Do Large Language Models Perform Latent Multi-Hop Reasoning without Exploiting Shortcuts?

We evaluate how well Large Language Models (LLMs) latently recall and compose facts to answer multi-hop queries like "In the year Scarlett Johansson was born, the Summer Olympics were hosted in the country of". One major challenge in evaluating this ability is that LLMs may have developed shortcuts by encounters of the head entity "Scarlett Johansson" and the answer entity "United States" in the same training sequences or merely guess the answer based on frequency-based priors. To prevent shortcuts, we exclude test queries where the head and answer entities co-appear in pretraining corpora. Through careful selection of relations and facts and systematic removal of cases where models might guess answers or exploit partial matches, we construct an evaluation dataset SOCRATES (ShOrtCut-fRee lATent rEaSoning). We observe that LLMs demonstrate promising latent multi-hop reasoning abilities without exploiting shortcuts, but only for certain types of queries. For queries requiring latent recall of countries as the intermediate answer, the best models achieve 80% latent composability, but this drops to just 5% for the recall of years. Comparisons with Chain-of-Thought composability highlight a significant gap between the ability of models to reason latently versus explicitly. Analysis reveals that latent representations of the intermediate answer are constructed more often in queries with higher latent composability, and shows the emergence of latent multi-hop reasoning during pretraining.

Hopping Too Late: Exploring the Limitations of Large Language Models on Multi-Hop Queries

Large language models (LLMs) can solve complex multi-step problems, but little is known about how these computations are implemented internally. Motivated by this, we study how LLMs answer multi-hop queries such as "The spouse of the performer of Imagine is". These queries require two information extraction steps: a latent one for resolving the first hop ("the performer of Imagine") into the bridge entity (John Lennon), and one for resolving the second hop ("the spouse of John Lennon") into the target entity (Yoko Ono). Understanding how the latent step is computed internally is key to understanding the overall computation. By carefully analyzing the internal computations of transformer-based LLMs, we discover that the bridge entity is resolved in the early layers of the model. Then, only after this resolution, the two-hop query is solved in the later layers. Because the second hop commences in later layers, there could be cases where these layers no longer encode the necessary knowledge for correctly predicting the answer. Motivated by this, we propose a novel "back-patching" analysis method whereby a hidden representation from a later layer is patched back to an earlier layer. We find that in up to 57% of previously incorrect cases there exists a back-patch that results in the correct generation of the answer, showing that the later layers indeed sometimes lack the needed functionality. Overall our methods and findings open further opportunities for understanding and improving latent reasoning in transformer-based LLMs.

How Do Large Language Models Acquire Factual Knowledge During Pretraining?

Despite the recent observation that large language models (LLMs) can store substantial factual knowledge, there is a limited understanding of the mechanisms of how they acquire factual knowledge through pretraining. This work addresses this gap by studying how LLMs acquire factual knowledge during pretraining. The findings reveal several important insights into the dynamics of factual knowledge acquisition during pretraining. First, counterintuitively, we observe that pretraining on more data shows no significant improvement in the model's capability to acquire and maintain factual knowledge. Next, there is a power-law relationship between training steps and forgetting of memorization and generalization of factual knowledge, and LLMs trained with duplicated training data exhibit faster forgetting. Third, training LLMs with larger batch sizes can enhance the models' robustness to forgetting. Overall, our observations suggest that factual knowledge acquisition in LLM pretraining occurs by progressively increasing the probability of factual knowledge presented in the pretraining data at each step. However, this increase is diluted by subsequent forgetting. Based on this interpretation, we demonstrate that we can provide plausible explanations for recently observed behaviors of LLMs, such as the poor performance of LLMs on long-tail knowledge and the benefits of deduplicating the pretraining corpus.

Do Large Language Models Latently Perform Multi-Hop Reasoning?

We study whether Large Language Models (LLMs) latently perform multi-hop reasoning with complex prompts such as "The mother of the singer of 'Superstition' is". We look for evidence of a latent reasoning pathway where an LLM (1) latently identifies "the singer of 'Superstition'" as Stevie Wonder, the bridge entity, and (2) uses its knowledge of Stevie Wonder's mother to complete the prompt. We analyze these two hops individually and consider their co-occurrence as indicative of latent multi-hop reasoning. For the first hop, we test if changing the prompt to indirectly mention the bridge entity instead of any other entity increases the LLM's internal recall of the bridge entity. For the second hop, we test if increasing this recall causes the LLM to better utilize what it knows about the bridge entity. We find strong evidence of latent multi-hop reasoning for the prompts of certain relation types, with the reasoning pathway used in more than 80% of the prompts. However, the utilization is highly contextual, varying across different types of prompts. Also, on average, the evidence for the second hop and the full multi-hop traversal is rather moderate and only substantial for the first hop. Moreover, we find a clear scaling trend with increasing model size for the first hop of reasoning but not for the second hop. Our experimental findings suggest potential challenges and opportunities for future development and applications of LLMs.

