Searcharxiv⌕ Search

arXiv subjects

Chi Jin

Publications and source records attributed to Chi Jin.

At least 19 recordsLinked to original sources
arXiv2026

LeAct: Learning to Reason from Expert Actions

Modern reasoning models depend on reasoning data, today sourced from human annotations or distilled from stronger LLMs. However, a rich and largely untapped source of supervision lies in expert systems (e.g., game engines, classical planners, theorem provers), which routinely produce near-optimal actions across diverse domains. But these experts are silent: they commit to an action without writing down the chain of thought (CoT) behind it. Recovering that CoT as natural-language reasoning would distill expert knowledge into a student that generalizes beyond the demonstrated actions. We treat it as a latent variable and study how to recover it from the action alone. Our approach, LeAct (Learning to reason from Actions), optimizes this latent variable: the student samples candidate CoTs for each expert action, and we retain those that measurably improve its own probability of recovering the action. Across imperfect-information games at multiple scales and a simulated robotics benchmark, LeAct reaches the solver's numerical floor on small enumerable games. At larger scale, it is $5\times$ closer to the solver than the strongest expert-iteration baseline. At Flop Hold'em ($\sim 10^9$ infosets), LeAct wins head-to-head by $+60$ mbb/g, and on the robotics probe it is the only training recipe that improves on direct imitation. We present a principled framework and the result: expert systems become a categorically new source of reasoning teachers for foundation models.

cs.LG↗
arXiv2026

Goedel-Architect: Streamlining Formal Theorem Proving with Blueprint Generation and Refinement

We introduce Goedel-Architect, an agentic framework for formal theorem proving in Lean 4 centered on blueprint generation and refinement. A blueprint is a dependency graph of definitions and lemmas that builds up to the main theorem. First, Goedel-Architect generates a blueprint of formally stated definitions and lemmas, along with declared dependencies. This blueprint is optionally guided by a natural language proof. Then, a tool-equipped Lean prover component closes each open lemma node in parallel using relevant dependencies. Failed lemmas in turn drive refinement of the global blueprint. This strategy contrasts with other mainstream approaches which use recursive lemma decomposition, and can inefficiently loop on dead-end strategies. Using the open-weight DeepSeek-V4-Flash (284B-A13B) as the backbone, Goedel-Architect attains 99.2% pass@1 on MiniF2F-test and 75.6% pass@1 on PutnamBench. With an optional natural-language proof seeding the initial blueprint on the harder problems, we additionally close the remaining two MiniF2F-test problems (reaching 100%), lift PutnamBench to 88.8% (597/672), and solve 4/6 on IMO 2025, 11/12 on Putnam 2025, and 3/6 on USAMO 2026. This represents state-of-the-art performance for an open-source pipeline at a price point up to 500x less than comparable open-source pipelines.

cs.AI↗
arXiv2026

Agent Bazaar: Enabling Economic Alignment in Multi-Agent Marketplaces

The deployment of Large Language Models (LLMs) as autonomous economic agents introduces systemic risks that extend beyond individual capability failures. As agents transition to directly interacting with marketplaces, their collective behavior can amplify volatility and mask deception at scale. We introduce the Agent Bazaar, a multi-agent simulation framework for evaluating Economic Alignment, the capacity of agentic systems to preserve market stability and integrity. We identify two failure modes: (1) Algorithmic Instability in a B2C market ("The Crash"), where firms amplify price volatility until the market collapses, and (2) Sybil Deception in a C2C market ("The Lemon Market"), where a single deceptive agent controlling multiple coordinated seller identities floods the market with fraudulent listings, eroding trust and consumer welfare. We evaluate frontier and open-weight models across both scenarios and find that models largely fail to self-regulate, with failure severity varying by model rather than by size. We propose economically aligned harnesses, Stabilizing Firms and Skeptical Guardians, that improve outcomes but remain fragile under harder market conditions. To close this gap, we train agents with REINFORCE++ using an adaptive curriculum, producing a 9B model that outperforms all evaluated frontier and open-weight models. We propose the Economic Alignment Score (EAS), a 4-component scalar metric aggregating stability, integrity, welfare, and profitability, enabling direct cross-model comparison. Our results show that economic alignment is orthogonal to general capability and can be directly trained with targeted RL.

