What are the root causes of hallucinations in large language models (LLMs)? We use Communication Complexity to prove that the Transformer layer is incapable of composing functions (e.g., identify a grandparent of a person in a genealogy) if the domains of the functions are large enough; we show through examples that this inability is already empirically present when the domains are quite small. We also point out that several mathematical tasks that are at the core of the so-called compositional tasks thought to be hard for LLMs are unlikely to be solvable by Transformers, for large enough instances and assuming that certain well accepted conjectures in the field of Computational Complexity are true.

Existing methods for controlling language models, such as RLHF and Constitutional AI, involve determining which LLM behaviors are desirable and training them into a language model. However, in many cases, it is desirable for LLMs to be controllable \textit{at inference time}, so that they can be used in multiple contexts with diverse needs. We illustrate this with the \textbf{Pink Elephant Problem}: instructing an LLM to avoid discussing a certain entity (a ``Pink Elephant''), and instead discuss a preferred entity (``Grey Elephant''). We apply a novel simplification of Constitutional AI, \textbf{Direct Principle Feedback}, which skips the ranking of responses and uses DPO directly on critiques and revisions. Our results show that after DPF fine-tuning on our synthetic Pink Elephants dataset, our 13B fine-tuned LLaMA 2 model significantly outperforms Llama-2-13B-Chat and a prompted baseline, and performs as well as GPT-4 in on our curated test set assessing the Pink Elephant Problem.

Understanding and adapting to climate change is paramount for professionals such as urban planners, emergency managers, and infrastructure operators, as it directly influences urban development, disaster response, and the maintenance of essential services. Nonetheless, this task presents a complex challenge that necessitates the integration of advanced technology and scientific insights. Recent advances in LLMs present an innovative solution, particularly in democratizing climate science. They possess the unique capability to interpret and explain technical aspects of climate change through conversations, making this crucial information accessible to people from all backgrounds Rillig et al. [2023], Bulian et al. [2023], Chen et al. [2023]. However, given that LLMs are generalized models, their performance can be improved by providing additional domain-specific information. Recent research has been focusing on augmenting LLMs with external tools and data sources to ensure that the information provided is scientifically accurate: for example, leveraging authoritative data sources such as ClimateWatch Kraus et al. [2023] and findings from the IPCC AR6 reports Vaghefi et al. [2023] helps in refining the LLM's outputs, ensuring that the information is grounded in the latest research.

We introduce Language Feedback Models (LFMs) that identify desirable behaviour - actions that help achieve tasks specified in the instruction - for imitation learning in instruction following. To train LFMs, we obtain feedback from Large Language Models (LLMs) on visual trajectories verbalized to language descriptions. First, by using LFMs to identify desirable behaviour to imitate, we improve in task-completion rate over strong behavioural cloning baselines on three distinct language grounding environments (Touchdown, ScienceWorld, and ALFWorld). Second, LFMs outperform using LLMs as experts to directly predict actions, when controlling for the number of LLM output tokens. Third, LFMs generalize to unseen environments, improving task-completion rate by 3.5-12.0% through one round of adaptation. Finally, LFM can be modified to provide human-interpretable feedback without performance loss, allowing human verification of desirable behaviour for imitation learning.

Vision language models (VLMs) have shown impressive capabilities across a variety of tasks, from logical reasoning to visual understanding. This opens the door to richer interaction with the world, for example robotic control. However, VLMs produce only textual outputs, while robotic control and other spatial tasks require outputting continuous coordinates, actions, or trajectories. How can we enable VLMs to handle such settings without fine-tuning on task-specific data? In this paper, we propose a novel visual prompting approach for VLMs that we call Prompting with Iterative Visual Optimization (PIVOT), which casts tasks as iterative visual question answering. In each iteration, the image is annotated with a visual representation of proposals that the VLM can refer to (e.g., candidate robot actions, localizations, or trajectories). The VLM then selects the best ones for the task. These proposals are iteratively refined, allowing the VLM to eventually zero in on the best available answer. We investigate PIVOT on real-world robotic navigation, real-world manipulation from images, instruction following in simulation, and additional spatial inference tasks such as localization. We find, perhaps surprisingly, that our approach enables zero-shot control of robotic systems without any robot training data, navigation in a variety of environments, and other capabilities. Although current performance is far from perfect, our work highlights potentials and limitations of this new regime and shows a promising approach for Internet-Scale VLMs in robotic and spatial reasoning domains. Website: pivot-prompt.github.io and HuggingFace: https://huggingface.co/spaces/pivot-prompt/pivot-prompt-demo.

