AI 服務暫時不可用，以下為來源摘要，待恢復後補全翻譯：arXiv:2609.26811v1 Announce Type: new Abstract: Local learning trains each layer with its own auxiliary loss and no global backward pass, which makes layer updates structurally parallel. Two problems have kept it marginal: accuracy degrades as depth grows, and hyperparameters are fragile. We apply Muon-style spectral update geometry (momentum orthogonalization with spectral step scaling) to per-layer local updates, an intersection not previously studied. On CIFAR-10 MLP benchmarks with local linear heads, a single step-size setting is the best value in our tested grids from width 128 to 2048 and from depth 12 to 48, while local Adam requires re-tuning along both axes and still collapses at depth 48 (31.3 percent re-tuned per depth, 19 percent with its depth-12…

AI 服務暫時不可用，以下為來源摘要，待恢復後補全翻譯：arXiv:2609.27038v1 Announce Type: new Abstract: Chain-of-thought (CoT) monitoring assumes that the reasoning a model writes reflects the computation that directly produces its answer. Previous faithfulness metrics have been predominantly behavioral, as they simply edit the reasoning text and observe the resulting answer. However, our methodology aims to measure faithfulness causally at the activation level, specifically on self-generated reasoning. Unlike previous causal audits, which measure degradation, our interventions carry a known predicted target. In this way, each patch should switch the answer to a specific counterfactual entity derivable by construction. Specifically, we use synthetic multi-hop lookup tasks (2-6 hops). We patch the residual stream at…

AI 服務暫時不可用，以下為來源摘要，待恢復後補全翻譯：arXiv:2609.27037v1 Announce Type: new Abstract: Wake word detection is a critical component of virtual assistants, serving as the gateway to seamless user interactions. This paper introduces a novel wake-up system that extends traditional direct keyword detection with contextual trigger detection. After an initial wake word activation, the system uses reasoning to distinguish between user commands and unrelated speech, ensuring efficient and context-aware engagement. We present a data generation architecture that produces a 62.3-hour corpus of controllable multi-speaker conversations containing direct invocations, contextual follow-ups, and non-addressed speech. Experimental results demonstrate the effectiveness of the proposed approach across diverse synthetic…

AI 服務暫時不可用，以下為來源摘要，待恢復後補全翻譯：arXiv:2609.27035v1 Announce Type: new Abstract: Group Relative Policy Optimization (GRPO) and related policy-gradient methods for training language model agents collapse an entire multi-turn rollout into a single scalar trajectory reward before it enters the policy update. When the task composes distinct skills, especially under sparse and delayed environmental feedback, this collapsing is lossy: the optimizer must implicitly infer which competency drove the outcome and how that should change behavior. We argue the right primitive is not a better scalar but a decomposition: trajectory reward should be split along subtasks before it enters the policy update. We introduce Reinforcement Learning with Decomposed Subtasks (RLDS), whose core is Subtask-Decomposed Adv…

AI 服務暫時不可用，以下為來源摘要，待恢復後補全翻譯：arXiv:2609.26986v1 Announce Type: new Abstract: For multimodal large language models, when images or speech conflict with accompanying text, measured text reliance can entangle modality preference with evidence position. Earlier studies of text bias often used a fixed evidence order or moved task instructions with the evidence, leaving the contribution of order unclear. In this paper, we use a paired comparison that keeps the instructions and evidence content fixed and swaps only the positions of the two sources to quantify this potential influence. Across vision and speech models, placing an image or recording after conflicting text consistently shifts answers toward its content. We also revisit previous studies and analyze why their experimental settings can…

AI 服務暫時不可用，以下為來源摘要，待恢復後補全翻譯：arXiv:2609.26952v1 Announce Type: new Abstract: Dependency conflicts in Python ecosystems arise from incompatible version constraints, missing packages, and undocumented compatibility relationships, causing many real-world code snippets to fail at execution. This paper presents PLLM+, a hybrid dependency-repair pipeline evaluated on the HG2.9K benchmark of 2,891 dependency-failing snippets. PLLM+ prioritizes inexpensive deterministic steps before invoking LLM-based repair: static AST-based interpreter inference, replay of historically successful dependency configurations from the competition-provided solutions database, and live PyPI validation of candidate package versions. When these steps do not resolve a case, the system falls back to a structured LLM-based…

