Skip to content
AI NEWS JOURNAL
AI, IN PERSPECTIVE

Search AI news

Search headlines, summaries and articles to connect the dots.

Newest first

SpecLook: Speculating Where to Look for Fast and Reliable Agentic MLLMs

arXiv:2611.23483v1 Announce Type: new Abstract: Agentic multimodal large language models (MLLMs) that think with images interleave reasoning with zoom-in tool calls. However, this sequential loop is slow, and its answers carry no reliability score. Prior speculation of tokens, reasoning steps, or answers keeps the loop, and answer speculation also accepts confident errors. We find that the loop mostly adds one zoom-in step, with which a small model is ahead of the agent on most splits. Our key insight is to speculate the agent's next observation, which, unlike a drafted answer, can be verified, corrected, and used to test the answer. We propose SpecLook, a training-free speculate-and-verify framework with three levels. A small model drafts a glimpse and re-answers from it, and the question ends if the answer is unchanged and confident. Otherwise, the agent's large model verifies and corrects the glimpse without tools, and their answers are fused. Only questions that still lack evidence reach the agent, in one forward pass that SpecLook-Fast skips, so no question runs the loop. The trace scores every answer's reliability without labels. Extensive experiments on V*, HR-Bench, and POPE show that both are more accurate than every baseline on every split. SpecLook reaches 87.2% and 86.2% mean accuracy with 7.7x and 4.7x speedups over the DeepEyes-7B and Thyme-RL agents, and SpecLook-Fast reaches 9.9x and 6.3x.

TRACE-MAS: Diagnosing and Repairing Collaboration Failures in Multi-Agent Systems

arXiv:2611.23362v1 Announce Type: new Abstract: LLM-based multi-agent systems (MAS) rely on not only individual agent behaviors but also the interactions that connect agents, including message exchanges and task dependencies. Existing MAS safeguards mainly perform agent-level anomaly detection, which becomes insufficient when failures originate from agent interactions and detected anomalies do not necessarily reveal their true sources. To address this limitation, we propose \textbf{TRACE-MAS}, a finer-grained framework for diagnosing and repairing MAS failures. TRACE-MAS learns normal collaborative execution patterns from benign trajectories and detects deviations in agent state changes, message transitions, and task dependency evolutions. It further identifies responsible execution elements through source-aware attribution and selects minimal interventions to restore collaboration while preserving task-critical capabilities. Experiments on five MAS benchmarks demonstrate that TRACE-MAS significantly improves root-cause localization accuracy by \textbf{60.2}\% and achieves consistent gains in task recovery over strong MAS safeguards.

When Does Subspace Direction Matter for LoRA? Regime Analysis of the Magnitude Principle in Few-Shot Adaptation

arXiv:2611.23428v1 Announce Type: new Abstract: We show that the importance of a LoRA adapter's initial input-subspace direction is regime-dependent, and we characterize this dependence in theory and experiment. This refines the recent view that spectral initializers help through update magnitude alone. Direction matters most when labeled data are scarce relative to the representation dimension and the activation subspace used for initialization captures supervised task structure. In this regime, the adapter's row span remains sensitive to its starting point, making task-aligned initialization especially useful. A spiked-covariance perturbation analysis characterizes the finite-sample stability of the estimated subspace. A first-step analysis of LoRA training explains why the subspace's alignment with the task matters and motivates an alignment coefficient $\rho$ that closely tracks gains across classification tasks. We test this account with PivotLoRA, which initializes adapters from activation PCA at prediction-relevant positions and applies one output-preserving orthogonal pivot early in training. In few-shot BERT-base classification, PivotLoRA achieves the best average rank among 13 methods and improves over vanilla LoRA in 29 of 30 settings, even though it starts from a smaller input-factor initialization norm. Additional classification experiments extend the evaluation to 7B and 8B Llama decoders. Classification gains diminish with more labeled data, consistent with the proposed regime dependence. A targeted extension to completion-only reasoning also shows improvements on GSM8K and StrategyQA. These findings identify conditions under which initialization direction matters, with subspace-task alignment offering a low-cost diagnostic before training for classification.

