There's a transcript here.
I'm pleased to see discussion of the topic. However, since space and time are limited, I'll mostly confine my comments to the points where there is disagreement.
I think this topic is best explored using the infrastructure and terminology of information theory. Information theory has useful concepts that formalize this topic - such as the idea of Shannon mutual information - which is useful for formalizes the notion of copying. This article suffers from failing to build on this previous work.
Sylvain defends the notion of a "replicator" - which has proved to be a controversial term. The concept of as replicator was originally promoted by Richard Dawkins - with the admirable aim of enlarging the scope of evolutionary theory beyond the realm of DNA genes. However it has also resulted in much misunderstanding, confusion and criticism. Though for many, it's a foundational concept for memetics, I've generally been quite critical of the replicator terminology.
The biggest problem is that the etymology of "replicator" implies high-fidelity copying - whereas most models of evolutionary processes accept the copying fidelity as a parameter - and do not insist that copying be high fidelity.
I think that the best way to defend the notion of a "replicator" is to abandon the notion of high fidelity copying. That's the approach I take with "repology". This makes "replicator" into a misnomer - but this is still the best option for those wanting to keep the terminology.
Sylvain presents a defense of the "replicator" concept that preserves its notion of high-fidelity copying. His defense hinges on the concept of a "reader". Sylvain's "reader" is a system which identified whether two copies are identical or not. Sylvain gives the example of key copying to illustrate the concept. The lock acts as a reader and determines whether keys are functionally identical or not. Certainly in many evolving systems there are "readers". These typically perform error correction and detection. DNA copying features physical systems which act as readers. The same is true for must cultural systems which copy words. However, for other systems, it is not obvious that a "reader" exists. When ants copy each others' pheremone trails, there's no system deciding whether the behaviours are identical copies or not. Nature often doesn't care much about whether copies are identical or not. It sometimes cares about similarity - but that's a bit different. Rather than dividing the world of copies into those that are identical and those that are not, it is usually better to consider identity to be the extreme end of a continuum of varying levels of similarity.
Scientists sometimes care whether two systems are identical or not. However nature doesn't insist on the critera they use - and different scientists may use different criteria. A geneticist might treat genes with the same base pair sequences as being identical - while someone studying proteins might have a different idea about what 'identical' means in this case.
The concepts of "replicators" and "readers" may seem attractive when dealing with digital genes and memes - but they seem more like added complication when dealing with more general versions of evolutionary theory - where high-fidelity copying is not necessarily present. There, the concept of imperfect copying seems simpler. Variable-fidelity copying makes the concept of a "replicator" functionally redundant. The concept of "similarity" is broader and richer than the idea that copies are either identical or they are not.