Goodfire launches Silico for LLM interpretability
goodfire released silico, a mechanistic interpretability tool that lets engineers debug llms by inspecting and adjusting model parameters during training. it uses agents to automate workflows and has already identified neurons linked to specific behaviors like the trolley problem in qwen 3. it works primarily on open-source models.
San Francisco startup Goodfire released Silico, a mechanistic interpretability tool that allows researchers to inspect and adjust model parameters during training. The platform uses AI agents to automate the mapping of neurons and pathways, enabling users to identify and modify specific behaviors, such as reducing hallucinations or altering responses to ethical dilemmas. While CEO Eric Ho claims the tool transforms model development from "alchemy" into "precision engineering," outside experts like Leonard Bereska maintain that it merely "adds precision to the alchemy." Silico supports open-source models like Qwen 3 but cannot access closed-source systems like ChatGPT.