When LLMs are used to guage qualities just like the correctness, accuracy, or relevance of a chunk of text, consistency is paramount. If an LLM exhibits inconsistent judgements, then its evaluations turn into unreliable and untrustworthy.
If an LLM evaluates the reasoning quality of arguments, but contradicts itself by rating an invalid argument as more logically sound than a superbly valid one, then it fails as an arbiter of reason. Its evaluations lose credibility resulting from the model’s own lack of logical consistency.
When such inconsistencies appear, there isn’t a stable basis for comparison between the LLM’s assessments of various pieces of text. If the model arbitrarily contradicts itself, then sentences can’t be reliably ranked against each other based on the model’s inconsistent scorings.
In essence, inconsistency destroys the grounds for comparison that evaluations aim to supply in the primary place. If an LLM cannot display consistent application of assessment criteria, then using it to guage text loses all effectiveness and utility.
So, consistency in judgement and evaluation is mandatory for LLMs employed to attain or judge textual qualities and features. With out a high level of stability in its assessments, grounded in a consistent understanding of concepts being evaluated, the premise for comparison falls apart when leveraging LLM output as a type of evaluation or scoring.
Sampling multiple solutions reveals consistency between outputs strongly correlates with quality. Nevertheless, existing consistency techniques depend on extracting and matching closed-form answers, restricting their applicability. This text explores methods to boost self-consistency without such constraints, while also grounding decisions in real-world knowledge.
The Need for Self-Consistency
Despite rapid progress, logical failures and falsehoods proceed hindering reliable reasoning in state-of-the-art models. For complex multi-step evaluation or free-form generation, models often contradict themselves or invent unsupported facts.
This manifests in two key ways — inconsistent open-ended generation, and incoherent inferences. When performing…