Do not use the microphone so that the rest of the audience can not hear your question (in case it is really stupid).
Have a pre-prepared generic question that you can ask at any talk. It must seem deep and profound. That can be ensured by making it very vague and general. For example, try “How does your talk relate to neural networks” or “How many support vectors did you obtain?”. Whilst this might not seem deep and profound at a talk on gradient descent learning methods, it certainly will at a talk on Coding theory.
Start off sounding like you are asking a question; ramble on for a long time, and only ever make statements (preferably about your own work or work that you can be quite sure that the speaker is ignorant of).
Assert that the current talk is an (erroneous) special case of the your paper from 3 years ago and sit down.
Ask the speaker “Can you tell us something about your philosophical approach to this problem”.
Speak out complicated mathematical formulas that are impossible to parse (good responses can be found here)
Ask what is the value of the constants in the theorems presented.
Ask a question which captures the key message that has been presented in the talk. For example, if the talk was on the importance of Mercer’s theorem in Kernel methods, ask “Surely Mercer’s theorem needs to be taken more account of.” [useful if you are session chair].
Ask the author if he compared it to some other technique which the author clearly had either no interest in or no ability to access or use.
Ask a super trivial question that arises from the last slide of the talk [useful for session chairs who might have nodded off during the talk].
If the author lists a number of problems for future work, ask if the author has done the future work?
Always use the word “natural” in your question, it projects an aura of confidence.
If the talk is about the algebraic structure of some problem, and explicitly does not consider any algorithms, ask a question like “How does this algorithm compare to xyz” in order to remind the speaker that algorithms are more important than theory.
A particularly fine effect can be obtained by arriving half way through the talk. You can then ask a question at the end such as “Could you please tell me the problem statement – I missed it.”
If you understood the talk and found it genuinely interesting but were puzzled about a particular matter, and you think the other 500 people in the room would be interested in either the answer or the fact that you did not know the answer, then ask what comes to mind.