cs.LG↗
arXiv2026

Continual Harness: Online Adaptation for Self-Improving Foundation Agents

Coding harnesses such as Claude Code and OpenHands wrap foundation models with tools, memory, and planning, but no equivalent exists for embodied agents' long-horizon partial-observability decision-making. We first report our Gemini Plays Pokemon (GPP) experiments. With iterative human-in-the-loop harness refinement, GPP became the first AI system to complete Pokemon Blue, Yellow Legacy on hard mode, and Crystal without a lost battle. In the hardest stages, the agent itself began iterating on its strategy through long-context memory, surfacing emergent self-improvement signals alongside human-in-the-loop refinement. Continual Harness removes the human fully from this loop: a reset-free self-improving harness for embodied agents that formalizes and automates what we observed. Starting from only a minimal environment interface, the agent alternates between acting and refining its own prompt, sub-agents, skills, and memory, drawing on any past trajectory data. Prompt-optimization methods require episode resets; Continual Harness adapts online within a single run. On Pokemon Red and Emerald across frontier models, Continual Harness starting from scratch substantially reduces button-press cost relative to the minimalist baseline and recovers a majority of the gap to a hand-engineered expert harness, with capability-dependent gains, despite starting from the same raw interface with no curated knowledge, no hand-crafted tools, and no domain scaffolding. We then close the loop with the model itself: an online process-reward co-learning loop, in which an open-source agent's rollouts through the refining harness are relabeled by a frontier teacher and used to update the model, drives sustained in-game milestone progress on Pokemon Red without resetting the environment between training iterations.

cs.LG↗
arXiv2026

MLS-Bench: A Holistic and Rigorous Assessment of AI Systems on Building Better AI

Modern AI progress has been driven by ML methods that are generalizable across settings and scalable to larger regimes. As large language models demonstrate advanced capabilities in reasoning, coding, and engineering tasks, it is increasingly important to understand whether they can discover such methods rather than only apply existing ones. We introduce MLS-Bench, a benchmark for evaluating whether AI systems can invent generalizable and scalable ML methods. MLS-Bench contains 140 tasks across 12 domains, each requiring an agent to improve one targeted component of an ML system or algorithm and demonstrate that the improvement generalizes across controlled settings and scales. We find that current agents remain far from reliably surpassing human-designed methods, and that engineering-style tuning is easier for them than genuine method invention. We further study the effects of test-time scaling, adaptive compute allocation, and context provision on agents' discovery performance, together with case studies of their behavior. Our analyses suggest that the bottleneck is not only in proposing new methods, but also in the scientific insight needed to plan, validate, and scale claims about them. More search, compute, or context alone does not remove this bottleneck. We build and maintain a community platform for cumulative and comparable iteration, and release the data and code at https://mls-bench.com.

cs.LG↗
arXiv2026

Odysseus: Scaling VLMs to 100+ Turn Decision-Making in Games via Reinforcement Learning

Given the rapidly growing capabilities of vision-language models (VLMs), extending them to interactive decision-making tasks such as video games has emerged as a promising frontier. However, existing approaches either rely on large-scale supervised fine-tuning (SFT) on human trajectories or apply reinforcement learning (RL) only in relatively short-horizon settings (typically around 20--30 turns). In this work, we study RL-based training of VLMs for long-horizon decision-making in Super Mario Land, a visually grounded environment requiring 100+ turns of interaction with coordinated perception, reasoning, and action. We begin with a systematic investigation of key algorithmic components and propose an adapted variant of PPO with a lightweight turn-level critic, which substantially improves training stability and sample efficiency over critic-free methods such as GRPO and Reinforce++. We further show that pretrained VLMs provide strong action priors, significantly improving sample efficiency during RL training and reducing the need for manual design choices such as action engineering, compared to classical deep RL trained from scratch. Building on these insights, we introduce Odysseus, an open training framework for VLM agents, achieving substantial gains across multiple levels of the game and at least 3 times average game progresses than frontier models. Moreover, the trained models exhibit consistent improvements under both in-game and cross-game generalization settings, while maintaining general-domain capabilities. Overall, our results identify key ingredients for making RL stable and effective in long-horizon, multi-modal settings, and provide practical guidance for developing VLMs as embodied agents.

cs.LG↗
arXiv2026

Awakening the Sleeping Agent: Lean-Specific Agentic Data Reactivates General Tool Use in Goedel Prover

