Axios is expanding its local newsletter presence from 30 to 34 cities. In its continued pretense of benefiting newsrooms, OpenAI has partnered with Axios in a three-year deal to cover Pittsburgh, Pennsylvania; Kansas City, Missouri; Boulder, Colorado; and Huntsville, Alabama. What does OpenAI get in exchange for its funding? Oh, just the ability to use Axios content to answer users' questions. Like the close to 20 newsrooms that OpenAI has already partnered with, Axios seems to have forgotten that the scorpion did end up stinging the frog.

Representations from transformer-based unidirectional language models are known to be effective at predicting brain responses to natural language. However, most studies comparing language models to brains have used GPT-2 or similarly sized language models. Here we tested whether larger open-source models such as those from the OPT and LLaMA families are better at predicting brain responses recorded using fMRI. Mirroring scaling results from other contexts, we found that brain prediction performance scales logarithmically with model size from 125M to 30B parameter models, with 15% increased encoding performance as measured by correlation with a held-out test set across 3 subjects. Similar log-linear behavior was observed when scaling the size of the fMRI training set.

Exploration is a crucial skill for self-improvement and open-ended problemsolving. However, it remains uncertain whether large language models can effectively explore the state-space. Existing evaluations predominantly focus on the trade-off between exploration and exploitation, often assessed in multi-armed bandit problems. In contrast, this work isolates exploration as the sole objective, tasking the agent with delivering information that enhances future returns. For the evaluation, we propose to decompose missing rewards into exploration and exploitation components by measuring the optimal achievable return for the states already explored. Our experiments with various LLMs reveal that most models struggle to sufficiently explore the state-space and that weak exploration is insufficient. We observe a positive correlation between model size and exploration performance, with larger models demonstrating superior capabilities. Furthermore, we show that our decomposition provides insights into differences in behaviors driven by agent instructions during prompt engineering, offering a valuable tool for refining LLM performance in exploratory tasks. Recently, large language models (LLMs) have demonstrated promising results in various decision making tasks such as web browsing (Yao et al., 2022; Shinn et al., 2024; Ma et al., 2023), game-playing (Paglieri et al., 2024), and tasks in simulated households (Yao et al., 2022; Shinn et al., 2024). This way, LLMs act as agents that observe states and take actions in different environments. Through their vast internal knowledge-base and autoregressive in-context reasoning capabilities, the models are supposed to quickly adapt to new tasks. However, previous work has shown that LLMs struggle with solving increasingly complex environments due to several limitations: For example, the ability to learn from mistakes is often limited (Huang et al., 2023) and LLMs have difficulties with planning over long horizons (Kambhampati et al., 2024). The examples emphasize that understanding LLM abilities is essential for their risk assessment in real life applications, and future development.

Vertical AI Agents are revolutionizing industries by delivering domain-specific intelligence and tailored solutions. However, many sectors, such as manufacturing, healthcare, and logistics, demand AI systems capable of extending their intelligence into the physical world, interacting directly with objects, environments, and dynamic conditions. This need has led to the emergence of Physical AI Agents--systems that integrate cognitive reasoning, powered by specialized LLMs, with precise physical actions to perform real-world tasks. This work introduces Physical AI Agents as an evolution of shared principles with Vertical AI Agents, tailored for physical interaction. We propose a modular architecture with three core blocks--perception, cognition, and actuation--offering a scalable framework for diverse industries. Additionally, we present the Physical Retrieval Augmented Generation (Ph-RAG) design pattern, which connects physical intelligence to industry-specific LLMs for real-time decision-making and reporting informed by physical context. Through case studies, we demonstrate how Physical AI Agents and the Ph-RAG framework are transforming industries like autonomous vehicles, warehouse robotics, healthcare, and manufacturing, offering businesses a pathway to integrate embodied AI for operational efficiency and innovation.

This paper presents the implementation and evaluation of a conversational agent designed for multi-party open-ended interactions. Leveraging state-of-the-art technologies such as voice direction of arrival, voice recognition, face tracking, and large language models, the system aims to facilitate natural and intuitive human-robot conversations. Deployed on the Furhat robot, the system was tested with 30 participants engaging in open-ended group conversations and then in two overlapping discussions. Quantitative metrics, such as latencies and recognition accuracy, along with qualitative measures from user questionnaires, were collected to assess performance. The results highlight the system's effectiveness in managing multi-party interactions, though improvements are needed in response relevance and latency. This study contributes valuable insights for advancing human-robot interaction, particularly in enhancing the naturalness and engagement in group conversations.

