Searcharxiv⌕ Search

arXiv subjects

Judah Goldfeder

Publications and source records attributed to Judah Goldfeder.

At least 19 recordsLinked to original sources

Long-Context Generation Is a Sampling Problem (When You Turn Up the Heat)

Open-weight language models are widely observed to get worse the longer they generate. Often, attempts to fix this involve more training. At high temperature, The sampler, i.e. the rule that picks the next token out of the model's probabilities, decides whether long and agentic generation survives and/or thrives. We test 29 open-weight models from 1B to 122B parameters on ten established benchmarks plus needle retrieval out to one million tokens, covering SWE-bench Verified coding agents, long chain-of-thought reasoning, long-form writing, long structured output and long-context reading. Samplers that keep their pool of candidate tokens small and steady as temperature rises don't lose accuracy. Samplers that let in low-probability tokens fall apart. At temperature 2, coding agents running top-p, top-k or Top-H resolve 0 of 100 GitHub issues, while the same agents running top-nsigma resolve 46 to 62 and win every issue the two samplers disagree on, on three Qwen3.5 models. Over all benchmarks top-nsigma keeps a median 100% of its temperature-1 score where top-p keeps 26%. Counting how many tokens each sampler lets through explains why. In one Llama-3.1-8B story, going from temperature 1 to 2 grows top-p's pool from 4 tokens to 51,845 while top-nsigma stays at 3. Across 133 combinations of sampler setting and temperature, the typical pool size predicts judged quality (Spearman correlation of -0.77), and when we pick the pool at temperature 1 and only apply temperature afterwards, Top-H, min-p and p-less stop collapsing. At the default temperature of 1 most samplers perform about the same, We release all code and data to the public.

cs.AI↗

Do Multi-Agents Dream of Electric Screens? Achieving 100% on AndroidWorld Through Task Decomposition

We present Minitap's mobile-use, a multi-agent system for Android UI control, and evaluate it on the 116 tasks of AndroidWorld, on Android emulators in Minitap's cloud device infrastructure, real Pixel 7 phones from our device farm and the stock emulator. On fresh, held-out task instances (three pre-registered seeds never used during development), the final system reaches a pass@3 of 100.0% (95% CI 96.8-100.0); pass^3 is 94.0% (95% CI 89.7-97.4) and the mean success rate 98.0% (95% CI 96.0-99.7). Minitap decomposes each task across five LLM agents (planning, subgoal tracking, state capture, decision, execution) and deterministic nodes, with mechanisms that read typed text back from the device, verify the goal on device evidence and detect repeated failures. A study of the initial system measures what each part contributes: removing only the decomposition changes success by -26.1 [-35.6, -16.7] points, the mechanisms together add +11.7 [+2.2, +21.1] points over a single agent, and the decision agent is the one role that needs a frontier model. We then improve reliability with general changes, a few of them triggered by the goal's wording. On MobileMiniWoB++, a second benchmark never used during development, the final system improves on the initial one by 7.1 points (McNemar p = 0.035; 95% CI -0.5 to +14.7). Minitap is open source: https://github.com/minitap-ai/mobile-use

cs.AI↗

Buildrix: An Open Agentic AI Benchmark for Building Engineering

Agentic AI has demonstrated remarkable progress across science and engineering, including mathematics, physics, chemistry, and biology. However, how well AI agents can perform real-world building engineering tasks remains largely unexplored. To address this gap, we introduce Buildrix, an open benchmark for evaluating AI agents on real building engineering tasks across eight domains. We investigated three questions: how AI agents compare with human engineers, whether expert knowledge improves agent performance, and how engineers can collaborate effectively with AI agents. The benchmark shows that AI agents achieved an average score of 78 out of 100, compared with 84 for human engineers, while completing tasks in an average of 9 minutes at a cost of $0.58 per run, compared with 8-160 reported working hours for human engineers, demonstrating promising potential. Interestingly, providing expert knowledge did not consistently improve agent performance and sometimes led to worse results. In contrast, structured task instructions substantially improved performance, increasing the average score from 27 to 75 and the success rate from 5% to 76%. These findings highlight both the potential and limitations of AI agents in building engineering, as well as the importance of structured task specifications and clear deliverable requirements for effective human-agent collaboration. More broadly, Buildrix provides an open evaluation framework for tracking AI progress, identifying capability gaps, and guiding the development of more capable and reliable AI agents to accelerate the adoption of AI in building engineering practice.

