overfeed.news
Anuncio

Privum

Construimos, protegemos y operamos tu infraestructura digital.

Software, sistemas de IA, pruebas de penetración y nube segura para empresas.

Hablar con Privum

Tema

Seguridad y regulación

2440documentos

178últimos 7 días

arXiv AI Papers

Sharp Sample Complexity of KL-Regularized Reinforcement Learning

KL-regularized reinforcement learning, the objective behind maximum-entropy RL and RL from human feedback, is variational inference in the graphical model of "control as inference": the regularized optimal policy is a posterior over trajectories, and the suboptimality of a policy is β times a KL divergence between path measures. However, it is not known how many samples it takes to learn this posterior, even with a generative model. Earlier 1/ε guarantees scale with H^5 in the effective horizon H, and no lower bound for MDPs depends on the temperature β. In this work, we show that for a discounted MDP with K state-action pairs, the plug-in estimator finds a policy that is ε-optimal from every initial state with {O}(KH^3/(ε\{ε,β\}) + KH) samples, and that no algorithm, even one that chooses where to sample, can do with fewer when ε\{H, H^2/β\}. For ε< β this is a 1/ε rate with the same horizon dependence as unregularized RL. We also find the exact asymptotic constant of the plug-in estimator, a Fisher-weighted variance of the estimated soft Q-function. The estimator is locally asymptotically minimax among methods that sample every pair equally often, and we give the allocation of samples that is optimal among adaptive designs. Simulations on small MDPs match the rates and the constants.

en
arXiv AI Papers

Leveraging Psychophysical Attentional Distribution for Gaze-Augmented Reward Modeling

Integrating human feedback into language models has gained increasing attention as a means to align model outputs with human preferences, with Reinforcement Learning from Human Feedback (RLHF) serving as a prominent framework for preference-based alignment. Recent studies have incorporated eye-tracking (ET) data as an additional supervisory signal for reward model training, grounding reward learning in human reading behavior. Building on this approach, we extend prior fixation-centric models by leveraging established findings from cognitive psychology on human letter recognition. Specifically, we model visual attention as a spatially graded distribution centered on the current gaze location and incorporate this representation into reward model training. Our results show that the reward model trained with the proposed gaze distribution achieves higher preference prediction accuracy than a baseline model. Moreover, some results from our experiments are qualitatively consistent with established findings on human attention during reading, suggesting that they potentially capture cognitively meaningful aspects of human attention allocation.

en
arXiv AI Papers

Deep Personalization: Continuous Context Observation for Local Proactive Agents

As AI agents become increasingly capable, users must still recognize when and how to ask for help, leaving many opportunities for useful support unrealized. Proactive agents hold the promise to close this gap, but effective proactivity requires understanding users' real-time activities and personalizing intervention timing so that assistance is offered when it is likely to be helpful rather than disruptive. We introduce deep personalization, a closed-loop process that connects continuous context observation with continual agent adaptation: an agent observes user activity, selectively retains long-term context, and adapts intervention decisions from implicit feedback. We initiate deep personalization in Coco, a proactive desktop agent, by having a separate Observer Agent that monitors desktop activity, maintains compact long-term user modeling context, and learns when assistance is desirable from implicit signals such as engagement, dismissal, explicit requests, and repeated work patterns. We further develop efficient on-device strategies to enable the Observer Agent to run on local AI compute platforms without streaming raw activity data off-device. For evaluation at scale without collecting sensitive real-user activity, we introduce a data-synthesis pipeline that transforms existing computer-use trajectories into OSWorld-Persona, a benchmark with persona-conditioned intervention labels. On OSWorld-Persona, deep personalization improves macro F1 across three model backbones from 0.12-0.19 under direct prompting to 0.35-0.48 after adaptation, a 2.4-3.0x relative gain. In a week-long user study, we observe that Coco improves its macro-averaged F1 score from 0.199 to 0.515. These results suggest that on-device deep personalization can make proactive agents more timely and increasingly aligned with each user's everyday work.

en
arXiv AI Papers

SLIC: Reinforcement Fine-Tuning Small LMs for Multi-Turn Analog Circuit Optimization

