 This is a short teaser presentation for the paper, Parallel Repetition of Special Sound, Multi-Round Interactive Proofs, by Thomas Aptema and myself. In this paper, we address and partly solve a very basic question in the context of interactive proofs of knowledge. What happens to the knowledge term, meaning to the probability to cheat, when you repeat such a proof of knowledge in parallel? Noifly, you expect the knowledge to go down to zero exponentially fast with a number of repetitions, but this is not always the case. I find it actually surprising that prior to our work, very little was known about this very natural question. Let me mention that one sometimes encounters variations of the definition of a proof of knowledge in the literature, this of course changes the game. In our work, we consider the original standard definition. Let me also stress that the situation is quite different when considering the ordinary soundness error, meaning the probability to make the verifier accept a no-instance. Here it is quite well understood how the soundness error behaves under parallel repetition. However, these results and techniques do not carry over to the knowledge error, where we require extractability of a witness from any sufficiently successful prooper. Indeed, prior to our work, only sort of two extreme cases were known. On the one hand, for a general proof of knowledge, it is known that the knowledge error does not go down to zero exponentially fast or even super-pollinomially fast. In general, a noticeable knowledge error remains. On the other end of the spectrum, it is known that a very special class of protocols behave as expected, namely the two special sound sigma protocols. Here the reasoning is very simple. It is known that the two special sound sigma protocol has a knowledge error one over the size of the challenge set, and then the exponential decrease of the knowledge error follows from the observation that the parallel repetition of two special sound sigma protocol is again a two special sound sigma protocol, but now with an exponentially large challenge set. Prior to our work, nothing was known in between these two extreme cases, and in our work, we start filling this gap by extending the simple result for two special sound sigma protocols to arbitrary case special sound sigma protocols, as well as to the appropriate notion of special sound multi-round protocols. Thus for all these protocols, which cover a large portion of all the practically relevant proofs of knowledge, including bullet proofs and variations thereof, we show that the knowledge error behaves as you would wish, meaning that it goes down to zero exponentially fast on the parallel repetition. Towards proving our results, one might hope that the simple argument for two special sound protocols can be recycled for these general special sound protocols, but that's not the case. The problem is that the parallel repetition of say a three special sound protocol is not three special sound anymore. It special sounds parameter is exponentially large, which then kills the argument, even worse so for multi-round protocols. Thus, we need an entirely new way to reason which we develop in the paper. At the core is the observation that the success probability of the cheating prover is not the right figure of Mary to work with in this context. Instead, we identify and study a fine grained sort of a punctured success probability which turns out to characterize extractability and second behave nicely on the parallel repetition. So putting these two together, then leads to our results. More details will be provided by Thomas in the conference presentation. So I hope to see you there.