Mixture of Experts (MoE) models have emerged as a primary solution for reducing the computational cost of Large Language Models. In this work, we analyze their scaling properties, incorporating an expanded range of variables. Specifically, we introduce a new hyperparameter, granularity, whose adjustment enables precise control over the size of the experts. Building on this, we establish scaling laws for fine-grained MoE, taking into account the number of training tokens, model size, and granularity. Leveraging these laws, we derive the optimal training configuration for a given computational budget. Our findings not only show that MoE models consistently outperform dense Transformers but also highlight that the efficiency gap between dense and MoE models widens as we scale up the model size and training budget. Furthermore, we demonstrate that the common practice of setting the size of experts in MoE to mirror the feed-forward layer is not optimal at almost any computational budget. In recent years, we have witnessed Large Language Models (LLMs) achieve exceptional performance in tasks across numerous domains (Chowdhery et al., 2022; Yin et al., 2023; Agostinelli et al., 2023). However, training those massive models incurs high computational costs, measured in millions of GPU-hours (Touvron et al., 2023b), enabled only by enormous budgets (Scao et al., 2023) and leading to non-negligible carbon footprints (Faiz et al., 2024). To combat these obstacles, the research community has been striving to increase the efficiency of LLMs.

Large language models (LLMs) have achieved remarkable success due to their exceptional generative capabilities. Despite their success, they also have inherent limitations such as a lack of up-to-date knowledge and hallucination. Retrieval-Augmented Generation (RAG) is a state-of-the-art technique to mitigate those limitations. In particular, given a question, RAG retrieves relevant knowledge from a knowledge database to augment the input of the LLM. For instance, the retrieved knowledge could be a set of top-k texts that are most semantically similar to the given question when the knowledge database contains millions of texts collected from Wikipedia. As a result, the LLM could utilize the retrieved knowledge as the context to generate an answer for the given question. Existing studies mainly focus on improving the accuracy or efficiency of RAG, leaving its security largely unexplored. We aim to bridge the gap in this work. Particularly, we propose PoisonedRAG , a set of knowledge poisoning attacks to RAG, where an attacker could inject a few poisoned texts into the knowledge database such that the LLM generates an attacker-chosen target answer for an attacker-chosen target question. We formulate knowledge poisoning attacks as an optimization problem, whose solution is a set of poisoned texts. Depending on the background knowledge (e.g., black-box and white-box settings) of an attacker on the RAG, we propose two solutions to solve the optimization problem, respectively. Our results on multiple benchmark datasets and LLMs show our attacks could achieve 90% attack success rates when injecting 5 poisoned texts for each target question into a database with millions of texts. We also evaluate recent defenses and our results show they are insufficient to defend against our attacks, highlighting the need for new defenses.

Visually-conditioned language models (VLMs) have seen growing adoption in applications such as visual dialogue, scene understanding, and robotic task planning; adoption that has fueled a wealth of new models such as LLaVa, InstructBLIP, and PaLI-3. Despite the volume of new releases, key design decisions around image preprocessing, architecture, and optimization are under-explored, making it challenging to understand what factors account for model performance $-$ a challenge further complicated by the lack of objective, consistent evaluations. To address these gaps, we first compile a suite of standardized evaluations spanning visual question answering, object localization from language, and targeted challenge sets that probe properties such as hallucination; evaluations that provide calibrated, fine-grained insight into a VLM's capabilities. Second, we rigorously investigate VLMs along key design axes, including pretrained visual representations and quantifying the tradeoffs of using base vs. instruct-tuned language models, amongst others. We couple our analysis with three resource contributions: (1) a unified framework for evaluating VLMs, (2) optimized, flexible code for VLM training, and (3) checkpoints for all models, including a family of VLMs at the 7-13B scale that strictly outperform InstructBLIP and LLaVa v1.5, the state-of-the-art in open-source VLMs.

This study explores the potential of LLMs to augment judgement in forecasting tasks. We evaluated the impact on forecasting accuracy of two GPT-4-Turbo assistants: one designed to provide high-quality advice ('superforecasting'), and the other designed to be overconfident and base-rate-neglecting. Participants (N = 991) had the option to consult their assigned LLM assistant throughout the study, in contrast to a control group that used a less advanced model (DaVinci-003) without direct forecasting support. Our preregistered analyses reveal that LLM augmentation significantly enhances forecasting accuracy by 23% across both types of assistants, compared to the control group. This improvement occurs despite the superforecasting assistant's higher accuracy in predictions, indicating the augmentation's benefit is not solely due to model prediction accuracy. Exploratory analyses showed a pronounced effect in one forecasting item, without which we find that the superforecasting assistant increased accuracy by 43%, compared with 28% for the biased assistant. We further examine whether LLM augmentation disproportionately benefits less skilled forecasters, degrades the wisdom-of-the-crowd by reducing prediction diversity, or varies in effectiveness with question difficulty. Our findings do not consistently support these hypotheses. Our results suggest that access to an LLM assistant, even a biased one, can be a helpful decision aid in cognitively demanding tasks where the answer is not known at the time of interaction.

The efficacy of language models, particularly in reasoning tasks, is significantly impacted by longer text lengths than those seen in training [19, 2, 15]. This phenomenon, referred to as "Length Generalization" or "Length Extrapolation" in the literature [25, 30], is also common in models based on the Transformer architecture [20, 16, 8, 32]. Notably, even Large Language Models (LLMs), known for their strong performance in a wide range of tasks and domains, are not immune to this problem [2, 5]. Recent research tried to address this challenge by modifications to the positional embeddings [25, 6, 7, 19, 13] or by using prompting strategies such as scratchpad [23] and chain-of-thought reasoning [28]. Nevertheless, there remains a lack of datasets specifically designed for the systematic evaluation of the problem.