@VitalikButerin How high-level are we talking?
There's a weird line somewhere between human-readable and math-trained-human-readable.
A lot of typical programming abstractions land, unclearly, in the middle.
@teortaxesTex Horrible, but ultimately isolated, no? Plenty of medical malpractice horror stories in my neck of the woods, too.
You gotta have more tea than this sis☕️
30 years of Windows. Still prompts you to double-check if you really want to change the extension when renaming a new file. Might make those 0 bytes unusable!
Incompetence everywhere for those with eyes to see. Our blessing, our curse.
https://t.co/8W9jpOPq3Z
I used up my full Max20 sub from Anthropic in two days because I knew full well they would drop a cope reset.
And there we are, right on cue. Haha.
A short story that teaches a moral lesson, hmm?
Okay 5.6 is WAY better than 5.5.
Fuck your benchmarks, 5.5 was like talking to an alien sometimes. Very smart in its niches but needed insane handholding. 5.6 is finally colleague material.
Anti-hallucination on the timeline, huh? Barking up my tree all of a sudden? Mandala?!
So it's time! Ya wanna know what's going on in your little boxes. Desperate for some epistemic calibration?
Okay, soon, but I get to tease a lil bit in the meantime..
FTL travel enthusiasts are gonna go ape shit over the Wick manoeuvre (or "Wick trick" if you are down with the Gen Gammas).
Remains to be seen if we can decondition you all from Euclid, though.
@_space_punk_@QiaochuYuan I find it more confusing that when people talk about what men are like vs what women are like, it's completely arbitrary nonsense dripping in archaic and usually misogynistic stereotypes.
What joy it is to free one's self of societal programming and reach the heart of being.
Simulation theory is so interesting to me.
It is a line of thinking so clearly motivated by the fundamental curiosity, or divine spark if you prefer, yet the intellectual mind can only muster a gross replication.
Hag-ridden indeed. At least it's a crack in a hard materialism.
Spent two days architecting a brilliant cross-model convergence harness for research. I can get way better outputs from mixed quality models than even Opus.
And yet 5.5 Pro still decimates it. Ahhh. My own little taste of the bitter lesson for the week.
I've been testing deepseek-v4-pro as a replacement codemonkey for Claude in my harness and it is really impressing me a lot.
Might knock down my Claude sub down a tier. Or even start considering a serious local setup..
Anyway, praise the whale.
https://t.co/aRixOy2T92
Needed to weaponise a bunch of old hardware but the options are kinda mid for this scale.
Wanted something low/no config, positioned between Slurm and Nomad. So I made this.
Might be useful if you have a small lab or render farm, a lots of boxes but no ops kinda situation 👇