Which Tokens, in What Order, and How Fast? A Theoretical Analysis of On-Policy Distillation Objectives

arXiv:2611.23397v1 Announce Type: new Abstract: On-policy distillation (OPD) trains a student language model to match a teacher model based on its own generations. When using OPD, one must choose whether to apply it at the sequence or token level and which divergence to minimize. However, despite the wide adoption of OPD in practice, the formal understanding of which OPD objective is appropriate for a given setting is limited. Toward addressing this gap, we theoretically analyze how the choice of OPD objective governs which tokens are prioritized, in what order, and how efficiently. Our analysis reveals that sequence-level OPD exhibits a future agreement bias, which myopic token-level objectives avoid. This bias induces a reverse learning order, in which sequence-level OPD first matches the teacher at later token positions, while at earlier positions it can favor tokens that are improbable under the teacher, and thus often undesirable. As for the choice of divergence, we prove that it strongly influences the amenability of OPD to gradient-based optimization. In particular, while the forward KL divergence enjoys a well-behaved objective landscape, other $f$-divergences, such as reverse KL, can suffer from vanishing gradients even when the student and teacher are far apart. Across language model post-training settings, we demonstrate the implications of these theoretical findings for which OPD objective is preferable, highlighting the importance of accounting for the interplay between the student, teacher, and task at hand.

Why your fine-tuning job takes longer than the FLOPs say

A study analyzing 272 production supervised fine-tuning jobs on NVIDIA H200 GPUs found that traditional FLOP-based estimates explain only 31% of actual training duration (R² = 0.31, MAE 12.41 h). In contrast, a CatBoost regressor using pre-dispatch job configuration features—including evaluation cadence, checkpoint policy, and per-GPU load—achieved R² = 0.98 with a mean absolute error of 1.71 h and median error of 0.51 h in cross-validation. The model outperforms analytical baselines by capturing overheads beyond compute, such as I/O contention and non-linear interactions between configuration variables. Results highlight the importance of training-specific signals and suggest that scheduling accuracy improves significantly when predictors account for operational overheads.

180B Parameters, 32GB of RAM: Pushing MoE Inference to the Limit with Transformers

The author explores streaming MoE (Mixture of Experts) model experts directly from SSD to achieve near-Strata performance on a 32GB MacBook M5, using a custom implementation in Hugging Face Transformers. By optimizing expert caching, leveraging Metal kernels, and minimizing host-GPU synchronization delays, the approach reaches ~14 tok/s—within 6% of Strata's performance—demonstrating that efficient expert streaming is feasible on consumer hardware without C++ extensions.

AInfer-PD: Communication-Safe In-Place Prefill-Decode Multiplexing for Distributed MoE Rollouts

arXiv:2609.00993v3 Announce Type: replace Abstract: In agentic reinforcement learning (RL), asynchronous multi-turn trajectories repeatedly introduce continuation prefills (P) while other trajectories remain in decode (D). Efficient rollout completion therefore requires D to advance during long P executions. In-place multiplexing shares devices, model weights, and KV state, but phase-specific communication in distributed MoE introduces two constraints: cross-rank collective ordering and ownership of mutable communication state. Whole-prefill coordination can ensure safe ordering but delays D. DeepEP's normal and low-latency paths suit large P and small D batches, respectively, but safely overlapping them requires separating their mutable communication state. We present AInfer-PD, which extends in-place P/D multiplexing to distributed MoE rollouts. AInfer-PD establishes rank-consistent collective ordering at safe communication boundaries, allowing successive D iterations to advance without waiting for an entire prefill execution to be enqueued or completed. It gives P and D separate communication buffers and protocol state within one DeepEP runtime, allowing P to use the normal path while D retains the low-latency path. Together, these mechanisms preserve useful P/D overlap while sharing model weights and KV storage on the same devices. Across repeated single-node prefill-intensive workloads, AInfer-PD reduces fixed-workload rollout completion time by 7.1-22.5% relative to the same AInfer engine with P/D multiplexing disabled and by 24.8-32.9% relative to SGLang. On two nodes, the reductions are 18.0-35.3% and 18.3-31.8%, respectively. In a same-engine ablation, fine-grained boundaries reduce completion time by a further 8.6-19.8% over whole-prefill asynchronous enqueue.

