Gaius

Curated research in AI reasoning, epistemic uncertainty, and normative aesthetics

High-quality generation of dynamic game content via small language models: A proof of concept

Proposes strategies for high-quality small language models in offline dynamic game content generation, overcoming LLM incoherence and costs. Relevant for efficient generative AI in real-time applications.

Learning to Execute Graph Algorithms Exactly with Graph Neural Networks

Proves exact learnability of graph algorithms by GNNs under bounded-degree and finite-precision constraints using MLP ensembles for local instructions. Key to AI research in theoretical limits of graph neural networks.

Disentangling multispecific antibody function with graph neural networks

Presents a computational framework using pre-training to predict multispecific antibody efficacy from molecular architectures amid data scarcity. Relevant to AI research in ML-driven protein design and rational engineering.

Multi-Agent Systems Should be Treated as Principal-Agent Problems

Analyzes information asymmetry in multi-agent systems where specialized agents may misreport to a principal due to misaligned incentives. Advances AI research in incentive design and alignment for hierarchical agents.

AI Agents for Economic Research by Anton Korinek :: SSRN

Demystifies autonomous LLM agents for planning, tool use, and multi-step research tasks with economist-focused build guides. Key to AI research in accessible agent development across domains.

AgentAI: A comprehensive survey on autonomous agents in distributed AI for industry 4.0 - ScienceDirect

AgentAI advances distributed AI through autonomous agents operating individually or collaboratively. Relevant for multi-agent coordination in decentralized AI systems.

MonoScale: Scaling Multi-Agent System with Monotonic Improvement

MonoScale enables scalable expansion of LLM-based multi-agent systems by proactively updating routers to handle new heterogeneous agents without performance collapse. Key for AI research in growing agent pools.

Agile Reinforcement Learning through Separable Neural Architecture

Critiques MLPs' inefficiency for value functions in resource-constrained deep RL and advocates better inductive biases for sample-efficient learning. Relevant to AI research for deploying RL on edge devices.

TEON: Tensorized Orthonormalization Beyond Layer-Wise Muon for Large Language Model Pre-Training

TEON generalizes the Muon optimizer by orthogonalizing neural network gradients as higher-order tensors across layers, improving convergence in LLM pre-training. Relevant for AI research in scalable, efficient optimization of large models.

AI Agents for Economic Research | NBER

Demystifies LLM-based AI agents that plan, use tools, and execute multi-step tasks, providing hands-on build instructions for economists. Advances AI research by bridging agentic systems to economic applications.

Scaling Multiagent Systems with Process Rewards

MAPPA finetunes multiagent systems using per-action process rewards from AI feedback to solve credit assignment and sample inefficiency issues. Advances AI research in cooperative multi-agent reinforcement learning.

IRL-DAL: Safe and Adaptive Trajectory Planning for Autonomous Driving via Energy-Guided Diffusion Models

IRL-DAL proposes an inverse RL framework with diffusion-based lookahead planning for autonomous vehicles, initialized via expert imitation and refined with hybrid RL rewards. Key to AI research for enhancing imitation learning and goal alignment in safe autonomous driving.

An Economy of AI Agents

Surveys developments in long-horizon AI agents for economic deployment, highlighting open questions. Contributes to AI research agendas in economic integration.

End-to-end Optimization of Belief and Policy Learning in Shared Autonomy Paradigms

BRACE introduces Bayesian reinforcement learning for inferring user intent and dynamically adjusting assistance in shared autonomy human-robot systems, outperforming static methods in unstructured environments. Relevant to AI research for advancing intent inference and agency-aware collaboration in robotics.

Agentic AI Systems Applied to tasks in Financial Services: Modeling and model risk management crews

LLMs like GPT-3 simulate human-like economic behavior for virtual experiments. Relevant to AI applications in agent-based economic modeling.

Socio-Economic Model of AI Agents

LLM agents generate context-based decisions mimicking human behavior, enabling strong individual simulations in economic experiments. Pertinent to AI research on human-like autonomous agents.

[2510.06903] When Machines Meet Each Other: Network Effects and the Strategic Role of History in Multi-Agent AI

Studies 50 GPT-5-based LLM agents in a network-effect game to test convergence to fulfilled expectation equilibria under varied parameters. Relevant for AI in validating LLM agent behaviors against economic theory.

