Convergent World Representations and Divergent Tasks
Core Francisco Park · 2026
Read Convergent World Representations and Divergent Tasks by Core Francisco Park (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2602-00533.txt.
Description
Core Francisco Park separates the world, the data process, and learned representations to study how their geometry forms and adapts. In a setup, 5,075 city points define the world and 7 shape tasks generate training data. Different tasks yield distinct world geometries, yet multi-task training aligns them, giving controlled evidence for the Multitask Scaling Hypothesis, part of the Platonic Representation Hypothesis. Training on all tasks then fine-tuning on new cities shows divergent tasks can harm new-entity uptake and generalisation, while multi-task training aligns outputs.
Catalog record
| CREATOR | Core Francisco Park |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 62563 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2602.00533