overfeed.news
Anuncio

NexTalk

Automatiza la primera respuesta. Mantén el toque humano.

Chat en vivo con agentes de IA en tu modelo y tu clave de API, con traspaso a tu equipo.

Empezar gratis

Tema

LLMs

2993documentos

239últimos 7 días

arXiv AI Papers

FlashScript: Compiling Recurring Agent Workflows into Reusable Neuro-Symbolic Programs

Many agent tasks recur with new data or small changes, yet agents regenerate their reasoning and actions on each run, producing different results. Distilling a prior run into a skill helps the agent avoid earlier mistakes, but it still re-runs the full process each time, so runs can vary and incur that cost. We compile a successful agent run into a reusable neuro-symbolic program (NSP) with a set of guidelines we call FlashScript. Actions and data transformations become code, instance-specific values become parameters, and decisions that need model judgment become LLM calls. We identify three layers where these calls belong and three techniques that replace most of them with code. The compilation does not remove nondeterminism; it localizes it, so the deterministic program and its neural calls can be reviewed and tested separately. Across 120 WorkBench analytics tasks with Claude Haiku 4.5, the NSPs were correct and made identical tool calls on every run, using about 70 times fewer input tokens and a third of the runtime of a skill-guided agent whose runs were inconsistent. In a composite version of the tasks, the skill-guided agent's accuracy fell to 23%, and its output inconsistency increased, while the NSPs remained correct and identical across all runs. The NSPs successfully generalized to all 108 unseen instances of the same templates. Our findings suggest that NSPs from agent trajectories can be considered cumulative software assets for end-users to test, maintain, and reuse in their recurring workflows.

en
arXiv AI Papers

Software Engineering LLMs: Why We Need Them and How Software Engineering Helps Build Them

Software engineering (SE) expertise can shape how coding large language models (LLMs) are trained and which engineering capabilities they acquire. We examine this role through an industrial case study of capability-focused training for a 92B-parameter coding LLM serving more than 100,000 developers. We demonstrate how established SE knowledge can guide the selection of engineering capabilities, the construction of verifiable training data, and the evaluation of model improvements. To target repository-level refactoring, we developed RefactorForge, a pipeline that combines refactoring knowledge, established SE tools, and verification techniques to construct training data. Across 306 Python repositories, RefactorForge generated 5,704 executable tasks and retained 4,572 verified teacher trajectories totaling 72.51 million tokens. Incremental supervised fine-tuning on these trajectories, accounting for 0.46% of the preceding coding mixture's tokens, improved RefactorBench performance by 18.00 percentage points and SWE-Bench ProMax performance by 7.06 points. Under OpenCode, these improvements extended to SWE issue resolution, with gains of 4.97 points on SWE-Bench Verified and 1.23 points on SWE-Bench Pro. Improvements also persisted across other coding-agent harnesses, while greenfield repository generation showed mixed results. We report lessons from constructing capability-focused training data and evaluating its effects across engineering workloads. Building on this industrial experience, we propose a research agenda for expanding SE's role in transforming coding LLMs into SE LLMs: incorporating SE knowledge into model training, applying engineering methods to model development and maintenance, and systematically defining the engineering capabilities these models should acquire.

en
arXiv AI Papers

Build, Don't Play: Self-Improving LLM Research Agents Develop a Hearthstone Battlegrounds AI That Rivals Top Players

Can LLM agents carry a hard real-world problem from raw data to a system that holds its own against people? We study this in Hearthstone Battlegrounds, an eight-player auto-battler with stochastic shops and battles, hidden information and nearly a thousand interacting cards. We present AutoTavern, a self-improving research system: specialised LLM research agents share a typed research memory, every claim passes verification gates, and live games close the loop, so the agents improve the simulator, opponents and policy and, from the same feedback, their own research process; a human sets goals and rules on ambiguous mechanics. In 20 days AutoTavern built TavernGym, the first faithful, full-content reproduction of the live game, and delivered 23 versions of TavernAgent, trained with PPO against opponents calibrated to real games. TavernAgent plays complete games on the live ladder by itself: its rating rose from 6,000 to about 15,000 and, on the public leaderboard's measure, its last 20 records average a placement of 1.60 with 80% first places, better than the server's ten highest-rated players on every statistic (2.34, 59%). TavernBench, 960 tasks in 40 scenarios, shows that the system the agents built is the strongest player: TavernAgent answers 91% of the play tasks against at most 63% for frontier LLMs, and decides in 0.03 s where they need up to 26 s; within the second a live turn allows, it keeps 90% and no LLM more than 26%. LLM agents build far better than they play. We will open-source TavernBench and TavernGym at https://github.com/MCG-NJU/AutoTavern.

