Towards System 2 Reasoning in LLMs: Learning How to Think With Meta Chain-of-Thought

We propose a novel framework, Meta Chain-of-Thought (Meta-CoT), which extends traditional Chain-of-Thought (CoT) by explicitly modeling the underlying reasoning required to arrive at a particular CoT. We present empirical evidence from state-of-the-art models exhibiting behaviors consistent with in-context search, and explore methods for producing Meta-CoT via process supervision, synthetic data generation, and search algorithms. Finally, we outline a concrete pipeline for training a model to produce Meta-CoTs, incorporating instruction tuning with linearized search traces and reinforcement learning post-training. Finally, we discuss open research questions, including scaling laws, verifier roles, and the potential for discovering novel reasoning algorithms. This work provides a theoretical and practical roadmap to enable Meta-CoT in LLMs, paving the way for more powerful and human-like reasoning in artificial intelligence.

Policy-shaped prediction: avoiding distractions in model-based reinforcement learning

Model-based reinforcement learning (MBRL) is a promising route to sample-efficient policy optimization. However, a known vulnerability of reconstruction-based MBRL consists of scenarios in which detailed aspects of the world are highly predictable, but irrelevant to learning a good policy. Such scenarios can lead the model to exhaust its capacity on meaningless content, at the cost of neglecting important environment dynamics. While existing approaches attempt to solve this problem, we highlight its continuing impact on leading MBRL methods -- including DreamerV3 and DreamerPro -- with a novel environment where background distractions are intricate, predictable, and useless for planning future actions. To address this challenge we develop a method for focusing the capacity of the world model through synergy of a pretrained segmentation model, a task-aware reconstruction loss, and adversarial learning. Our method outperforms a variety of other approaches designed to reduce the impact of distractors, and is an advance towards robust model-based reinforcement learning.

Scaffold or Crutch? Examining College Students' Use and Views of Generative AI Tools for STEM Education

Developing problem-solving competency is central to Science, Technology, Engineering, and Mathematics (STEM) education, yet translating this priority into effective approaches to problem-solving instruction and assessment remain a significant challenge. The recent proliferation of generative artificial intelligence (genAI) tools like ChatGPT in higher education introduces new considerations about how these tools can help or hinder students' development of STEM problem-solving competency. Our research examines these considerations by studying how and why college students use genAI tools in their STEM coursework, focusing on their problem-solving support. We surveyed 40 STEM college students from diverse U.S. institutions and 28 STEM faculty to understand instructor perspectives on effective genAI tool use and guidance in STEM courses. Our findings reveal high adoption rates and diverse applications of genAI tools among STEM students. The most common use cases include finding explanations, exploring related topics, summarizing readings, and helping with problem-set questions. The primary motivation for using genAI tools was to save time. Moreover, over half of student participants reported simply inputting problems for AI to generate solutions, potentially bypassing their own problem-solving processes. These findings indicate that despite high adoption rates, students' current approaches to utilizing genAI tools often fall short in enhancing their own STEM problem-solving competencies. The study also explored students' and STEM instructors' perceptions of the benefits and risks associated with using genAI tools in STEM education. Our findings provide insights into how to guide students on appropriate genAI use in STEM courses and how to design genAI-based tools to foster students' problem-solving competency.

LayoutVLM: Differentiable Optimization of 3D Layout via Vision-Language Models

Open-universe 3D layout generation arranges unlabeled 3D assets conditioned on language instruction. Large language models (LLMs) struggle with generating physically plausible 3D scenes and adherence to input instructions, particularly in cluttered scenes. We introduce LayoutVLM, a framework and scene layout representation that exploits the semantic knowledge of Vision-Language Models (VLMs) and supports differentiable optimization to ensure physical plausibility. LayoutVLM employs VLMs to generate two mutually reinforcing representations from visually marked images, and a self-consistent decoding process to improve VLMs spatial planning. Our experiments show that LayoutVLM addresses the limitations of existing LLM and constraint-based approaches, producing physically plausible 3D layouts better aligned with the semantic intent of input language instructions. We also demonstrate that fine-tuning VLMs with the proposed scene layout representation extracted from existing scene datasets can improve performance.

ChatCollab: Exploring Collaboration Between Humans and AI Agents in Software Teams

We explore the potential for productive team-based collaboration between humans and Artificial Intelligence (AI) by presenting and conducting initial tests with a general framework that enables multiple human and AI agents to work together as peers. ChatCollab's novel architecture allows agents - human or AI - to join collaborations in any role, autonomously engage in tasks and communication within Slack, and remain agnostic to whether their collaborators are human or AI. Using software engineering as a case study, we find that our AI agents successfully identify their roles and responsibilities, coordinate with other agents, and await requested inputs or deliverables before proceeding. In relation to three prior multi-agent AI systems for software development, we find ChatCollab AI agents produce comparable or better software in an interactive game development task. We also propose an automated method for analyzing collaboration dynamics that effectively identifies behavioral characteristics of agents with distinct roles, allowing us to quantitatively compare collaboration dynamics in a range of experimental conditions. For example, in comparing ChatCollab AI agents, we find that an AI CEO agent generally provides suggestions 2-4 times more often than an AI product manager or AI developer, suggesting agents within ChatCollab can meaningfully adopt differentiated collaborative roles. Our code and data can be found at: https://github.com/ChatCollab.

