Week 12Mechanistic interpretability, robustness, and research literacy
0/6
- Active project
- Research Track (S10 Interpretability, Days 67–70) & Vulnerability Analysis Track (S12 CVEs/Teardowns, Days 71–72)
- Interview focus · 2 h/day outside the 555
- Explaining correlation vs. causal intervention in interpretability, why activation patching doesn't replace application security, and how to analyze real-world AI CVEs.
This week
6 study days- Day 67Reproduce a documented logit-lens experimentR34S10
- Day 68Activation Patching ExperimentR34 R45S10
- Day 69Study alignment, sycophancy, refusal robustness, and behavioral…R31 R42S10
- Day 70Write a short reproducibility reportR34 R42S10
- Day 71Public AI CVE Selection & AnalysisR05 R35S12
- Day 72Write a vulnerability report with severity rationaleR35 R40S12