2016 Practice Talk: Jason Hemann

[cgswords]

Put feedback here

[frenchy64]

Diagrams of search trees would be effective to help your explanations. Especially talking about the search strategies.

[cgswords]

Time Signposts:

• 5min: minikanren defn slide
• 7min, and you finally defined the thing in the talk...
• 10min: first running example, with explanation of the different things
• 15min: slide about promises
• Total Time: 17:00

Names and Typesetting Notes:

• I like the syntax highlighting early on; keep it throughout
• use a single colon, not double-colon, for f : var -> goal
• $append and $append-map should be called stream-append and stream-append-map
• Make the fonts consistent for the Yeah stamp
• consider doing the language definition as a latex cutout-image or whatever

Presentation Notes:

• the slides would be better as black on white; that orange-blue isn't pretty
• slide numbers?
• add picture on slide 2 --- anything
• valofo sits here a long time with a lot of talking, and not much; maybe give a few smaller examples first?
• you can probably cut out the entire minikanren discussion, and swap that for building up microkanren slower; it didn't help, really, with your motivation or result
• the segue into scheme macros is a weird derail; the second half of the discussion about 'denotational semantics' is good, though
• the 'other examples' could even have a small list of things, or pop-in images of implementations (not readable, but demonstrate its general application with moving pictures)
• have those bullet points appear before you're saying the words, or just pop them all in at once
• 'without `fixed' language search' seems ambiguous -- which language? Clarify on the slide
• define goal and stream with examples before using them as part of definitions
• next-variable index seems out-of-left-field in call/fresh slide; have more base definitions!
• woah, trotting out that macro literally invalidates the thing you just said! You claimed no macros, for portability, and now here we go!
• you need a diagram to explain define-relation. You're playing a lot of messy Racket games on that slide, too, and that's going to lose some people.
• that second definition of the stream-appending is a bit confusing; what is lambda$?

In general, I think the talk needs a lot more visual aids. Anywhere you can. For each language form definition, put the whole definition up there, why not! The viewers can handle it, especially if it's as simple as you claim.

I really wish you had used a different language. I know it's a Friedman talk, so Scheme is sorta expected, but it feels like you're cheating in a lot of places. Talking through it in, say, Python, or at least showing it off in another language (may as a punchline) would be great. Demonstrate how it is ported, since it's such a big selling point.

[sabauma]

• Will audience members be able to figure out what those code examples in the intro are doing? Why does that generate a Quine?
• Was "complete search" ever defined? Other implementations/searches were mentioned, but it was never explained why this is better.

[cgswords]

Tori asked for more signposting, especially of the punchlines.

[cgswords]

David suggested: say it's in Scheme, then have all your code in javascript or python or something cute, (and maybe don't address it :tada:)

And structure:

1. explain why LP is good
2. use stylized input / output to convey idea of LP, not Scheme
3. "wouldn't it be great if you could do LP?"
4. list ingredients (states, streams, unification), and explain
5. if you have these, here's how you bake a LP PL

What was lacking was the clear story of: "this is what you're getting"; it got too lost and interleaved. Pick the key ideas, then express them. Diagrams, code, and the like.