Possible structures for a technical talk

Edited version of a private email I sent about linux.conf.au 2010 preparation, but not conference-specific. The email discussion led to me enumerating the kind of talks that I think you can give about technical subjects at a linux-conf.au-style (non-academic developer-oriented) conference.

Recitation of facts; or the architecture diagram talk

In this kind of talk, the speaker walks the audience through, say, the architecture of a project. The talk contents are dictated by the architecture. Five minutes on the input module, five minutes on preprocessing, five minutes on transforming into a new set of matrix bases, five minutes on postprocessing, five minutes on sanity checking, five minutes on rendering.

This is the most obvious structure for a technical talk. It’s also the driest structure. It is useful mostly for audience members who want to work on or with the project code, which is for most talks a small fraction of the audience. It best suits very major projects with years of cruft in their architecture and a wide potential userbase.

Insane hack adventure; or "hacking the Tivo", Andrew Tridgell, linux.conf.au 2001

In this kind of talk, the speaker usually talks about a project he or she personally undertook and structures it as a battle against the odds. So of course Y is the most obvious solution, although not trivial. So I tried Y for a couple of months. And then it turns out that the X protocol absolutely makes Y impossible for these reasons [dramatic pause] and at this point any normal human being would have given up, but I retired to a monastery for six months and returned refreshed and a lotus blossom I saw on my return inspired me to try Q…

This is probably the most effective structure for a technical talk and it’s surprisingly widely applicable (you can frame a lot of things in terms of here’s the current problem, here’s what you think the obvious solution is, here’s what happened when I tried it). The biggest limitation is that it’s very hard to do it about work you didn’t do yourself so it’s not available to all speakers. And you can misjudge it: so of course I tried Y says the speaker uh, Y is obviously dumb, why didn’t you go straight to Q? asks the audience (at some conferences, they might well ask it out loud, repeatedly). It doesn’t work for speakers who don’t have expertise (at least narrowly on this one topic) over that the audience has.

Demonstration of surprising ease

In this talk, the speaker proposes a task which sounds too formidable to complete in the talk slot. (Pretty much any task, for some talk slots.) They then proceed to show how it can be done despite the odds using their tool. Edward Hervey edited a video PiTiVi at linux.conf.au 2008. At OSDC 2008 Thomas Lee added a ‘unless’ keyword to the Python language.

This structure can be fun but it has a lot of pitfalls. Edward was continually battling to type while dealing with a handheld microphone. Speakers should recruit an assistant for these talks. The Python keyword talk was kind of fun, but it was also hard to follow: and now I am going over to this totally other tree and adding several lines to this other file at line 1099 and 1346 and is obviously required: he did not succeed in making me think I could easily edit the Python language. In all cases, it needs a decent amount of rehearsal: even more so than other talks.

Dropping of wisdoms

These talks are generally (at linux.conf.au anyway) talks about the social/community/artful aspects of coding. For example, Jonathan Corbet’s talk about how kernel hacking actually works in terms of the trees and maintainers etc at 2009. (One of Linus Torvalds’s bigger 2009 contributions was being in the audience for that talk and adding occasional commentary on his various prejudices.) Andrew Bennetts has had some good luck with a series of talks that began at OSDC 2008, in which he essentially lists every tool he can think of to help you make your Python programs faster. I didn’t see it but I would guess Paul Fenwick’s talk about awesome things you’ve never heard of in Perl was the linux.conf.au equivalent in 2009.

These are useful talks, the main pitfall is that they require considerable expertise over the bulk of the audience. Otherwise someone is just listing a bunch of stuff you do every day back to you. They also need to be about tasks or tools that a lot of people actually use or want to use. No one wants the hidden tricks to using libnousersexceptme.