The profound success of transformer-based language models can largely be attributed to their ability to integrate relevant contextual information from an input sequence in order to generate a response or complete a task. However, we know very little about the algorithms that a model employs to implement this capability, nor do we understand their failure modes. For example, given the prompt "John is going fishing, so he walks over to the bank. Can he make an ATM transaction?", a model may incorrectly respond "Yes" if it has not properly contextualized "bank" as a geographical feature, rather than a financial institution. We propose the LLM Race Conditions Hypothesis as an explanation of contextualization errors of this form. This hypothesis identifies dependencies between tokens (e.g., "bank" must be properly contextualized before the final token, "?", integrates information from "bank"), and claims that contextualization errors are a result of violating these dependencies. Using a variety of techniques from mechanistic intepretability, we provide correlational and causal evidence in support of the hypothesis, and suggest inference-time interventions to address it.

Watermarking has recently emerged as an effective strategy for detecting the outputs of large language models (LLMs). Most existing schemes require whitebox access to the model's next-token probability distribution, which is typically not accessible to downstream users of an LLM API. In this work, we propose a principled watermarking scheme that requires only the ability to sample sequences from the LLM (i.e. We provide performance guarantees, demonstrate how it can be leveraged when white-box access is available, and show when it can outperform existing white-box schemes via comprehensive experiments. It can be critical to understand whether a piece of text is generated by a large language model (LLM). For instance, one often wants to know how trustworthy a piece of text is, and those written by an LLM may be deemed untrustworthy as these models can hallucinate. This problem comes in different flavors -- one may want to detect whether it was generated by a specific model or by any model. Furthermore, the detecting party may or may not have white-box access (e.g. an ability to compute log-probabilities) to the generator they wish to test against. Typically, parties that have white-box access are the owners of the model so we refer to this case as first-party detection and the counterpart as third-party detection. The goal of watermarking is to cleverly bias the generator so that first-party detection becomes easier. Most proposed techniques do not modify the underlying LLM's model weights or its training procedure but rather inject the watermark during autoregressive decoding at inference time. They require access to the next-token logits and inject the watermark every step of the sampling loop. This required access prevents third-party users of an LLM from applying their own watermark as proprietary APIs currently do not support this option. Supporting this functionality presents a security risk in addition to significant engineering considerations.

Generative artificial intelligence (AI) has opened the possibility of automated content production, including coding in software development, which can significantly influence the participation and performance of software developers. To explore this impact, we investigate the role of GitHub Copilot, a generative AI pair programmer, on software development in open-source community, where multiple developers voluntarily collaborate on software projects. Using GitHub's dataset for open-source repositories and a generalized synthetic control method, we find that Copilot significantly enhances project-level productivity by 6.5%. Delving deeper, we dissect the key mechanisms driving this improvement. Our findings reveal a 5.5% increase in individual productivity and a 5.4% increase in participation. However, this is accompanied with a 41.6% increase in integration time, potentially due to higher coordination costs. Interestingly, we also observe the differential effects among developers. We discover that core developers achieve greater project-level productivity gains from using Copilot, benefiting more in terms of individual productivity and participation compared to peripheral developers, plausibly due to their deeper familiarity with software projects. We also find that the increase in project-level productivity is accompanied with no change in code quality. We conclude that AI pair programmers bring benefits to developers to automate and augment their code, but human developers' knowledge of software projects can enhance the benefits. In summary, our research underscores the role of AI pair programmers in impacting project-level productivity within the open-source community and suggests potential implications for the structure of open-source software projects.

Large language models (LLMs) deployed as agents solve user-specified tasks over multiple steps while keeping the required manual engagement to a minimum. Crucially, such LLMs need to ground their generations in any feedback obtained to reliably achieve desired outcomes. We propose an end-to-end reinforcement learning method for teaching models to leverage execution feedback in the realm of code synthesis, where state-of-the-art LLMs struggle to improve code iteratively compared to independent sampling. We benchmark on competitive programming tasks, where we achieve new start-of-the art results with both small (8B parameters) and large (70B) models while reducing the amount of samples required by an order of magnitude. Our analysis of inference-time behavior demonstrates that our method produces LLMs that effectively leverage automatic feedback over multiple steps.

Multi-modal Large Language Models (MLLMs) have recently exhibited impressive general-purpose capabilities by leveraging vision foundation models to encode the core concepts of images into representations. These are then combined with instructions and processed by the language model to generate high-quality responses. Despite significant progress in enhancing the language component, challenges persist in optimally fusing visual encodings within the language model for task-specific adaptability. Recent research has focused on improving this fusion through modality adaptation modules but at the cost of significantly increased model complexity and training data needs. In this paper, we propose EMMA (Efficient Multi-Modal Adaptation), a lightweight cross-modality module designed to efficiently fuse visual and textual encodings, generating instruction-aware visual representations for the language model. Our key contributions include: (1) an efficient early fusion mechanism that integrates vision and language representations with minimal added parameters (less than 0.2% increase in model size), (2) an in-depth interpretability analysis that sheds light on the internal mechanisms of the proposed method; (3) comprehensive experiments that demonstrate notable improvements on both specialized and general benchmarks for MLLMs. Empirical results show that EMMA boosts performance across multiple tasks by up to 9.3% while significantly improving robustness against hallucinations. Our code is available at https://github.com/SaraGhazanfari/EMMA

We present a foundation model for zero-shot metric monocular depth estimation. The predictions are metric, with absolute scale, without relying on the availability of metadata such as camera intrinsics. And the model is fast, producing a 2.25-megapixel depth map in 0.3 seconds on a standard GPU. These characteristics are enabled by a number of technical contributions, including an efficient multi-scale vision transformer for dense prediction, a training protocol that combines real and synthetic datasets to achieve high metric accuracy alongside fine boundary tracing, dedicated evaluation metrics for boundary accuracy in estimated depth maps, and state-of-the-art focal length estimation from a single image. Extensive experiments analyze specific design choices and demonstrate that Depth Pro outperforms prior work along multiple dimensions. Zero-shot monocular depth estimation underpins a growing variety of applications, such as advanced image editing, view synthesis, and conditional ...

