I've actually used this as an interview question for several years (and has been my first/only "interview-GPT" question), as it's a problem I ran into at work,

I've actually used this as an interview question for several years (and has been my first/only "interview-GPT" question), as it's a problem I ran into at work, can be solved by someone new to programming, and has lots of headroom and alternates for efficiency, optimization, and alternative implementation discussions. The use case I ran into was an initial + updated "dictionary of values" use case. Imagine: `{ "foo:bar:baz": 1.23, "bleep:bloop:blah": 4.56, ...etc.. }`. I wanted to send the first batch as a "full dictionary", and then send updated batches as: `{ "aa": 1.23, "bb": 4.56 }` whereby the client should already be able to reverse: `aa` => `foo:bar:baz`, and `bb`: `bleep:bloop:blah` since they could md5 the original key, and then match that to the "compressed" unique key. For the interview question, asking "what's the minimum unique prefix length", and exactly in the context of "how short can you abbreviate a git commit" was an excellent, low-friction, easily understandable problem. I was shocked when the initial GPT-3 nailed the naive implementation, and in retrospect, not-surprised when its superficially-correct "optimized" solution was only superficially correct. ;-) Better than 50% of the interviewees I'd asked this question of.