Posts
All the articles I've posted.
-
My AI agent fact-checked a blog draft against a memory summary and nearly rewrote a true story
I had an AI agent fact-check a blog draft, and it used its summary memory to change a correct anecdote into a wrong one. Raw Honcho conversation logs caught it.
-
Judging broken sentences with Decisions API and Jev gave four times the false positives
I had OpenAI Decisions API and Jev judge LLM-edited notes for broken sentences. They flagged ~4x more normal notes than gpt-6-luna; split questions backfired.
-
When an LLM removed duplicates from my notes, the remaining sentences lost what they pointed to
Removing note duplicates with an LLM left 'this technology stack' and 'also' pointing at nothing. I stopped adding code rules and had an LLM judge each edit.
-
My LLM note-merge test passed, but the fixed code never ran once
I fixed an LLM note-merge bug in my Obsidian plugin and reran it on real notes. It passed, but the fixed code never ran until I restored the pre-merge state.
-
I deleted the wrong example from my LLM note-merge rule, but the rule stayed
My Obsidian plugin's LLM rule for merging duplicate notes blocked even same-product notes. I traced it, tested real merges, and adopted Karpathy's LLM Wiki.
-
Choosing a self-hosted Git server and CI/CD: GitLab, Gitea, OneDev and GitHub compared
With code that can't leave the company, I compared self-hosted GitLab, Gitea, OneDev and GitHub Enterprise Server on specs, SSO, approvals and cost.
-
In LLM summaries, a claim made once disappeared and one made twice survived
Chasing an LLM summary that dropped a source's causal claim, I ruled out grammar and counterintuitiveness with controls. A claim stated once vanished.
-
I found a rule missing from the LLM merge prompt, and filling it in changed nothing
Chasing an LLM note merge that dropped one source's content, I found the merge prompt lacked an information-density rule. Adding it changed nothing.
-
I was designing a guard against LLM note over-splitting and didn't see the one that already existed
An LLM proposed splitting one note into five pieces of 20–30 words each. Designing a guard against over-splitting, I found one built the day before.
-
The code I carefully built to catch LLM language mixing turned out to have done nothing
I built code to detect and fix Korean sentences leaking into LLM-written English notes. It fixed zero. The real cause was a prompt missing a language rule.