4 ms·
> Circuit Tracer on Gemma-2-2b decoderesearch/circuit-tracer: https://github.com/decoderesearch/circuit-tracer https://github.com/decoderesearch/circuit-trace
by westurner 4mo ago
> Circuit Tracer on Gemma-2-2b
decoderesearch/circuit-tracer:
https://github.com/decoderesearch/circuit-tracer https://github.com/decoderesearch/circuit-tracer
ScholarlyArticle: "Dissecting the Black Box: Circuit-Level Analysis of LLM Vulnerability Detection" (2026-05)
https://arxiv.org/abs/2605.29901v1 https://arxiv.org/abs/2605.29901v1
- westurner 4mo agoExplainable AI: https://en.wikipedia.org/wiki/Explainable_artificial_intelligence https://en.wikipedia.org/wiki/Explainable_artificial_intelli... "Harmonic Loss Trains Interpretable AI Models" (2025) https://news.ycombinator.com/item?id=42941954 https://news.ycombinator.com/item?id=42941954 : > Harmonic loss enables improved interpretability and faster convergence, owing to its scale invariance and finite convergence point by design,
- westurner 4mo agoWould harmonic loss train a more explainable model for e.g. CircuitTracer to trace?