en
arXiv AI Papers

The Language of Instruction: Genre and Generative AI

This article argues that instruction-fine-tuned language models should be understood as implementing a genre, what is proposed as the instructional-conversational genre. This genre is an amalgamation of discursive language from prior conversations between generative models and humans recursively run through these massive neural networks. It is instantiated in applications through instruction fine-tuning, the inclusion of a pretext known as a system prompt, and the formalization of inputs in mark-up languages characterized by the use of special tokens and role assignment in turntaking structures.

en
arXiv AI Papers

Measure Before You Train: Typed Decision Models vs. a Classical Baseline vs. a Generative LLM for Routing a Personal Multilingual Inbox

Personal email agents decide, for every message, what it is, whether it is urgent, whether to draft a reply, create a task, or ignore it. Their error rate is rarely measured: the stored labels were written by the system itself. We present Foyer, a self-hosted personal inbox agent in daily use, and a measurement protocol (stratified sampling, frozen snapshots, thread-grouped splits, pre-registered hypotheses, human labels in the agent's review loop) that yields ground truth before any training. On 739 real Italian/English emails labelled by the owner (231 unbiased, 500 adversarially sampled), we compare nine systems under thread-grouped cross-validation: the production keyword rules, a TF-IDF + logistic-regression baseline, local typed decision models (Tev1, Nimble), a hosted decision model (Nimble v4-max), and hosted generative LLMs (Qwen2.5-7B, Gemma 4 26B/31B). Four results. The production rules score 28% on message type against a 58% majority class: the field shown most prominently carried no information. With ~200 labels, logistic regression has the highest accuracy on every field (type 0.79, urgency 0.74, needs-reply 0.83, is-noise 0.90), though it never predicts the two rarest classes and is statistically tied with Gemma 4 on type; few-shot prompting does not close the gap. The sender's relationship (client, supplier, service) is recovered from other emails of the same contact with 98.6% accuracy where the contact has been seen, and is not inferable from a single message by any model: it belongs in the contact graph, not the prompt. Finally, accuracy is not the product metric: composed into actions, the most accurate system most often ignores emails needing action (18% vs 4% for the best LLM); a cost-chosen threshold on its noise probability halves that rate and makes it the cheapest system.

en
arXiv AI Papers

Where Rounding Happens Matters: Rounding-Location Alignment Stabilizes LLM Reinforcement Learning

Reinforcement learning for large language models often uses separate systems for inference and training. Even with identical weights and BF16 tensors, implementation differences can make an intended on-policy update off-policy and destabilize training. Existing remedies align the entire inference and training stacks, change global precision, or modify the learning algorithm. We ask whether aligning selected parts of the execution stack can stabilize BF16 RL while other implementation differences remain. We align training-side rounding locations with inference in three regions---RoPE, SwiGLU, and residual RMSNorm---by retaining selected intermediates in FP32 until their BF16 outputs. Across four Qwen-family models, aligning all three regions prevents the collapse observed under the default BF16 baseline; RoPE-path alignment alone also suffices in the studied runs, despite remaining numerical discrepancies. The interventions keep the inference engine, matrix-multiplication and attention kernels, and learning objective unchanged.

en
arXiv AI Papers

Beyond Discrete Tokens: Preserving Numerical Continuity in Medical Vision-Language Models

