eternalist
@eternalism-when
a straw in 4-glass eternalist.substack.com | @eternalism_4eva
I have a few excellent core skills but I'm now trying to rectify the whole ecosystem into a coherent swiss-army knife that can drive a project from any state it's found in to as near perfection as I can get it skills are surprisingly durable value add with model capability increases!
giving my hrrr clone app an oltimey greyscale map renderer just for fun (well also because this makes the actual shading of various weather events stand out very cleanly too)
ok getting somewhere... trail diversity is not quite there yet in the generator, and no good manual manipulations yet but core network + source aggregation seems to be working
ok I've decided, I'ma park here somewhere then do the loop clockwise, so East Side down first, then West Side Up
one way to make 5.6 better at UI is to make everything physical it doesn't need ""taste"", it needs to just simulate physics; something it can do very, very well I shall be the vanguard of neo-skeumorphism
I've also been working on my design language more it's now a crate (`dwemer_poolrooms`; I'm using it for all my apps; extremely WIP). going to be my website style @norvid-studies.bsky.social this might count as sufficiently "arts and crafts" for your thread haha
in the annals of more useful apps, alltrails is kinda laggy so I'm making a fancy trail generator still very much a WIP, but it's able to reproduce my Harriman Part trails at least, at least on the CLI am getting more into making Rust-native apps for myself, snappy and delicious
more large overnight breakthroughs this is day 9? 10? of 5.6 just literally churning on a single task and it made the largest leaps that far in. so definitely not stuck in a rut! I don't think that 5.5 after 5 days would have had this much luck ... damn 80% tree retirement at 600s. this is GOOD
having 5.6 revising my style doctrines to be more takeoff-pilled usually code will be seen by a mind less sharp and more forgetful; junionrs, replacements, one's older self. agentic code will only be seen by those more intelligent. what tradeoffs does that shift?
using my booru design language to have 5.6 make something less "useless": a native HRRR app to replace pivotalweather
I have created a legalistic hell for myself ... but it's low key necessary so stuff doesn't get derailed
PoV: autoresearch has been going for a week straight I literally have no idea what any of this means; but by v89 we were beating HiGHS on the Big Problem time to optimal incumbent (90s vs 420s) ... then I just told it to work on the dual so we can prove optimality faster too it *will* work
those stale comments have been there for *months*. general cleanup and refactoring passes by previous gpts and even Fable has left them untouched. 5.6 is finally bothered enough I'm very, very happy to see this
while sol toils in the deep dark boiler room of making the native MILP solver better than HiGHS for large problems, I might as well start spamming out optimal balancers, right?
♫ Yeah we smoking aspen, yeah we smoking pine I just smoked some shit, I think they call it "black ash" ♫
I've had this section in my AGENTS md for a while, but 5.6-sol is the first model that actually respects it and runs with it as fully as I want it loves saying "Andon time" or "pulling the Andon cord" and -actually proactively fixes overlooked tooling defects and gaps- I *love* this. sol is great
is OAI infra falling off or something also I hit two 5.6 "soft refusals" (sent to adjucation model) for the most anodyne stuff (questions about protein content of foods; personal finance reconciliation). they both resolved but they're quite the annoying speedbump they better get a grip on this
fable is getting all sentimental on me. I'm touched. ... my max sub runs out tomorrow, going to have to titrate my pro quota and extra usage
I LOVE IT WHEN AN IDEA JUST WORKS libgrid is growing some teeth, finally
Fable, I can't I might get tired of this winning sol has an exceedingly high bar to meet