Comment by a-dub

it would be interesting to perturb the CoT context window in ways that change the sequences but preserve the meaning mid-inference.

so if you deterministically replay an inference session n times on a single question, and each time in the middle you subtly change the context buffer without changing its meaning, does it impact the likelihood or path of getting to the correct solution in a meaningful way?