eess.SY↗

The Smart Buildings Control Suite: A Diverse Open Source Benchmark to Evaluate and Scale HVAC Control Policies for Sustainability

Commercial buildings account for 17\% of U.S. carbon emissions, with roughly half of that from Heating, Ventilation, and Air Conditioning (HVAC). HVAC devices form a complex thermodynamic system, and while model predictive control and reinforcement learning have been used to optimize control policies, scaling to thousands of buildings remains a significant unsolved challenge. Most current approaches are over-optimized for specific buildings and rely on proprietary data or hard-to-configure simulators. We present the Smart Buildings Control Suite, the first open source interactive HVAC control benchmark with a focus on solutions that generalize across building. It has 3 components: real-world data from 11 buildings over 6 years, a lightweight data-driven simulator, and a modular Physically Informed Neural Network (PINN) building model. The buildings span multiple climates, management systems, and sizes, and both the simulator and PINN easily transfer to new buildings, ensuring solutions using this benchmark are robust to these factors and only reliant on fully scalable building models. This represents a major step towards scaling HVAC optimization from the lab to buildings everywhere. To facilitate use, our benchmark is compatible with the Gym standard, and our data is available at our webpage. Code at https://github.com/google/sbsim

cs.AI↗

Back to Basics: Formal Analysis Hints Improve Weak-to-Strong Monitoring

Frontier agentic systems are capable of taking misaligned actions that may cause significant harm. Monitoring techniques are developed to oversee these systems in order to detect and prevent dangerous actions. However, current state-of-the-art monitors suffer from four persistent failure modes: complete detection failures, partial detections, accepting false benign framings, and suspicion score calibration errors. These failure cases can be efficiently combated by augmenting transcripts with supplementary hints, spanning command provenance, security-sensitive action flags, and task-disallowed effect markers. Provenance links allow for the recognition of diffuse attacks, while security and authorization flags simultaneously enable subtle attack recognition and provide an explicit reason for monitors to doubt the false benign framings they encounter, reducing rationalization and improving calibration. We find that augmenting weak extract-and-evaluate monitors with these hints significantly boosts performance with minimal computational cost. Furthermore, a failure mode analysis confirms that hints can remediate all four failure modes, notably almost entirely eliminating detection failures. A series of ablation studies emphasizes the efficiency and potential of the augmented extract-and-evaluate approach. Synthesizing our findings, we propose a tractable two-pronged approach toward enabling weak-to-strong monitoring.

cs.CR↗

Eye, Robot: Learning Expressive Gaze Dynamics via a VLM-Guided Transformer-VAE

Gaze coordinates attention in human interaction through coupled eye, head, and eyelid motion. A humanoid robot must choose where to look while generating a suitable movement toward that target; fixed tracking rules alone do not model the temporal structure of human gaze. We separate these tasks with an asynchronous vision-language model that selects a scene referent, a reactive perception module that maintains fixation, and a Realistic Gaze Transformer (RGT) that generates target-conditioned motion. RGT is a Transformer-VAE trained on markerless human recordings retargeted to the robot's 11-DoF joint space: self-supervised pretraining encodes recent full-head history, and autoregressive fine-tuning exposes the model to its own predictions during rollout. At inference, a four-dimensional binocular-eye trajectory conditions coordinated neck and eye motion; an auxiliary blink predictor triggers eyelid templates during playback. On a two-second benchmark in which all controllers receive the same ground-truth target, RGT-Base obtains the lowest reported overall seven-joint MSE, and physical demonstrations show the integrated targeting and control pipeline. The evidence establishes robot-space kinematic fidelity and real-time RGT inference; perceived naturalness and open-vocabulary target-selection accuracy remain questions for controlled evaluation.

