minot (part 4)

Show hard words

yesterday:

  • three new "greenland" benchmarks: letter count, unit conversion, part-of-speech detection. 💡 it should not be a surprise to the contemporary reader that the models struggle the most with "letter count" - how many "r"s are in strawberry.

still to do:

  • code cleanup 💡 the "run" method is written in slightly different form seven times
  • more benchmarks
  • dashboard UI improvements