L3 Lab at COLM 2026

L3 member at COLM: Weihua Du

Reinforcement learning (RL) has become a central post-training paradigm for large language models (LLMs), but its performance is highly sensitive to the quality of training problems. This sensitivity stems from the non-stationarity of RL: rollouts are generated by an evolving policy, and learning is shaped by exploration and reward feedback, unlike supervised fine-tuning (SFT) with fixed trajectories. As a result, prior work often relies on manual curation or simple heuristic filters (e.g., accuracy), which can admit incorrect or low-utility problems. We propose GradAlign, a gradient-aligned data selection method for LLM reinforcement learning that uses a small, trusted validation set to prioritize training problems whose policy gradients align with validation gradients, yielding an adaptive curriculum. We evaluate GradAlign across three challenging data regimes: unreliable reward signals, distribution imbalance, and low-utility training corpus, showing that GradAlign consistently outperforms existing baselines, underscoring the importance of directional gradient signals in navigating non-stationary policy optimization and yielding more stable training and improved final performance. We release our implementation at https://github.com/StigLidu/GradAlign

Abstract source: arXiv · CC BY 4.0

L3 member at COLM: Weihua Du

As NLP evaluation shifts from static benchmarks to multi-turn interactive settings, LLM-based simulators have become widely used as user proxies, serving two roles: generating user turns and providing evaluation signals. Yet, these simulations are frequently assumed to be faithful to real human behaviors, often without rigorous verification. We formalize the Sim2Real gap in user simulation and present the first study running the full \(\tau\)-bench protocol with real humans (451 participants, 165 tasks), benchmarking 31 LLM simulators across proprietary, open-source, and specialized families using the User-Sim Index (USI), a metric we introduce to quantify how well LLM simulators resemble real user interactive behaviors and feedback. Behaviorally, LLM simulators are excessively cooperative, stylistically uniform, and lack realistic frustration or ambiguity, creating an "easy mode" that inflates agent success rates above the human baseline. In evaluations, real humans provide nuanced judgments across eight quality dimensions while simulated users produce uniformly more positive feedback; rule-based rewards are failing to capture rich feedback signals generated by human users. Overall, higher general model capability does not necessarily yield more faithful user simulation. These findings highlight the importance of human validation when using LLM-based user simulators in the agent development cycle and motivate improved models for user simulation.

Abstract source: arXiv · CC BY 4.0

L3 member at COLM: Weihua Du

Despite rapid progress, current AI agents are primarily optimized for isolated task completion. We argue for a paradigm shift toward training agents as collaborators that communicate and adapt to people. To facilitate this shift in real-world complex applications, we first formalize three dimensions of collaborative AI agents: Productivity, Proactivity, and Personalization (PPP). We introduce UserVille, an interactive environment with configurable LLM-based user simulators and user-centric feedback to evaluate these dimensions, and propose a multi-objective reinforcement learning framework that optimizes them using rewards from task outcomes, question effort, and preference adherence. On two real-world agentic tasks (SWE-Bench and BrowseComp-Plus), PPP-trained agents outperform strong LLM baselines (including GPT-5) by an average of 16.7 points, ask more targeted questions, and generalize to unseen preferences and tasks. A follow-up user study further highlights the importance of user-centric feedback for training collaborative agents that are both more effective and easier to supervise.

Abstract source: arXiv

L3 member at COLM: Tianran Sun

Program verification relies on loop invariants, yet automatically discovering strong invariants remains a long-standing challenge. We investigate whether large language models (LLMs) can accelerate program verification by generating useful loop invariants. We introduce Quokka, a framework for LLM-based invariant synthesis with soundness guarantees and state-of-the-art performance. Unlike prior work that treats LLM outputs as noisy symbolic material requiring substantial post-processing, Quokka adopts a simpler algorithm design that directly validates whether each LLM-generated invariant helps prove the target assertion. We construct a benchmark of 866 evaluation instances and 3589 training instances derived from SV-COMP, and evaluate 9 LLMs spanning multiple model families. We demonstrate that supervised fine-tuning and Best-of-N sampling yield measurable improvements, and we show that Quokka consistently outperforms prior LLM-based verifiers. Our code and data are publicly available at https://github.com/Anjiang-Wei/Quokka

