We lifted 5 functions out of this paper's own repositories and ran 3 of them in a sandbox. "Ran" means the function executed on a synthesized input and returned a value. It is not a reproduction of the paper's results.
| Repository | Role | Ran |
|---|---|---|
| cheongwoong/impact_of_cooccurrence | canonical | 3 of 5 |
| Function | Status | Where it lives |
|---|---|---|
| get_masks_for_baseline | Ran | cheongwoong/impact_of_cooccurrence/src/utils/evaluation.py code served (permissive licence) · get_code("6b674ce8cc2cbafd") |
| is_pure_alphanumeric | Ran | cheongwoong/impact_of_cooccurrence/src/data_statistics/precompute/extract_entity_set.py code served (permissive licence) · get_code("1b79f5ed386cf90e") |
| postprocess_single_prediction_for_baseline | Ran | cheongwoong/impact_of_cooccurrence/src/utils/evaluation.py code served (permissive licence) · get_code("c3a69f7597cbf30f") |
| filtering | Not yet run | cheongwoong/impact_of_cooccurrence/src/utils/text_processing.py code served (permissive licence) · get_code("91161371fcb17d46") |
| text_normalization_without_lemmatization | Not yet run | cheongwoong/impact_of_cooccurrence/src/utils/text_processing.py code served (permissive licence) · get_code("bf4d06c6c913353a") |
Some links come from the archived Papers with Code dataset (CC BY-SA 4.0): attribution and licence.
Large language models (LLMs) often make factually incorrect responses despite their success in various applications. In this paper, we hypothesize that relying heavily on simple co-occurrence statistics of the pre-training corpora is one of the main factors that cause factual errors. Our results reveal that LLMs are vulnerable to the co-occurrence bias, defined as preferring frequently co-occurred words over the correct answer. Consequently, LLMs struggle to recall facts whose subject and object rarely co-occur in the pre-training dataset although they are seen during finetuning. We show that co-occurrence bias remains despite scaling up model sizes or finetuning. Therefore, we suggest finetuning on a debiased dataset to mitigate the bias by filtering out biased samples whose subject-object co-occurrence count is high. Although debiased finetuning allows LLMs to memorize rare facts in the training set, it is not effective in recalling rare facts unseen during finetuning. Further research in mitigation will help build reliable language models by preventing potential errors. The code is available at \url{https://github.com/CheongWoong/impact_of_cooccurrence}.
The same record, over MCP at https://syntology.ai/mcp:
get_harvested_code_for_paper("2310.08256")
get_code_for_paper("2310.08256")
have("2310.08256")
Connect an agent — have() is free.