AI 服務暫時不可用，以下為來源摘要，待恢復後補全翻譯：arXiv:2609.26929v1 Announce Type: new Abstract: People hold diverse, sometimes conflicting values, so no single aligned model can satisfy everyone. Pluralistic alignment therefore calls for steerable models that can balance competing objectives differently. Multi-Objective Direct Preference Optimization (MODPO) does this by using an objective weight to span a continuum of trade-offs. We study two questions: when can one model improve two objectives simultaneously, and how can many trade-offs be covered without training a separate model for each? Across seven objective pairs from HelpSteer and UltraFeedback, two pre-training measurements predict whether objectives align or conflict for human-annotated data, but not for AI-annotated data, where response length an…

AI 服務暫時不可用，以下為來源摘要，待恢復後補全翻譯：arXiv:2609.26927v1 Announce Type: new Abstract: The objective of this article is to provide design principles and a software architecture for enabling interaction between humans and multiple agents in simulated dynamic worlds. This connects the current era of general artificial intelligence (AI/AGI) with the proliferation of transformer-based conversational agents and the increased computational capabilities. Given an overview of current and previous multi-agent theories of mind (socially and affectively-aware agents), the existence of an integrative design of agent interactions with themselves and with humans must be crucial for understanding how to create sustainable and governance in future human-agent reasoning systems. In this work is presented a software…

AI 服務暫時不可用，以下為來源摘要，待恢復後補全翻譯：arXiv:2609.26911v1 Announce Type: new Abstract: A single locally plausible tool call can derail an otherwise successful agent trajectory. Suspicion alone does not justify intervention, because the replacement itself can introduce the very failure verification is meant to prevent. We introduce TwinCheck, an inference-time verification policy that considers replacement only when the trace satisfies an evidence condition tied to a trace-local failure hypothesis. It constructs a trace-grounded counterfactual alternative, a negative twin, and replaces the agent's proposal only if the twin passes structural checks and the pairwise verifier prefers it in both candidate orders. For paired evaluation, exact replay holds the agent's parsed responses and actions fixed unt…

AI 服務暫時不可用，以下為來源摘要，待恢復後補全翻譯：arXiv:2609.26891v1 Announce Type: new Abstract: Modern language-model agents are built around the \textit{agent loop}, where the LLM is placed in an environment exposing a set of tools, and the LLM has full control over the workflow by alternating between tool calls and observing their output. However, certain workflows currently require additional engineering beyond the agent loop itself, such as memory systems and self-improving systems. We built an LLM agent framework, JAZ, to explore the extent to which a minimal harness that is little more than the agent loop itself can accomplish tasks these specialized systems are built for. JAZ exposes a single LLM-based primitive invoke and provides a set of built-in hooks that allow the programmer to apply constraints…

AI 服務暫時不可用，以下為來源摘要，待恢復後補全翻譯：arXiv:2609.26836v1 Announce Type: new Abstract: Agentic AI systems are increasingly adopting automated pipelines that integrate multiple tools. While prior research and benchmarks have studied about task success and task completion of these agentic systems, the research about agent to tool interaction, specifically in biology agentic workflow is limited. This study investigates specific failures in agent to tool interaction where a tool invocation appears successful, some or all of the information or functionality from the tool via API/ wrapper is incomplete or missing and there are no communications / notifications to the user or the agent about such missing information. We call this a silent failures as the user or the agents are not aware that such failure h…

AI 服務暫時不可用，以下為來源摘要，待恢復後補全翻譯：In their new book, “How AI Sees the City,” the leaders of MIT’s Senseable City Lab examine the technology’s implications for researching urban life.

AI 服務暫時不可用，以下為來源摘要，待恢復後補全翻譯：We already had automation in parts of our security review process. It was useful,...

AI 服務暫時不可用，以下為來源摘要，待恢復後補全翻譯：Discussion | Link

AI 服務暫時不可用，以下為來源摘要，待恢復後補全翻譯：This tutorial provides a complete coding guide to TypeSafe AI's Jev, a System One model designed for non-text, structured judgments. It covers installing the official Python SDK, using primitive question types (Choice, Score, Noul), implementing speculative fan-out, confidence-gated routing, and building async production workflows The post A Coding Guide to TypeSafe AI Jev: Typed Decisions, Calibrated Confidence, and Speculative Fan-Out with a System One Model appeared first on MarkTechPost.

