@dwarkesh_sp@ericjang11 if nearby things matter more because of convolution, could creating the same effect using patch embeddings with board state like mirroring ViT with images solve this?
@boazbaraktcs > Projects will expected to be research paper quality - do you mean publishable? if so then students would likely not be able to have large commitments or more than another class outside of this course.
I also wondered the same thing when deciding what to major in- but in 2026, the reality is the current hype, so the most popular papers + startup funding, is focused entirely on the learning aspect, which is entirely computer science based.
That is not to say there aren't areas of research/companies that aren't focused on the mechanical/electrical aspect.
imo its a very good indicator of intelligence - its not a surprise to me how LLM math benchmarks use such competitions and quant companies specifically recruit from high performers there. In my experience, in class, the best performers always come from a competition background.
i think there needs to be said difference between doing well in these comps and being top. doing well means you're partially seeking validation from these organizations who value them. being top means you're doing these competitions purely for the true love of the game
@danveloper I'm a bit confused - what do you mean by approximate a graph here? Forgive me, MoE is a topic still new to me, but I don't understand your analogy.
@LeeMcClymont@BrandonLuuMD I’m talking about revising decade later not 2 months later. You cannot physically index a decade of hadwrittwn notes better than Command F will digitally