This column was first written during the 2009 Cornell Probability Summer School, and is now revised, based on talks at the 2012 school. There are now five commandments:
1. Less is more: keep it simple.
2. Thou shalt not prove.
3. Do unto others as you would have them do unto you.
4. If you are going to use slides, then only use slides.
5. Make good use of your time, but when it is over, STOP!

Expanding on these Rules:

1. All you should try to do in the talk is to explain and motivate the problem, state the main result, relate it to the literature, and explain why it is interesting.

2. It is good to describe the key ideas that make the proof work, but you should not go into much detail. To gauge the right level, suppose that you are drinking a beer and talking to a friend of yours about your research.

3. I know we have all experienced pain at the talks of others. It is important to remember your reactions to other people’s talks when you give yours. When the speaker rushes headlong into his favorite proof, we often scratch our heads and say, “What the heck is he talking about?” We look at slides jammed full of text and equations and think, “I can’t read that.”

4. In the Cornell Summer School lecture room, the boards have to be lowered to reveal the screen, which is a wall painted white. Conversely, if you want to use the board then you have to raise it up. At this point you have two bad choices: write on top of the displayed slide or at the ends of the board which are dark. It is much better to have everything you want to say in the slides themselves. You can always skip them if you want. In the old days when one wrote with pens on transparencies it was difficult to change your talk at the last minute, but now all you have to do is re-TeX the file, so you can make changes up to the last minute.

5. As in hypothesis testing, there are two types of errors. The second type of error, which is rarer, but should also be avoided, is to speak for ten of your 25 minutes and then sit down. If you don’t have much to say, then don’t sign up to give a talk. The first, and more serious, error is to run over your allotted time. At conferences with parallel sessions, you need to keep within time limits so that the whole process can function smoothly, but one should do this even when there is not that type of time constraint. Returning to Rule 3, as you have probably experienced at seminars, once the allotted time is up, people’s interest in what you are saying declines faster than the tail of the normal distribution.

Since I have a little more space left I will include some wisdom I found on the internet:

The Question: How many slides for a presentation?
The Answer: It does not matter. Have the right content, say one thing per slide and finish before the allotted time. Do not worry how many slides your presentation runs into. But I will still give you a ball park number. Normally in the business presentations I have seen in my career, the presenter spends two to three minutes per slide. Hence: No. of slides = (No. of minutes) / 2
It is better to have lesser number of slides and finish before than to exceed time. By finishing early you also get more time for question and answers.”

The Cornell CS department is more poetic:

A short talk is a long abstract. If a one-hour talk corresponds to a paper, then a short talk corresponds to a long abstract. It should communicate without distracting detail.

A short talk is a captivating lead paragraph. For a reporter, the quality of that lead paragraph will determine who reads on. A successful short talk will encourage the listener to ask questions, or to buttonhole you outside the lecture hall, or to pursue your literature pointers.

A short talk is a commercial. It is a great occasion to advertise your research accomplishments and/or expository skills. But unlike a prime time spot on network TV, it should give an accurate picture of the product!

I am running out of time = space so I should stop (see Rule 5). William T. Ross has some more advice in his blog. In essence: Prepare. Practice. Less is more. Don’t include any proofs. End on time!