oh my god

Wow. Glad it ended up mostly slapstick and not harm.

The build looks fun tho.

Building a REPL is hard

Wow! This is pixel art, but harder 🤩

That's funny. Compared to folks getting their first "Senior" title 3 years into their career in late 2010s
😁

It's also possible he's considering it part of his reproductive strategy

They have an ep about Elon but it's from 2y ago and I can't recommend what I didn't listen to 😉

Oh it was so much weirder than that. Check out "Behind the bastards" podcast about Himmler.
Bonus - latest eps are about Peter Thiel 😅

We've got an experimental tool that does the reverse - protects a string of text that's visible to the user from being accessible via any reference to DOM elsewhere. It's called LavaDome.

But the reverse has more practical use-cases.

Ultimately the goal is to create an environment in which a UI component (that someone installed from npm and never read its code) can't modify the rest of the document, only the node it's meant to render to and its children.

I'm working on LavaMoat (runtime protections against malware - a tool that stopped the recent npm malware by default) lavamoat.github.io

One gap that I've yet to close is that once you get a reference to one DOM node, you get them all.

I'd need a DOM node that doesn't expose parent and ownerDoc.

The only thing I could spot is legends in charts are too bright with their fff background. And fixing that counts as perfectionism already. 😁

I think you could inline the CSS declaration for body background to avoid the initial flash of white.

Creating an architecture upfront where interpretation of various inputs is independent. Would be harder to fool.

With the existing architecture you could put commands in as extra layer - like retraining - and data as input.

We're not ready to do that efficiently but intuitively seems possible