For some reason, the code completions given by “code_davinci_002” are significantly worse than the ones given by “text_davinci_002”. I am using the default settings for both of them.
Here is an example:
query to text_davinci_002: “write a find the smallest element within two lists in java”
answer:
public static int getMin(int array1, int array2) {
int min1 = array1[0];
int min2 = array2[0];
for (int i = 1; i < array1.length; i++) {
if (array1[i] < min1) {
min1 = array1[i];
}
}
for (int j = 1; j < array2.length; j++) {
if (array2[j] < min2) {
min2 = array2[j];
}
}
return Math.min(min1, min2);
}
But when I provide this same prompt to code_davinci_002, it malfunctions and just repeats the prompt over and over again. This example is not an isolated incident as I am experiencing this problem for most prompts I provide. However, the javascript sandbox works very well for me so I am pretty sure I am doing something wrong with the settings.
Can anyone give any suggestions?
Codex works really good, but it does have a problem with knowing when to stop
You definitely need to use the stop tokens for this, and the prompts are not quite the same
here are some ideas on how to avoid that problem 
Thank you! I realized that I have to use comments instead of just raw text
. Silly me!