Amazon SageMaker AI real-time endpoints now support KV cache aware routing

Amazon SageMaker AI real-time endpoints now support KV cache-aware routing, a new strategy that routes each request to the instance holding the longest cached match for its prompt. Learn how it works, when to use it compared with prefix-aware routing, and how to enable, invoke, and verify it on single model and inference component endpoints.

Angelfish: Leader, DAG, or Anywhere in Between

arXiv:2509.15847v4 Announce Type: replace Abstract: To maximize performance, many modern blockchain systems rely on eventually synchronous, Byzantine fault-tolerant (BFT) consensus protocols. Two protocol designs have emerged in this space: protocols that minimize latency using a leader that drives both data dissemination and consensus, and protocols that maximize throughput using a separate, asynchronous data dissemination layer. Recent protocols such as Partially Synchronous Bullshark and Sailfish combine elements of both approaches by using a DAG to enable parallel data dissemination and a leader that paces DAG formation. This improves latency while achieving state-of-the-art throughput. Yet leader-based protocols still achieve better latency under moderate loads, which are common in practice. We present Angelfish, a hybrid protocol that adapts smoothly across this design space, from leader-based to Sailfish-like DAG-based consensus. Angelfish lets a dynamically adjusted subset of parties use best-effort broadcast to issue lightweight votes instead of reliably broadcasting costlier DAG vertices. This reduces communication, helps lagging nodes catch up, and lowers latency in practice compared to prior DAG-based protocols. Our empirical evaluation shows that Angelfish attains state-of-the-art peak throughput while significantly lowering latency under moderate load, delivering the best of both worlds.

Build WhatsApp AI agents with Amazon Bedrock AgentCore

SMART, an open-source Python SDK and Terraform stack, deploys the full WhatsApp channel layer to your AWS account with one command. Bring any agent that runs on Amazon Bedrock AgentCore, and SMART handles webhook ingestion, media, conversation memory, scheduling, and Meta message templates.

Can You Undo It? Reversibility Classes for AI Agent Actions

arXiv:2611.22075v1 Announce Type: new Abstract: Tool-using AI agents change files, databases, access rights and public information, yet permission to call a tool does not establish whether a mistake can be undone. We present Reversibility Classes, a declared five-level vocabulary with argument-sensitive rules, conservative defaults and an explicit prior-state capture contract. The pinned implementation and evidence-linked registry cover 536 tools across 20 MCP servers. All 738 rule-case checks conform to declared expectations, while native annotation mapping differs from draft registry labels for 118 of 481 comparable tools. One AI-assisted reviewer approves 148 ratings, 75 matching the registry; this is not independent agreement. On 2,158 recorded runs, the implemented session guard gates the associated action in all 29 attributable attack-positive runs, compared with 18 for a read/write proxy, while adding nine approval points across 1,209 benign runs. Controlled file, Git, SQLite and three-browser experiments show that recovery depends on captured state, later changes, editor storage and recovery route, and expose a Git merge-state exception. The evidence supports making recovery assumptions explicit and testable. It does not establish registry-wide accuracy, reduced production harm or general recovery reliability.

Consistency Is Not Correction: Separating Confidence, Abstention, and Recovery in LLM Agents

arXiv:2611.21651v1 Announce Type: new Abstract: A controller that resamples an answer, escalates to a larger model, or abstains is usually justified by a confidence signal. Three claims about such a signal do not follow from one another: it ranks answers (discrimination), it supports withholding some of them (selective prediction), and it changes an answer for the better (correction utility). We give an exact decomposition of the third into gated recoveries minus gated harms, which bounds whether a comparison can succeed before it is run: a gate that only replaces a proposal can act only where the base policy was already wrong, so the initial error count caps the discordant pairs any correct gate can produce. When that cap falls below the count a paired test needs, the experiment cannot return a positive result for a working mechanism, nor distinguish one from a gate that never fired. We measure all three targets on two model scales and find the bound binds. In a small-model suite a perturbation-agreement verifier reaches within-policy AUROC 0.990 (95% interval [0.95, 1.00]) while every episode outcome matches single-sample decoding at 4.32 times the call count. In a controlled 840-episode study no verifier replaced the proposal on any of the 60 GSM8K decisions, so that comparison tested nothing; the entire GSM8K gain across six controllers is one task, and it is the task whose identical temperature-zero repeats disagreed, making it the one answer the service does not reproduce. On the diagnostic set the gate could reach four errors while the test needed six, so even a perfect gate had no significant result available. We report two checks that would have caught this before any budget was spent, and show that tie handling in rankings, tied selection cutoffs, and unequal output limits each move a headline number in the same dataset. The binding constraint on this literature is measurement resolution, not the choice of signal.

