New approach enables physically consistent AI models of chaotic systems
Enforcing physics-based constraints prevents AI models from generating unreliable predictions for complex nonlinear systems in weather, climate, and turbulent flows.
Search headlines, summaries and articles to connect the dots.
Enforcing physics-based constraints prevents AI models from generating unreliable predictions for complex nonlinear systems in weather, climate, and turbulent flows.
arXiv:2611.25171v1 Announce Type: new Abstract: Test-time compute (TTC) improves language models by spending additional inference compute to generate, verify, and select among candidate outputs, but this recipe has not transferred directly to time series forecasting. A language model can verify a proof or vote across samples, but a forecaster must be causal; it cannot use future information to score a candidate. We propose Backtest-Gated Augmentation Averaging (BGAA), a TTC algorithm for time series forecasting that builds a causal verifier from the series's own history, using rolling-origin backtests to score structured views of the context, discard those that underperform the untransformed context, and average the survivors' predictive distributions. BGAA improves every tested TSFM-benchmark combination, reducing MASE by 0.5-1.0% on GIFT-Eval, 1.3-5.0% on fev-bench, and 1.7-5.9% on BOOM, where a 313M model with TTC outperforms the 1B and 2.5B models in its family. We further show that useful views have structured, model-dependent effects consistent with a prior-transport interpretation. On the 53 cross-fittable GIFT-Eval configs, the deployed policy gains only 0.4-1.3% CRPS, versus 1.4-4.1% when choosing the best view for each dataset configuration and 17-19% when choosing the best view separately for each test window. This gap suggests that a major remaining challenge is window-level view selection.
arXiv:2611.23484v1 Announce Type: new Abstract: Training 3D vision-language models (VLMs) on large medical imaging datasets collected across institutions is constrained by data quality. Vast, uncurated pools of image-report pairs are characterized by redundancy, class imbalance, and cross-institutional heterogeneity. We address this challenge of data curation for improving 3D VLM pretraining by proposing an offline, training-free, reference-guided data curation recipe. Our approach leverages a small, high-quality reference set to independently extract quality signals for each modality of a large, uncurated pool and merges them through a rank-aggregation step into a single ranking that guides sample selection. We validate this method on datasets comprising 35k 3D abdominal CT volume-report pairs from two institutions and 58k 2D chest X-ray image-report pairs. On 3D CT, using our reference-guided data curation strategy, our curated model matches the zero-shot disease classification performance of a state-of-the-art baseline (Merlin) while using $3.3\times$ less training data and a reference set of 5k samples. On 2D chest X-rays, our method yields a curated model that matches or exceeds the zero-shot disease classification performance of a model trained on the full pool while using up to $10\times$ less training data and a reference set of only 74 samples. Ablations show that the selection is stable under resampling of the reference, that smaller reference sets still outperform random sampling, and that label rebalancing does not improve performance. To support reproducibility, we release our code.
OpenAI explains how it manages frontier AI risk from training through deployment with model alignment, sandboxing, blocking safeguards, and monitoring.
arXiv:2611.17495v1 Announce Type: new Abstract: Chain-of-thought (CoT) is currently a valuable tool for overseeing AI models, but some architectural shifts could significantly reduce CoT monitorability. We operationalize a measure that serves as a proxy for the amount of unverbalized serial cognition a model can perform, as a specific instantiation of the notion of opaque serial depth: roughly, the length of the longest path through a model's computation that does not pass through an interpretable bottleneck. Our main contribution is a particular standard for what counts as such a bottleneck. We define natural-language-rooted (NL-rooted) nodes as nodes that were initialized from a pre-training prior to produce human-understandable outputs, whose output space has not been expanded beyond that prior, and whose outputs the model has not been modified to interpret as a different type of data. By construction, the resulting measure, NLS depth, can be adjudicated before training from an architecture, training recipe, and deployment configuration alone, and is insensitive to ordinary post-training such as prompting, distillation, and RL. We find that, across open-weight models, NLS depth correlates strongly with layer count, with some exceptions, and has grown slowly over recent years. It would likely grow slowly under continued scaling of standard transformers, while latent-reasoning architectures such as full-bandwidth transformers could increase it by a large amount. We work through how NLS depth applies to a range of existing and proposed architectures, discuss alternative operationalizations, and give a worked example bounding the depth of Gemma3-1B. We argue that AI developers should track and report the NLS depth of their models as a complement to direct measurements of CoT monitorability.