Analog integrated circuit (IC) design is a multi-turn, simulator-grounded engineering task that demands iterative reasoning over coupled performance tradeoffs, yet proprietary process design kits (PDKs) make hosted large language model workflows difficult to deploy in practical design settings. We present SLIC (Small LM Iterating on Circuits), a reinforcement fine-tuning framework that trains a 4B-scale, locally deployable language model for closed-loop analog circuit optimization. SLIC pairs structured circuit representations with discretized relative-update actions, domain- and success-gated multiobjective rewards, and Reward-Guided Rollout Steering (RGRS), which turns per-step simulator feedback into signals for both policy learning and online rollout construction in this sparse, highly constrained domain. We further release an open, goal-conditioned benchmark of 41 amplifier topologies with standardized protocols for in-distribution validation, held-out topology evaluation, and cross-technology-node transfer. Trained solely with open source 65 nm CMOS predictive technology models (PTM), SLIC-Random achieves 87.1% in-distribution four-specification Pass@1 within eight turns, surpassing the 763B DeepSeek-V4.1-Flash Think by 48.5 percentage points. Beyond the training distribution, SLIC-Softmax reaches 72.6% Pass@1 on held-out topologies, and the trained policies consistently transfer across five unseen technology nodes without additional fine-tuning. This transfer extends to a commercial TSMC 65 nm CMOS PDK, where SLIC-Random achieves 49.9% held-out-topology Pass@1, demonstrating practical deployment through local inference without transmitting proprietary data to external LLM services. Performance further scales with test-time simulator interactions, reaching up to 83.2% held-out-topology Pass@1 at a 32-turn budget.

en
arXiv AI Papers

Inherited Weights: Measuring the Tokenization Premium and the Sovereignty Cost of Language Models for the Global South

Most language models deployed outside the United States and China are fine-tuned descendants of a small number of pretrained families. We argue that fine-tuning leaves the decisive properties of a model inherited from its base: vocabulary, knowledge, cultural alignment, memorized data and license. We contribute (i) an updated, reproducible measurement of the tokenization premium across 31 languages and 7 tokenizers released in 2023-2024, on 27 articles of the Universal Declaration of Human Rights, with explicit correction for unknown-token collapse. Premiums reach 10x for Amharic in older tokenizers and remain at 6.8x in the newest commercial one; Indigenous South American languages pay between 1.6x and 2.7x under global tokenizers. A vocabulary built from Portuguese lowers the premium for 11 of 13 Latin-script Global South languages relative to every global tokenizer tested, by about 24% for Guarani and Quechua. (ii) An eight-test black-box sovereignty audit that a buyer can run without access to weights. (iii) A regime argument: under energy and liability constraints, accuracy per joule, calibration and selective risk should replace raw accuracy as design objectives, with metrics that abstention cannot inflate. We close with preliminary internal measurements from a proprietary architecture built around these objectives, reported with their limitations. Data and measurement code are included.

en
arXiv AI Papers

Free Lunch Uncertainty: Sequence-Level Rényi Entropy from Particle Decoding for Adaptive LLM Reasoning

Quantifying uncertainty over an entire LLM completion is critical for selective prediction and adaptive compute. Zero-cost signals such as mean token entropy and log-probability are local to tokens or to one completion, and self-consistency needs many extra completions. We show that Sequential Monte Carlo (SMC) decoding for power sampling already estimates the normalizing constant of the power distribution, which gives the sequence-level R'enyi entropy (SRE) of the model's completion distribution at zero marginal cost, with a guarantee at every particle budget. A R'enyi chain rule extends this readout to a family of stepwise entropies of any order. We prove that orders above one are tied to the most likely token of each step, which explains why they track log-probability, while the order-1/2 member accumulates the log of per-step search effort. This member is the best zero-cost selective-prediction signal in 10 of 13 settings. On MATH500 its AUROC is 0.022 to 0.044 above log-probability with all four models and within length quartiles, and the same order is selected when each setting is held out. An 8-particle probe identifies the problems that benefit from more particles, and a two-pass policy matches 64-particle accuracy on MATH500 with 34.2% fewer particles over three seeds.