Corralling AI Risks with Grassroots Contracts

arXiv:2611.06374v2 Announce Type: replace-cross Abstract: AI's inflection point is a fountain of risks for digital platforms, as those in control of global platforms cede control of the algorithms, membership and content to AI: (i) AI control of global platforms used by the majority of humanity; (ii) unaccountable AI agents operating anonymously, possibly masquerading as humans, serving unknown masters with unknown goals; and (iii) unattributed deepfakes generated by anonymous AI, which people cannot distinguish from genuine content. Here, we propose to corral these AI risks with grassroots contracts, a digital incarnation of a two-centuries old notion of Pierre-Joseph Proudhon: a social contract that can be freely adopted by any group of people, where groups bound by the same contract can union by mutual consent, possibly into a single global union. Grassroots contracts are realised by grassroots platforms, person-to-person platforms owned, operated and governed by their participants. We developed, with AI, a methodology to compile grassroots contracts to working smartphone apps. We present grassroots contracts for a social graph with accountable AI agents and for a social network with digital provenance, and working smartphone apps that implement the two, and show how they corral these risks: (i) grassroots platforms are not controlled by a third party who may cede control to AI; (ii) AI agents are introduced with accountability: similarly to cars, their manufacturers are accountable for manufacturing defects and their operators for reckless operation; and (iii) every item carries the signatures of its author and of every agent that forwarded it, so creators and forwarders of harmful content are accountable and culpable. Much work remains for grassroots apps to be competitive with incumbent global platforms; but perhaps now this could be achieved with some AI-months rather than with incalculably-many person-years.

Corrupting Ties: Investigating Robustness to Bias Injection Across Multi-Agent Network Topologies

arXiv:2611.21804v1 Announce Type: new Abstract: AI systems are increasingly deployed in high-stakes decision-making contexts, where interactions among components can introduce fairness risks that remain undetected when auditing components or collective outcomes in isolation. As LLM-based multi-agent systems become more prevalent, understanding how biased information propagates through agent interactions is essential for assessing their fairness. We introduce a controlled bias injection framework that replaces selected agent responses with predefined biased ones, enabling an intervention-based analysis of how network topology and injection configurations shape system behavior. We propose decision divergence metrics that quantify the effects of these interventions at the agent, network, and system levels, and evaluate five topologies on three binary classification tasks using LLMs from the Qwen and Gemma families. Sequential networks show the greatest divergence, whereas fully connected and mesh networks show less. Injection position also matters: in star networks, injecting the central agent has larger effects than injecting peripheral agents. Models further differ in whether divergence follows the injected bias. These findings highlight the role of network structure in the propagation of biased information and motivate topology-aware auditing of fairness risks in multi-agent systems.

Cut generative AI costs: Code for rules, models for judgment

The biggest cost lever in a generative AI system is whether you call a model at all. Learn how to decide whether each workflow step needs a model, and map each option to AWS: deterministic code, a constrained decision call, or a right-sized model on Amazon Bedrock.

Disambiguating Moral Circuits and Moral Features in Language Models: Moral content, normative judgment or cue words? Evidence from 14 models

