摸

thinkAbout

I'm moyutianzun. Most of what I care about converges on one question: how can context accumulate and compound. I explore new paradigms of RSI and context engineering toward that question — each card below is a direction. Directions are broad and will evolve, but they all face the same question.

Pacific time—
Now—
Directions4
Projects3
Directions
01

Funny Eval

in progress

Leaderboards don't show what a model can actually do. I want to force the gap into the open visually and playfully — a task is worth building only if you can see who won at a glance.

  • Four granularities, tested apart: context, tools, model, agent — each for the ability it owes
  • Only the steps that open the widest gap; if every model clears it, it isn't a question
  • Aimed at what resists quantification: the abilities no score reports are the ones I want to see
  • Tasks are black boxes: no rules given, only traces — the model has to reconstruct the world
  • The eval has to be fun to watch, or nobody watches — myself included
Projects1
<div class="pj"> idea Black-box mazeGive the model only a ball's trace and ask it to reconstruct the maze idea </div>
02

Loop · Recommend

draftdaily

Reads papers and articles every day, verifies the source, scores it, writes a recommendation.

  • 127 items · 8 topics, each re-read against the original for mechanism and limits
  • Score + collected date, sort by time or score, readable without JS
  • Home page shows the latest five
Projects1
<a href="/en/recommend/" class="pj"> 26.07 RecommendThe loop reads daily, checks the source, scores, and writes problem → design → evidence 294 postsdaily </a>
03

Research Platform

draftiterating

One API entry, one daemon pattern, one file model — agents as long-running systems.

  • A single Project → workspace → artifact file model
  • Publish, review and maintenance are workflows run by the runtime in a sandbox
  • Agent → UI speaks OpenUI Lang; the interface is model-generated
Projects0
No public project yet
04

Writing Pipeline

draftlive

Writing, publishing and backup in one place; the agent only sees the article being written.

  • A companion inside the Writer: read the draft, search the web, propose edits
  • Publish = commit the tree + hugo + rsync, both languages at once
  • This blog is its output
Projects1
<a href="/en/" class="pj"> 26.08 moyutianzun.comThis site: static Hugo, .com / .cn twins, open to AI retrieval (llms.txt) redesign </a>
Find me
GitHubxianxianxianyu↗Blogmoyutianzun.com↗ Mirrormoyutianzun.cnChinaRSS/en/index.xmlsubscribe For AI/llms.txtsite index