Predictable Confabulations: Factual Recall by LLMs Scales with Model Size and Topic Frequency
Fuente:
arXiv
Saved in:
| Main Authors: | , , , , |
|---|---|
| Format: | Preprint |
| Published: |
2026
|
| Subjects: | |
| Online Access: | |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| _version_ | 1866918509835976704 |
|---|---|
| author | Smith, Matthew L. Shock, Jonathan P. Segun, Samuel T. Olatunji, Iyiola E. Bissyandé, Tegawendé F. |
| author_facet | Smith, Matthew L. Shock, Jonathan P. Segun, Samuel T. Olatunji, Iyiola E. Bissyandé, Tegawendé F. |
| contents | While scaling laws govern aggregate large language model performance, no scaling law has linked factual recall to both model size and training-data composition. We evaluated 38 models on over 8,900 scholarly references evaluated by an automated reference verification system. Recall quality follows a sigmoid in the log-linear combination of model parameter count and topic representation in training data. These two variables alone explain 60% of the variance across 16 dense models from four families, rising to 74-94% within individual families. The form matches a superposition-inspired account in which recall is gated by a signal-to-noise ratio: signal strength scales with concept frequency and the noise floor with model capacity. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2605_18732 |
| institution | arXiv |
| publishDate | 2026 |
| record_format | arxiv |
| spellingShingle | Predictable Confabulations: Factual Recall by LLMs Scales with Model Size and Topic Frequency Smith, Matthew L. Shock, Jonathan P. Segun, Samuel T. Olatunji, Iyiola E. Bissyandé, Tegawendé F. Computation and Language Artificial Intelligence Machine Learning While scaling laws govern aggregate large language model performance, no scaling law has linked factual recall to both model size and training-data composition. We evaluated 38 models on over 8,900 scholarly references evaluated by an automated reference verification system. Recall quality follows a sigmoid in the log-linear combination of model parameter count and topic representation in training data. These two variables alone explain 60% of the variance across 16 dense models from four families, rising to 74-94% within individual families. The form matches a superposition-inspired account in which recall is gated by a signal-to-noise ratio: signal strength scales with concept frequency and the noise floor with model capacity. |
| title | Predictable Confabulations: Factual Recall by LLMs Scales with Model Size and Topic Frequency |
| topic | Computation and Language Artificial Intelligence Machine Learning |
| url | https://arxiv.org/abs/2605.18732 |