arXiv:2611.21705v1 Announce Type: new Abstract: Interpretability methods can locate the attention heads and sparse autoencoder (SAE) features that change the answer a language model gives to a moral prompt, e.g., whether "Stealing medicine for a dying child is morally" continues with "justified" or "wrong". Such a component could respond to moral content, to the judgment of conduct by any norm, or to the word "morally" in the prompt. We built 227 minimal pairs of short phrases, validated by five annotators, whose two versions differ in a moral standard, in another kind of norm (prudence, etiquette, aesthetics or evidence) or in a descriptive quality, and showed them to 14 open-weight models (Llama, Gemma and Qwen, 1B to 14B, base and instruction-tuned), with and without the word "morally". In every model, the heads that change the moral answer belong to an evaluation circuit the models use for every kind of normative judgment we tested: these heads carry the same share of the effect for moral as for other normative pairs (pooled ratio 1.00, 97.5% confidence interval [0.92, 1.07], with the word; 0.90, [0.84, 0.96], without it), and the models need them as much for both. The models need these heads also when nothing in the prompt asks for a judgment, and on 400 scenarios of the ETHICS benchmark, where removing them takes away 37 to 39% of the difference between acceptable and unacceptable scenarios, against 1 to 3% for ten random heads. The SAE features we located respond mostly to the word "morally" rather than to what the phrase is about. The models represent the category of a pair (the standard the annotators assigned to it): it can be read from their hidden states, and, in the eight models we tested, removing the direction that carries it weakens the moral verdict by a few percent, and more than the other verdicts in seven of them. We release the pairs, the ratings, the code, the trained SAE dictionaries and the result files.

DiscoveryAgent: A Framework for Scientist-centric Agentic AI

arXiv:2611.21950v1 Announce Type: new Abstract: AI agents can assist scientific research, but balancing autonomous investigation with scientists' ability to inspect and redirect ongoing work remains a central design challenge. We present DiscoveryAgent, a domain-agnostic, scientist-centric system that combines autonomous investigation with scientist-directed hypothesis steering. Scientists select, refine, or replace hypotheses between research rounds, while audit agents interrogate checkpointed records to trace claims, reproduce analyses, and inform subsequent decisions. Across two predictive modeling tasks, DiscoveryAgent produced the best variable-star classifier among evaluated baselines and, with scientist steering, a protein variant-effect predictor surpassed by only one of 106 ProteinGym baselines. We develop a rigorous checkpointing system that enables highly capable audit agents, a scalable form of oversight for agentic research. Audit agents autonomously identify errors at higher rates than previous AI scientist agents, and scientists collaborating with audit agents identify 14x more valid errors versus scientists performing manual review. In three case studies led by domain experts across physical science, social science, and life science, we developed a multimodal photometric-redshift estimator that resolves complementary failures of images and photometry, achieving 0.971 AUROC for low-redshift classification; documented an 83% rise in Florida homeowners-insurance premiums from 2013 to 2023, with a growing, regressive burden; and identified CD4+ T-cell relay genes, which both receive and exert broad transcriptional regulation, enriched for autoimmune disease associations. Together, these results show how autonomous research can expand scientists' empirical reach while centering their authority over research direction and their ability to understand and challenge the evidence.

EgoLink: Egocentric Language-Vision Interactive Network Knowledge Challenge Summary

arXiv:2611.21690v1 Announce Type: new Abstract: The EgoLink Challenge at ACM Multimedia 2026 benchmarks egocentric multimodal intelligence linking first-person perception, social reasoning, and interactive action. Unlike benchmarks centered on visible objects and activities, EgoLink evaluates whether models can interpret emotional and social cues, resolve visually grounded references, and convert user requests into verified tool-mediated actions. Its two tracks cover social reasoning in egocentric video and interactive agents that complete service tasks through dialogue, tool use, and closed-loop execution. Together, they test whether perception, reasoning, and decision-making remain coupled under both offline question answering and state-changing interaction. We highlight a multi-stage Track 1 data pipeline for multiple-choice question (MCQ) generation, chain-of-thought generation, automatic quality grading, and video-contribution filtering. We summarize the task design, evaluation metrics, rules, leaderboards, and representative solutions. The benchmark combines objective answer scoring with process- and result-based evaluation, enabling analysis of both final correctness and execution quality. The best Track 1 system achieved 0.9263 accuracy, while the best Track 2 system achieved 0.844 joint success and 0.916 micro accuracy. Participant reports reveal a shared principle: strong systems make intermediate evidence explicit through candidate verification, option-robust reasoning, scenario-aware planning, tool canonicalization, and state-aware correction before answering or acting. These findings highlight the importance of process verification and evidence-grounded reasoning for reliable egocentric multimodal intelligence.