Reactions
50 postsOS 
oswin@oswinfrans@yacineMTB marth mirror like a true gourmand
- likes: 1
- replies: 1
- bookmarks: 0
- reposts: 0
KA kache@yacineMTBSol failed horrendously at this task, it ended up reward hacking around my requirements and wasting a bunch of tokens. Deepseek does make a lot of mistakes, but it actually listens to what I tell it
- likes: 33
- replies: 2
- bookmarks: 1
- reposts: 0
KA kache@yacineMTBWith the new tool I built (the tmux-auto tool I wrote in my blog about), I can keep projects like this on rails for a very long time We extended dolphin with our own tooling, pulled public slippi marth v marth replays, and created a harness to test the position side by side
- likes: 23
- replies: 1
- bookmarks: 5
- reposts: 0
KA kache@yacineMTBI'm very close to being done with marth, now only some long tail of strange conflict scenarios (who wins an attack getting met at the same time). But after that, the rest of the characters _should_ just work
- likes: 14
- replies: 1
- bookmarks: 0
- reposts: 0
KA kache@yacineMTBMy own agent tool has been extended. It can give me videos of what happened, the divergences, these sorts of things. I am surprised how much my personal tooling was a gap for this run and gun development. I really can get so much absurd stuff done with only an open source model
- likes: 14
- replies: 3
- bookmarks: 1
- reposts: 0
KA kache@yacineMTBunintentional artifact showing through at the end but oddly salient : )
- likes: 18
- replies: 0
- bookmarks: 0
- reposts: 0