Objective: To develop a unified vision-language model for prompt-conditioned quantitative ultrasound measurement that addresses the mismatch between continuous measurement targets and discrete token-based numerical prediction. Methods: This article presents HAD-VLM, a hybrid autoregressive-diffusion framework that combines autoregressive text generation with continuous measurement prediction within a shared backbone. Visual prompt tuning adapts the vision encoder to ultrasound images. A diffusion process predicts a five-component representation comprising caliper endpoint coordinates and measurement length. A post-processing strategy combining the proposed coordinate-length consistency mechanism with ensemble prediction is introduced to refine measurement estimates. We evaluate four echocardiographic measurement tasks on EchoNet-LVH, with additional assessments of medical question answering and X-ray vertebral measurement. Results: HAD-VLM achieves mean absolute errors of 1.36, 2.77, 3.05, and 1.16 mm for IVSd, LVIDd, LVIDs, and LVPWd, respectively, reducing errors by 21.4-37.6\% relative to the Vicuna-based autoregressive baseline. It also outperforms fine-tuned Qwen3-VL and direct MLP regression across all four tasks. Medical question answering performance is preserved on the evaluated dataset. For X-ray vertebral measurement, HAD-VLM reduces mean absolute error from 11.4 to 6.7 pixels. Conclusion: Integrating autoregressive language generation with diffusion-based continuous prediction improves quantitative medical image measurement while retaining prompt-conditioned task flexibility. Significance: HAD-VLM provides a unified approach to combining language interaction with anatomically grounded numerical prediction for quantitative medical image analysis.

en
arXiv AI Papers

Monetizing Intelligence: Advertising and Subscription Design on LLM Platforms

Large language model (LLM) platforms introducing advertising must decide how much capability ad revenue should fund and what consumers should pay to avoid ads. We study a monopolistic platform that jointly chooses costly response quality, advertising intensity, and subscription prices when advertising reduces user value but better service raises advertising returns. Three findings follow. First, when partly ad-supported and ad-free plans coexist, advertising can finance useful service for consumers who receive none under pure subscription. Lower effective quality in middle plans weakens the alternatives available to premium users, allowing higher prices for unchanged ad-free service. Some fully ad-supported users gain utility, while partly ad-supported and ad-free users lose: broader access can coexist with utility polarization. Second, an intermediate advertising cap can maximize consumer surplus by preserving some revenue that finances service while limiting degradation. Third, cheaper compute can raise quality and reduce ad intensity, yet sufficiently large cost reductions can increase aggregate advertising revenue above its initial level because better service raises the value of the remaining advertised interactions. An extension shows how advertising-funded upgrades can bring plans to the same maximum capability and advertising exposure, eliminating the basis for different prices. These results make menu design central to evaluating LLM advertising: the consequences extend beyond users who see ads to the alternatives and prices faced by those who do not.

en
arXiv AI Papers

Adaptive Frequency Threshold Byte Pair Encoding for Mixed Natural Language and Structured Symbolic Corpora

Modern language models rely on subwording tokenization to define the discrete units of input processing, typically using methods like Byte Pair Encoding (BPE) that treat whitespace as a fixed semantic boundary. This design is inherently limited, as many meaningful concepts---including multiword expressions and the structured environments of formal languages like LaTeX---frequently span across whitespaces. While recent superword tokenization methods attempt to address this by relaxing whitespace constraints, they often rely on greedy frequency-driven merging that fails to account for token length or the specific needs of underrepresented domain-specific languages. In this paper, we introduce Adaptive Frequency Threshold Byte Pair Encoding (AdaBPE), a new tokenization method that leverages an adaptive frequent-pattern mining approach. AdaBPE features a twofold dynamicity: it adapts minimum frequency thresholds based on both the length of a candidate token, to prevent favoring frequent short patterns, and its type, to distinguish between general and formal language patterns, using LaTeX as case in point. This allows the model to efficiently allocate vocabulary capacity to domain-specific notations, such as LaTeX, even when they represent a small fraction of the training data. We evaluate AdaBPE across five datasets spanning LaTeX and natural language, showing consistent gains in encoding efficiency and downstream language-model performance on domain-specific content without compromising general-language performance.