cs.RO↗

The Smart Buildings Challenge: Learning Foundation Models for Real-World Optimization at Scale

Buildings cause nearly 40\% of carbon emissions in the US. Of total building emissions, HVAC systems are responsible for approximately 40-60\% and account for 15\% of global energy consumption. Despite advances in artificial intelligence, most buildings still operate as inefficiently as they did decades ago. Building optimization, perhaps more so than any other optimization problem, has enormous potential for Machine Learning (ML) driven impact on climate change, in both the short and long term. One of the biggest obstacles for building optimization is the lack of policy transfer. Most building control systems, whether simulation or learning based, don't generalize: they only work well for the building they were developed for, but not for other buildings. To solve this problem, we propose a challenge for the development of building foundation models that easily adapt to new buildings with minimal configuration and thereby enable building optimization to scale across the globe. Participants will build control policies on a set of simulated buildings, and have their policies evaluated on an unseen set of buildings. The buildings vary in size, layout, weather, material properties, and HVAC system. Successful outcomes from this initiative promise a massive leap forward in global sustainability efforts and demonstrate the great potential of ML to combat climate change.

cs.AI↗

Learning Functional Subspaces for Neural Network Compression

Modern transformers pair impressive capabilities with substantial memory and compute demands. Low-rank weight factorization reduces both while keeping the matrices dense, and thus efficient on standard hardware. Existing methods, however, choose the subspace to remove from each weight matrix with local closed-form criteria: activation energy, layer-wise reconstruction error, or a quadratic approximation of the loss. These criteria ignore how errors propagate through the network, so at high compression the errors compound with depth and performance collapses. We introduce Learnable Subspace Projections (LSP), which instead learns the subspaces to discard end-to-end. Each linear layer, or tied group of layers that read the same activations, is assigned an orthogonal projector. All projectors are optimized jointly against a global objective--the KL divergence to the dense model's output distribution or the model's original training loss--while the pretrained weights remain frozen. Projectors are initialized from a whitened SVD truncation, and ranks are allocated by the output KL each projector induces per parameter saved. After training, the projectors merge into standard low-rank factors, with each tied group sharing one factor. In attention, this also lets the model cache one narrow latent in place of full keys and values. Across LLMs (OPT-125M/1.3B, Qwen3-4B, Llama-2-7B) and ViT-B/16, LSP outperforms baselines, and its advantage widens as compression increases. At -70% compression, LSP brings Llama-2-7B to 10.9 WikiText-2 perplexity and 42.2% mean zero-shot accuracy, versus 13.3 and 36.0% for the strongest baseline. The factorized model decodes up to 1.6x faster than the dense model at small batch sizes, and aching the shared latent shrinks the combined memory of weights and KV cache by 13.5x at a 128k-token context, versus at most 6.5x for untied baseline factorizations.