What's new in research at Microsoft? Host Anson Ho explores the latest, including AI-powered universal translation, a breakthrough in a 31-year-old mathematics challenge, and advances making AI coding agents faster and more efficient. The post Microsoft Research Focus – Edition #98 appeared first on Microsoft Research.
OpenAI begins embedded assessments with Apollo Research, METR, and Transluce, and Goodfire to independently evaluate model alignment, safeguards, and monitoring systems.
arXiv:2611.14748v1 Announce Type: new Abstract: DNN surrogates are increasingly used to accelerate scientific workflows that must process high-rate sensor streams in near real time, exemplified by Ptychographic imaging at next-generation X-ray light sources. Such surrogates deliver 100-1000x higher throughput than iterative solvers, enabling faster steering of experiments. To preserve surrogate accuracy during an experiment, as the input distribution evolves, state-of-the-art deployments freeze the pretrained surrogate or run a separate CUDA context that fine-tunes concurrently with serving. However, multi-context concurrency causes significant contention on the GPU, inflating both the mean and the variance of inference latency, triggering deadline misses in workflows where sensor data has to be processed within a hard per-unit deadline. To address this challenge, we propose UniPipe, a single-context system that unifies inference serving and fine-tuning of a DNN surrogate on a single edge GPU. UniPipe is driven by three key insights: both serving and fine-tuning require a forward pass and can be fused into one; the per-sample latency of modern GPUs is non-linear in batch size; and the workload is repetitive across intervals, enabling offline scheduling. Specifically, UniPipe fuses training and inference data in a single forward pass with a selective backward pass on the training subset, and uses a polynomial-time scheduler that exploits the latency curve to compute the batch composition. We evaluate UniPipe on a real-world Ptychographic reconstruction workflow with sensor rates up to 5000 patterns/sec.; UniPipe lowers reconstruction MSE by up to 30% and inference MSE by up to 50% over a pretrained baseline, and reduces the deadline-miss rate by 13%-22% over a multi-context baseline. UniPipe's unified deployment reduces the latency spread from 106% to 57% of the mean, and increases total throughput compared to multi-context baseline.
Health & Bioscience
arXiv:2611.13003v1 Announce Type: new Abstract: Reinforcement learning agents are often deployed with policies that cannot be retrained or modified. In digital environments whose states can be copied and resumed, an operator has another option: sample several continuations from the current state, evaluate them, and continue from the endpoint of the selected branch. We call this procedure branch-and-commit. We study when it improves return, what produces the improvement, and where its computational budget should be spent. Across four environments, two frozen policies, five continuation verifiers, and a model-free selection rule, we find that the benefit of branching depends on dispersion: the difference between the best candidate return and the mean candidate return. Branching provides no benefit in low-dispersion settings, while increasing return by 12.8% in our main Craftax-Medium setting. Most of this gain comes not from selecting better actions, but from selecting among different stochastic realizations of the same action. Several competent verifiers achieve similar episode returns, suggesting that verifier quality is not the main constraint at our operating budget. Instead, computation is better allocated to states where useful dispersion is likely. On held-out seeds, gating branching using predicted dispersion retains 69% of the improvement from uniform branching while using 48% as many branching calls. These results characterize when branch-and-commit is useful and clarify what its gains represent.
arXiv:2611.12937v1 Announce Type: new Abstract: Do large language models represent psychological constructs internally, or do they only generate outputs that resemble psychometric structure? We address this question using a cross-persona paradigm across 14 model variants from two families (Llama 3 and Qwen 2.5, 0.5B-14B, base and instruct). Given responses on one psychological scale from real individuals (N = 272 Taiwanese university students), models predict responses on six other scales. At the behavioral level, larger variants reproduce and amplify human cross-scale correlations; amplification persists after correcting the seven-scale human correlations for measurement attenuation. We then examine whether this structure is reflected internally. Contrastive direction analysis finds geometry aligned with the human matrix in larger instruct variants, but little alignment in base variants under this assay. Across models, geometry strength is associated with behavioral amplification (r = 0.68, p = 0.008; partial r = 0.69 controlling for log size). To relate internal structure to output behavior, a matched activation-probing paradigm shows that probe slopes vary less than behavioral slopes (1.38-1.77 vs 0.52-1.53). A synthetic control shows that ridge probing can inflate these slopes; after subtracting a baseline based on a reliability proxy from another analysis, adjusted slopes remain associated with behavior (r = 0.88). We term the measured probe-behavior gap readout attenuation. These results establish decodability and geometric alignment under the tested assays, without showing that the measured directions mediate the cross-persona answers.
