WHAT NEW IN X10.28 WED

X 上的 AI 合集档案 · 22 个合集 · 58,925 条

minimax-m3.1VIDEO

用 M3.1 Flash 一晚把 400+ 条 Obsidian 笔记整理成 wiki

作者在 OpenRouter 上测过匿名 Space Bunny(连续 10 天第一),正式版跑 mcode init/exec 全权限整理笔记库:建实体页、双链、索引与操作日志,170+ tok/s,最后人工核对 diff 通过。

Mr. Buzzoni
@Mr_Buzzoni7 天前收录原推发布于 2026.10.20
Mr. Buzzoni @Mr_Buzzoni原推发布于 2026.10.20
在 X 上打开原推

I had 400+ half-finished notes rotting in my Obsidian vault. MiniMax M3.1 Flash turned them into a working wiki in one evening M3.1 Flash is the official release of "Space Bunny", and it's even more powerful. It was the stealth model I was testing on OpenRouter, where it held #1 for 10 days straight ran mcode init first, so it read the whole vault and wrote its own guidance file before editing a page then gave it the vault plus a doc on how Karpathy structures his second brain, ran exec with full permissions and left the room > read every stale note before changing anything > built entity pages and cross-linked them with [[wikilinks]] > kept an index page and an append-only log of every action > flagged notes that disagree, in both pages and in the log median output speed is 170+ tok/s, so the log never stalls 73.7% on DeepSWE v1.1, right next to the top coding models $0.15 per million input tokens, $0.60 output, and agent tasks cost 70% less than on the MiniMax model before it still read the diff by hand -> the wiki held up > the line that stuck with me: Obsidian is the IDE, the LLM is the programmer, the wiki is the codebase mcode init mcode exec "process my vault against my second-brain schema" --permission full --cwd ~/vault

x.com/mr_buzzoni/status/2112564618786652484采集于 2026.10.21 12:04