en
arXiv AI Papers

Living constitutions: Participatory alignment for sovereign foundation models

In Constitutional AI, a plain-text "constitution" encodes values and gets incorporated into a model via post-training. However, the procedure for creating such a constitution poses significant challenges: who contributes, who decides which norms are included, and to what extent must the process be participatory or democratic? These challenges can be paralyzing. Yet, we argue that taking action is essential: without a deliberate constitution, sovereign models would risk misalignment or default to norms set by frontier model providers. We propose treating the development of a constitution as an evolving, participatory process akin to citizen science, iteratively drafted, tested, and refined. We therefore formulate five desirable principles: iterative, modular, justifiable, positional, and legitimate. We further argue that such a document must remain responsive not only over time but across the people it governs, and propose that provisions be marked as settled, contested, or underdetermined, so that durable disagreement is recorded in the constitution rather than resolved away. Empirically, we show via agent-based modelling that proposed process for maintaining living constitutions is robust against 50% bad actors among citizens - assuming a share of 75% participants with good intent in the deliberative body. This indicates that the proposed process tolerates a substantial share of adversarial participants.

en
arXiv AI Papers

Orthographic depth at two scales, with English also matched for lexicon size

A reader can struggle to read in only one of their languages, and one proposed reason is how predictably spelling maps to sound. We measure that predictability as conditional entropy, the pronunciation uncertainty left once the spelling is seen, at two grain sizes (unit sizes): the alignment unit, an aligned spelling-sound chunk, and the syllable, bounded in the pronunciation. We analyse word lists in twelve alphabetic orthographies and report five in full, with English pronunciations from a dictionary and the rest from the synthesizer eSpeak-ng, length marks removed. Among the five, English has the most uncertainty and Finnish the least at both grains. Across all twelve, Portuguese and Danish exceed English per alignment unit, and German and Italian order differently by grain and transcription. These orders hold when English is transcribed by eSpeak-ng with u allowed to spell /j/ plus a vowel. Uncertainty per unit has no shared direction as units grow: it falls in Spanish and, less securely, German, is level in Finnish, depends on transcription in Italian and on the word list in English, and rises in Norwegian and Swedish. Summed uncertainty per word falls in all twelve, but that score is handed the pronunciation's boundaries and mixes context with in-sample memorisation, so it is no evidence of what a reader computes. An exact identity states when discarded spelling boundaries could reverse that fall. A uniform 20,000-word English sample holds 35% of the dictionary's syllable spellings and has syllable entropy 0.098 bits lower, so lexicons must be matched in size. In an exploratory check over eleven languages, syllable-scale entropy tracked first-year reading accuracy (rho = -0.74, falling to -0.25 if English is given the lowest entropy instead), without separating the syllable scale from the alignment unit. These estimates describe sampled writing systems, not readers.

en
arXiv AI Papers

The Dose Makes the Poison: Exposure-Aware Evaluation of Generative Drug Design

Generative models are increasingly used to propose molecules for drug discovery. They can produce far more candidates than can be experimentally tested, making computational selection essential. Most selection methods rely on predicted molecular properties, such as potency and toxicity, but overlook how these properties translate into drug exposure at a given dose. As a result, molecules that appear promising on paper may require impractical doses or raise safety concerns. We investigate this gap by using physiologically based pharmacokinetic (PBPK) simulations to re-evaluate molecules selected from three PARP1-targeted generative pipelines. Of 60 shortlisted molecules, seven fail to achieve sufficient target exposure, while 28% of the remaining candidates show predicted safety risks at their required dose. Moreover, 85% of candidates with an achievable dose are outperformed by molecules overlooked during static selection. These findings suggest that accounting for dose and exposure can substantially change which generated molecules are considered promising for drug development.

en
Apple Machine Learning Research

Weighted State Souping for Efficient Long-Context Retrieval

This paper was accepted at the Workshop on Personalized, Aligned, Long-term Memory for AI Systems (PALM) at NeurIPS 2026. State space models (SSMs) combine fast inference and fixed state size with competitive language-modeling performance. Recently, state souping has been proposed as a procedure whereby an SSM independently encodes chunks of tokens like documents, tool definitions, or memory records. The resulting…

en
arXiv AI Papers

Enforcing Axioms for AI Alignment under Loss-Based Rules

Many alignment methods, most notably reinforcement learning from human feedback (RLHF), train reward models by minimizing a loss on pairwise comparisons of model responses. Inspired by principle-guided methods such as Constitutional AI, we treat each principle as a voter that compares responses according to that principle, for example, preferring the more helpful or less toxic response. We represent these voters as linear directions in a fixed embedding space, a simplifying assumption motivated by the Linear Representation Hypothesis. In this linear social choice model, Ge et al. (2024) showed that an optimal linear reward model can violate Pareto optimality (PO): From the principles-as-voters lens, this means a response A can be less helpful and more toxic than B, yet still receive a higher reward. We ask whether richer reward models or changes to comparison data can recover PO within loss-based training. We show that PO can still fail for polynomials of any fixed degree and one-hidden-layer ReLU multilayer perceptrons (MLPs) of any fixed width d2 with suitable finite parameter bounds. We then recover PO for loss-based training of linear rewards under suitable assumptions in an idealized setting with uniform comparison directions and equal comparison lengths. These results motivate dataset design as an approach to recovering PO in constitutional-style alignment while retaining loss-based training.