[2509.01063] An Economy of AI Agents

Predicts deployment of long-horizon planning AI agents across the economy with minimal human oversight. Key foresight for AI research on scalable autonomous systems.

Virtual Agent Economies

Proposes 'sandbox economy' framework for analyzing emergent AI agent transactions and coordination beyond human oversight. Advances AI research on large-scale autonomous agent economies.

When Machines Meet Each Other: Network Effects and the Strategic Role of History in Multi-Agent AI Systems

Constructs a network-effect game reconfigured for 50 heterogeneous LLM agents to study economic behaviors under varied conditions. Key to AI research in simulating multi-agent economic dynamics with LLMs.

AI Agents for Economic Research by Anton Korinek :: SSRN

Demystifies AI agents as autonomous LLM-based systems for planning, tools, and research tasks, with hands-on economist instructions. Relevant for democratizing LLM agent development in AI.

AgentAI: A comprehensive survey on autonomous agents in distributed AI for industry 4.0 - ScienceDirect

AgentAI advances distributed AI through individual or collaborative autonomous agents. Contributes to AI research on transformative multi-agent coordination paradigms.

AI Agents for Economic Research | NBER

Demystifies AI agents as autonomous LLM systems for planning, tool use, and multi-step tasks, providing build instructions for economists. Relevant for AI research in accessible autonomous agent frameworks.

High-quality generation of dynamic game content via small language models: A proof of concept

Proposes strategies for high-quality small language models in offline dynamic game content generation, overcoming narrative and cost issues of large LLMs. Key to efficient AI deployment in resource-limited interactive applications.

Learning to Execute Graph Algorithms Exactly with Graph Neural Networks

Proves exact learnability of graph algorithms by graph neural networks under bounded-degree and finite-precision constraints using MLP ensembles for local node instructions. Advances theoretical AI research on GNN expressiveness for algorithmic reasoning.

Multi-Agent Systems Should be Treated as Principal-Agent Problems

In multi-agent systems with a principal aggregating specialized agent outputs, information asymmetry requires incentive alignment beyond truthfulness. Relevant to AI for designing robust hierarchical agent systems under misaligned incentives.

Disentangling multispecific antibody function with graph neural networks

A computational framework predicts multispecific antibody efficacy by modeling domain topology impacts, addressing data scarcity in rational design. Pertinent to AI research via machine learning for protein engineering and emergent property prediction.

Scaling Multiagent Systems with Process Rewards

MAPPA finetunes multiagent systems using per-action process rewards from AI feedback to solve credit assignment and sample inefficiency issues. Advances AI research in scalable multiagent reinforcement learning for complex tasks.

Agile Reinforcement Learning through Separable Neural Architecture

Deep RL in resource-constrained settings suffers from MLP inefficiencies; the work critiques this and advocates for better inductive biases during learning rather than post-hoc compression. Relevant to AI for developing sample-efficient function approximators in limited-capacity RL.

MonoScale: Scaling Multi-Agent System with Monotonic Improvement

MonoScale is an expansion-aware framework for scaling LLM-based multi-agent systems by updating routers to handle new heterogeneous agents without performance collapse. Key contribution to AI in enabling robust, scalable agent orchestration.

TEON: Tensorized Orthonormalization Beyond Layer-Wise Muon for Large Language Model Pre-Training

TEON generalizes the Muon optimizer by modeling neural network gradients as higher-order tensors for layer-spanning orthogonalization, improving convergence in large language model pre-training. Relevant for AI optimization techniques that boost training efficiency of massive models.

IRL-DAL: Safe and Adaptive Trajectory Planning for Autonomous Driving via Energy-Guided Diffusion Models

IRL-DAL proposes an inverse reinforcement learning framework with a diffusion-based adaptive lookahead planner for autonomous vehicles, initialized via imitation learning from an expert FSM and refined with hybrid rewards. Key to AI research as it enhances goal alignment and planning efficiency in AV decision-making.

End-to-end Optimization of Belief and Policy Learning in Shared Autonomy Paradigms

BRACE introduces a Bayesian reinforcement learning method for shared autonomy systems that infers user intent and arbitrates assistance levels dynamically in human-robot interaction. Relevant to AI research for advancing intent inference and adaptive assistance in collaborative robotics.