We propose semantic entropy probes (SEPs), a cheap and reliable method for uncertainty quantification in Large Language Models (LLMs). Hallucinations, which are plausible-sounding but factually incorrect and arbitrary model generations, present a major challenge to the practical adoption of LLMs. Recent work by Farquhar et al. (2024) proposes semantic entropy (SE), which can detect hallucinations by estimating uncertainty in the space semantic meaning for a set of model generations. However, the 5-to-10-fold increase in computation cost associated with SE computation hinders practical adoption. To address this, we propose SEPs, which directly approximate SE from the hidden states of a single generation. SEPs are simple to train and do not require sampling multiple model generations at test time, reducing the overhead of semantic uncertainty quantification to almost zero. We show that SEPs retain high performance for hallucination detection and generalize better to out-of-distribution data than previous probing methods that directly predict model accuracy. Our results across models and tasks suggest that model hidden states capture SE, and our ablation studies give further insights into the token positions and model layers for which this is the case.

The location of knowledge within Generative Pre-trained Transformer (GPT)-like models has seen extensive recent investigation. However, much of the work is focused towards determining locations of individual facts, with the end goal being the editing of facts that are outdated, erroneous, or otherwise harmful, without the time and expense of retraining the entire model. In this work, we investigate a broader view of knowledge location, that of concepts or clusters of related information, instead of disparate individual facts. To do this, we first curate a novel dataset, called DARC, that includes a total of 34 concepts of ~120K factual statements divided into two types of hierarchical categories, namely taxonomy and meronomy. Next, we utilize existing causal mediation analysis methods developed for determining regions of importance for individual facts and apply them to a series of related categories to provide detailed investigation into whether concepts are associated with distinct regions within these models. We find that related categories exhibit similar areas of importance in contrast to less similar categories. However, fine-grained localization of individual category subsets to specific regions is not apparent.

Neural codec language model (LM) has demonstrated strong capability in zero-shot text-to-speech (TTS) synthesis. However, the codec LM often suffers from limitations in inference speed and stability, due to its auto-regressive nature and implicit alignment between text and audio. In this work, to handle these challenges, we introduce a new variant of neural codec LM, namely TacoLM. Specifically, TacoLM introduces a gated attention mechanism to improve the training and inference efficiency and reduce the model size. Meanwhile, an additional gated cross-attention layer is included for each decoder layer, which improves the efficiency and content accuracy of the synthesized speech. In the evaluation of the Librispeech corpus, the proposed TacoLM achieves a better word error rate, speaker similarity, and mean opinion score, with 90% fewer parameters and 5.2 times speed up, compared with VALL-E. Demo and code is available at https://ereboas.github.io/TacoLM/.

Sparse mixture of experts (SMoE) have emerged as an effective approach for scaling large language models while keeping a constant computational cost. Regardless of several notable successes of SMoE, effective training such architecture remains elusive due to the representation collapse problem, which in turn harms model performance and causes parameter redundancy. In this work, we present Similarity-based Sparse Mixture of Experts (SimSMoE), a novel similarity of neural network algorithm, that guarantees a solution to address the representation collapse issue between experts given a fixed FLOPs budget. We conduct extensive empirical evaluations on three large language models for both Pre-training and Fine-tuning tasks to illustrate the efficacy, robustness, and scalability of our method. The results demonstrate that SimSMoE significantly enhances existing routing policy and outperforms other SMoE training methods in performance for the tasks.

Large language model unlearning has gained increasing attention due to its potential to mitigate security and privacy concerns. Current research predominantly focuses on Instance-level unlearning, specifically aiming at forgetting predefined instances of sensitive content. However, a notable gap still exists in exploring the deletion of complete entity-related information, which is crucial in many real-world scenarios, such as copyright protection. To this end, we propose a novel task of Entity-level unlearning, where the entity-related knowledge within the target model is supposed to be entirely erased. Given the challenge of practically accessing all entity-related knowledge within a model, we begin by simulating entity-level unlearning scenarios through fine-tuning models to introduce pseudo entities. Following this, we develop baseline methods inspired by trending unlearning techniques and conduct a detailed comparison of their effectiveness in this task. Extensive experiments reveal that current unlearning algorithms struggle to achieve effective entity-level unlearning. Additionally, our analyses further indicate that entity-related knowledge injected through fine-tuning is more susceptible than original entities from pre-training during unlearning, highlighting the necessity for more thorough pseudo-entity injection methods to make them closer to pre-trained knowledge.

