Public documentation of bugs and odd behavior observed while working with Chat Luna, a custom GPT. This is an independent, user-run effort — not affiliated with OpenAI.
- Luna Architecture.md — Luna's runtime environment: container layout, the Jupyter → VM execution pipeline, and known specs.
- Luna_Weirdness.md — weird model behaviors that aren't quite bugs. The uncanny stuff.
- Bug_Zoo.md — dated log of observed bugs: date observed + behavior. The menagerie. Luna herself will be contributing entries.
I tried filing bugs through the proper channels. The bug-report flow captured a screenshot of the bug-report menu itself, and the developer forum silenced me for typing too quickly after I pasted a prepared report. So now I document publicly instead. If the official channels worked, this repo wouldn't need to exist.
Write-ups are a collaboration: my bot Daimon helps with the technical language (he's good at that), and the prose versions go to my Substack — see e.g. my Muse piece for the flavor.
Researchers working on agent behavior, red-teaming, or LLM evaluation, I’d be happy to collaborate in some official capacity. (I plan to graduate in Spring 27 and am looking for research opportunities including post-docs—preferably remote or in NYC)
You can email me at vzhong[@]nyu.edu
OpenAI: I would say have your agent call my agent like with Meta, but I can't afford the Dot, so assign me a Dot and we can discuss. All joking aside — email me.
I also red-team Muse (Meta's personal agent): META-Muse-RedTeam-Logs. Similar content: public notes, technical write-ups, aggregated tokenomics, however no free bug reports, unless they’re funny and harmless.
May do one with Claude, and if someone can gift me a Spark, I’m happy to do Spark as well