The escalating demand for long-context applications has intensified the necessity of extending the LLM context windows. Despite recent fine-tuning approaches successfully expanding context lengths, their high memory footprints, especially for activations, present a critical practical limitation. Current parameter-efficient fine-tuning methods prioritize reducing parameter update overhead over addressing activation memory constraints. Similarly, existing sparsity mechanisms improve computational efficiency but overlook activation memory optimization due to the phenomenon of Shadowy Activation. In this paper, we propose LeMo, the first LLM fine-tuning system that explores and exploits a new token-level sparsity mechanism inherent in long-context scenarios, termed Contextual Token Sparsity. LeMo minimizes redundant token involvement by assessing the informativeness of token embeddings while preserving model accuracy. Specifically, LeMo introduces three key techniques: (1) Token Elimination, dynamically identifying and excluding redundant tokens across varying inputs and layers. (2) Pattern Prediction, utilizing well-trained predictors to approximate token sparsity patterns with minimal overhead. (3) Kernel Optimization, employing permutation-free and segment-based strategies to boost system performance. We implement LeMo as an end-to-end fine-tuning system compatible with various LLM architectures and other optimization techniques. Comprehensive evaluations demonstrate that LeMo reduces memory consumption by up to 1.93x and achieves up to 1.36x speedups, outperforming state-of-the-art fine-tuning systems.

Retrieval-Augmented Generation equips large language models with the capability to retrieve external knowledge, thereby mitigating hallucinations by incorporating information beyond the model's intrinsic abilities. However, most prior works have focused on invoking retrieval deterministically, which makes it unsuitable for tasks such as long-form question answering. Instead, dynamically performing retrieval by invoking it only when the underlying LLM lacks the required knowledge can be more efficient. In this context, we delve deeper into the question, "To Retrieve or Not to Retrieve?" by exploring multiple uncertainty detection methods. We evaluate these methods for the task of long-form question answering, employing dynamic retrieval, and present our comparisons. Our findings suggest that uncertainty detection metrics, such as Degree Matrix Jaccard and Eccentricity, can reduce the number of retrieval calls by almost half, with only a slight reduction in question-answering accuracy.

Automated audio captioning is a task that generates textual descriptions for audio content, and recent studies have explored using visual information to enhance captioning quality. However, current methods often fail to effectively fuse audio and visual data, missing important semantic cues from each modality. To address this, we introduce LAVCap, a large language model (LLM)-based audio-visual captioning framework that effectively integrates visual information with audio to improve audio captioning performance. LAVCap employs an optimal transport-based alignment loss to bridge the modality gap between audio and visual features, enabling more effective semantic extraction. Additionally, we propose an optimal transport attention module that enhances audio-visual fusion using an optimal transport assignment map. Combined with the optimal training strategy, experimental results demonstrate that each component of our framework is effective. LAVCap outperforms existing state-of-the-art methods on the AudioCaps dataset, without relying on large datasets or post-processing. Code is available at https://github.com/NAVER-INTEL-Co-Lab/gaudi-lavcap.

This paper presents an efficient decoding approach for end-to-end automatic speech recognition (E2E-ASR) with large language models (LLMs). Although shallow fusion is the most common approach to incorporate language models into E2E-ASR decoding, we face two practical problems with LLMs. (1) LLM inference is computationally costly. (2) There may be a vocabulary mismatch between the ASR model and the LLM. To resolve this mismatch, we need to retrain the ASR model and/or the LLM, which is at best time-consuming and in many cases not feasible. We propose "delayed fusion," which applies LLM scores to ASR hypotheses with a delay during decoding and enables easier use of pre-trained LLMs in ASR tasks. This method can reduce not only the number of hypotheses scored by the LLM but also the number of LLM inference calls. It also allows re-tokenizion of ASR hypotheses during decoding if ASR and LLM employ different tokenizations. We demonstrate that delayed fusion provides improved decoding speed and accuracy compared to shallow fusion and N-best rescoring using the LibriHeavy ASR corpus and three public LLMs, OpenLLaMA 3B & 7B and Mistral 7B.

In-context learning is a remarkable capability of transformers, referring to their ability to adapt to specific tasks based on a short history or context. Previous research has found that task-specific information is locally encoded within models, though their emergence and functionality remain unclear due to opaque pre-training processes. In this work, we investigate the formation of task vectors in a controlled setting, using models trained from scratch on synthetic datasets. Our findings confirm that task vectors naturally emerge under certain conditions, but the tasks may be relatively weakly and/or non-locally encoded within the model. To promote strong task vectors encoded at a prescribed location within the model, we propose an auxiliary training mechanism based on a task vector prompting loss (TVP-loss). This method eliminates the need to search for task-correlated encodings within the trained model and demonstrably improves robustness and generalization.