News
The raw AI news stream — everything we ingest across RSS, Hacker News, arXiv, GitHub, Bluesky, YouTube and more, newest first.
| Source | Age | Title | Summary | Status |
|---|---|---|---|---|
arxiv arxiv-ai | 29d | MIPS: Mitigating Mode Collapse in Reinforcement Learning for Large Outcome Spaces | Expected-reward maximization in reinforcement learning concentrates probability mass on a few outcomes, even in tasks that require man… | ↗Not briefed |
bluesky bluesky-ai | 29d | did they ever get around to adding watermarks to LLM outputs? what ever happened to that | did they ever get around to adding watermarks to LLM outputs? what ever happened to that | ↗Not briefed |
arxiv arxiv-ai | 29d | AdCoEvo: A Co-Evolving Multi-Agent Advertising Delivery System | Digital advertising is an important means for platforms to reach potential users and drive business growth. Advertising delivery is th… | ↗Not briefed |
hn hn-ai | 29d | Fastest Vector Index for Cold Queries | — | ↗Not briefed |
arxiv arxiv-ai | 29d | Are You Reading the Right Thing? Rethinking Internal Signals for Prompt-Injection Detection | Several recent studies propose detecting prompt injections from a language model's internal state. However, these methods do not measu… | ↗Not briefed |
bluesky bluesky-ai | 29d | oh yeah i am not imagining this translation path involving an llm in its internals at all. | oh yeah i am not imagining this translation path involving an llm in its internals at all. | ↗Not briefed |
bluesky bluesky-ai | 29d | I'm increasingly fascinated (and frustrated!) by "disclosure statements" that do the bare minimum of acknowledging the use of an LLM but do… | I'm increasingly fascinated (and frustrated!) by "disclosure statements" that do the bare minimum of acknowledging the use of an LLM b… | ↗Not briefed |
arxiv arxiv-ai | 29d | xLightGCN: Explainable Graph Neural Network Based Recommendation Systems with Most Influential Peers | Recommendation systems, systems that predict a user's preference for items in order to curate that user's feed, are a ubiquitous and c… | ↗Not briefed |
bluesky bluesky-ai | 29d | I mean I feel like the way to go there is to design a representation that could be translated INTO a Real(tm) programming language rather t… | I mean I feel like the way to go there is to design a representation that could be translated INTO a Real(tm) programming language rat… | ↗Not briefed |
arxiv arxiv-ai | 29d | Large-Step Last-Iterate Convergence of Random-Reshuffling Stochastic Gradient Descent | Random reshuffling stochastic gradient descent (RR-SGD) is widely used for finite-sum optimization in machine learning. For convex $L$… | ↗Not briefed |
hn hn-ai | 29d | Membrane – Structured persistent memory for local LLMs | — | ↗Not briefed |
arxiv arxiv-ai | 29d | More Than Seeing: Vision as Scalable Anchors in Native Multimodal Pre-Training | Native multimodal pre-training trains a single model from scratch on text and multimodal data, enabling it to understand both language… | ↗Not briefed |
arxiv arxiv-ai | 29d | Integrating Simulation and Chain-of-thought Reasoning in Multimodal-Language Models For Physical Reasoning | Multimodal language models often need to reason about the physical dynamics of the world in order to answer questions appropriately. V… | ↗Not briefed |
arxiv arxiv-ai | 29d | Progress-Aware Distillation for Mitigating Stagnation in Small Language Model Agents | While LLM-based agent systems have demonstrated remarkable proficiency in reasoning and solving complex real-world tasks, they also co… | ↗Not briefed |
arxiv arxiv-ai | 29d | Learning to Select Features with Manifold Deep Unfolding PCA | Unsupervised feature selection (UFS) is of great significance in image processing and pattern recognition. Nevertheless, existing mode… | ↗Not briefed |
bluesky bluesky-ai | 29d | Saudi employee of OpenAI arrested after allegedly planting a recording device in one of the company's San Francisco offices u.afp.com/SCQo | Saudi employee of OpenAI arrested after allegedly planting a recording device in one of the company's San Francisco offices u.afp.com… | ↗Not briefed |
