Roadmap priority: B16
Summary
Re-add the baba-is-ai demo to the bundled library. It was pulled during the v1.0.0 library submission over a config-file issue flagged in review.
Why it matters
It is one of the more research-grounded demos — it reproduces the knowing-doing gap from BALROG (arXiv:2411.13543), where LLMs score near zero on spatial puzzle games despite being able to state the correct rule. Good demonstration material and a genuine research result.
What needs to be done
- Recover the exact reviewer note from the NetLogo-Libraries submission — the issue was an extensionless config file that did not fit the bundle conventions.
- Rename or restructure the config so it conforms (compare
demos/tests/config.txt.example and the other bundled demos).
- Verify the demo runs against a current default model — it was built against Ollama
qwen2.5:7b.
- Re-run and confirm behaviour still reproduces the intended result.
- Re-submit to the library bundle.
Open questions
- Is the config issue the only blocker, or were there other reviewer notes?
- Should it ship with an Ollama config (free, offline, matches the original result) or a cloud config (faster to demo)?
Roadmap priority: B16
Summary
Re-add the baba-is-ai demo to the bundled library. It was pulled during the v1.0.0 library submission over a config-file issue flagged in review.
Why it matters
It is one of the more research-grounded demos — it reproduces the knowing-doing gap from BALROG (arXiv:2411.13543), where LLMs score near zero on spatial puzzle games despite being able to state the correct rule. Good demonstration material and a genuine research result.
What needs to be done
demos/tests/config.txt.exampleand the other bundled demos).qwen2.5:7b.Open questions