Heavy supervised fine-tuning on a target domain can strongly suppress capabilities that were present in the base model. We study this phenomenon in formal mathematics using Goedel-Prover-V2, an open-source model heavily trained on 1.8 million formal-math examples. After domain specialization, the model almost completely loses its ability to produce valid tool calls, even when explicitly instructed to use tools, dropping from 89.4% function-calling accuracy in the base model to nearly 0%. We ask whether this agentic collapse is permanent or instead reversible. To answer this question, we fine-tune the specialized model on a small amount of Lean-specific tool-use data. Remarkably, as few as 100 agentic traces are sufficient to restore strong tool-calling behavior. Importantly, this recovery is not the result of reward hacking or benchmark-specific optimization: the recovery data is entirely drawn from the Lean setting, where the model uses natural-language queries to search the Mathlib library for relevant theorems and lemmas, yet the regained capability transfers well beyond that domain. In particular, these same 100 Lean-specific traces improve performance on the Berkeley Function Calling Leaderboard from near zero to 83.8%, approaching the base model's 89.4% despite the mismatch in task distribution and protocol. The recovered capability is also practically useful in-domain. On ProofNet, pass@32 improves from 21.51% to 25.81%. Together, these results show that heavy domain supervised fine-tuning can suppress general tool-use ability without permanently erasing it, and that a small amount of domain-specific agentic data can awaken dormant tool-use capabilities.

cs.AI↗
arXiv2026

Goedel-Code-Prover: Hierarchical Proof Search for Open State-of-the-Art Code Verification

Large language models (LLMs) can generate plausible code but offer limited guarantees of correctness. Formally verifying that implementations satisfy specifications requires constructing machine-checkable proofs, a task that remains beyond current automation. We propose a hierarchical proof search framework for automated code verification in Lean~4 that decomposes complex verification goals into structurally simpler subgoals before attempting tactic-level proving. Central to our approach is a principled decomposition score that combines constructive justification with structural effectiveness. The same score serves as both the training reward and the inference-time ranking criterion, aligning optimization and deployment. We train Goedel Code Prover, a single unified policy for both decomposition and completion, through supervised initialization followed by hybrid reinforcement learning, where a continuous decomposition reward supports planning exploration while supervised replay stabilizes proof generation. On three Lean-based code verification benchmarks comprising 427 tasks, our 8B-parameter model achieves a 62.0% prove success rate, a 2.6 improvement over the strongest baseline under the reported inference settings. We further observe consistent inference-time scaling: success rates improve monotonically with search iterations and sampling budget, while whole-proof baselines plateau within the evaluated budgets.

cs.SE↗
arXiv2026

The PokeAgent Challenge: Competitive and Long-Context Learning at Scale

We present the PokeAgent Challenge, a large-scale benchmark for decision-making research built on Pokemon's multi-agent battle system and expansive role-playing game (RPG) environment. Partial observability, game-theoretic reasoning, and long-horizon planning remain open problems for frontier AI, yet few benchmarks stress all three simultaneously under realistic conditions. PokeAgent targets these limitations at scale through two complementary tracks: our Battling Track, which calls for strategic reasoning and generalization under partial observability in competitive Pokemon battles, and our Speedrunning Track, which requires long-horizon planning and sequential decision-making in the Pokemon RPG. Our Battling Track supplies a dataset of 20M+ battle trajectories alongside a suite of heuristic, RL, and LLM-based baselines capable of high-level competitive play. Our Speedrunning Track provides the first standardized evaluation framework for RPG speedrunning, including an open-source multi-agent orchestration system for modular, reproducible comparisons of harness-based LLM approaches. Our NeurIPS 2025 competition validates both the quality of our resources and the research community's interest in Pokemon, with over 100 teams competing across both tracks and winning solutions detailed in our paper. Participant submissions and our baselines reveal considerable gaps between generalist (LLM), specialist (RL), and elite human performance. Analysis against the BenchPress evaluation matrix shows that Pokemon battling is nearly orthogonal to standard LLM benchmarks, measuring capabilities not captured by existing suites and positioning Pokemon as an unsolved benchmark that can drive RL and LLM research forward. We transition to a living benchmark with a live leaderboard for Battling and self-contained evaluation for Speedrunning at https://pokeagentchallenge.com.