bluesky bluesky-ai | 29d | “Anthropic is warning that some of its artificial intelligence models could soon become “self-replicating” and find ways to infiltrate inte… | “Anthropic is warning that some of its artificial intelligence models could soon become “self-replicating” and find ways to infiltrate… | ↗Not briefed |
arxiv arxiv-ai | 29d | EvoProbe: Evaluation-Driven Benchmark Evolution for LLM Agents | Static benchmarks for large language model (LLM) agents face saturation and data contamination risks, motivating continual updates to… | ↗Not briefed |
bluesky bluesky-ai | 29d | OpenAI :: Who planted a recording device in my business ChatGPT: That sounds important! Let's get started. First: did your security person…🔧 ChatGPT | OpenAI :: Who planted a recording device in my business ChatGPT: That sounds important! Let's get started. First: did your security p… | ↗Not briefed |
bluesky bluesky-ai | 29d | “Perhaps the most relevant issue is that by releasing this trove of material in this manner, the companies are in the process of destroying… | “Perhaps the most relevant issue is that by releasing this trove of material in this manner, the companies are in the process of destr… | ↗Not briefed |
arxiv arxiv-ai | 29d | Recognition Is Not Grounding: Benchmarking Multilingual Financial Evidence Extraction | Comparing financial disclosures across languages requires identifying shared accounting concepts and recovering all numerical facts th… | ↗Not briefed |
arxiv arxiv-ai | 29d | Comprehensive Comparisons of Reinforcement Learning Algorithms in Autonomous Driving | This paper presents a comprehensive comparison of four representative reinforcement learning (RL) methods in autonomous vehicle (AV) d… | ↗Not briefed |
rss techmeme | 29d | Microsoft releases updates fixing 1,400+ security vulnerabilities, surpassing September's record of 972, which was already ~10x the normal monthly patch load (Brian Krebs/Krebs on Security) | Brian Krebs / Krebs on Security: Microsoft releases updates fixing 1,400+ security vulnerabilities, surpassing September's record of 9… | ↗Not briefed |
arxiv arxiv-ai | 29d | How Do LLMs Understand Legal Philosophy? Diagnosing Legal-Philosophical Reasoning in LLMs | Legal reasoning is a fundamental capability for large language models (LLMs) to complete legal AI tasks such as legal judgment predict… | ↗Not briefed |
bluesky bluesky-ai | 29d | Even the damn DOJ press release about the accused OpenAI insider threat actor doesn't link to the damn complaint. Here's the @free.law dock… | Even the damn DOJ press release about the accused OpenAI insider threat actor doesn't link to the damn complaint. Here's the @free.law… | ↗Not briefed |
bluesky bluesky-ai | 29d | I have questions | I have questions | ↗Not briefed |
arxiv arxiv-ai | 29d | Evaluating Agentic Capabilities Before Post-Training: An End-to-End Approach | Agentic capabilities are increasingly important for modern LLMs, yet remain difficult to evaluate during pretraining. Existing approac… | ↗Not briefed |
bluesky bluesky-ai | 29d | Got kicked out of Anthropic because I taught it to throw up | Got kicked out of Anthropic because I taught it to throw up | ↗Not briefed |
bluesky bluesky-ai | 29d | #SchlagZeilenSalat 🍊Es🤡: Richterin verbietet Bäume Abholzen für Golfplatz Bank of America bekommt Milliardenstrafe aufgebrummt Ingineur h… | #SchlagZeilenSalat 🍊Es🤡: Richterin verbietet Bäume Abholzen für Golfplatz Bank of America bekommt Milliardenstrafe aufgebrummt Ingin… | ↗Not briefed |
bluesky bluesky-ai | 29d | they're running a cult over at Anthropic | they're running a cult over at Anthropic | ↗Not briefed |
arxiv arxiv-ai | 29d | VideoHarness-SRSI: Structured Recursive Self-Improvement for Video Understanding Harness | LLM system capabilities depend not only on model parameters but also on the executable harnesses that organize information acquisition… | ↗Not briefed |
bluesky bluesky-ai | 29d | The thumb-sized device, which had a microphone and could connect to a camera, was concealed in toilet paper, according to an FBI complaint… | The thumb-sized device, which had a microphone and could connect to a camera, was concealed in toilet paper, according to an FBI comp… | ↗Not briefed |