FactorSim: Generative Simulation via Factorized Representation

Generating simulations to train intelligent agents in game-playing and robotics from natural language input, from user input or task documentation, remains an open-ended challenge. Existing approaches focus on parts of this challenge, such as generating reward functions or task hyperparameters. Unlike previous work, we introduce FACTORSIM that generates full simulations in code from language input that can be used to train agents. Exploiting the structural modularity specific to coded simulations, we propose to use a factored partially observable Markov decision process representation that allows us to reduce context dependence during each step of the generation. For evaluation, we introduce a generative simulation benchmark that assesses the generated simulation code's accuracy and effectiveness in facilitating zero-shot transfers in reinforcement learning settings. We show that FACTORSIM outperforms existing methods in generating simulations regarding prompt alignment (e.g., accuracy), zero-shot transfer abilities, and human evaluation. We also demonstrate its effectiveness in generating robotic tasks.

Building a Domain-specific Guardrail Model in Production

Generative AI holds the promise of enabling a range of sought-after capabilities and revolutionizing workflows in various consumer and enterprise verticals. However, putting a model in production involves much more than just generating an output. It involves ensuring the model is reliable, safe, performant and also adheres to the policy of operation in a particular domain. Guardrails as a necessity for models has evolved around the need to enforce appropriate behavior of models, especially when they are in production. In this paper, we use education as a use case, given its stringent requirements of the appropriateness of content in the domain, to demonstrate how a guardrail model can be trained and deployed in production. Specifically, we describe our experience in building a production-grade guardrail model for a K-12 educational platform. We begin by formulating the requirements for deployment to this sensitive domain. We then describe the training and benchmarking of our domain-specific guardrail model, which outperforms competing open- and closed- instruction-tuned models of similar and larger size, on proprietary education-related benchmarks and public benchmarks related to general aspects of safety. Finally, we detail the choices we made on architecture and the optimizations for deploying this service in production; these range across the stack from the hardware infrastructure to the serving layer to language model inference optimizations. We hope this paper will be instructive to other practitioners looking to create production-grade domain-specific services based on generative AI and large language models.

Hypothetical Minds: Scaffolding Theory of Mind for Multi-Agent Tasks with Large Language Models

Multi-agent reinforcement learning (MARL) methods struggle with the non-stationarity of multi-agent systems and fail to adaptively learn online when tested with novel agents. Here, we leverage large language models (LLMs) to create an autonomous agent that can handle these challenges. Our agent, Hypothetical Minds, consists of a cognitively-inspired architecture, featuring modular components for perception, memory, and hierarchical planning over two levels of abstraction. We introduce the Theory of Mind module that scaffolds the high-level planning process by generating hypotheses about other agents' strategies in natural language. It then evaluates and iteratively refines these hypotheses by reinforcing hypotheses that make correct predictions about the other agents' behavior. Hypothetical Minds significantly improves performance over previous LLM-agent and RL baselines on a range of competitive, mixed motive, and collaborative domains in the Melting Pot benchmark, including both dyadic and population-based environments. Additionally, comparisons against LLM-agent baselines and ablations reveal the importance of hypothesis evaluation and refinement for succeeding on complex scenarios.

Learning Formal Mathematics From Intrinsic Motivation

How did humanity coax mathematics from the aether? We explore the Platonic view that mathematics can be discovered from its axioms - a game of conjecture and proof. We describe Minimo (Mathematics from Intrinsic Motivation): an agent that jointly learns to pose challenging problems for itself (conjecturing) and solve them (theorem proving). Given a mathematical domain axiomatized in dependent type theory, we first combine methods for constrained decoding and type-directed synthesis to sample valid conjectures from a language model. Our method guarantees well-formed conjectures by construction, even as we start with a randomly initialized model. We use the same model to represent a policy and value function for guiding proof search. Our agent targets generating hard but provable conjectures - a moving target, since its own theorem proving ability also improves as it trains. We propose novel methods for hindsight relabeling on proof search trees to significantly improve the agent's sample efficiency in both tasks. Experiments on 3 axiomatic domains (propositional logic, arithmetic and group theory) demonstrate that our agent can bootstrap from only the axioms, self-improving in generating true and challenging conjectures and in finding proofs.

Tutorly: Turning Programming Videos Into Apprenticeship Learning Environments with LLMs

Online programming videos, including tutorials and streamcasts, are widely popular and contain a wealth of expert knowledge. However, effectively utilizing these resources to achieve targeted learning goals can be challenging. Unlike direct tutoring, video content lacks tailored guidance based on individual learning paces, personalized feedback, and interactive engagement necessary for support and monitoring. Our work transforms programming videos into one-on-one tutoring experiences using the cognitive apprenticeship framework. Tutorly, developed as a JupyterLab Plugin, allows learners to (1) set personalized learning goals, (2) engage in learning-by-doing through a conversational LLM-based mentor agent, (3) receive guidance and feedback based on a student model that steers the mentor moves. In a within-subject study with 16 participants learning exploratory data analysis from a streamcast, Tutorly significantly improved their performance from 61.9% to 76.6% based on a post-test questionnaire. Tutorly demonstrates the potential for enhancing programming video learning experiences with LLM and learner modeling.