Congratulations!

In every scientific field I know, the amount of published stuff exceeds the community's collective bandwidth to absorb and build upon ideas by 100x or more. Inevitably, the vast majority of what's published makes zero impact. Yet we pretend that publication itself has some value. It doesn't.

What makes a pack full?

AlphaCode clusters candidates based on similarity of outputs given (synthetically generated from problem description) inputs iirc. This would be natural here as well - same benchmark, same domain. Doesn't generalize to other domains though.

Also need a resettable environment/emulator, at least for training (and at test time if using search). “Pure logic” (math, side-effects-free code) is a self contained environment. Now, you of course know that 😀 What research do you suggest to pay attention to, which can work around this constraint?

Also, selling weird/niche things on Marketplace - interesting people just show up at the doorstep.

Art classes, small music venues, camping in popular NP campgrounds, burning man, doing small group workshops for local companies. What doesn't work (for me): co-working spaces, conferences, online communities, professional classes/courses, gym, hiking trails.

Very cool project Alex! How compatible do you think it is (or can potentially be) with iOS? `BackendCoreML` and `BackendCPU` should "just work" I believe, looking at the code. I am not familiar with MPS to tell and I wonder if there is anything else in Honeycrisp requiring macOS-only frameworks.

Thank you for the list Lily!