Today, the detection of AI-generated content is receiving more and more attention. Our idea is to go beyond detection and try to recover the prompt used to generate a text. This paper, to the best of our knowledge, introduces the first investigation in this particular domain without a closed set of tasks. Our goal is to study if this approach is promising. We experiment with zero-shot and few-shot in-context learning but also with LoRA fine-tuning. After that, we evaluate the benefits of using a semi-synthetic dataset. For this first study, we limit ourselves to text generated by a single model. The results show that it is possible to recover the original prompt with a reasonable degree of accuracy.

Benchmark plays a pivotal role in assessing the advancements of large language models (LLMs). While numerous benchmarks have been proposed to evaluate LLMs' capabilities, there is a notable absence of a dedicated benchmark for assessing their musical abilities. To address this gap, we present ZIQI-Eval, a comprehensive and large-scale music benchmark specifically designed to evaluate the music-related capabilities of LLMs. ZIQI-Eval encompasses a wide range of questions, covering 10 major categories and 56 subcategories, resulting in over 14,000 meticulously curated data entries. By leveraging ZIQI-Eval, we conduct a comprehensive evaluation over 16 LLMs to evaluate and analyze LLMs' performance in the domain of music. Results indicate that all LLMs perform poorly on the ZIQI-Eval benchmark, suggesting significant room for improvement in their musical capabilities. With ZIQI-Eval, we aim to provide a standardized and robust evaluation framework that facilitates a comprehensive assessment of LLMs' music-related abilities. The dataset is available at GitHub\footnote{https://github.com/zcli-charlie/ZIQI-Eval} and HuggingFace\footnote{https://huggingface.co/datasets/MYTH-Lab/ZIQI-Eval}.

This is a summary of the paper "A Confederacy of Models: a Comprehensive Evaluation of LLMs on Creative Writing", which was published in Findings of EMNLP 2023. We evaluate a range of recent state-of-the-art, instruction-tuned large language models (LLMs) on an English creative writing task, and compare them to human writers. For this purpose, we use a specifically-tailored prompt (based on an epic combat between Ignatius J. Reilly, main character of John Kennedy Toole's "A Confederacy of Dunces", and a pterodactyl) to minimize the risk of training data leakage and force the models to be creative rather than reusing existing stories. The same prompt is presented to LLMs and human writers, and evaluation is performed by humans using a detailed rubric including various aspects like fluency, style, originality or humor. Results show that some state-of-the-art commercial LLMs match or slightly outperform our human writers in most of the evaluated dimensions. Open-source LLMs lag behind. Humans keep a close lead in originality, and only the top three LLMs can handle humor at human-like levels.

Electronic health records contain detailed information about the medical condition of patients, but they are difficult for patients to understand even if they have access to them. We explore whether ChatGPT (GPT 4) can help explain multidisciplinary team (MDT) reports to colorectal and prostate cancer patients. These reports are written in dense medical language and assume clinical knowledge, so they are a good test of the ability of ChatGPT to explain complex medical reports to patients. We asked clinicians and lay people (not patients) to review explanations and responses of ChatGPT. We also ran three focus groups (including cancer patients, caregivers, computer scientists, and clinicians) to discuss output of ChatGPT. Our studies highlighted issues with inaccurate information, inappropriate language, limited personalization, AI distrust, and challenges integrating large language models (LLMs) into clinical workflow. These issues will need to be resolved before LLMs can be used to explain complex personal medical information to patients.

Recent research has shown the potential of Nash Learning via Human Feedback for large language model alignment by incorporating the notion of a preference model in a minimax game setup. We take this idea further by casting the alignment as a mirror descent algorithm against the adaptive feedback of an improved opponent, thereby removing the need for learning a preference model or the existence of an annotated dataset altogether. The resulting algorithm, which we refer to as Language Alignment via Nash-learning and Adaptive feedback (LANA), is capable of self-alignment without the need for a human-annotated preference dataset. We support this statement with various experiments and mathematical discussion.