When your slides have animations, the PDF will likely be screwed up. For example, here is the only animated slide I had in the talk. You can see that the PDF doesn't represent the beginning or the ending state of the animation. Were I posting this one, I'd change the slide so that it had the end of the animation and looked good in the PDF.
Here's the 2015 slide where I convert the christmas tree to a standard graph:
And here's the 2016 slide:
I'm sure I was writing the slides quickly in 2015, and simply wrote what I was thinking. Remember the advice from BVZ -- keep the text pithy, and you don't have to write complete sentences.
Similarly, in 2015, I put the whole problem description on one slide, and animated the bullets to present them one-by-one. Here's the final slide:
In 2016, I split these over two slides, and I think it presents more clearly that way. One thing to remember, which is hard to remember, is that when you're writing a paper, you often want to maximize information-per-space, and you want the final result to be a reference material. With a slide, you are maximizing clarity of presentation, which means that slides don't need to be encyclopedic, and you don't need to pack too much information into each slide. If you post the PDF, the readers will understand that they often need to synthesize information across multiple slides.
Similarly, in 2015, I was too chatty about the alternative of disjoint sets, making for a cluttered slide that was hard to read:
In 2016, I made it much pithier, and instead simply described the implementations verbally. Both of covered in detail in CS302, so I could assume my audience knew about them:
I put the details of the implementations on a separate slide:
The last major slide change from 2015 to 2016 had to do with the Minimum Spanning Tree implementations:
It's a junky slide with too much clutter. In 2015, I presented it as one slide. In 2016, I presented it as three, adding the Prim box second, and the Kruskal box third. That way, I could keep my audience's focus on what I was talking about, rather than reading text and trying to figure out what it's saying. Each of Prim and Kruskal has some subtletly in begin able to remove the log component from the algorithm, and it's better to have the audience focus on me, rather than reading ahead.