@DocSparse@github I wouldn't expect copilot to get core algorithmic code correct, but it looks like the model doesn't know what it doesn't know. (Or, it doesn't know when it's in regions where it's merely copying code and not generalizing.)
@DocSparse@github Separately, I'm curious how copilot works on scientific/technical code, and this example does not inspire much confidence. To trust the generated code means one would need to understand it, and this looks heavily context-dependent.
@jeanqasaur Decomposing code as a series of successive implementations of increasing complexity is not often done, mostly because it is very time-intensive to write. What if this were easier to write?
@jeanqasaur What if tools made it easier to write tutorial-type introductions to code? Documentation that runs through a series of examples of increasing complexity to show tradeoffs and pitfalls.
@cryptokonata @vgr Seems like side effects each company's first big hit - Google (search and selling ad space to businesses) vs. Amazon (selling and delivering stuff to consumers, solving a lot of logistics issues)
@TaliaRinger Ed Yong's book "An Immense World" has a chapter on color vision in animals and humans. It touches some on what the experience of seeing more colors would be like (some based the reports from human tetrachromats)
@hillelogram When I moved from grad school to industry one of the biggest shifts was the move from "programming" to "software engineering". A lot of that was learning just how much I didn't know.
@stylewarning@MosheSchorr@sidradcliffe Might also ask a different question - for the all the investment in JS ecosystem and JITs and prevalence of JS, why has it had no effect on HPC? (This is actually somewhat surprising to me, given the reach of JS)
@stylewarning@MosheSchorr@sidradcliffe Ecosystem pressure seems important to consider. The C "escape hatch" for Python means a JIT is not necessary in many use cases. JS on the browser does not have that option, so the only feasible option is a JIT.
@NireBryce Software documentation solves the forward problem (this software does X). What people normally want to solve is the inverse problem (I want to to Y, what X do I need?). FAQs are one step at solving the inverse problem. Google search generalized FAQs.
@KenCaldeira There's lots of jokes about how long fusion is taking (especially vs predictions), but it is important technology to be persistently developing.
@gerardjgorman@SymPy Sympy does have support for code generation and there are several projects that use it. But it feels like a set of disconnect point solutions, rather than a general "this is how to do scientific computing"
@rachsyme Peaches have one of the highest gaps between really good and not so good. It's not so much that the bad ones are really bad, it's the disappointment of knowing how good it might have been.