bluesky bluesky-ai | 29d | The more I’ve learned about the people behind OpenAI and Anthropic, the more stupid and pathetic they seem. Navel-gazing Silicon Valley bub… | The more I’ve learned about the people behind OpenAI and Anthropic, the more stupid and pathetic they seem. Navel-gazing Silicon Valle… | ↗Not briefed |
arxiv arxiv-ai | 29d | Beyond the Trial-and-Error Loop: Hybrid Projection and Automated Tuning for Distributed Training | Configuring distributed LLM training across hundreds of GPUs requires jointly optimizing tensor, pipeline, expert, context, and data p… | ↗Not briefed |
arxiv arxiv-ai | 29d | AgentSafeEvo: Adversarial Co-Evolution for LLM Agent Safety | Large language model (LLM) agents accomplish complex tasks through multi-turn interactions with external tools. However, malicious con… | ↗Not briefed |
hn hn-ai | 29d | Elm-Toolkit – Collection of Elm Tools for Devs in the Node and TypeScript World | — | ↗Not briefed |
bluesky bluesky-ai | 29d | If you want me to sit in thinking and produce literature on this all day ? put 160k and benefits on somebody’s books for a salary that… | If you want me to sit in thinking and produce literature on this all day ? put 160k and benefits on somebody’s books for a salary… | ↗Not briefed |
bluesky bluesky-ai | 29d | Anthropicが言う「IoTデバイスでLLM」シナリオ、速度の遅さはさほど問題にならなそう。デバイスのパフォーマンス低下による露見を防ぐためにメモリ使用量や動作速度も抑制するだろし。 個々のIoTデバイス上で動作するbotにLLMとしての性能を持たせるうえでの障害は、侵入… | Anthropicが言う「IoTデバイスでLLM」シナリオ、速度の遅さはさほど問題にならなそう。デバイスのパフォーマンス低下による露見を防ぐためにメモリ使用量や動作速度も抑制するだろし。 個々のIoTデバイス上で動作するbotにLLMとしての性能を持たせるうえでの障… | ↗Not briefed |
arxiv arxiv-ai | 29d | From Failures to Learnable Worlds: Policy-Conditioned Environment Generation for Navigation Agents | Simulated environments enable scalable training data generation for embodied agents, but more data does not necessarily provide what a… | ↗Not briefed |
bluesky bluesky-ai | 29d | www.anthropic.com/engineering/... こういうのを読むと、人間がレビューで責任を持つべきだ! みたいのはただのジョブセキュリティなんでは、ってなる | www.anthropic.com/engineering/... こういうのを読むと、人間がレビューで責任を持つべきだ! みたいのはただのジョブセキュリティなんでは、ってなる | ↗Not briefed |
bluesky bluesky-ai | 29d | I don't know if it was asked of an LLM. What was originally claimed was that a GC didn't have to be a lawyer. | I don't know if it was asked of an LLM. What was originally claimed was that a GC didn't have to be a lawyer. | ↗Not briefed |
bluesky bluesky-ai | 29d | (protestant voice) jesus did not assume the LLM nature so therefore LLMs cannot receive justification | (protestant voice) jesus did not assume the LLM nature so therefore LLMs cannot receive justification | ↗Not briefed |
bluesky bluesky-ai | 29d | Meta announced Muse Wonder yesterday. It scored 56 on Artificial Analysis, close to Claude Opus 5.5’s 58. Meta now has both a leading AI a… | Meta announced Muse Wonder yesterday. It scored 56 on Artificial Analysis, close to Claude Opus 5.5’s 58. Meta now has both a leading… | ↗Not briefed |
hn hn-ai | 29d | Early results show 65% fewer input tokens overall | — | ↗Not briefed |
bluesky bluesky-ai | 29d | With regard to AI specifically, there is a lot of talking past one another on here because the AI community isn’t here. Unless you’re moni… | With regard to AI specifically, there is a lot of talking past one another on here because the AI community isn’t here. Unless you’re… | ↗Not briefed |
hn hn-ai | 29d | Visa gets Beijing's green light for bank card business | — | ↗Not briefed |
bluesky bluesky-ai | 29d | I think the president’s brain has been replaced with a 2023 vintage LLM whose next word is always the one most likely to please his immedia… | I think the president’s brain has been replaced with a 2023 vintage LLM whose next word is always the one most likely to please his im… | ↗Not briefed |
hn hn-ai | 29d | Show HN: Brutus CLI – Tiny Dagger to Kill Your Mac's Caesar | — | ↗Not briefed |
arxiv arxiv-ai | 29d | Recursive Formation and First-Class Reuse of Compositional Capabilities | We introduce FormCraft, a framework for representing acquired operations as first-class capabilities that can be recursively composed,… | ↗Not briefed |
bluesky bluesky-ai | 29d | "We expect a future awash with agents running on all types of hardware." | "We expect a future awash with agents running on all types of hardware." | ↗Not briefed |