Minimizing Factual Inconsistency and Hallucination in Large Language Models

Large Language Models (LLMs) are widely used in critical fields such as healthcare, education, and finance due to their remarkable proficiency in various language-related tasks. However, LLMs are prone to generating factually incorrect responses or "hallucinations," which can lead to a loss of credibility and trust among users. To address this issue, we propose a multi-stage framework that generates the rationale first, verifies and refines incorrect ones, and uses them as supporting references to generate the answer. The generated rationale enhances the transparency of the answer and our framework provides insights into how the model arrived at this answer, by using this rationale and the references to the context. In this paper, we demonstrate its effectiveness in improving the quality of responses to drug-related inquiries in the life sciences industry. Our framework improves traditional Retrieval Augmented Generation (RAG) by enabling OpenAI GPT-3.5-turbo to be 14-25% more faithful and 16-22% more accurate on two datasets. Furthermore, fine-tuning samples based on our framework improves the accuracy of smaller open-access LLMs by 33-42% and competes with RAG on commercial models.

Knowledge Capture and Replay for Continual Learning

Deep neural networks have shown promise in several domains, and the learned task-specific information is implicitly stored in the network parameters. It will be vital to utilize representations from these networks for downstream tasks such as continual learning. In this paper, we introduce the notion of {\em flashcards} that are visual representations to {\em capture} the encoded knowledge of a network, as a function of random image patterns. We demonstrate the effectiveness of flashcards in capturing representations and show that they are efficient replay methods for general and task agnostic continual learning setting. Thus, while adapting to a new task, a limited number of constructed flashcards, help to prevent catastrophic forgetting of the previously learned tasks. Most interestingly, such flashcards neither require external memory storage nor need to be accumulated over multiple tasks and only need to be constructed just before learning the subsequent new task, irrespective of the number of tasks trained before and are hence task agnostic. We first demonstrate the efficacy of flashcards in capturing knowledge representation from a trained network, and empirically validate the efficacy of flashcards on a variety of continual learning tasks: continual unsupervised reconstruction, continual denoising, and new-instance learning classification, using a number of heterogeneous benchmark datasets. These studies also indicate that continual learning algorithms with flashcards as the replay strategy perform better than other state-of-the-art replay methods, and exhibits on par performance with the best possible baseline using coreset sampling, with the least additional computational complexity and storage.

Classification Representations Can be Reused for Downstream Generations

Contrary to the convention of using supervision for class-conditioned $\it{generative}$ $\it{modeling}$, this work explores and demonstrates the feasibility of a learned supervised representation space trained on a discriminative classifier for the $\it{downstream}$ task of sample generation. Unlike generative modeling approaches that aim to $\it{model}$ the manifold distribution, we directly $\it{represent}$ the given data manifold in the classification space and leverage properties of latent space representations to generate new representations that are guaranteed to be in the same class. Interestingly, such representations allow for controlled sample generations for any given class from existing samples and do not require enforcing prior distribution. We show that these latent space representations can be smartly manipulated (using convex combinations of $n$ samples, $n\geq2$) to yield meaningful sample generations. Experiments on image datasets of varying resolutions demonstrate that downstream generations have higher classification accuracy than existing conditional generative models while being competitive in terms of FID.