Once you figure that out, then you go to booked shows (shows where you are specifically selected to perform) in front of big audiences like at the Improv or the Comedy Store. This is where you use your best jokes. But you can't know your best jokes before you try them out. You never know if something will be funny until you try it.
These are the fuzzy rules. Some people use improvisation at booked shows and some people perform the same jokes over and over again at open mics without experimenting but I'd say this is the standard path to take.
True, but becoming less true in these days of ubiquitous high-res cameraphones. Used to be that if you tried out some risky or potentially offensive new material in front of ten drunks, the worst-case scenario was that you pissed off the ten drunks. Now you have to worry about pissing off the whole world, because any one of the ten drunks can record you and upload the video to YouTube for the world to see. Which makes it harder to "A/B test" new material than it used to be.
I don't know of a solution to this. No one wants to hear the same jokes for a decade. To get new "best stuff" you have to experiment. When you experiment, it may not be funny/ready. And with today's technology, it can be recorded at any time.
Similarly, we have a Circus Open Mic in Seattle (https://www.facebook.com/events/533760116677373/) where ground rules are that admission is free and open to the public but it's generally a supportive and in-community audience. We've had comedians, magicians, aerialists, acrobats, jugglers, extremely weird people, clowns, and strip history lessons.
I present both of these as ideas for ways to develop and grow without being subject to eternal Google search indexing and YouTubing by random civilians. I think both of these seem to come down to controlling the venue and rules. If you experiment in a bar or club, you're in the wild.
We do these in theatres, black boxes, and dance/performance spaces. This allows us to control the door.
Tiny onions are the most funny food.
* Getting a joke to work best often requires A/B testing
* I tried to choose my best jokes objectively. I'd watch my videos and keep track of the jokes that got the laughs consistently. I then wrote a program to sort jokes by the best ratio.
* There are lots of rules (though they are fuzzy). This kind of reminds me of programming language syntax. Like you should attempt to make the punchline the last part of the sentence. Start with your funniest joke, end with your second funniest (this rule might be reversed, I can't remember).
* Try to be as concise as possible. Usually if you can cut out words your joke will be more effective. This reminds me of removing DRY violations in code.
* Some jokes are really technical. They'd consistently get a laugh if every word was said in the exact tone and order they're supposed to, but if anything about the joke changed, it wouldn't "compile".
* Like programming, it'll probably take 10 years to get good at it (see http://norvig.com/21-days.html )
* Sometimes you'd think of a tag for an already existing joke. A tag is a joke that only makes sense in the context of the immediately preceding joke. This is like adding a feature to a legacy code base. It often requires refactoring and maintenance.
I like the way Marc Maron put it: One liners are the Math majors of comedy. Story tellers are the English majors.