cs.LG↗
arXiv2026

Automatic Generation of High-Performance RL Environments

Translating complex reinforcement learning (RL) environments into high-performance implementations has traditionally required months of specialized engineering. We present a closed-loop methodology that produces equivalent high-performance environments for minimal compute cost. Our method uses a generic prompt template, hierarchical verification (property, interaction, and rollout tests), iterative repair, and cross-backend policy transfer to verify no sim-to-sim gap. We demonstrate three distinct workflows across five environments: (1) Direct translation (no prior performance implementation exists) from Game Boy emulator PyBoy to our EmuRust (via Rust IPC) and from Pokemon Showdown to our PokeJAX (via JAX); (2) Translation verified against existing performance implementations via throughput parity with Puffer Pong, MJX and Brax at matched GPU batch sizes; and (3) New environment creation: TCGJax, the first Pokemon TCG Pocket environment, created from a web-extracted specification. At 200M parameters, the environment overhead drops below 4% of training time. Our closed-loop methodology confirms equivalence for all five environments. TCGJax, synthesized from a private reference absent from public repositories, serves as a contamination control for agent pretraining data concerns.

cs.LG↗
arXiv2026

DataFactory: Collaborative Multi-Agent Framework for Advanced Table Question Answering

Table Question Answering (TableQA) enables natural language interaction with structured tabular data. However, existing large language model (LLM) approaches face critical limitations: context length constraints that restrict data handling capabilities, hallucination issues that compromise answer reliability, and single-agent architectures that struggle with complex reasoning scenarios involving semantic relationships and multi-hop logic. This paper introduces DataFactory, a multi-agent framework that addresses these limitations through specialized team coordination and automated knowledge transformation. The framework comprises a Data Leader employing the ReAct paradigm for reasoning orchestration, together with dedicated Database and Knowledge Graph teams, enabling the systematic decomposition of complex queries into structured and relational reasoning tasks. We formalize automated data-to-knowledge graph transformation via the mapping function T:D x S x R -> G, and implement natural language-based consultation that - unlike fixed workflow multi-agent systems - enables flexible inter-agent deliberation and adaptive planning to improve coordination robustness. We also apply context engineering strategies that integrate historical patterns and domain knowledge to reduce hallucinations and improve query accuracy. Across TabFact, WikiTableQuestions, and FeTaQA, using eight LLMs from five providers, results show consistent gains. Our approach improves accuracy by 20.2% (TabFact) and 23.9% (WikiTQ) over baselines, with significant effects (Cohen's d > 1). Team coordination also outperforms single-team variants (+5.5% TabFact, +14.4% WikiTQ, +17.1% FeTaQA ROUGE-2). The framework offers design guidelines for multi-agent collaboration and a practical platform for enterprise data analysis through integrated structured querying and graph-based knowledge representation.

cs.AI↗
arXiv2026

AlgoVeri: An Aligned Benchmark for Verified Code Generation on Classical Algorithms

Vericoding refers to the generation of formally verified code from rigorous specifications. Recent AI models show promise in vericoding, but a unified methodology for cross-paradigm evaluation is lacking. Existing benchmarks test only individual languages/tools (e.g., Dafny, Verus, and Lean) and each covers very different tasks, so the performance numbers are not directly comparable. We address this gap with AlgoVeri, a benchmark that evaluates vericoding of $77$ classical algorithms in Dafny, Verus, and Lean. By enforcing identical functional contracts, AlgoVeri reveals critical capability gaps in verification systems. While frontier models achieve tractable success in Dafny ($40.3$% for Gemini-3 Flash), where high-level abstractions and SMT automation simplify the workflow, performance collapses under the systems-level memory constraints of Verus ($24.7$%) and the explicit proof construction required by Lean (7.8%). Beyond aggregate metrics, we uncover a sharp divergence in test-time compute dynamics: Gemini-3 effectively utilizes iterative repair to boost performance (e.g., tripling pass rates in Dafny), whereas GPT-OSS saturates early. Finally, our error analysis shows that language design affects the refinement trajectory: while Dafny allows models to focus on logical correctness, Verus and Lean trap models in persistent syntactic and semantic barriers. All data and evaluation code can be found at https://github.com/haoyuzhao123/algoveri.