en
arXiv AI Papers

SemantEq: Faithful Bidirectional Equivalence of Formal Statements

Statement autoformalization translates mathematics from natural language into a proof assistant such as Lean 4, and models now do it at scale. As provers get stronger, the burden of correctness shifts from the proof to the statement, that is, to whether the formal statement faithfully translates its source. Given a trusted reference formalization, the natural test is logical equivalence, and it proves too much. P Q holds trivially for any two statements P and Q that the library proves, because a prover can prove each separately and pass through P True Q. The stronger the prover, the more unrelated pairs pass. Earlier methods restrict the prover and check that each proof names the other statement. We introduce SemantEq, a more principled relation that instead requires each statement to turn into the other only through a chain of allowed steps, such as logical rearrangement and rewriting with the library's own equations. Our programs judge a proof by the proof term Lean checks, not by its tactic script. SemantEq-Eval, a search without a model, reaches 100% precision and 65.6% recall on the ProofNetVerif test split under our adjudicated labels, against 95.8% and 61.8% for BEq+, and an 8.9-fold increase in its CPU time adds at most three pairs. A language model finds more proofs, but under two of four prompts it proves both directions of about two in five of the non-equivalent pairs. SemantEq-ProofAudit audits each model-written proof under a policy derived from the relation and accepts none of the 104 non-equivalent pairs, and it raises recall on 407 equivalent pairs from 65.4% to 81.1% under the best prompt. Separating the generation of equivalence proofs from their acceptance lets a stronger prover raise recall without changing what an acceptance means. We release code and data at: https://github.com/hlxr89/SemantEq

en
arXiv AI Papers

Green Code or Costly Code? Evaluating the Energy Costs of Iterative LLM-Based Code Optimizations

Large Language Models (LLMs) are increasingly used to optimize code for performance. This practice often assumes that improvements in execution speed naturally translate into overall lower energy consumption. We challenge this assumption empirically by measuring whether performance-optimized code produced by LLM self-refinement pipelines is also energy-efficient relative to canonical human reference solutions, and whether the energy cost of the optimization process can be offset by subsequent execution-time savings. We evaluate 11 open-source models across four decoding configurations on all 1,000 problems of EffiBench, a state-of-the-art performance benchmark, while collecting per-process, hardware-level energy measurements throughout. Inference energy spans 0.79 to 8.68 MJ across model-variant pairs, which is roughly equivalent to the energy of 250 to 2800 GPT-4.1 queries. Limiting the maximum number of output tokens has the largest impact, reducing energy use by up to 69% compared to the default configuration. On the execution side, 42 of 44 configurations produce code that consumes more energy per problem than the canonical human reference, with overheads ranging from +4% to +886%. The two exceptions achieve only marginal gains, with break-even points on the order of 10^6 to 10^7 executions. Mixture-of-Experts architectures consume substantially less inference energy than equivalently-sized dense models without sacrificing correctness, making architectural choice a more consequential sustainability lever than decoding parameters. Performance optimization and energy efficiency are not always aligned goals: LLM self-refinement consistently produces faster-seeking but energy-heavier code relative to what a human engineer would write. All artifacts are available at https://doi.org/10.5281/zenodo.23598170.

en
arXiv AI Papers

NgramRL: Accelerating RL Post-Training for Structured Generation via Rollout-Aware N-gram Speculative Decoding

Reinforcement learning (RL) post-training improves the reasoning capabilities of large language models, but its cost is dominated by autoregressive rollout generation, which can exceed 70% of training time. Existing speculative-decoding methods are poorly suited to this setting: model-based drafters consume scarce GPU memory and may become misaligned as the policy evolves, while existing model-free methods do not explicitly exploit repetition across training epochs. We observe that recent rollouts of the same prompt provide a high-quality, self-updating draft source. Based on this insight, we propose NgramRL, a per-prompt n-gram drafter constructed from bounded recent rollout history. NgramRL requires no auxiliary model, drafter training, or persistent GPU-resident drafter state, while target-model verification preserves the rollout distribution. To hide CPU cache lookup and scheduling, we further introduce an adaptive fan-out prefetcher that predicts likely verification outcomes (j, b), consisting of the accepted draft length j and target-emitted continuation token b. A causal budget allocator concentrates prefetches on high-probability outcomes, while uncovered outcomes fall back to sequential lookup without changing committed tokens. Across two model families and three reasoning and structured-generation workloads, NgramRL commits 3.60-7.77 tokens per target-model forward in the warm-cache regime, up to 1.99x more than SuffixDecoding and 7.0x more than prompt-lookup decoding. On Spider, the prefetcher covers 96.6% of Qwen3-8B outcomes at a candidate budget of 16 and improves measured rollout throughput by 1.38x over synchronous NgramRL, approaching its 1.40x overlap ceiling.

