When we started @moment_dev, I could not stop thinking about a lingering, deep sense that software teams can do much better than gdocs/notion/etc. I wanted something rich, collaborative, but “just” Markdown files. Something hackable. A true offline mode, git merging/branching. So we built Moment for software teams. Everything is Markdown files and code that can be edited and customized. Buttons, tables, charts—none of this built into the product. Everything you see can be forked, changed, shared. Want (e.g.) MUI? Simply import it. No package.json reqd.

95% of what we are doing in AI is stuff that's simple enough to explain to a child, made way over-complicated by mathematical-looking notation and unclear thinking. A short rant 🧵 using attention as an example. What's attention? Say you have a bunch of input/output examples of a function. You want to predict the function output for some new point. The DUMBEST thing you can do is just find the closest input point to your point, and report its corresponding output as your prediction

k8s folks alway say "infrastructure should be boring", but we spend a lot of time on fancy non-boring stuff, like (eg) horizontal pod autoscaling.

So here's a counterfactual thread: what if k8s WAS boring?

What if `kubectl` proactively told you how your deployment was going? What if `kubectl` proactively gave you specific, information about what caused your deployment to fail? What if that information DID NOT require wiring up complex monitoring services or buying some SaaS?