It has been reported that LLMs can recognize their own writing. As this has potential implications for AI safety, yet is relatively understudied, we investigate the phenomenon, seeking to establish whether it robustly occurs at the behavioral level, how the observed behavior is achieved, and whether it can be controlled. First, we find that the Llama3-8b-Instruct chat model - but not the base Llama3-8b model - can reliably distinguish its own outputs from those of humans, and present evidence that the chat model is likely using its experience with its own outputs, acquired during post-training, to succeed at the writing recognition task. Second, we identify a vector in the residual stream of the model that is differentially activated when the model makes a correct self-written-text recognition judgment, show that the vector activates in response to information relevant to self-authorship, present evidence that the vector is related to the concept of "self" in the model, and demonstrate that the vector is causally related to the model's ability to perceive and assert self-authorship. Finally, we show that the vector can be used to control both the model's behavior and its perception, steering the model to claim or disclaim authorship by applying the vector to the model's output as it generates it, and steering the model to believe or disbelieve it wrote arbitrary texts by applying the vector to them as the model reads them.

Temporal point processes (TPPs) are widely used to model the timing and occurrence of events in domains such as social networks, transportation systems, and e-commerce. In this paper, we introduce TPP-LLM, a novel framework that integrates large language models (LLMs) with TPPs to capture both the semantic and temporal aspects of event sequences. Unlike traditional methods that rely on categorical event type representations, TPP-LLM directly utilizes the textual descriptions of event types, enabling the model to capture rich semantic information embedded in the text. While LLMs excel at understanding event semantics, they are less adept at capturing temporal patterns. To address this, TPP-LLM incorporates temporal embeddings and employs parameter-efficient fine-tuning (PEFT) methods to effectively learn temporal dynamics without extensive retraining. This approach improves both predictive accuracy and computational efficiency. Experimental results across diverse real-world datasets demonstrate that TPP-LLM outperforms state-of-the-art baselines in sequence modeling and event prediction, highlighting the benefits of combining LLMs with TPPs. Temporal point processes (TPPs) (Shchur et al., 2021) are powerful tools for modeling the occurrence of events over time, with widespread applications in domains such as social networks, urban dynamics, transportation, natural disasters, and e-commerce. The challenge of predicting both the type and timing of future events based on historical sequences has led to the development of increasingly sophisticated models. Traditional TPP models often rely on handcrafted features or specific assumptions about temporal dependencies, which limit their ability to capture complex event patterns in real-world datasets.

We present Synthio, a novel approach for augmenting small-scale audio classification datasets with synthetic data. Our goal is to improve audio classification accuracy with limited labeled data. Traditional data augmentation techniques, which apply artificial transformations (e.g., adding random noise or masking segments), struggle to create data that captures the true diversity present in real-world audios. To address this shortcoming, we propose to augment the dataset with synthetic audio generated from text-to-audio (T2A) diffusion models. However, synthesizing effective augmentations is challenging because not only should the generated data be acoustically consistent with the underlying small-scale dataset, but they should also have sufficient compositional diversity. To overcome the first challenge, we align the generations of the T2A model with the small-scale dataset using preference optimization. This ensures that the acoustic characteristics of the generated data remain consistent with the small-scale dataset. To address the second challenge, we propose a novel caption generation technique that leverages the reasoning capabilities of Large Language Models to (1) generate diverse and meaningful audio captions and (2) iteratively refine their quality. The generated captions are then used to prompt the aligned T2A model. We extensively evaluate Synthio on ten datasets and four simulated limited-data settings. Results indicate our method consistently outperforms all baselines by 0.1%-39% using a T2A model trained only on weakly-captioned AudioSet.

Existing 3D facial emotion modeling have been constrained by limited emotion classes and insufficient datasets. This paper introduces "Emo3D", an extensive "Text-Image-Expression dataset" spanning a wide spectrum of human emotions, each paired with images and 3D blendshapes. Leveraging Large Language Models (LLMs), we generate a diverse array of textual descriptions, facilitating the capture of a broad spectrum of emotional expressions. Using this unique dataset, we conduct a comprehensive evaluation of language-based models' fine-tuning and vision-language models like Contranstive Language Image Pretraining (CLIP) for 3D facial expression synthesis. We also introduce a new evaluation metric for this task to more directly measure the conveyed emotion. Our new evaluation metric, Emo3D, demonstrates its superiority over Mean Squared Error (MSE) metrics in assessing visual-text alignment and semantic richness in 3D facial expressions associated with human emotions. "Emo3D" has great applications in animation design, virtual reality, and emotional human-computer interaction.