But the sampler still guesses a witness that can tell it about the hash values of a few codewords. The key is to use codes with near optimal list recovery so that knowing the colors of a few seeds doesn't tell you much about the others!

What happens when you apply this idea to the code intersection oracle? Well you get someone who can pick out sesame seeds and tell you their color, while blindfolded!

Jonas, Chinmay, Mark and I made a really powerful observation: the clonability of a classical witness means QCMA verifiers imply samplers. With spectral Forrelation, this means sampling points from a sparse subset, given its Fourier transform. (Picture from Chinmay Nirkhe)

The oracle: Basically the intersection of hash functions, a linear code, and a mask. I like to think of the code as cuts in a block of tofu, and the hash function as different colored sesame seeds (Shoutout to Kunal Marwaha), and we ask if most of the sesame seeds were removed.

Finally finished my 9+1, super excited to run the NY marathon next year! 😃

Time to start training 😅