Wittgenstein Apartment: an open behavioral predicate ontology for consistent AI characters
Kon-tiki-ship · reddit · 2026-10-03
A developer releases Wittgenstein Apartment, a dataset built on a behavioral predicate ontology: instead of defining characters with adjectives, it models them via the actions they can and cannot access.
- Represents intentional human actions at the sense level (not just lemma level), related to WordNet/VerbNet/FrameNet
- Action access is constrained by skills, profession, social role, authority, biography — tiered as Allowed / Not Allowed / Conditional / High-Cost Conditional
- Includes a Goal Ontology and task-structure info; intended for character generation, narrative planning, agents, simulators, and action validation
- Open for academic/non-commercial research use; AI training and commercial use require permission
- The author seeks criticism, especially on whether an explicit action-access layer makes sense in LLM-based systems vs prompting/runtime logic
More from Research
- NVIDIA's SoL-Pi uses AI agents to cut coding agent API costs by ~50% vs Codex — dr_cintas · 2026-10-03
- MIT's Cathy Wu boosts RL training efficiency 30x to design smarter transportation systems — MIT News AI · 2026-10-03
- Economist calls for mirror life regulations; researchers say the threat is wildly premature — anshulkundaje · 2026-10-03
- Benchmarking long-range ML interatomic potentials on SrTiO3 defect energetics — rbhar90 · 2026-10-03
- Meta AI's RL-XAR fixes AI slop via expert-aligned rubrics, plus 4 more papers explained — idanbeck · 2026-10-03
- Neuralink pretrained AI models on 50,000 hours of unlabeled brain data, betting on the scaling playbook — Dr_Singularity · 2026-10-03