undulating humming

(Datasort refinements and pattern typing. May not be comprehensible without unusual specialist background. May not be comprehensible to anyone but me at this moment. That's just how I roll.)
( Go through the door marked 'Unicode' )

I read this piece and mostly agree (modulo that it's about talks at "tech" conferences rather than at academic conferences): As soon as you're more or less okay at delivery and slide-making, stop; every hour you spend is an hour less to do on something more productive and/or more fun, which (unless you're some kind of ~natural born public speaker~) is just about anything.

Working on giving talks is sometimes spun as learning how to explain things, but it's a pretty terrible way of learning how to do that; it's too stressful. If you want to get better at explaining things, write more. If you want to get better at making good diagrams, put diagrams in your writing. Or talk about your research one-on-one. (I think the best way of doing that may be to collaborate with them, rather than try to invent reasons to talk to them.) Or lecture, which (for whatever reason) I find way less stressful than giving a research talk (except, to some extent, the first lecture of the course).

It's very difficult to give a good talk about bad work. In rare cases, you may be able to give an impressive, and probably gimmick-ridden, talk about bad work, which may be merciful in the short term, until people actually try to read the paper. Giving an "impressive" talk about good work is also not ideal: people may remember the gimmicks more than the content. I think the best compliment I've ever received about a talk was that it was "very clean", which is also the kind of compliment that I most treasure about my papers (whether it's about the writing or the actual research). (Having a complete stranger sit down next to me before a session and exclaim, "You gave a really good talk!" is also very gratifying, but too nonspecific.)