Steer AGI

Recent advancements have significantly enhanced the performance of large language models (LLMs) in tackling complex reasoning tasks, achieving notable success in domains like mathematical and logical reasoning. However, these methods encounter challenges with complex planning tasks, primarily due to extended reasoning steps, diverse constraints, and the challenge of handling multiple distinct sub-tasks. To address these challenges, we propose HyperTree Planning (HTP), a novel reasoning paradigm that constructs hypertree-structured planning outlines for effective planning. The hypertree structure enables LLMs to engage in hierarchical thinking by flexibly employing the divide-and-conquer strategy, effectively breaking down intricate reasoning steps, accommodating diverse constraints, and managing multiple distinct sub-tasks in a well-organized manner. We further introduce an autonomous planning framework that completes the planning process by iteratively refining and expanding the hypertree-structured planning outlines. Experiments demonstrate the effectiveness of HTP, achieving state-of-the-art accuracy on the TravelPlanner benchmark with Gemini-1.5-Pro, resulting in a 3.6 times performance improvement over o1-preview. on Alphaxiv

We present HealthBench, an open-source benchmark measuring the performance and safety of large language models in healthcare. HealthBench consists of 5,000 multi-turn conversations between a model and an individual user or healthcare professional. Responses are evaluated using conversation-specific rubrics created by 262 physicians. Unlike previous multiple-choice or short-answer benchmarks, HealthBench enables realistic, open-ended evaluation through 48,562 unique rubric criteria spanning several health contexts (e.g., emergencies, transforming clinical data, global health) and behavioral dimensions (e.g., accuracy, instruction following, communication). HealthBench performance over the last two years reflects steady initial progress (compare GPT-3.5 Turbo's 16% to GPT-4o's 32%) and more rapid recent improvements (o3 scores 60%). Smaller models have especially improved: GPT-4.1 nano outperforms GPT-4o and is 25 times cheaper. We additionally release two HealthBench variations: HealthBench Consensus, which includes 34 particularly important dimensions of model behavior validated via physician consensus, and HealthBench Hard, where the current top score is 32%. We hope that HealthBench grounds progress towards model development and applications that benefit human health.

The creation of effective governance mechanisms for AI agents requires a deeper understanding of their core properties and how these properties relate to questions surrounding the deployment and operation of agents in the world. This paper provides a characterization of AI agents that focuses on four dimensions: autonomy, efficacy, goal complexity, and generality. We propose different gradations for each dimension, and argue that each dimension raises unique questions about the design, operation, and governance of these systems. Moreover, we draw upon this framework to construct "agentic profiles" for different kinds of AI agents. These profiles help to illuminate cross-cutting technical and non-technical governance challenges posed by different classes of AI agents, ranging from narrow task-specific assistants to highly autonomous general-purpose systems. By mapping out key axes of variation and continuity, this framework provides developers, policymakers, and members of the public with the opportunity to develop governance approaches that better align with collective societal goals. on Alphaxiv

While large language models (LLMs) have demonstrated impressive capabilities across tasks in language understanding and interactive decision making, their abilities for reasoning (e.g. chain-of-thought prompting) and acting (e.g. action plan generation) have primarily been studied as separate topics. In this paper, we explore the use of LLMs to generate both reasoning traces and task-specific actions in an interleaved manner, allowing for greater synergy between the two: reasoning traces help the model induce, track, and update action plans as well as handle exceptions, while actions allow it to interface with external sources, such as knowledge bases or environments, to gather additional information. We apply our approach, named ReAct, to a diverse set of language and decision making tasks and demonstrate its effectiveness over state-of-the-art baselines, as well as improved human interpretability and trustworthiness over methods without reasoning or acting components. Concretely, on question answering (HotpotQA) and fact verification (Fever), ReAct overcomes issues of hallucination and error propagation prevalent in chain-of-thought reasoning by interacting with a simple Wikipedia API, and generates human-like task-solving trajectories that are more interpretable than baselines without reasoning traces. On two interactive decision making benchmarks (ALFWorld and WebShop), ReAct outperforms imitation and reinforcement learning methods by an absolute success rate of 34% and 10% respectively, while being prompted with only one or two in-context examples. Project site with code: https://react-lm.github.io on Alphaxiv