Abstract source: arXiv · CC BY 4.0

L3 member at COLM: Pranjal Aggarwal

Computer-use agents hold the promise of assisting in a wide range of digital economic activities. However, current research has largely focused on short-horizon tasks over a limited set of software with limited economic value, such as basic e-commerce and OS-configuration tasks. A key reason is that creating environments for complex software requires significant time and human effort, and therefore does not scale. To address this, we introduce Gym-Anything, a framework for converting any software into an interactive computer-use environment. We frame environment creation itself as a multi-agent task: a coding agent writes setup scripts, downloads real-world data, and configures the software, while producing evidence of correct setup. An independent audit agent then verifies evidence for the environment setup against a quality checklist. Using a taxonomy of economically valuable occupations grounded in U.S. GDP data, we apply this pipeline to 200 software applications with broad occupational coverage. The result is CUA-World, a collection of over 10K long-horizon tasks spanning domains from medical science and astronomy to engineering and enterprise systems, each configured with realistic data along with train and test splits. CUA-World also includes CUA-World-Long, a challenging long-horizon benchmark with tasks often requiring over 500 steps, far exceeding existing benchmarks. Distilling successful trajectories from the training split into a 2B vision-language model outperforms models 2\(\times\) its size. We also apply the same auditing principle at test time: a separate VLM reviews completed trajectories and provides feedback on what remains, improving Gemini-3-Flash on CUA-World-Long from 11.5% to 14.0%. We release all code, infrastructure, and benchmark data to facilitate future research in realistic computer-use agents.

Abstract source: arXiv · CC BY 4.0

L3 member at COLM: Weihua Du

Recent large language model (LLM) agents have shown promise in using execution feedback for test-time adaptation. However, robust self-improvement remains far from solved: most approaches still treat each problem instance independently, without accumulating reusable knowledge. This limitation is particularly pronounced in domain-specific languages such as Triton, which are underrepresented in LLM pretraining data. Their strict constraints and non-linear optimization landscape further make naive generation and local refinement unreliable. We propose AdaExplore, an agent framework that enables self-improvement via accumulated execution feedback for performance-critical kernel code generation through two complementary stages: failure-driven adaptation and diversity-preserving search, jointly improving correctness and optimization performance without additional fine-tuning or external knowledge. In the adaptation stage, the agent synthesizes tasks and converts recurring failures into a reusable memory of validity rules, helping subsequent generations remain within the feasible set. In the search stage, the agent organizes candidate kernels as a tree and alternates between small local refinements and larger structural regeneration, allowing it to explore the optimization landscape beyond local optima. Experiments on kernel runtime optimization benchmarks validate these gains: AdaExplore achieves 3.11x and 1.62x speedups on KernelBench Level-2 and Level-3, respectively, within 100 steps, and continues to improve with additional computation.

Abstract source: arXiv · CC BY 4.0

L3 member at COLM: Weihua Du

Autoregressive language models remain the dominant language modeling paradigm, yet their sequential decoding limits efficiency. While recent efforts have explored discrete diffusion language models for parallel decoding, these approaches suffer from severe degradation in text quality when decoding multiple tokens simultaneously due to token-wise independence assumptions. Continuous-space diffusion and flow models offer a compelling alternative, but existing methods are often restricted to unconditional full-sequence generation, and some rely heavily on learned embedding encoders to map discrete tokens into continuous latent states, requiring additional pretrained encoders. In this work, we propose SimpLM, a continuous flow language modeling framework that operates directly on the probability simplex in the vocabulary space. By applying Gaussian interpolation to the probability states, we derive a closed-form posterior logit correction that yields an analytic skip connection for logit transfer, bypassing the hidden-state information bottleneck in Transformer backbones. Beyond this, SimpLM natively supports conditional and block-based semi-autoregressive generation via a vectorized time formulation that assigns separate time variables to block positions, providing a principled foundation for continuous-time conditional generation. Furthermore, SimpLM can adapt a pretrained autoregressive model into a continuous flow model, reducing the cost of training non-autoregressive models from scratch. Empirically, with only 5B training tokens, SimpLM achieves a generative perplexity below 30 at a block size of 32 and a generation length of 128.

Abstract source: OpenReview