The Inference Report

Continuous self-improvement requires an ever-expanding pool of self-generated, diverse, adaptive goals. For language agents, existing training environment pools (hand-curated, statically synthesized, or frozen-verifier) keep the goal distribution fixed as the learner scales. We introduce SPADE (Self-Play in Adaptive Synthetic Executable Environments), a self-play RL framework in which a single LLM plays two roles: an Environment Designer that writes complete, long-horizon training environments as executable code with an OpenAI Gym-style reset()/step() interface, and a Reasoning Agent that learns to act in them. Each is a stateful, multi-turn environment (state transitions, reward functions, and verification code), so one interface spans reasoning problems and multi-step agentic tool use. The Reasoning Agent's regret is estimated using the gap between its reward with and without privileged hints; in optimizing this regret signal the Environment Designer learns to target environments at the edge of the agent's capabilities while keeping them feasible. Through extensive experimentation, we find several components critical to success: grounding the Environment Designer on documents sampled from a large pretraining corpus, and giving it an accumulated environment memory. Scaling to 30B-parameter models, SPADE improves over the strongest fixed-environment baseline by +5.3 on average across eight held-out math, science, code, and reasoning benchmarks, and lifts the tool-use setting by +5.7 on BFCL-v4 multi-turn and +13.9 on ACEBench-Agent; on the games setting, the margin over the strongest baseline grows with model scale. By making environment design itself a learnable component, SPADE takes a concrete step toward open-ended self-improvement.

In today's era of rapid advancements in artificial intelligence, computer security and online safeguarding measures have undergone significant improvements. However, malicious websites continue to facilitate the spread of phishing schemes, fraudulent activities and unsolicited communications. Conventional methodologies in machine learning, deep learning and counterfeit website detection predominantly depend on static data analysis, which frequently proves ineffective against the evolving nature of malicious online entities. In response to these challenges, in this work, we propose a signed network-based approach for malicious URL identification, SiNMULI. We introduce an innovative framework that conceptualises the identification of harmful URLs as a signed network-based binary classification problem strongly rooted in the fundamental principles of social network analysis and social balance theory. In this approach, a signed network is constructed based on the backlinks, i.e., external hyperlinks of URLs, wherein each node symbolises a URL and the hyperlinks function as signed edges. Utilising a balance-theoretic inference mechanism, our methodology propagates edge signs and classifies unlabeled domains by employing a 51% majority rule across incoming links. Experimental results on this real-world dataset demonstrate that SiNMULI achieves 99.89% accuracy, 99.62% precision, and 99.80% F1-score, outperforming traditional ML and deep learning baseline models. Beyond high accuracy, SiNMULI offers interpretability, resilience against adversarial obfuscation, and independence from training data, making it a lightweight and scalable solution for real-world cyber defence.

We introduce Accelerating Dexterity via Pre-Training (ADEPT), a large-scale reinforcement learning (RL) framework for learning sim-to-real transferable dexterity across high degree-of-freedom (DoF) robot embodiments that can solve long-horizon tasks directly from raw visuo-tactile perception. ADEPT pretrains a dexterous policy on a generic object reposing task, then post-trains downstream policies with this pretrained behavior as a prior. ADEPT enables learning new behaviors that are otherwise difficult to discover from scratch on multi-fingered robots and avoids learning the same set of skills over again for every new downstream task. The pretrained policy zero-shots the reposing phase of downstream tasks, but naïve RL fine-tuning rapidly degrades this capability during transfer. We address this with a stable post-training recipe combining behavior-cloning distillation, critic warm-up, and conservative on-policy updates. To safely exploit the full kinematic dexterity, we introduce a joint-space Geometric Fabric that mediates between the RL policy and the robot. We distill post-trained teachers into perceptive students that zero-shot sim-to-real transfer on two embodiments: a 23 DoF Kuka-Allegro with two RGB cameras, and a 29 DoF Flexiv-Sharpa with two RGB cameras and five vision-based tactile sensors, and can solve long-horizon tasks from challenging initial states with dexterity at human-level speed.

On-policy distillation (OPD) trains a student on its own responses using dense token-level guidance from a stronger teacher. In long-context tasks, however, token-level teacher support can favor locally plausible responses that omit evidence distributed across the input or violate global task constraints. Task-specific verifiers, in contrast, evaluate task completion at the response level and may return graded rewards that reflect partial success. We diagnose this mismatch on fixed responses from two representative long-context evidence-aggregation tasks. Across longer input ranges, trajectory-level OPD scores become progressively less aligned with verifier rewards, indicating teacher-verifier disagreement. Motivated by this observation, we introduce Group-Calibrated On-Policy Distillation (GC-OPD). GC-OPD separately normalizes verifier rewards and trajectory-level OPD scores within each rollout group and uses their difference as a signed teacher-verifier disagreement residual. Relative-advantage-based credit assignment (RACA) distributes this trajectory-level residual across tokens according to their relative OPD advantages while preserving the original OPD signal. Across five long-context benchmarks, post-training with GC-OPD raises the five-benchmark averages of the official Qwen3-4B and Qwen3-8B checkpoints from 29.08 to 40.47 and from 35.12 to 44.65, respectively. Vanilla OPD reaches 39.31 and 43.56 under the same setup. Controlled ablations show that the signed residual is more effective than either an additional OPD-derived term or direct group-normalized verifier reward addition, while RACA further improves over uniform token allocation. Together, these results demonstrate that group-relative residual calibration can incorporate verifier outcomes without discarding dense token-level guidance. Code is available at https://github.com/SolereZhang/GC-OPD.

This technical report describes our winning submission to the AES AIMLA 2025 Challenge on querying sound effects by vocal imitation. We investigate two complementary fine-tuning strategies: contrastive learning with a frozen, pretrained CED encoder, and joint contrastive-triplet learning with semi-hard negatives using a MobileNetV3 encoder. This report has been updated for posterity to include details released after the challenge.

Deep models for irregularly-sampled time series answer queries at arbitrary continuous timestamps, yet report nothing about how far each answer should be trusted. We show the attention layer itself can close that gap: with the right stochastic formulation, the pass that makes each prediction also reports, in closed form and at no extra cost, how far it should be trusted. We introduce Lévy Attention, a cross-attention operator whose output is a stochastic integral against an inhomogeneous Poisson random measure: query-key compatibilities assemble an intensity over a continuous (time x channel) index space, the measure scatters atoms under it, and the output averages an interpolated value field at those atoms. In expectation it reduces to a mollified cosine-kernel attention, so it replaces a softmax layer and trains with exact gradients. What softmax discards, the Poisson construction preserves in closed form: the evidence $Λ_q$ (total compatibility mass) and the disagreement $\mathrm{tr}\,Σ_V(q)$ (value spread). An exact variance identity makes their combination $\hatσ(q)=\sqrt{\mathrm{tr}\,Σ_V(q)\,\varphi(Λ_q)}$ the root-mean-square deviation of the sampled operator, emitted by the deterministic pass with no trained head. Empirically, disagreement carries the signal, while the evidence factor swings from uninformative on dense data to strongly informative on sparse. On t-PatchGNN the operator swap costs at most 5.6% accuracy against a matched control and nothing on the sparsest dataset. The free disagreement signal improves on 20-pass MC dropout across matched five-seed suites, and $\hatσ$ scales a calibrated Gaussian whose zero-sample CRPS beats a fifty-draw sampler; a split-conformal wrapper reaches nominal coverage at every level, and one pass ranks 3,383 unseen patients by trust in 1.4 seconds.