Conversation

Ben Kuhn
@benskuhn
Replying to
@benskuhn
So, you should be able to really speed up tacit knowledge acquisition if, instead of observing all that training data first-hand, you can instead learn about it from books/articles/whatever. This should allow you to observe the "important parts" of many more datapoints, faster.
1
7
Ben Kuhn
@benskuhn
Unfortunately, when people write down stories about... anything, they usually elide so many details that it's impossible to extract valid, meaningful tacit knowledge from them. For example, most articles about software design use extremely simplified toy example code.
2
22
Ben Kuhn
@benskuhn
Simple toy examples will show you roughly what a software design might look like, but they will do almost nothing to help you understand "if I implement this idea in the real world in my 100k-LOC codebase, what will it look like 2y later?"
1
8
Ben Kuhn
@benskuhn
If you're trying to get people to use a software design idea, simply explaining what it *is* is only like 10% of your problem. The other 90% is stuff like: - When is it reasonable to apply the idea? - What are the trade-offs vs. alternative ideas? - How large are the benefits?
1
14
Ben Kuhn
@benskuhn
Those are questions that you basically can't answer only with abstract, explicit reasoning. Instead, you need to digest a lot of training data and then answer them intuitively. (This isn't just about software design, the same things apply to any tacit knowledge heavy field!)
2
9
Ben Kuhn
@benskuhn
But in many fields, people don't seem to put nearly enough value on case studies that are detailed enough to extract real training data from.
1
7
Ben Kuhn
@benskuhn
In software, I think (good) outage postmortems are a counterexample to this—they have enough detail that you can use them to speedrun tacit knowledge acquisition.
2
15
Ben Kuhn
@benskuhn
It's interesting to think about why postmortems are particularly good at including enough detail to train intuition well. Maybe because people realize it's important to learn the maximum amount from each datapoint, because the data points (incidents) are obviously very costly?
1
10
Ben Kuhn
@benskuhn
Also maybe because there's an obvious point at which to write up a postmortem of an outage (right after the outage), whereas there's not an obvious point at which to write up a story about many other types of management decision, or software design decision, etc.
1
11
Cedric Chin
@ejames_c
Replying to
@ejames_c
and
@benskuhn
The property you’re describing is something called ‘cognitive fidelity’, as in ‘this simulation or case study has high cognitive fidelity to the real world task, and therefore leads to good skill transfer’. I’ve yet to find a generalised explication of principles though.
2