@Adám it seems that ≢∪ got a speed bump sometime since Lesson 46.

      n←?1e6⍴1e6
      ]runtime -c ≢∪n +/≠n

  ≢∪n  → 6.0E¯3 |   0% ⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕
  +/≠n → 7.1E¯3 | +19% ⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕

vs

      n←?1e6⍴1e6
      ]runtime -c ≢∪n +/≠n

  ≢∪n  → 4.9E¯3 |   0% ⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕
  +/≠n → 3.1E¯3 | -38% ⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕

@xpqz What version are you using?

@TRiG Hi there. Interested in APL?

@Adám Top one is mine (latest 18.2), bottom one whichever you used at the time, I guess 18.0?

So I realise that "got a speed bump" is potentially ambiguous language, but 18.2 seems to have a special case for ≢∪.

      n←?1e6⍴1e6
      cmpx '≢∪n' '{a←∪n⋄≢a}⍬' '+/≠n'
  ≢∪n        → 5.4E¯3 |   0% ⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕
  {a←∪n⋄≢a}⍬ → 5.3E¯3 |  -2% ⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕
  +/≠n       → 6.7E¯3 | +24% ⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕

This I find curious.

@xpqz it's probably the opposite, that ≠ was improved but then they removed that optimisation in 18.2

@rak1507 Yes, that seems consistent.

      ]runtime -c ≢∪n ≢⍸≠n

  ≢∪n  → 5.6E¯3 |   0% ⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕
  ≢⍸≠n → 9.2E¯3 | +65% ⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕

So the last expression has slowed down dramatically, too. I think you're spot on.

Shame.

@Adám I know almost nothing about it. Arrived here from the APL Orchard, and got there from Explain xkcd.

Though I have seen it in use before to answer one of my questions on this site.

@TRiG Welcome.

@TRiG Ah, cool. If you have any questions, feel free to ask. Btw, this is the APL Orchard, but Explain xkcd links to the its APL Wiki article. In general, apl.wiki is a great resource.