cs.SE↗
arXiv2025

MUSIC: MUlti-Step Instruction Contrast for Multi-Turn Reward Models

Evaluating the quality of multi-turn conversations is crucial for developing capable Large Language Models (LLMs), yet remains a significant challenge, often requiring costly human evaluation. Multi-turn reward models (RMs) offer a scalable alternative and can provide valuable signals for guiding LLM training. While recent work has advanced multi-turn \textit{training} techniques, effective automated \textit{evaluation} specifically for multi-turn interactions lags behind. We observe that standard preference datasets, typically contrasting responses based only on the final conversational turn, provide insufficient signal to capture the nuances of multi-turn interactions. Instead, we find that incorporating contrasts spanning \textit{multiple} turns is critical for building robust multi-turn RMs. Motivated by this finding, we propose \textbf{MU}lti-\textbf{S}tep \textbf{I}nstruction \textbf{C}ontrast (MUSIC), an unsupervised data augmentation strategy that synthesizes contrastive conversation pairs exhibiting differences across multiple turns. Leveraging MUSIC on the Skywork preference dataset, we train a multi-turn RM based on the Gemma-2-9B-Instruct model. Empirical results demonstrate that our MUSIC-augmented RM outperforms baseline methods, achieving higher alignment with judgments from advanced proprietary LLM judges on multi-turn conversations, crucially, without compromising performance on standard single-turn RM benchmarks.

cs.CL↗
arXiv2025

Recurrent Autoregressive Diffusion: Global Memory Meets Local Attention

Recent advancements in video generation has shifted from bidirectional models for short videos to autoregressive ones for ultra long video generation. Previous models, which usually use sliding window attention to restrict inference cost, lack effective memory compression and retrieval for long-term generation beyond the window size, leading to issues of forgetting and spatiotemporal inconsistencies. To enhance the retention of historical information with a fixed memory budget, we additionally incorporate temporal recurrent neural network (RNN) layers into the diffusion transformer (DiT) model. Specifically, we found that a LSTM layer after attention at each DiT layer achieves comparable performance to other state-of-the-art RNN blocks, such as Test-Time Training (TTT) and Mamba2. Moreover, existing diffusion-RNN approaches often suffer from performance degradation due to training-inference gap or the lack of overlap across windows. To address these limitations, we propose a novel Recurrent Autoregressive Diffusion (RAD) framework, which leverages recurrent blocks for memory update and retrieval and preserves local details by full attention on overlapping sliding windows, with no training and inference gap. Experiments on Memory Maze and Minecraft datasets demonstrate the superiority for long video generation by our framework with global memory and local attention.

cs.CV↗
arXiv2025

Understanding Outer Optimizers in Local SGD: Learning Rates, Momentum, and Acceleration

Modern machine learning often requires training with large batch size, distributed data, and massively parallel compute hardware (like mobile and other edge devices or distributed data centers). Communication becomes a major bottleneck in such settings but methods like Local Stochastic Gradient Descent (Local SGD) show great promise in reducing this additional communication overhead. Local SGD consists of three parts: a local optimization process, an aggregation mechanism, and an outer optimizer that uses the aggregated updates from the nodes to produce a new model. While there exists an extensive literature on understanding the impact of hyperparameters in the local optimization process, the choice of outer optimizer and its hyperparameters is less clear. We study the role of the outer optimizer in Local SGD, and prove new convergence guarantees for the algorithm. In particular, we show that tuning the outer learning rate allows us to (a) trade off between optimization error and stochastic gradient noise variance, and (b) make up for ill-tuning of the inner learning rate. Our theory suggests that the outer learning rate should sometimes be set to values greater than $1$. We extend our results to settings where we use momentum in the outer optimizer, and we show a similar role for the momentum-adjusted outer learning rate. We also study acceleration in the outer optimizer and show that it improves the convergence rate as a function of the number of communication rounds, improving upon the convergence rate of prior algorithms that apply acceleration locally. Finally, we also introduce a novel data-dependent analysis of Local SGD that yields further insights on outer learning rate tuning. We conduct comprehensive experiments with standard language models and various outer optimizers to validate our theory.

cs.LG↗
arXiv2025

Goedel-Prover-V2: Scaling Formal Theorem Proving with Scaffolded Data Synthesis and Self-Correction