arXiv:2611.09964v1 Announce Type: new Abstract: Pretrained vision language action (VLA) models show strong zero-shot performance, but are rarely deployed directly and are often first post-trained on demonstrations from the target setting. Yet VLAs are typically compared by zero-shot or full-data fine-tuned success, revealing little about the data and compute required for adaptation. We study adaptation efficiency for six VLAs ($\pi_{0}$, $\pi_{0.5}$, and GR00T N1-N1.7) on four LIBERO suites, varying both demonstrations and training steps. This reveals differences hidden by full-data fine-tuning: models within two points on the full LIBERO-Spatial dataset differ by up to $3\times$ in success with 50 demonstrations. $\pi_{0.5}$ needs $2$--$5\times$ fewer demonstrations than $\pi_{0}$ and is the most sample-efficient model at every budget, with the largest gap on the long-horizon suite. Within GR00T, N1.5 adapts more efficiently than N1.6 and N1.7 on three of four suites, despite their stronger zero-shot performance. We study one pretraining factor behind these differences by freezing the VLM during finetuning: this severely degrades the $\pi$ models, whose VLM is jointly trained with the action expert during pretraining, but barely affects GR00T models. Finally, validation loss on held-out demonstrations tracks closed-loop success within a suite, but understates how much more sample-efficient $\pi_{0.5}$ is than $\pi_{0}$.
arXiv:2611.09993v1 Announce Type: new Abstract: Chest radiograph report generation is hampered by retrieved reports that resemble the current image but are clinically inconsistent with it, and by diagnostic-label imbalance. We present RG-MVR, a retrieval-grounded framework comprising Retrieval-Augmented Evidence Grounding, which combines dense memory summaries with visually gated sparse retrieved evidence; Adaptive Multi-view Evidential Reasoning, which weights image, memory, and retrieved evidence per case; and a Clinical Prompt Learner, which models finding mention separately from positive, negative, or uncertain status. On MIMIC-CXR, RG-MVR achieved a clinical-label F1 of 0.483 and a BLEU-4 of 0.118 (IU X-ray: 0.202 and 0.110). In an exploratory external cohort of 22 examinations, clinical-label agreement declined markedly (F1, 0.071) and unsupported statements remained common, although physicians found fewer unsupported temporal comparisons (9 vs. 20 cases) and device-related statements (7 vs. 15) than with the baseline. Evidence-aware retrieval and structured diagnostic prompting may reduce specific reporting errors, but broader external validation is required.
New research from OpenAI and Upwork finds shifting demand for freelance work, new skill combinations, and larger budgets and more hours for AI-related projects.
arXiv:2611.08864v1 Announce Type: new Abstract: We present the Adaptive Scheduling Pipeline (ASP), an offline rollout scheduler for multi-instance asynchronous reinforcement learning (RL) of large language models (LLMs). ASP jointly reduces three sources of rollout underutilization - padding bubbles from intra-minibatch length heterogeneity, skewness bubbles from long-tail outputs, and sorting bubbles from fluctuations in training-queue occupancy induced by length-sorted dispatch - through a single offline preprocessing pass. This enables efficient multi-instance rollout that overcomes the early scaling plateau of single-instance inference while smoothing the arrival of generated samples at the trainer. We integrate the scheduler into an asynchronous architecture that decouples rollout from training while keeping policy lag observable and bounded. Complete ASP improves end-to-end throughput by up to 3.22 times and reduces iteration time by up to 76.8% relative to MindSpeed-RL, while reducing iteration time by up to 62.5% relative to verl. Within ASP, scheduling reduces mean iteration time by 10.9%. ASP also scales more robustly with additional generation resources than these baselines and shows no visible convergence degradation in the measured bounded-lag regime. ASP has since been used in an internal telecom-domain RL post-training workflow, reducing the reported training period from more than three months to about two months and producing a domain-enhanced model for knowledge answering and tool calling.