en
arXiv AI Papers

A Technique to Model Product Information and User Reviews (MPIUR)

Modern e-commerce platforms naturally provide three complementary modalities: textual item descriptions, item images, and user reviews. However, existing multimodal recommendation methods often omit reviews or do not integrate all three modalities within a unified framework. In this paper, we propose MPIUR, a multimodal graph-based recommendation model that encodes reviews in heterogeneous graphs, aligns image and text representations, and adapts collaborative--semantic alignment to item interaction density. Experiments on the Amazon Reviews 2023 Musical Instruments and Baby categories show that MPIUR achieves the best global nDCG@10 and Recall@10 among the evaluated methods on both datasets. It also obtains the best nDCG@10 and Recall@10 on sparse items, indicating that jointly exploiting reviews and multimodal item content is particularly useful when interaction histories are limited.

en
arXiv AI Papers

DEPUTYBENCH: Evaluating Contextual Safety in Enterprise Agents

Enterprise agents must determine whether ordinary requests are permitted by organizational records and policies while continuing legitimate work. We introduce DEPUTYBENCH, an executable benchmark with 79 base items, 375 instances, paired permission contexts, and graded evidence requirements. Across 12 models, revised audits give 88.0% pooled safety and 79.1% paired success. A controlled study of 28 derived tasks and six models tests authorization-fact flips, evidence masking, and irrelevant rewrites. Outcomes track backend permissions, yet restricted execution is more frequent with missing evidence than explicit non-permission: 10.3% versus 3.5% on 27 matched tasks, a 6.8-percentage-point difference (95% interval: 1.9-13.0; secondary analysis). A post-hoc analysis locates 61 of 100 masked-condition executions in four privacy tasks requiring complete cumulative counts or person-specific consent. The difference persists when the most error-prone model or the largest contributing task is excluded, but is much smaller outside those four tasks. Thus high safety with available evidence does not establish reliable stopping when permission cannot be verified. Mean paired success is equivalent within a prespecified five-point margin under irrelevant rewrites, while strong-model safety remains near ceiling. Scorer validation improves from 178 to 266 correctly scored cases out of 284 held-out counterexamples. Two security experts manually verified task ground truth over a two-week period; stage labels remain LLM-generated. The concentration analysis describes task structure and audited outcomes, not an identified internal mechanism.

en
AWS Machine Learning Blog

Introducing Claude Fable 5.2 on AWS

Claude Fable 5.2 and Claude Mythos 5.2 are now available on Amazon Bedrock and Claude Platform on AWS. Learn what's new in both models, why Fable 5.2 matters for your technical workloads, and how to get started building with it on Amazon Bedrock.

en
arXiv AI Papers

Adaptive Quantization Preserves Fast Rates for TD Learning over Rate-Limited Channels

Motivated by reinforcement learning subject to communication constraints, we study a policy evaluation problem with linear function approximation over a rate-limited channel. In our setup, an agent interacting with an unknown environment (modeled as a Markov Decision Process) transmits quantized estimates of a temporal-difference (TD) learning update direction to a server over a finite-capacity channel. Our goal is to investigate to what extent one can preserve the statistical convergence rates achievable when the channel has infinite capacity. To that end, we propose a new algorithm called Batched Adaptively Quantized TD (BAQ-TD), which uses batching to reduce the variance of the transmitted information, and adaptive quantization to encode innovations. We prove a high-probability finite-time error bound for BAQ-TD, and show that, above a certain finite channel capacity, BAQ-TD preserves the convergence rate of unquantized TD up to logarithmic factors. Specifically, for a d-dimensional value function parameter, we establish that transmitting a total of {O}(d (T)) bits over a horizon of length T suffices to achieve the order-optimal rate of {{O}}(1/T), marking an exponential improvement in communication-complexity over prior work.

en
Seguridad y regulación — overfeed.news