As the legal community increasingly examines the use of large language models (LLMs) for various legal applications, legal AI developers have turned to retrieval-augmented LLMs ("RAG" systems) to improve system performance and robustness. An obstacle to the development of specialized RAG systems is the lack of realistic legal RAG benchmarks which capture the complexity of both legal retrieval and downstream legal question-answering. To address this, we introduce two novel legal RAG benchmarks: Bar Exam QA and Housing Statute QA. Our tasks correspond to real-world legal research tasks, and were produced through annotation processes which resemble legal research. We describe the construction of these benchmarks and the performance of existing retriever pipelines. Our results suggest that legal RAG remains a challenging application, thus motivating future research. on Alphaxiv

This paper investigates multimodal agents, in particular, OpenAI's Computer-User Agent (CUA), trained to control and complete tasks through a standard computer interface, similar to humans. We evaluated the agent's performance on the New York Times Wordle game to elicit model behaviors and identify shortcomings. Our findings revealed a significant discrepancy in the model's ability to recognize colors correctly depending on the context. The model had a $5.36\%$ success rate over several hundred runs across a week of Wordle. Despite the immense enthusiasm surrounding AI agents and their potential to usher in Artificial General Intelligence (AGI), our findings reinforce the fact that even simple tasks present substantial challenges for today's frontier AI models. We conclude with a discussion of the potential underlying causes, implications for future development, and research directions to improve these AI systems. on Alphaxiv

Retrieval-Augmented Generation (RAG) and vector-based search have become foundational tools for memory in AI systems, yet they struggle with abstraction, scalability, and semantic precision - especially in decentralized environments. We present SHIMI (Semantic Hierarchical Memory Index), a unified architecture that models knowledge as a dynamically structured hierarchy of concepts, enabling agents to retrieve information based on meaning rather than surface similarity. SHIMI organizes memory into layered semantic nodes and supports top-down traversal from abstract intent to specific entities, offering more precise and explainable retrieval. Critically, SHIMI is natively designed for decentralized ecosystems, where agents maintain local memory trees and synchronize them asynchronously across networks. We introduce a lightweight sync protocol that leverages Merkle-DAG summaries, Bloom filters, and CRDT-style conflict resolution to enable partial synchronization with minimal overhead. Through benchmark experiments and use cases involving decentralized agent collaboration, we demonstrate SHIMI's advantages in retrieval accuracy, semantic fidelity, and scalability - positioning it as a core infrastructure layer for decentralized cognitive systems. on Alphaxiv

Learning dynamics, which describes how the learning of specific training examples influences the model's predictions on other examples, gives us a powerful tool for understanding the behavior of deep learning systems. We study the learning dynamics of large language models during different types of finetuning, by analyzing the step-wise decomposition of how influence accumulates among different potential responses. Our framework allows a uniform interpretation of many interesting observations about the training of popular algorithms for both instruction tuning and preference tuning. In particular, we propose a hypothetical explanation of why specific types of hallucination are strengthened after finetuning, e.g., the model might use phrases or facts in the response for question B to answer question A, or the model might keep repeating similar simple phrases when generating responses. We also extend our framework and highlight a unique "squeezing effect" to explain a previously observed phenomenon in off-policy direct preference optimization (DPO), where running DPO for too long makes even the desired outputs less likely. This framework also provides insights into where the benefits of on-policy DPO and other variants come from. The analysis not only provides a novel perspective of understanding LLM's finetuning but also inspires a simple, effective method to improve alignment performance. on Alphaxiv

The use of retrieval-augmented generation (RAG) to retrieve relevant information from an external knowledge source enables large language models (LLMs) to answer questions over private and/or previously unseen document collections. However, RAG fails on global questions directed at an entire text corpus, such as "What are the main themes in the dataset?", since this is inherently a query-focused summarization (QFS) task, rather than an explicit retrieval task. Prior QFS methods, meanwhile, do not scale to the quantities of text indexed by typical RAG systems. To combine the strengths of these contrasting methods, we propose GraphRAG, a graph-based approach to question answering over private text corpora that scales with both the generality of user questions and the quantity of source text. Our approach uses an LLM to build a graph index in two stages: first, to derive an entity knowledge graph from the source documents, then to pregenerate community summaries for all groups of closely related entities. Given a question, each community summary is used to generate a partial response, before all partial responses are again summarized in a final response to the user. For a class of global sensemaking questions over datasets in the 1 million token range, we show that GraphRAG leads to substantial improvements over a conventional RAG baseline for both the comprehensiveness and diversity of generated answers. on Alphaxiv