en
arXiv AI Papers

Advancing SVD-based LLM Compression via Layer-Wise Error Model Search

Low-rank SVD-based compression offers a powerful strategy to reduce the computational costs of LLMs. However, existing methods face two key limitations: (i) global rank allocation, where uncalibrated error proxies fail to capture complex error propagation, and (ii) decomposition quality, where Fisher-based estimators suffer from severe rank collapse. In this work, we address these limitations by introducing Layer-wise Error Modeling Search (LEMS) and KFAC-SVD. LEMS advances rank allocation by introducing a layer-wise error surrogate that integrates local and global layer importance alongside a propagation bias, enabling effective global rank allocation via an ILP formulation. KFAC-SVD improves decomposition quality by utilizing token-wise statistics, mitigating the rank deficiency observed in prior Fisher-based SVD approaches. Across Mistral, Qwen3, and Llama 3 model families, we show that LEMS consistently outperforms existing search strategies, delivering significant zero-shot accuracy gains of up to 4.8 p.p. that generalize to model sizes of 70B parameters, while KFAC-SVD achieves an average perplexity improvement of 15%. Project page and code: https://lems-svd.github.io

en
arXiv AI Papers

Trusting the Tool: Asymmetric Deference of LLM Vulnerability Auditors to Static-Analysis Evidence

Pipelines that combine large language models (LLMs) with static analysis let the model consume the analyser's output as evidence, but whether the model distinguishes true from false evidence, and whether its scrutiny depends on what the evidence claims, has not been measured. We built an intraprocedural analyser for C/C++ that produces proof obligations and audited vulnerable and patched functions from PrimeVul under six evidence conditions that differ only in the notes, including fabricated assurances that a missing check exists and fabricated alarms that an added check is missing. The study covers 916 pairs in four disjoint sets, eleven model configurations from four vendors and 16,572 audits; no model is trained or fine-tuned. True evidence barely improved accuracy (+0.02). False evidence acted asymmetrically: relative to true notes about the same operations, false assurances lowered the detection of real vulnerabilities by up to 19 points, whereas false alarms never significantly lowered accuracy and were disputed far more often. An internally registered replication on seven new models confirmed that false assurances move verdicts towards their claim more than false alarms do (mean asymmetry +0.16, 95% CI +0.10 to +0.24). Instructions in the audit prompt did not reliably reduce deference; the best one failed to replicate. Verifying each claim in a separate call and showing the refuted claims raised accuracy under false assurances by 12 points in a pre-specified test (p=0.002) without loss under true evidence; for a second model the gain was smaller and not significant. Pipelines should therefore verify a tool's guarantees before the model relies on them, and studies should measure deference in a direction-aware way.

en
arXiv AI Papers

Do "True" and "False" Add Up to One? Complement Consistency in LLM-Agent Opinion Simulations

In LLM-agent opinion simulations, each agent reports the probability that a claim is true after seeing its own previous probability and the probabilities reported by its neighbors, and that number becomes its state in the next round. If the number describes the agent's position rather than the sentence it was shown, then rewriting the claim as its complement and replacing every probability p with 1 - p should return the complementary answer, so the two reports should sum to one. We test this with three small open models: Qwen3-4B, Llama-3.2-3B and Gemma-3-4B. On 2,304 agent states taken from a running network simulation, Qwen's paired reports sum to 1.18 on average, and on 48 real fact-checking claims from AVeriTeC they again exceed one. A controlled benchmark of 24 worlds, each with exactly two possible outcomes, shows the same failure in all three models. The discrepancy is two to ten times larger than the change produced by a same-meaning paraphrase, and it persists under repeated sampling and across peer group sizes and report strengths. When the reports are fed back into two populations that start from complementary states, the populations do not remain mirror images: their divergence is positive after five rounds for every model. Averaging each answer with the complement of its paired answer restores consistency by construction and lowers Brier loss averaged over the two wordings, but at the final round it beats the better of the two original wordings for only one of the three models. A schedule-by-targeting experiment on the same model gives a small, stable interaction because such sweeps keep the claim wording fixed. Consistency under rewording therefore needs its own test before simulated probabilities are read as beliefs

en