arXiv:2611.08640v1 Announce Type: new Abstract: Scaling test-time compute has proven highly effective at improving LLM reasoning. A popular method is best-of-n sampling, where a model generates many candidate solutions and one is selected. This motivates the question: how can we select the best candidate? When humans make choices, we find it more natural to compare two options than judge each in isolation. Additionally, when the options become complex, we like to judge the options one aspect at a time, such as comparing two job candidates on experience, communication, etc. In the test-time setting, candidate solutions should be judged in the same way. Here we introduce Aspect-wise Pairwise Tournaments (APT), which verifies by comparison and compares by aspect. Each matchup of a single elimination tournament is decided by a vote over multiple off-the-shelf aspect verifiers, each expressing a pairwise preference along its own aspect. Across five domains and three generator LLMs, APT improves over binary multi-aspect approval, self-consistency and reward model verification, with the largest gains on open-ended tasks. We find that both components matter; neither binary approval over the same aspects nor a generic pairwise judge at a matched budget matches APT. Because aspects are specified in natural language, APT also lets users steer selection toward properties they care about, better aligning with user preferences without needing to regenerate solutions. No component is trained, so APT transfers to new domains by simply rewriting prompts.
Learn how OpenAI’s Control Room tracks model workloads across our research infrastructure and can rapidly stop them when safety or security concerns arise.
arXiv:2611.08837v1 Announce Type: new Abstract: Equivalent encodings of an optimization problem can produce different learned predictions. Interpreting these differences requires accounting for numerical error and multiple optima. We construct metamorphic tests for optimization solver surrogates with explicit eligibility conditions, output alignment, and numerical thresholds. Reference-solver agreement and independent constraint checks determine which sampled pairs enter the audit. We study alternating-current optimal power flow (AC-OPF), the traveling salesman problem (TSP), the capacitated vehicle routing problem (CVRP), and maximum independent set (MIS). Across five grids, GridSFM-Open violates each of 13 relations on at least one grid, and 12 by more than twice the threshold. Randomly initialized copies of the grid models fail nearly the same relations, so the failures do not require training. Source patches and input canonicalization remove selected discrepancies, but no intervention establishes preserved accuracy and feasibility within the declared margin on every grid. In routing, Policy Optimization with Multiple Optima (POMO) has a single-pass rotation/reflection violation rate that rises from 12.7% to 89.5% between 20 and 100 cities. Graph-Based Diffusion Solvers for Combinatorial Optimization (DIFUSCO) violates the MIS disjoint-union relation on 35.6% of draws. Post hoc normalization within connected components removes these violations on the tested graphs. These findings support reporting metamorphic consistency alongside optimality, feasibility, and accuracy, and validating repair quality separately from consistency.
arXiv:2611.07069v1 Announce Type: new Abstract: Reinforcement learning (RL) in Partially Observable Markov Decision Processes (POMDPs) is computationally challenging due to the curse of long-range dependence on history. Fortunately, recent studies have shown that a range of POMDPs that satisfy the filter stability assumption can be well approximated by a superstate Markov Decision Process (MDP) based on a truncated history in a finite-memory window, with a value gap that decays with the window length. However, existing finite-time convergence analyses on such superstate MDP surrogates are either limited to over-simplified linear function approximation or rely on an i.i.d. sampling oracle that is violated by POMDP trajectories, both of which severely limit the applicability of these theoretical results in practice. This motivates us to bridge the gap by investigating RL in POMDPs with nonlinear neural value function approximation. Our main contributions are threefold: i) We first establish a new finite-state MDP point-wise approximation error bound, which lays the foundation for our RL algorithm design and analysis in POMDPs; ii) By applying Temporal-Difference (TD) learning to POMDP-generated superstate transitions and targeting the Bellman fixed point of the superstate MDP, we prove a finite-time critic error bound with nonlinear neural value function approximation; and iii) By combining the neural critic function approximation error bound with the policy mirror descent framework, we obtain a finite-time sublinear regret against the optimal POMDP value. To the best of our knowledge, such a finite-time sublinear regret based on neural function approximation is the first in the literature for finite-memory POMDPs. Our numerical experiments also verify our theoretical results.
Maximum inner product search (MIPS) sits at the core of recommendation systems, search, and retrieval-augmented generation (RAG). Given a query vector, MIPS systems search within a database of millions of candidate key vectors for the one that best matches the query. The original data point associated with that best-matching key is then used to handle the query: in RAG, the document that is best suited to answer the question is typically prefixed to the query. As the number of keys grows to millions and vector dimensions reach thousands, the computational cost of running an exhaustive…