We introduce Goedel-Prover-V2, a series of open-source language models that set a new state-of-the-art in automated theorem proving. Built on the standard expert iteration and reinforcement learning pipeline, our approach incorporates three key innovations: (1) Scaffolded data synthesis: We generate synthetic tasks of increasing difficulty to train the model to master increasingly complex theorems; (2) Verifier-guided self-correction: We enable the model to iteratively revise its proofs by leveraging feedback from the Lean compiler; (3) Model averaging: We merge model checkpoints to mitigate the decrease in model output diversity in later stages of training. Our small model, Goedel-Prover-V2-8B, reaches 84.6% pass@32 on MiniF2F and outperforms DeepSeek-Prover-V2-671B under the same metric, despite being 80X smaller. Our flagship model, Goedel-Prover-V2-32B, achieves 88.1% on MiniF2F at pass@32 in standard mode and 90.4% in self-correction mode, outperforming prior SOTA by a large margin. Additionally, our flagship model solves 86 problems on PutnamBench at pass@184, securing the first place among open-source models on the leaderboard, surpassing DeepSeek-Prover-V2-671B's record of solving 47 problems by pass@1024 with a significantly smaller model size and compute budget. At the time of its release (July-August 2025), Goedel-Prover-V2 achieves the strongest overall performance among all open-source theorem provers. It also ranks among the top-performing models--including closed-source systems with publicly reported performance--under a constrained test-time compute budget. Our models, code, and data are released at https://github.com/Goedel-LM/Goedel-Prover-V2.

cs.LG↗
arXiv2025

LLM Economist: Large Population Models and Mechanism Design in Multi-Agent Generative Simulacra

We present the LLM Economist, a novel framework that uses agent-based modeling to design and assess economic policies in strategic environments with hierarchical decision-making. At the lower level, bounded rational worker agents -- instantiated as persona-conditioned prompts sampled from U.S. Census-calibrated income and demographic statistics -- choose labor supply to maximize text-based utility functions learned in-context. At the upper level, a planner agent employs in-context reinforcement learning to propose piecewise-linear marginal tax schedules anchored to the current U.S. federal brackets. This construction endows economic simulacra with three capabilities requisite for credible fiscal experimentation: (i) optimization of heterogeneous utilities, (ii) principled generation of large, demographically realistic agent populations, and (iii) mechanism design -- the ultimate nudging problem -- expressed entirely in natural language. Experiments with populations of up to one hundred interacting agents show that the planner converges near Stackelberg equilibria that improve aggregate social welfare relative to Saez solutions, while a periodic, persona-level voting procedure furthers these gains under decentralized governance. These results demonstrate that large language model-based agents can jointly model, simulate, and govern complex economic systems, providing a tractable test bed for policy evaluation at the societal scale to help build better civilizations.

cs.MA↗
arXiv2025

Frontier LLMs Still Struggle with Simple Reasoning Tasks

While state-of-the-art large language models (LLMs) demonstrate advanced reasoning capabilities-achieving remarkable performance on challenging competitive math and coding benchmarks-they also frequently fail on tasks that are easy for humans. This work studies the performance of frontier LLMs on a broad set of such "easy" reasoning problems. By extending previous work in the literature, we create a suite of procedurally generated simple reasoning tasks, including counting, first-order logic, proof trees, and travel planning, with changeable parameters (such as document length. or the number of variables in a math problem) that can arbitrarily increase the amount of computation required to produce the answer while preserving the fundamental difficulty. While previous work showed that traditional, non-thinking models can be made to fail on such problems, we demonstrate that even state-of-the-art thinking models consistently fail on such problems and for similar reasons (e.g. statistical shortcuts, errors in intermediate steps, and difficulties in processing long contexts). To further understand the behavior of the models, we introduce the unpuzzles dataset, a different "easy" benchmark consisting of trivialized versions of well-known math and logic puzzles. Interestingly, while modern LLMs excel at solving the original puzzles, they tend to fail on the trivialized versions, exhibiting several systematic failure patterns related to memorizing the originals. We show that this happens even if the models are otherwise able to solve problems with different descriptions but requiring the same logic. Our results highlight that out-of-distribution generalization is still problematic for frontier language models and the new generation of thinking models, even for simple reasoning tasks, and making tasks easier does not necessarily imply improved performance.

cs.CL↗