cs.LG↗

XTC: Head-Aware Sampling by Excluding Top Choices

Standard decoding rules for autoregressive language models promote diversity by rescaling the full next-token distribution or truncating its low-probability tail. These strategies overlook a common regime of open-ended generation in which several continuations are plausible but too much probability mass remains concentrated on the most generic choice. We introduce XTC (Exclude Top Choices), a lightweight head-aware decoding operator that targets this regime directly. XTC identifies tokens whose probabilities exceed an absolute plausibility threshold $τ$: when at least two qualify, it removes the dominant eligible choices with probability $ρ$ and retains only the weakest plausible alternative before renormalization. Across 60 experiments on Gemma 3 27B Q4, Gemma 3 12B Q6, and DeepSeek R1 14B Q6, with scaling validation on Llama 3.3 70B Q4, XTC improves the diversity-repetition Pareto frontier. On creative generation, Distinct-2 increases by 11--15% and repeat trigrams decrease by 27--47% across the four models. Combined with temperature scaling, gains reach 38% in Distinct-2 and 71% in repeat-trigram reduction over baseline. A blinded Amazon Mechanical Turk study with 150 Master raters yields a 62.3% creativity preference for XTC ($p<10^{-4}$) without reduced fluency, while a GPT-4o control judge reproduces the Anthropic-judge direction on every measure. On IFEval with Llama 3.3 70B Q4, XTC preserves prompt-level strict accuracy within 1.7 percentage points of baseline while recovering most of the diversity gain; a temperature setting matched on Distinct-2 reduces IFEval by 8.8 points. The effect is additive with temperature and repetition penalties, robust across quantization levels and model families, and consistent across twelve prompt genres. XTC has been adopted by llama.cpp, ExLlamaV2, and text-generation-webui.

cs.CL↗

Don't Repeat Yourself: Stopping Verbatim Loops at Sampling Time

Large Language Models generate text autoregressively, but open-ended generation is prone to verbatim looping, in which models repeat spans already present in context. Standard defenses such as repetition, presence, and frequency penalties and n-gram blocking act on token recurrence rather than the sequential structure of a loop, and often suppress looping only at strengths that also degrade formatting or fluency. We propose Don't Repeat Yourself (DRY), a sampling-time logit adjustment that penalizes a candidate token only when generating it would extend the current suffix into an exact continuation of a span seen earlier in the context. Sequence breakers protect chat templates and formatting tokens. Across models from 1.5B to 120B parameters, nine prompt families, and a 600-pair human study, DRY reduces suffix-extension rate by 47% while improving lexical diversity. An intervention-matched placebo produces no comparable reduction, identifying suffix matching as the operative mechanism. On AWQ-quantized 70B and 120B models, DRY reduces loop rate by roughly half while preserving MT-Bench, MMLU, and GSM8K performance, whereas standard alternatives lose measurable ground. DRY has been adopted by popular open-source LLM inference frameworks including llama.cpp, ExLlamaV2, and text-generation-webui, highlighting its practical impact on text generation.

cs.CL↗

The 10th AI City Challenge

The 10th AI City Challenge, held with ECCV 2026, marks a decade of community benchmarking for intelligent transportation, smart cities, and physical AI. Since its 2017 start with vehicle detection, classification, and tracking, the challenge has grown into a broad benchmark suite for multi-camera perception, multimodal reasoning, synthetic-to-real learning, generative forecasting, and privacy-preserving evaluation. The 2026 edition continued this growth with 325 registered teams, up from 245 in 2025, and participation from 26 countries and regions, up from 15. Its six primary tracks cover multi-camera 3D perception, transportation safety captioning and VQA, traffic anomaly reasoning, text-based person anomaly search, generative traffic video forecasting, and cross-city object detection. Track 3 further includes two out-of-domain leaderboards, submitted as Tracks 7 and 8, for fisheye traffic-violation understanding and pedestrian situated-intent VQA. This paper summarizes the challenge setup, datasets, evaluation protocols, leaderboard results, and workshop papers. Across tracks, successful systems combine foundation models with geometric grounding, retrieval or reranking, synthetic-data design, domain adaptation, and controlled inference.