Improving Probability-based Prompt Selection Through Unified Evaluation and Analysis

Large Language Models (LLMs) have demonstrated great capabilities in solving a wide range of tasks in a resource-efficient manner through prompting, which does not require task-specific training, but suffers from performance fluctuation when there are multiple prompt candidates. Previous works have introduced gradient-free probability-based prompt selection methods that aim to choose the optimal prompt among the candidates for a given task but fail to provide a comprehensive and fair comparison between each other. In this paper, we propose a unified framework to interpret and evaluate the existing probability-based prompt selection methods by performing extensive experiments on 13 common NLP tasks. We find that all existing methods can be unified into some variant of the method that maximizes the mutual information between the input and the corresponding model output (denoted as MI). Using the finding, we develop several variants of MI and increases the effectiveness of the best prompt selection method from 87.79% to 94.98%, measured as the ratio of the performance of the selected prompt to that of the optimal oracle prompt. Furthermore, we propose a novel calibration method called Calibration by Marginalization (CBM) that is orthogonal to existing methods and helps increase the prompt selection effectiveness of the best method by 99.44%. The code and datasets used in our work will be released at https://github.com/soheeyang/unified-prompt-selection.

In-Context Instruction Learning

Instruction learning of Large Language Models (LLMs) has enabled zero-shot task generalization. However, instruction learning has been predominantly approached as a fine-tuning problem, including instruction tuning and reinforcement learning from human feedback, where LLMs are multi-task fine-tuned on various tasks with instructions. In this paper, we present a surprising finding that applying in-context learning to instruction learning, referred to as In-Context Instruction Learning (ICIL), significantly improves the zero-shot task generalization performance for both pretrained and instruction-fine-tuned models. One of the core advantages of ICIL is that it uses a single fixed prompt to evaluate all tasks, which is a concatenation of cross-task demonstrations. In particular, we demonstrate that the most powerful instruction-fine-tuned baseline (text-davinci-003) also benefits from ICIL by 9.3%, indicating that the effect of ICIL is complementary to instruction-based fine-tuning.

Contextualized Generative Retrieval

The text retrieval task is mainly performed in two ways: the bi-encoder approach and the generative approach. The bi-encoder approach maps the document and query embeddings to common vector space and performs a nearest neighbor search. It stably shows high performance and efficiency across different domains but has an embedding space bottleneck as it interacts in L2 or inner product space. The generative retrieval model retrieves by generating a target sequence and overcomes the embedding space bottleneck by interacting in the parametric space. However, it fails to retrieve the information it has not seen during the training process as it depends solely on the information encoded in its own model parameters. To leverage the advantages of both approaches, we propose Contextualized Generative Retrieval model, which uses contextualized embeddings (output embeddings of a language model encoder) as vocab embeddings at the decoding step of generative retrieval. The model uses information encoded in both the non-parametric space of contextualized token embeddings and the parametric space of the generative retrieval model. Our approach of generative retrieval with contextualized vocab embeddings shows higher performance than generative retrieval with only vanilla vocab embeddings in the document retrieval task, an average of 6% higher performance in KILT (NQ, TQA) and 2X higher in NQ-320k, suggesting the benefits of using contextualized embedding in generative retrieval models.

Knowledge Unlearning for Mitigating Privacy Risks in Language Models

Pretrained Language Models (LMs) memorize a vast amount of knowledge during initial pretraining, including information that may violate the privacy of personal lives and identities. Previous work addressing privacy issues for language models has mostly focused on data preprocessing and differential privacy methods, both requiring re-training the underlying LM. We propose knowledge unlearning as an alternative method to reduce privacy risks for LMs post hoc. We show that simply applying the unlikelihood training objective to target token sequences is effective at forgetting them with little to no degradation of general language modeling performances; it sometimes even substantially improves the underlying LM with just a few iterations. We also find that sequential unlearning is better than trying to unlearn all the data at once and that unlearning is highly dependent on which kind of data (domain) is forgotten. By showing comparisons with a previous data preprocessing method known to mitigate privacy risks for LMs, we show that unlearning can give a stronger empirical privacy guarantee in scenarios where the data vulnerable to extraction attacks are known a priori while being orders of magnitude more computationally efficient. We release the code and dataset needed to replicate our results at https://github.com/joeljang/knowledge-unlearning .

TemporalWiki: A Lifelong Benchmark for Training and Evaluating Ever-Evolving Language Models

Language Models (LMs) become outdated as the world changes; they often fail to perform tasks requiring recent factual information which was absent or different during training, a phenomenon called temporal misalignment. This is especially a challenging problem because the research community still lacks a coherent dataset for assessing the adaptability of LMs to frequently-updated knowledge corpus such as Wikipedia. To this end, we introduce TemporalWiki, a lifelong benchmark for ever-evolving LMs that utilizes the difference between consecutive snapshots of English Wikipedia and English Wikidata for training and evaluation, respectively. The benchmark hence allows researchers to periodically track an LM's ability to retain previous knowledge and acquire updated/new knowledge at each point in time. We also find that training an LM on the diff data through continual learning methods achieves similar or better perplexity than on the entire snapshot in our benchmark with 12 times less computational cost, which verifies that factual knowledge in LMs can be safely updated with minimal training data via continual learning. The dataset and the code are available at https://github.com/joeljang/temporalwiki .

Generative Retrieval for Long Sequences

Text retrieval is often formulated as mapping the query and the target items (e.g., passages) to the same vector space and finding the item whose embedding is closest to that of the query. In this paper, we explore a generative approach as an alternative, where we use an encoder-decoder model to memorize the target corpus in a generative manner and then finetune it on query-to-passage generation. As GENRE(Cao et al., 2021) has shown that entities can be retrieved in a generative way, our work can be considered as its generalization to longer text. We show that it consistently achieves comparable performance to traditional bi-encoder retrieval on diverse datasets and is especially strong at retrieving highly structured items, such as reasoning chains and graph relations, while demonstrating superior GPU memory and time complexity. We also conjecture that generative retrieval is complementary to traditional retrieval, as we find that an ensemble of both outperforms homogeneous ensembles.