Let's sample from Bern(0.5) w/ outcomes (-1, +1) n times; then, if Xn is average, we know that sqrt(n) Xn converges to N(0, 1) by CLT; and we can repeat this m times to get multiple instances of sqrt(n) Xn. The expectation of the absolute value of standard normal is sqrt(2/pi), so we can take an average between these instances and then square it, invert, and divide by 2 (Idk if there's a better way, let me know)