From 409bfe3cce2c8d3b0143e2cbd15bc6ba61a5113e Mon Sep 17 00:00:00 2001 From: Zen Date: Thu, 11 Jan 2024 12:04:00 -0800 Subject: [PATCH] Update README.md --- README.md | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/README.md b/README.md index 1e9f36c0..57f76593 100644 --- a/README.md +++ b/README.md @@ -84,7 +84,7 @@ We see interventions are knobs that can mount on models. And people can share th ## Causal Abstraction: From Interventions to Gain Interpretability Insights -Basic interventions are fun we cannot make any causal claim systematically. To gain actual interpretability insights, we want to measure the counterfactual behaviors of a model in a data-driven fashion. In other words, if the model responds systematically to your interventions, then you start to associate certain regions in the network with a high-level concept. We also call this alignment search process with model internals. +Basic interventions are fun but we cannot make any causal claim systematically. To gain actual interpretability insights, we want to measure the counterfactual behaviors of a model in a data-driven fashion. In other words, if the model responds systematically to your interventions, then you start to associate certain regions in the network with a high-level concept. We also call this alignment search process with model internals. ### Understanding Causal Mechanisms with Static Interventions Here is a more concrete example,