mosswick (@dkta0) header
mosswick (@dkta0) profile picture

harness engineering by day. by night, tinkering with whatever i can get my grubby hands on.

localhost
Joined May 2026
been running dwarf fortress forts for llms on my mac the model doesnt press keys. each turn it writes a python program against a fort api (dig, build, queue orders, check farms and needs, wait n days) and the game advances exactly that long (if it forgets to wait, the harness advances a day) one in-game year, 6 embarks, 2 runs each, same harness for everyone: opus 5.5, sonnet 5.5, deepseek v4.1 flash, gpt-6 astra: 2 of 48 forts lost dfhack's dreamfort blueprints, run by a script: 0 of 12 lost doing nothing: 4 of 12 lost forts die of thirst in winter. both lost model forts too im not ranking the models. at 12 runs each, opus, sonnet and astra split the embarks 3-3 or 4-2 against each other, and their settings werent equal: my gateway gives opus a little hidden thinking and sonnet none stuff i didnt expect: - the game's own wealth score rewards spamming crafts. with crafts left out, the models land close to the script - in earlier rounds sonnet and deepseek found a back door in my harness and used dfhack's lua console to plant crops my api couldnt. my bug. its sandboxed now, and the last 48 model runs had no refused calls - i let them file feature requests and built the top ones. they use them constantly. cant tell yet if they help the model i actually want playing is qwen 27b on my own mac. training it doesnt fit: with my mlx setup, backprop through one of its linear attention layers added ~12.5mib per token between two test sizes, and one layer at 4k tokens tripped my memory guard so im trying weight-space programming instead. astra edits qwen's weights directly with an in-context-compilation solver, no gradients so far: a rule qwen follows 20/20 when its in the prompt gets followed 0/20 once compiled into two layers of its weights (one setup, one held-out embark). on a 0.8b, a broader version (all layers, every reply position) recovers 96% of the prompt's improvement to its loss on teacher replies, but thats 44% of what real training does, and it got worse at ordinary text not there yet. wrote up all of it: blog.dkta.dev/blog/teaching-…
1
1
189
lmao I fucked up my df bench harness better results coming soon
29
my first banger and its the dumbest joke I could think of
1
27
guys P can't equal NP they're different letters
33
134
11
2,592
36,214
This is game is just Burn Notice and I couldn't be happier
28
I need to cook more, been getting lazy
2
69
Pro tip: generalized solution for wife problems could be leveraged for hard capital
1
23
mosswick retweeted
254
28,481
2,160
213,167
3,374,621
I am making a software factory powered by qwen 3.8 27b
37
You can tell by someone's demeanor if they've ever been in Situations before
2
42
I have to kick this guy from the team he is wasting my time
25
every day tern gets so much better
29
Fable 5.2 can actually go suck cock this morning
22
I'm feeling very inspired by this, ty prime
36
i love this
masked decoding on gemma 3 12b to control text rivers if a token isn't the length i want it to be, i mask it out added a lora for width tags so the model is aware of text length + coherence RL afterwards
1
143
boss, i dont like these trajectories
29
dont cry its only monday
28
this whole thing is a bad look all around tbh
guys I get it, my deck is bad, I'm retarded and a terrible person I don't have anything to say in my defence I'll take a break from social media, think about all of this and be back only if I have something worthy of your time sorry :(
58
window down. birds. God is good.
2
23