Even better than a set<string> would be a map<string, int> to store the number of occurrences of each word.

To solve any problem the first step is to ask questions. Therefore:

Step # 1 is to fully define the problem:
A. Is 1976 a word? Are aol and AOL different words? The answer to both questions is probably no given your problem description.
B. What should the program do if the input file doesn't exist?
C. Does the instant message log require parsing to only return the words you've typed? From the description you don't want to count the other person's words.
etc.

Step #2 is to break the problem into, potentially reusable, parts:
I'd suggest something like:
1. Open file
2. Read line
3. Break line into words
4. Store word into appropriate data structure (this depends on what the language you're using supports: A SET is the obvious answer for this problem
5. Close input file
6. Count # of unique words
7. Output the count

Was there any unexpected challenges you have to overcome given the log size or was it trivial?

once i switched to perl it was no problem. trying to use powershell to read files in chunks was lolbad.

This will be a hard class for OP then imo.

edit: my 2c - c# string class has a lot of options
http://msdn.microsoft.com/en-us/library/7wtc81z6

Using a Python dictionary was the first thought I had reading the OP, but after seeing he somehow didn't care about the frequency, the -u unique flag is clearly the simplest way to do it.