A review of methods for alleviating hallucination issues in large language models
Bibliographic record
Abstract
Large language models have demonstrated impressive language processing capabilities in recent years, exhibiting unparalleled excellence in the field of natural language processing. However, the generated text sometimes contains hallucinations, which is the text that contradicts the knowledge in the real world, the context, and the user input. This problem is mainly due to the inherent limitations of the method itself in aspects such as data quality, the model training process, and the model generation process. The issue of hallucinations has always been closely monitored by the academic community. It is widely recognized that its potential consequences should not be underestimated. This paper systematically summarizes the research on the causes of hallucinations in large language models, and introduces mainstream classification methods as well as current measures to address the issue of hallucinations. To be more specific, the article divides the causes of hallucinations into two categories: 1. hallucinations come from the training process and 2. hallucinations come from the generation process. Also, 4 typical types of causes for the former and 5 typical types of causes for the latter are provided. Simultaneously, a detailed discussion of 16 methods to mitigate hallucinations that arise in the generation process is offered. Finally, this paper also discusses inherent flaws that may exist in large language models, aiming to help people gain a more comprehensive understanding and research into hallucinations and large language models. In general, the text details about the hallucinations that exist in the large language model. Meanwhile, according to the previous research, it is pointed out that it is difficult for the large language model based on autoregressive method for token prediction to avoid the hallucinations completely.
Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.
How this classification was reachedexpand
Full frame machine prediction
Teacher imitationNot calibrated prevalence, not ground truth. Human validation pending. The Gemma side is a direct model label for every work in the frame, read from the title-only record. The Codex side is a classifier learned from the 10,348 direct Codex labels and calibrated to design-weighted sample rates; fields without enough sample support carry no Codex call. Candidate is the union of the two sides; consensus is their intersection. These outputs are machine_predicted_unvalidated and are not human labels.
Distilled classifier scores by category (both heads)
| Category | Codex | Gemma |
|---|---|---|
| Metaresearch | 0.003 | 0.011 |
| Meta-epidemiology (narrow) | 0.002 | 0.001 |
| Meta-epidemiology (broad) | 0.001 | 0.002 |
| Bibliometrics | 0.002 | 0.003 |
| Science and technology studies | 0.001 | 0.001 |
| Scholarly communication | 0.002 | 0.004 |
| Open science | 0.002 | 0.002 |
| Research integrity | 0.001 | 0.003 |
| Insufficient payload (model declined to judge) | 0.004 | 0.002 |
Machine scores (provisional)
The two teacher heads of the student model, read on this work. A score orders the frame for review; it never asserts a category, and the validation status ships verbatim with every row.
Baseline scores from an immature model (maturity gate not passed, 7 training rounds). Scores rank; they never assert a category.
score_only:v0-immature-baseline · verbatim from the scoring run: score_only means the number may rank works, and no category label ships from itClassification
machine, unvalidatedMachine predicted; a candidate call from one source (direct Gemma or distilled Codex), not a consensus.
How this classification was reached, model by model and score by score, is at the end of the page under "How this classification was reached".