The Inference Report

We introduce the Multilevel Euler-Maruyama (ML-EM) method compute solutions of SDEs and ODEs using a range of approximators $f^1,\dots,f^k$ to the drift $f$ with increasing accuracy and computational cost, only requiring a few evaluations of the most accurate $f^k$ and many evaluations of the less costly $f^1,\dots,f^{k-1}$. If the drift lies in the so-called Harder than Monte Carlo (HTMC) regime, i.e. it requires $ε^{-γ}$ compute to be $ε$-approximated for some $γ>2$, then ML-EM $ε$-approximates the solution of the SDE with $ε^{-γ}$ compute, improving over the traditional EM rate of $ε^{-γ-1}$. In other terms it allows us to solve the SDE at the same cost as a single evaluation of the drift. In the context of diffusion models, the different levels $f^{1},\dots,f^{k}$ are obtained by training UNets of increasing sizes, and ML-EM allows us to perform sampling with the equivalent of a single evaluation of the largest UNet. Our numerical experiments confirm our theory: we obtain up to fourfold speedups for image generation on the CelebA dataset downscaled to 64x64, where we measure a $γ\approx2.5$. Given that this is a polynomial speedup, we expect even stronger speedups in practical applications which involve orders of magnitude larger networks.

We introduce DreamerAD, the first latent world model framework that enables efficient reinforcement learning for autonomous driving by compressing diffusion sampling from 100 steps to 1 - achieving 80x speedup while maintaining visual interpretability. Training RL policies on real-world driving data incurs prohibitive costs and safety risks. While existing pixel-level diffusion world models enable safe imagination-based training, they suffer from multi-step diffusion inference latency (2s/frame) that prevents high-frequency RL interaction. Our approach leverages denoised latent features from video generation models through three key mechanisms: (1) shortcut forcing that reduces sampling complexity via recursive multi-resolution step compression, (2) an autoregressive dense reward model operating directly on latent representations for fine-grained credit assignment, and (3) Gaussian vocabulary sampling for GRPO that constrains exploration to physically plausible trajectories. DreamerAD achieves 87.7 EPDMS on NavSim v2, establishing state-of-the-art performance and demonstrating that latent-space RL is effective for autonomous driving.

As LLMs are increasingly used as judges in code applications, they should be evaluated in realistic interactive settings that capture partial context and ambiguous intent. We present TRACE (Tool for Rubric Analysis in Code Evaluation), a framework that evaluates LLM judges' ability to predict human preferences and automatically extracts rubric items to reveal systematic biases in how humans and models weigh each item. Across three modalities -- chat-based programming, IDE autocompletion, and instructed code editing -- we use TRACE to measure how well LLM judges align with developer preferences. Among 13 different models, the best judges underperform human annotators by 12-23%. TRACE identifies 35 significant sources of misalignment between humans and judges across interaction modalities, the majority of which correspond to existing software engineering code quality criteria. For example, in chat-based coding, judges are biased towards longer code explanations while humans prefer shorter ones. We find significant misalignment on the majority of existing code quality dimensions, showing alignment gaps between LLM judges and human preference in realistic coding applications.

Agentic artificial intelligence (AI) in organizations is a sequential decision problem constrained by reliability and oversight cost. When deterministic workflows are replaced by stochastic policies over actions and tool calls, the key question is not whether a next step appears plausible, but whether the resulting trajectory remains statistically supported, locally unambiguous, and economically governable. We develop a measure-theoretic Markov framework for this setting. The core quantities are state blind-spot mass B_n(tau), state-action blind mass B^SA_{pi,n}(tau), an entropy-based human-in-the-loop escalation gate, and an expected oversight-cost identity over the workflow visitation measure. We instantiate the framework on the Business Process Intelligence Challenge 2019 purchase-to-pay log (251,734 cases, 1,595,923 events, 42 distinct workflow actions) and construct a log-driven simulated agent from a chronological 80/20 split of the same process. The main empirical finding is that a large workflow can appear well supported at the state level while retaining substantial blind mass over next-step decisions: refining the operational state to include case context, economic magnitude, and actor class expands the state space from 42 to 668 and raises state-action blind mass from 0.0165 at tau=50 to 0.1253 at tau=1000. On the held-out split, m(s) = max_a pi-hat(a|s) tracks realized autonomous step accuracy within 3.4 percentage points on average. The same quantities that delimit statistically credible autonomy also determine expected oversight burden. The framework is demonstrated on a large-scale enterprise procurement workflow and is designed for direct application to engineering processes for which operational event logs are available.

Retrieval-augmented generation (RAG) systems are increasingly used to analyze complex policy documents, but achieving sufficient reliability for expert usage remains challenging in domains characterized by dense legal language and evolving, overlapping regulatory frameworks. We study the application of RAG to AI governance and policy analysis using the AI Governance and Regulatory Archive (AGORA) corpus, a curated collection of 947 AI policy documents. Our system combines a ColBERT-based retriever fine-tuned with contrastive learning and a generator aligned to human preferences using Direct Preference Optimization (DPO). We construct synthetic queries and collect pairwise preferences to adapt the system to the policy domain. Through experiments evaluating retrieval quality, answer relevance, and faithfulness, we find that domain-specific fine-tuning improves retrieval metrics but does not consistently improve end-to-end question answering performance. In some cases, stronger retrieval counterintuitively leads to more confident hallucinations when relevant documents are absent from the corpus. These results highlight a key concern for those building policy-focused RAG systems: improvements to individual components do not necessarily translate to more reliable answers. Our findings provide practical insights for designing grounded question-answering systems over dynamic regulatory corpora.

Hallucination remains a critical bottleneck for large language models (LLMs), undermining their reliability in real-world applications, especially in Retrieval-Augmented Generation (RAG) systems. While existing hallucination detection methods employ LLM-as-a-judge to verify LLM outputs against retrieved evidence, they suffer from inherent confirmation bias, where the verifier inadvertently reproduces the errors of the original generation. To address this, we introduce Multi-Agent Reinforced Self-Check for Hallucination (MARCH), a framework that enforces rigorous factual alignment by leveraging deliberate information asymmetry. MARCH orchestrates a collaborative pipeline of three specialized agents: a Solver, a Proposer, and a Checker. The Solver generates an initial RAG response, which the Proposer decomposes into claim-level verifiable atomic propositions. Crucially, the Checker validates these propositions against retrieved evidence in isolation, deprived of the Solver's original output. This well-crafted information asymmetry scheme breaks the cycle of self-confirmation bias. By training this pipeline with multi-agent reinforcement learning (MARL), we enable the agents to co-evolve and optimize factual adherence. Extensive experiments across hallucination benchmarks demonstrate that MARCH substantially reduces hallucination rates. Notably, an 8B-parameter LLM equipped with MARCH achieves performance competitive with powerful closed-source models. MARCH paves a scalable path for factual self-improvement of LLMs through co-evolution. The code is at https://github.com/Qwen-Applications/MARCH.