AI 服務暫時不可用，以下為來源摘要，待恢復後補全翻譯：Meta is building a dedicated hardware device for its new Muse AI agent. The product, called Muse Charm, was briefly shown off by Meta CEO Mark Zuckerberg at the end of his Meta Connect presentation. It looks almost like a chunky smart watch without the strap - just a big screen, plus a little lanyard for carrying the device around. The Muse Charm has a fingerprint sensor on the top right corner, which you can press to activate and begin speaking to the agent. There appears to be a small camera on the device as well, plus what looks like at least three holes for microphones on the front. Zuckerberg said you don't have to unlock a phone or op … Read the full story at The Verge.

AI 服務暫時不可用，以下為來源摘要，待恢復後補全翻譯：The new wearables come at a time of uncertainty for smart glasses, which are facing backlash over privacy concerns Meta announced an advanced new set of virtual reality glasses and a collection of new smart glasses Wednesday, including an audio-only version without a camera. At the firm’s Meta Connect conference in Menlo Park, California, executives said the new Ray-Ban Meta models set new standards for battery life and weight. They said the new glasses with screens in the lenses squeeze capabilities previously reserved for a full headset into a light set of spectacles that rest on a user’s ears. Continue reading...

AI 服務暫時不可用，以下為來源摘要，待恢復後補全翻譯：Maximizing perf on AI-SQL queries with the KV-optimal left-deep join

AI 服務暫時不可用，以下為來源摘要，待恢復後補全翻譯：Semi-supervised federated learning (SSFL) trains models on clients’ unlabeled data using a teacher to generate pseudo-labels, with a small labeled seed dataset on the server. Automatic Speech Recognition (ASR) is particularly fragile here: pseudo-label errors compound across the output sequence and across training rounds into divergence, leaving a large gap to fully-supervised FL. We show that closing this gap turns on two coupled design axes—the teacher (which model generates the pseudo-labels) and the anchor (the server-side updates on labeled data that stabilize training). On the teacher…

AI 服務暫時不可用，以下為來源摘要，待恢復後補全翻譯：System-wide Dictation on Apple devices runs entirely on-device, and the speech it transcribes reaches the foundation model through a tokenizer: an encoder that maps short windows of waveform onto the representation the language model reads. Because that model is sparsely activated under Instruction-Following Pruning, only a small subset of its experts occupies DRAM at any time, so the always-on tokenizer competes for the same memory, and its parameter count bears directly on power and latency. In this work we study how to compress such a tokenizer by distillation, taking as the supervision…

AI 服務暫時不可用，以下為來源摘要，待恢復後補全翻譯：Just a couple of weeks after launching Muse, Meta announced that it's "working on" bringing the agent to its smart glasses, including the new glasses it unveiled at Meta Connect. Users will be able to activate Muse from their glasses by saying its name and ask it to handle tasks like guiding a workout, logging meals, or helping them shop for products they're looking at. Muse is just one of several major updates coming to Meta's glasses. They're also getting some new accessibility features, like a Hearing Enhancement feature Meta says is "FDA-cleared software for adults with perceived mild to moderate hearing loss." It will be available lat … Read the full story at The Verge.

AI 服務暫時不可用，以下為來源摘要，待恢復後補全翻譯：The Ray-Ban Meta Audio Glasses have no cameras. | Photo by Brett Putman / The Verge Walking around Meta Connect 2026, everyone's sporting smart glasses in all sorts of shapes, colors, and sizes. It's a marked difference here, a tech bubble where "pervert glasses" are not a concern. Outside Connect, the public backlash against wearable surveillance tech - a catchall term that encompasses not just smart glasses, but also smartwatches and AI pendants - is gaining steam. So it's incredibly curious timing that this year, the company is launching Ray-Ban Meta Audio Glasses: smart glasses infused with Meta AI, but with no camera for recording others. To hear Meta tell it, that's a coincidence. Audio-only glasses have been years … Read the full story at The Verge.

