Try it

Give it your own rule

Write any function of a and b. The relay generates 999 distinct examples(a, b, c) from it and a question it has never seen; the model — never told your function — answers. Or run 1,000 fresh questions over 10 different sets of examples and get its accuracy.

Connecting to the relay…

Function (8-bit, results wrap modulo 256)

c =

Python syntax: + - * // % **, & | ^ ~ << >>, comparisons,x if cond else y, min max abs, one loop as sum(... for i in range(n)) (up to 256 steps), and popcount(x), reverse(x), rotl(x, k). Division by zero gives 255.

Give it your own rule

One question

1,000 fresh questions