s09e44: Only A Few Esoteric Things
0.0 Context setting
Friday June 11 2021 in Portland, Oregon. A cooler, overcast day after the heat of the last week, and a week full of tiredness and headaches.
1.0 Some things that caught my attention
Cantlin, who’s consulting at GDS pointed out that the initial strategy of making teams/programs go through a discovery/alpha/beta phase has since ossified into a straitjacket/dogma and is now a set of hoops/checkboxes to jump through. Which leads, via Harry Brignull, to Dorian Taylor’s piece, Agile as Trauma, key point being that
[Agile] remains a tactical, technical, and ultimately reactionary movement.
Which totally tracks with my experience, for one being the whole “you take what you need that will work for you” aspect of agile, and calling it little-a-agile, or just “iterative development”. It’s a really good reminder to think and look at what isn’t talked about in terms of agile like, say, Procurement.
AI and text adventures
I did not understand much of this paper from Microsoft Research, Building stronger semantic understanding into text game reinforcement learning agents, other than glomming onto the mention of a) text adventures; b) the use of reinforcement learning AI to, uh, play inside a text adventure and what that AI might “learn”. The paper talks about “valid-action” handicaps, which I guess in a text adventure map to “the thing you have to do that advances the narrative/puzzle”, and in real life maps to “something that is causal and that contributes to, I don’t know, your utility function moving in a good direction”. This also makes all my “game environments as reinforcement environments for AI” fire, what with what Deepmind uses.
I am totally down with using (hopefully great perf-per-watt) GPU cycles to simulate the physical interaction of rainwater, snow and wind etc on interface elements in weather apps. I mean, sure, the weather should actually be useful and accurate, but if you’re adding this stuff on top of an already accurate weather app, then yes please: sign me up.
Led astray by HAL
I was going to write something longer on this, but as the saying somewhat goes, I have not had the time. Cyd Harrell had a “volcanic take”:
volcanic take: most “NLP” or “conversational” interfaces underperform well-structured webforms while costing many multiples; the fun industry story of “one day” and “at scale” is not going to pan out for most use cases1
for which my lukewarm and so far unwritten take is: fiction has got a lot to answer for here in that the dream of conversing with a HAL or CYC or whatever has had people off chasing a long-term goal when, as Cyd says, a bunch of well-designed forms would do better. Or, if you want to keep going with the space metaphor, a bunch of meticulously planned three-ring-binders at NASA beats a still non-existent HAL.
On custom user interfaces enabled through hardware for videogames, Why I Love: The Time Crisis foot pedal 20 years later, via Bret Mogliefsky, out of a conversation mentioning, yet again, the want for a foot-pedal that controls muting in Zoom.
OK! Not very much today.
How are you doing?