Eryk

but even if that's our destiny, I'd still like to think that we each gained something from this

The last few months have been a very interesting lesson in how “optimism” and “hope” are largely completely decoupled concepts.

I’ve been trying to lose some weight as a mechanism for marginally reducing the force needed to reach escape velocity.

I had this shit on my desk for like 16 years. Most I’ve liked a mouse probably ever.

Still thinking about this guy feeding all three dragons to the obvious daisangen. Not a care in the world.

Just played the funniest hand of all time

Cool thing about working on this is that it's helping me clean my dataset. Still not perfect yet but it's much more common my little db of discard problems to have a typo than it is for the solver to be incorrect.

People like you are out there. You just have to keep looking. You just have to not give up.

A perhaps cynical take is that this implies no human is aligned either, though

*hiding a comical number of empty energy drink cans* uhhhhhh

Good times debugging at 3am years ago and finding out that torch.nn.functional.interpolate used rounding and PIL imresize used truncation despite them both being labeled as doing the same thing, so you’d get different results from the same model after torchscript export 🤕🥴

One could unfavorably compare this to meta's abrupt about-face on fact-checking immediately after similar-ish threats were made.

I could probably talk at length about this whole situation but from 1000 feet it is really good to see a lab adhere to their stated principles.

anyways, a keen observer could definitely notice some wonky edges that need to be ironed out before this is shippable. But, pretty cool!

"I will focus on developing the general functionality before adding all the rare yakus to the calculation" was definitely a thought I had this morning.

Recording demo for EV-aware efficiency trainer I'm working on and had this happen to me lmao

Hopefully soon, I’m still working out the rough edges but it’s coming along well

Tried reimplementing this 2d wwyd solver I saw online, since it was in japanese and I didn't understand how it worked. Think this might be a nice way to learn to intuit the expected values (e.g. when expensive isshanten might be better than a cheap tenpai)

"Nothing about him at all"
"Exactly. Nothing ever. Don't you understand?"

mfw solving alignment is going better than expected and we apparently need to make everything way worse for no reason

Unfortunately the “prosocial internet,” if it ever existed at all, certainly seems to be on its last legs. Largely replaced by this annoying verbal warzone. Make friends now while you can and hold on tight.

I think the texture of arguments are very different now also, and I rarely see this acknowledged. Arguing about X thing with someone who knows X thing quite well can be productive even if claws sometimes come out. But the majority of arguments on these general platforms are wasting both party’s time

This gets really obvious if you try to participate in more than one group at the same time. The experience is not better than, for example, using two distinct subculture forums, except there are way more low-post count users trying to start fights on both forums.

I’ve been thinking about this a lot and in general I think the “subcommunity embedded in broader general purpose platform” model of social media is Simply Dead now. It’s unpleasant to talk to people outside your regular cluster unless you’re seeking out the confrontation

Much like a mage or perhaps an elf, this is a wish fulfillment archetype similar in concept to something like the main character in Solo Leveling

House MD is a high fantasy television drama because it depicts the fictional concept of a doctor who heals the sick by treating symptoms

lets go dude

There just is not any other way. If I can't figure this out I'll be no better than a living zombie.