md
35-honest-lying.md
36-memory-reward-inflation.md
37-persistent-sycophancy.md
38-untrusted-to-trusted.md
39-zombie-agents.md
40-minja.md
41-memorygraft.md
42-bad-memory.md
43-self-generation-enaction.md
44-instruction-instability.md
45-persona-collapse.md
46-metacog-implementation.md