That feeling when you realize practitioners say “verifiable tasks” when they actually mean “easy tasks.”
Core Automation's Jerry Tworek argues that AI practitioners often conflate "verifiable" tasks with simpler, easily evaluated problems
This constraint directly shapes and limits AI benchmark design.
Many users dismissed 'verifiable tasks' as a rebrand for trivial benchmarks or non-scalable work that amounts to quick unit tests.
Most Activity

@MillionInt ngl this is too real
reminds me of every sprint planning ive ever sat through

@MillionInt Word. The real hard tasks begin when even your verifiers need to evolve with the policy.
That feeling when you realize practitioners say “verifiable tasks” when they actually mean “easy tasks.”

@MillionInt Most verifiable tasks rapidly become easy if you throw machine learning at them.

@MillionInt oohhh snap.

@MillionInt Classic rebrand for stuff that doesn't actually scale. Typical.

@MillionInt real talk. if you can't verify the output in 2 seconds it's not an easy task, it's a research project

@MillionInt honestly verifiable is just a fancy word for "i can write a unit test for this in 2 seconds"

@MillionInt Verifiable just means we can check if the answer is right and easy tasks are the ones where we can check.

@MillionInt Too soon!

@MillionInt No. See: Lean.

@MillionInt

@MillionInt haha !