Author here. I will take the opportunity to advertise my blog at https://fredrikj.net/blog/ where I post occasional development updates about Arb and other projects.
I dream of a programming language where I would be able to specify a computation and maybe error bounds on input/output and get efficient code respecting the bounds automatically. Arb takes us close to that, but I still have to handle recomputing with greater precision by hand, and things get complicated with multiple inputs. What are your thoughts on making arbitrary precision/rigorous error bounds more ergonomic to use for programmers?
That's a great question. Of course, you can express the computation as a function or a symbolic expression and evaluate it using a function that recomputes with greater precision automatically (or automatically compiles efficient code with bounds). Integrating this kind of functionality into the core semantics of a programming language is a very interesting problem.
Good point. I added an RSS feed link to https://fredrikj.net/blog/. I just hacked together some quick code to generate it, so let me know if it doesn't work.
Thanks! Just tried to add it to feedly but unfortunately it still gives me an error, reporting that it can't find feed. Will have a look at the source to see if I can help figuring it out, once I am back at my computer.