Saved in:
| Main Author: | |
|---|---|
| Format: | Recurso digital |
| Language: | |
| Published: |
Zenodo
2026
|
| Subjects: | |
| Online Access: | https://doi.org/10.5281/zenodo.20429921 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| _version_ | 1866902159029698560 |
|---|---|
| author | Lovell, Jason |
| author_facet | Lovell, Jason |
| contents | <p>First public release of nanoAWM: a tiny, from-scratch (pure-Python, no PyTorch/NumPy) action-conditioned world model for tool-using agents, plus the MiniOS task suite it learns on.</p> <p>Honest headline: on a genuinely held-out split with object and action vocabulary disjoint from training, the learned world-model planner scores 0.524, versus 0.067 for the best non-world-model baseline and 1.000 for an oracle upper bound. The in-distribution 1.000 is a sanity check, not the result.</p> <p>Documented negative results are included by design (action-paraphrase collapse, cross-surface OOD, lexicon holdout). Code, the trained model, the dataset, the paper, and all audits are in the repository.</p> |
| format | Recurso digital |
| id | zenodo_https___doi_org_10_5281_zenodo_20429921 |
| institution | Zenodo |
| language | |
| publishDate | 2026 |
| publisher | Zenodo |
| record_format | zenodo |
| spellingShingle | nanoAWM: A Tiny Agent World Model Lab Lovell, Jason agent world models consequence simulation computer-use agents consequence aliasing <p>First public release of nanoAWM: a tiny, from-scratch (pure-Python, no PyTorch/NumPy) action-conditioned world model for tool-using agents, plus the MiniOS task suite it learns on.</p> <p>Honest headline: on a genuinely held-out split with object and action vocabulary disjoint from training, the learned world-model planner scores 0.524, versus 0.067 for the best non-world-model baseline and 1.000 for an oracle upper bound. The in-distribution 1.000 is a sanity check, not the result.</p> <p>Documented negative results are included by design (action-paraphrase collapse, cross-surface OOD, lexicon holdout). Code, the trained model, the dataset, the paper, and all audits are in the repository.</p> |
| title | nanoAWM: A Tiny Agent World Model Lab |
| topic | agent world models consequence simulation computer-use agents consequence aliasing |
| url | https://doi.org/10.5281/zenodo.20429921 |