PRIMETIME: Limits of LLMs in Temporal Primitives
Edward Gaere, Florian von Wangenheim · 2025
Read PRIMETIME: Limits of LLMs in Temporal Primitives by Edward Gaere, Florian von Wangenheim (2025) 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-2504-16155.txt.
Description
Edward Gaere and Florian von Wangenheim test how language models handle dates and times in 2025. Time reasoning rests on two basic skills: reading a datetime and doing arithmetic on it. Existing tests blur those skills with world knowledge into one score. Their tool, PRIMETIME, generates clean synthetic examples without limit.
Each skill can then be graded on its own, free of contamination. Results show the basic skills themselves wobble, from near zero to perfect depending on form. Because the generator also makes training data, weak spots can be drilled directly. The work reframes time errors as basic skill gaps, not reasoning depth.
Catalog record
| CREATOR | Edward Gaere, Florian von Wangenheim |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 239 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2504-16155