Large Language Models (LLM) It aims to align with human preferences, ensuring reliable and trustworthy decision making. However, these models acquire biases, logical leaps, and hallucinations, making them invalid and harmless for critical tasks involving logical thinking. Logical coherence problems make it impossible to develop logically coherent LLMs. They also use temporal reasoning, optimization, and automated systems, resulting in less reliable conclusions.
Current methods for aligning large language models (LLMs) with human preferences are based on supervised training with instruction-following data and reinforcement learning from human feedback. However, these methods suffer from problems such as hallucinations, biases, and logical inconsistency, which undermine the validity of LLMs. Therefore, most improvements in LLM coherence have been made based on simple factual knowledge or a simple link between a few statements, while neglecting other decision-making scenarios or more complex tasks that involve more of an element. This gap limits its ability to provide coherent and reliable reasoning in real-world applications where consistency is essential.
To evaluate logical consistency in large language models (LLM), researchers from cambridge university and Monash University proposed a universal framework for quantifying logical coherence by evaluating three key properties: transitivity, commutativityand negation invariance. Transitivity guaranteed that if a model determined that one element was preferred over a second and the second over a third, it also concluded that the first element was chosen over the third. Commutativity ensured that the model's judgments remained the same regardless of the order in which the items were compared.
At the same time, the consistency of negation invariance in the handling of relational negations was tested. These properties formed the basis of reliable reasoning in the models. The researchers formalized the evaluation process by treating an LLM as an operator role. FFF that compared pairs of items and assigned relational decisions. Logical consistency was measured using metrics such as trans(K)s_{tran}(K)stran(K) for transitivity and scomms_comm_scommfor commutativity. Side (K)s_{tran}(K)side(K) Quantized transitivity by sampling subsets of elements and detecting cycles in the relation graph. At the same time, scomms_{comm}scomm evaluated whether the model's judgments remained stable when the order of the items in the comparisons was reversed. Both metrics ranged between 0 to 1where higher values indicate better performance.
The researchers applied these metrics to several LLMs, revealing vulnerabilities to biases such as permutation bias and positional bias. To address this, they introduced a data augmentation and refinement technique using rank aggregation methods to estimate partial or ordered preference rankings from noisy or sparse pairwise comparisons. This improved logical consistency without compromising alignment with human preferences and emphasized the essential role of logical consistency in improving the performance of logic-dependent algorithms.
The researchers tested three tasks to assess logical coherence in LLMs: abstractive summary, reclassification of documents, and temporal ordering of events using data sets like SumEval, NovelEvaland catering. They evaluated transitivity, commutativity, negation invariance, and human and self-agreement. The results showed that newer models such as Deepseek-chat, Phi-3-medium, and Gemma-2-9B had greater logical coherence, although this did not correlate strongly with the accuracy of human agreement. He catering The data set showed greater coherence, focusing on temporal and causal relationships. Chain-of-thought prompts had mixed results, sometimes reducing transitivity due to additional reasoning tokens. Self-agreement was related to transitivity; This shows that coherent reasoning is essential for logical coherence, and models like Phi-3-medium and Gemma-2-9B have the same reliability for each task, emphasizing the need for cleaner training data.
In the end, the researchers demonstrated the importance of logical consistency in improving the reliability of large language models. They presented a method for measuring key aspects of consistency and explained a data cleaning process that reduces the number of default values while remaining relevant to humans. This framework can further be used to guide further research to improve the consistency of LLMs and to continue efforts to implement LLMs into decision-making systems to improve effectiveness and productivity.
Verify he Paper. All credit for this research goes to the researchers of this project. Also, don't forget to follow us on <a target="_blank" href="https://x.com/intent/follow?screen_name=marktechpost” target=”_blank” rel=”noreferrer noopener”>twitter and join our Telegram channel and LinkedIn Grabove. Don't forget to join our SubReddit over 60,000 ml.
UPCOMING FREE ai WEBINAR (JANUARY 15, 2025): <a target="_blank" href="https://info.gretel.ai/boost-llm-accuracy-with-sd-and-evaluation-intelligence?utm_source=marktechpost&utm_medium=newsletter&utm_campaign=202501_gretel_galileo_webinar”>Increase LLM Accuracy with Synthetic Data and Assessment Intelligence–<a target="_blank" href="https://info.gretel.ai/boost-llm-accuracy-with-sd-and-evaluation-intelligence?utm_source=marktechpost&utm_medium=newsletter&utm_campaign=202501_gretel_galileo_webinar”>Join this webinar to learn practical information to improve LLM model performance and accuracy while protecting data privacy..
Divyesh is a Consulting Intern at Marktechpost. He is pursuing a BTech in Agricultural and Food Engineering from the Indian Institute of technology Kharagpur. He is a data science and machine learning enthusiast who wants to integrate these leading technologies in agriculture and solve challenges.