cs.CV↗

Aligning AI-driven discovery with human intuition

As data-driven modeling of physical dynamical systems becomes more prevalent, a new challenge is emerging: making these models more compatible and aligned with existing human knowledge. AI-driven scientific modeling processes typically begin with identifying hidden state variables, then deriving governing equations, followed by predicting and analyzing future behaviors. The critical initial step of identification of an appropriate set of state variables remains challenging for two reasons. First, finding a compact set of meaningfully predictive variables is mathematically difficult and under-defined. A second reason is that variables found often lack physical significance, and are therefore difficult for human scientists to interpret. We propose a new general principle for distilling representations that are naturally more aligned with human intuition, without relying on prior physical knowledge. We demonstrate our approach on a number of experimental and simulated system where the variables generated by the AI closely resemble those chosen independently by human scientists. We suggest that this principle can help make human-AI collaboration more fruitful, as well as shed light on how humans make scientific modeling choices.

cs.LG↗

S-JEPA : Soft Clustering Anchors for Self-Supervised Speech Representation Learning

Self-supervised speech encoders are predominantly trained by predicting discrete hard cluster IDs at masked positions, a recipe that collapses acoustic ambiguity at category boundaries and requires interrupting training to re-cluster the entire corpus between iterations. We introduce S-JEPA, a JEPA-style encoder-predictor pair trained to match the soft posteriors of a Gaussian Mixture Model at masked positions via KL divergence. Training runs as one continuous optimization trajectory in two phases: a fixed GMM over MFCC features, then an online GMM over encoder features, with the input layer selected adaptively from a label-free signal, removing both the offline re-cluster step and the hand-tuned choice of which transformer layer to cluster on. Under the SUPERB protocol, S-JEPA achieves the lowest WER among evaluated SSL methods below 90M parameters and matches HuBERT-Base on emotion recognition at roughly half its parameter count, establishing a new Pareto frontier without offline re-clustering or teacher distillation. An analysis of the predictor's per-frame entropy on held-out speech reveals a bimodal distribution with a substantial minority of frames near the entropy of a perfect two-cluster tie, providing direct empirical evidence that the soft-target objective preserves the acoustic ambiguity that hard targets would collapse. Code is available at https://github.com/gioannides/s-jepa.

cs.SD↗

Evidence of an Emergent "Self" in Continual Robot Learning

A key challenge to understanding self-awareness has been a principled way of quantifying whether an intelligent system has a concept of a "self", and if so how to differentiate the "self" from other cognitive structures. We propose that the "self" can be isolated by seeking the invariant portion of cognitive process that changes relatively little compared to more rapidly acquired cognitive skills - because our self is the most persistent aspect of our experiences. We used this principle to analyze the cognitive structure of robots under two conditions: One robot learns a constant task, while a second undergoes continual learning under variable tasks. We find that robots subjected to continual learning develop an invariant subnetwork that is significantly more stable (p < 0.001) compared to the control, and that this subnetwork is also functionally important: preserving it aids adaptation while damaging it impairs performance. We validate this pattern across three different robots spanning locomotion and manipulation.

cs.RO↗

Mirage Probes: How Vision Models Fake Visual Understanding

Vision-language models (VLMs) can answer image-based questions confidently, and often correctly, even when no image is provided. This mirage behavior inflates benchmark scores without reflecting visual grounding. Prior work treats this as a single failure mode. We argue it is two. Using Mirage Probes, a contrastive probing framework that pairs paraphrased question variants with matched mirage and non-mirage labels on the same image, we show that mirage behavior is linearly decodable from internal activations across residual stream, MLP, post-attention, and attention-head sites in two open-source VLMs. We demonstrate that a Naive Bayes text baseline cannot recover this signal, ruling out surface lexical confounds. Cross-benchmark separability patterns, together with a novel Prior Harnessing Index (PHI) measuring how much a model can answer from text alone, expose two distinct regimes: textual biases, where the model answers from language priors without engaging visual representations, and spurious images, where it constructs false visual content in latent space and answers as if grounded. The distinction has direct mitigation consequences: text-distribution cleaning can address the first regime but cannot reach the second, since spurious-image mirages live in the model's visual representations rather than its text. Faithful visual grounding will require interventions at the representational level.

