I personally found almost all points they were talking about interesting. Sure, I also often get the feeling that they should dive deeper on a particular topic, or that a certain point raises more questions than it answers but then again we are clocking in at almost 3h already and I am not sure how far the term „long form content“ can (or should) be stretched
My critic has always been that Lex is superficial, which is understandable, because he is targeting casual audiences.
Like some other comments suggest, he is squandering the opportunity in a certain degree. All his question can be answered by someone who is not Chris Lattner, and the content would not be too different from what's we see right now.
I personally place high regards on Chris Lattner's technical prowess in the compiler and machine-program interfacing space. That's something that he can deliver content that no other I am aware of can do better.
Yet, all I have been seeing is something that have been repeated everywhere else, and I am too tired of this blandness.
I actually watched a couple of videos by Yannic Kilcher in the recent days, but that is a completely different format. There is probably a reason he does videos and not audio only, why he is not interviewing (and interviewing a different person with different expertise each episode).
What I value in Lex's content is that it is not complete pop-science level, that he is very neutral and especially that he leads his guest to make opinionated statements. The last point is usually the single value nugget I can extract from the episodes. It is basically a shortcut for me watching Yannic Kilcher like videos and getting deep into a topic through the proxy of a person who is not rarely one of the best in his field. I do not have the time / energy to get deeply into compilers and programming language design I just want to get a feeling in broad strokes where the field is heading, what are the key developments and bottlenecks that have shaped the past and especially future of certain technologies and applications.
I will probably dig a (tiny) bit deeper into MLIR after listening to the podcast because it seems to me to be the 80/20 kind of way to get a better feeling for the developments happening in compilers and how ML workloads are mapped to accelerators.
