Post B82ppAnDo9xhKt97Ka by ariadne@social.treehouse.systems
(DIR) More posts by ariadne@social.treehouse.systems
(DIR) Post #B82evzqhnQftvRpJUu by ariadne@social.treehouse.systems
0 likes, 0 repeats
last night, i found out about /goal. let me tell you, that is a game changer.but... did you know /goal can simply be "Read ./explorations/000-base.txt"?and if /goal can be a file in a repo, then the bot can update its own goal target and run fully independently.so now, i am having it structure its plans into a series of explorations and work through them sequentially. goal is satisfied when no further work exists.
(DIR) Post #B82fHqZWy73PWZDELg by ariadne@social.treehouse.systems
0 likes, 0 repeats
token consumption seems way down, because pre-planning into explorations allows it to focus on microtasks.once the testsuite is parallelized, this will unlock a lot of performance gains
(DIR) Post #B82fXv0EGODRtnMcF6 by valpackett@social.treehouse.systems
0 likes, 0 repeats
@ariadne why does this sound like discovering some drug dosage trick
(DIR) Post #B82fwUrkZmy44p7Mbw by ariadne@social.treehouse.systems
0 likes, 0 repeats
@valpackett i mean it is basically a context dosage trick
(DIR) Post #B82gXrjiVWGLisBAGm by tknarr@mstdn.social
0 likes, 0 repeats
@ariadne I'd be interested to see how the self-updating of the goal goes. I sense great potential to go really off-the-rails there without too much effort.
(DIR) Post #B82ppAnDo9xhKt97Ka by ariadne@social.treehouse.systems
0 likes, 0 repeats
@tknarr https://github.com/kaniini/bunix/commit/b494c6c32d1717ab1d557f552d2cc149015b60d5
(DIR) Post #B82wDxUN0bMl2SWvvE by ariadne@social.treehouse.systems
0 likes, 0 repeats
@imrehg codex
(DIR) Post #B82wEa49a0ZIgRZ2yO by imrehg@fosstodon.org
0 likes, 0 repeats
@ariadnewhich harness this is?(I have seen it in qwen code, but haven't had a good case to try yet -- I guess I haven't thought about it hard enough)