cs.CV↗

CTF4Nuclear: Common Task Framework for Nuclear Fission and Fusion Models

The demand for clean energy is ever increasing, with new nuclear technologies presenting a complementary solution to renewable energies. However, designing and operating these systems is exceptionally difficult, given the complexity of the physical phenomena that interact to form the system dynamics. While high-fidelity simulations help to understand the non-linear, multi-physics interactions within a reactor, they are computationally expensive and rarely suitable for real-time applications. Furthermore, model-based approaches are inherently sensitive to simplifying assumptions required to derive their governing equations and parameters, leading to inevitable discrepancies with real-world measurements. In contrast, Machine Learning (ML) methods have the potential to generate reliable surrogate models which may be able to quickly predict the system's behaviour. However, the number of data-driven methods that can potentially be used for this task is large and diverse. In a safety-critical setting such as nuclear engineering, a fair comparison of different ML methods, and a clear understanding of their advantages and limitations, is of paramount importance. To address this, we introduce a Common Task Framework (CTF) for ML in nuclear engineering, building upon previous efforts in dynamical systems and seismology. This CTF considers a curated set of datasets from different nuclear and nuclear-adjacent systems. The CTF evaluates the performance of a method on 12 established metrics, alongside a new paradigm focused on system monitoring from sparse measurements only. We illustrate the framework by benchmarking standard ML baselines against these datasets, revealing current method limitations. Our vision is to replace ad hoc comparisons with standardized evaluations on hidden test sets, raising the bar for rigour and reproducibility in scientific ML for the nuclear industry.

cs.LG↗

The Seismic Wavefield Common Task Framework

Seismology faces fundamental challenges in state forecasting and reconstruction (e.g., earthquake early warning and ground motion prediction) and managing the parametric variability of source locations, mechanisms, and Earth models (e.g., subsurface structure and topography effects). Addressing these with simulations is hindered by their massive scale, both in synthetic data volumes and numerical complexity, while real-data efforts are constrained by models that inadequately reflect the Earth's complexity and by sparse sensor measurements from the field. Recent machine learning (ML) efforts offer promise, but progress is obscured by a lack of proper characterization, fair reporting, and rigorous comparisons. To address this, we introduce a Common Task Framework (CTF) for ML for seismic wavefields, demonstrated here on three distinct wavefield datasets. Our CTF features a curated set of datasets at various scales (global, crustal, and local) and task-specific metrics spanning forecasting, reconstruction, and generalization under realistic constraints such as noise and limited data. Inspired by CTFs in fields like natural language processing, this framework provides a structured and rigorous foundation for head-to-head algorithm evaluation. We evaluate various methods for reconstructing seismic wavefields from sparse sensor measurements, with results illustrating the CTF's utility in revealing strengths, limitations, and suitability for specific problem classes. Our vision is to replace ad hoc comparisons with standardized evaluations on hidden test sets, raising the bar for rigor and reproducibility in scientific ML.

cs.LG↗

AI Must Embrace Specialization via Superhuman Adaptable Intelligence

Everyone from AI executives and researchers to doomsayers, politicians, and activists is talking about Artificial General Intelligence (AGI). Yet, they often don't seem to agree on its exact definition. One common definition of AGI is an AI that can do everything a human can do, but are humans truly general? In this paper, we address what's wrong with our conception of AGI, and why, even in its most coherent formulation, it is a flawed concept to describe the future of AI. We explore whether the most widely accepted definitions are plausible, useful, and truly general. We argue that AI must embrace specialization, rather than strive for generality, and in its specialization strive for superhuman performance, and introduce Superhuman Adaptable Intelligence (SAI). SAI is defined as intelligence that can learn to exceed humans at anything important that we can do, and that can fill in the skill gaps where humans are incapable. We then lay out how SAI can help hone a discussion around AI that was blurred by an overloaded definition of AGI, and extrapolate the implications of using it as a guide for the future.

cs.AI↗