New Appoach for Random Number Generation???

[ttw]

You could sample the last bit or byte of two separate computations and combine them by adding or differincing mod 2 or mod 256. These should be in floating point numbers where some things get discarded. There's no reason to use the same computation. I don't know how good it would be but one could pick off the last few bits of the some numbers being used in a matrix computation (which uses lots of floating point); then add them mod 2^k with k bits. The last few bits should be almost independent and adding several of them smooths out the distribution. Note that using the most significant bits or bytes doesn't work because of Benford's law (discovered by Simon Newcomb).