Agentic Platform Silico Autonomously Replicates Classic Interpretability Papers

mathildepapillo · x · 2026-07-30

GoodfireAI tested its agentic research platform Silico by tasking it with replicating three classic causal interpretability papers (Interpretability in the Wild, ROME, and Mixing Mechanisms) without access to the original authors' code.

The team provided only a single prompt per task, allowing the agent to run autonomously to complete the replications.

Original post →

More from coding & agent

coding & agent channel →