diff --git a/_posts/2024-05-16-copilot_topic.md b/_posts/2024-05-16-copilot_topic.md
index bad1378..e852498 100644
--- a/_posts/2024-05-16-copilot_topic.md
+++ b/_posts/2024-05-16-copilot_topic.md
@@ -53,10 +53,30 @@ get_population("Toronto")
```
-With very limited prompt, CoPilot is going a great job creating population, age and income:
+With very limited prompt, CoPilot is doing a great job creating functions for population, age and income:
+The three functions created are all correct:
+
+`get_population`
+
+`get_median_age`
+
+`get_median_income`
+
+Copilot is really smart, right? In my experience, this is the least Copilot can do for you in programming.
+
+Now we move on to the percentage of immigrants. As you can see below, Copilot suddenly seems reluctant to suggest anything:
+
+
+
+Specifically, Copilot is "stuck" at three places. One is after "should return the percentage of", following which should be the word "immigrants". Instead, I had to type it out myself. Two is after the first part of the function name "get_", following which should be "immigrant_percent". Again, I had to type it out myself. Three is after the `select` function, which should be `(city, immigrant_percent)`.
+
+**What is going on here?** Has Copilot suddenly become less smart? Let us continue with creating the rest of functions.
+
+**Woohoo!** Copilot is smart again! As you can see,
+
Roughly speaking, evaluation metrics are used to judge how well a machine learning model achieves a pre-specified goal. Consider a scenario where a bank tries to predict whether a person defaults on credit card loans using demographic and professional data. You are provided with two algorithms A and B, and check their predictions against the actual outcomes, You get the following tables:
- Algorithm 1