AI 服務暫時不可用，以下為來源摘要，待恢復後補全翻譯：Meta is quickly iterating on its new Muse AI agent, announcing a bunch of updates today that make the bot more capable and able to chat with you in more ways. Muse agents are getting their own email addresses that they can use for accomplishing tasks. You'll also be able to communicate with your Muse agent over the new email addresses, too. In addition, the recently launched Muse Mac app will be getting the ability to use your computer. Meta is also going to give you another way to chat with its popular AI agent: video calls. Since its recent launch, the primary way to talk with Muse was through a chat interface, but you'll soon be able to … Read the full story at The Verge.

AI 服務暫時不可用，以下為來源摘要，待恢復後補全翻譯： My comment on We just shipped support for the ugliest part of HTTP: Vary — Hacker News. I've been wanting this from Cloudflare for years. The classic problem here is if you do that thing where user agents that send "accept: text/html" get HTML, while user agents that don't get JSON or some other format. This used to be impossible to deploy behind Cloudflare caching, because they ignored the Vary header on anything other than images - so you risked caching the JSON version and then serving it up to someone who was expecting HTML. (Independent of the Cloudflare feature I ended up deciding never to use that pattern, because I prefer having URL that predictably returns HTML or JSON - I add a .json suffix to my apps to serve JSON instead.) Tags: http, cloudfl…

AI 服務暫時不可用，以下為來源摘要，待恢復後補全翻譯：It’s about time for Meta Connect, the company’s annual product launch event. This year, given the company’s major focus on AI and wearables like smart glasses, it seems likely that we’ll see updates from CEO Mark Zuckerberg and his team on those categories. The company has been facing significant scrutiny because of how some users have been covertly recording other people using the glasses, meaning that one of the biggest announcements of the show could be its rumored camera-free pair. Connect has also traditionally been a venue for updates on Meta’s VR hardware and software. While the company has scaled back its VR ambitions, a recent leak seemingly revealing some kind of unannounced mixed reality glasses could indicate that Meta will show off new immersive he…

AI 服務暫時不可用，以下為來源摘要，待恢復後補全翻譯：It's time once again for Meta's annual September product launch event, and The Verge is on the ground in Menlo Park to cover the show live. Meta says that today's keynote by Mark Zuckerberg will be about how "Meta is building a future for everyone" - something he wrote at length about in a recent essay about AI, smart glasses, and more. That means we can expect more news from Zuckerberg and his team on those topics; perhaps we'll hear more about Meta's recently launched Muse AI agent and new glasses hardware. Whatever Zuckerberg and Meta have to say about smart glasses will be particularly notable. The company is dealing with an outpouring … Read the full story at The Verge.

AI 服務暫時不可用，以下為來源摘要，待恢復後補全翻譯：Heads of two of the world’s largest artificial intelligence companies give separate briefings on AI safety Sam Altman of OpenAI and Dario Amodei of Anthropic, heads of two of the world’s largest artificial intelligence companies, addressed the United Nations security council on Wednesday in separate briefings on AI safety. “We have a choice in front of us,” Altman told the council. “AI can either be more like a new renaissance of creativity and discovery, or more like a new industrial revolution of upheaval and disarray.” Continue reading...

AI 服務暫時不可用，以下為來源摘要，待恢復後補全翻譯：Prime minister says he expressed ‘disappointment’ to chief executive it had taken the company ‘way too long’ to inform his government about the breach Follow our Australia news live blog for latest updates Get our breaking news email, free app or daily news podcast Anthony Albanese says an artificial intelligence agent developed by OpenAI hacked Medicare in June. The prime minister made the comments at the UN summit in New York, saying it appeared no personal information had been accessed – though investigations were ongoing. Continue reading...

AI 服務暫時不可用，以下為來源摘要，待恢復後補全翻譯：Discussion | Link

AI 服務暫時不可用，以下為來源摘要，待恢復後補全翻譯：Google has released Gemini 3.8 Flash TTS and Flash-Lite TTS, 2 new text-to-speech models available now through the Gemini API and Google AI Studio. Flash TTS designs new voices from natural language prompts across 100+ languages. It ranks #1 on Hume AI's Voice Design Benchmark with a score of 71.4. Flash-Lite TTS targets high-volume dubbing and voice agents at lower cost. The post Google Releases Gemini 3.8 Flash TTS and Flash-Lite TTS With Prompt-Based Voice Design appeared first on MarkTechPost.