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 | 17d | The Instruction Tax on Factual Recall: Base Models Can Beat Their Instruction-Tuned Counterparts | We present our solution to the AKBC LM-KBC 2026 Shared Task, where we achieved a score of 0.6836 on the official leaderboard, more tha… | ↗Not briefed |
bluesky bluesky-ai | 17d | Bugged conference room, packed bags for Saudi Arabia. Ex-OpenAI staffer caught in a spying scandal. Who was he really working for? | Bugged conference room, packed bags for Saudi Arabia. Ex-OpenAI staffer caught in a spying scandal. Who was he really working for? | ↗Not briefed |
arxiv arxiv-ai | 17d | DIVE: Benchmarking Full-Duplex Interaction with Visual Engagement in Continuous Streams | Recent advances in multimodal large language models (MLLMs) have substantially improved spoken interaction and video understanding. Ex… | ↗Not briefed |
bluesky bluesky-ai | 17d | “The DHS raised concerns about DeepSeek’s handling of user data, warning that sensitive information from US government users could potentia… | “The DHS raised concerns about DeepSeek’s handling of user data, warning that sensitive information from US government users could pot… | ↗Not briefed |
hn hn-ai | 17d | 12.8B destructible voxels rendered in the browser with Rust WASM custom engine | — | ↗Not briefed |
arxiv arxiv-ai | 17d | MAS-Guard: Unified Guardrails from Individual Agents to Multi-Agent LLM Systems | Agentic AI extends large language models (LLMs) from generating responses to planning and executing complex workflows, substantially e… | ↗Not briefed |
bluesky bluesky-ai | 17d | Former OpenAI worker accused of bugging conference room planned travel to Saudi Arabia: Prosecutors #TheHill https://twp.ai/4hwIL6 | Former OpenAI worker accused of bugging conference room planned travel to Saudi Arabia: Prosecutors #TheHill https://twp.ai/4hwIL6 | ↗Not briefed |
hn hn-ai | 17d | Show HN: Stop Claude Code from compacting mid-task | I run long sessions with Claude. For a long time I've had Claude in the middle of some difficult coding task, and then I see the… | ↗Not briefed |
arxiv arxiv-ai | 17d | When Copying Is Hard: Copy-Constrained Decoding for Exact Span Reproduction | Precisely reproducing a target span verbatim within a long context is a fundamental capability test for LLMs: even when the target str… | ↗Not briefed |
arxiv arxiv-ai | 17d | Self-Credit: Credit Assignment for Long-Horizon Agentic RL via Intra-group Rescaling | LLM agents are commonly trained with RLVR, but the sparse nature of typical RLVR rewards makes credit assignment challenging, especial… | ↗Not briefed |
arxiv arxiv-ai | 17d | VisualEvolve: A Skill Self-Evolution Framework for Visual Reasoning | Visual tool use enables vision-language models (VLMs) to acquire task-relevant evidence, but its effectiveness can be limited by prede… | ↗Not briefed |
hn hn-ai | 17d | I built Dowse, free open-source way to find any file on your PC by describing it | — | ↗Not briefed |
arxiv arxiv-ai | 17d | RACE: RoPE-Aligned Attention Compression Based on Query-Key Activation Energies | Head-dimension compression can reduce the memory required by the key-value (KV) cache, but existing methods often reconstruct full-dim… | ↗Not briefed |
arxiv arxiv-ai | 17d | CAPSafe: A Benchmark for Contextual Action and Plan Safety Reasoning in Embodied Agents | Reasoning about safety is essential for embodied agents powered by large language models (LLMs). We introduce the Contextual Action an… | ↗Not briefed |
hn hn-ai | 17d | Show HN: TinyML AI – A New Architecture to Control Embedded Intelligence | — | ↗Not briefed |
arxiv arxiv-ai | 17d | Object-Oriented Harness: Evolving Agent Harnesses through Object-Oriented Design | Agent harnesses are executable code that specifies the prompts, tools, and control flow around a language model. A single global harne… | ↗Not briefed |
bluesky bluesky-ai | 17d | there's something really funny about people who try to educate you on using LLM tools where 90% of the time what they're telling you to do … | there's something really funny about people who try to educate you on using LLM tools where 90% of the time what they're telling you t… | ↗Not briefed |
hn hn-ai | 17d | Feedback at Every Step | — | ↗Not briefed |
hn hn-ai | 17d | Republic 0.6.0 | — | ↗Not briefed |
hn hn-ai | 17d | Ask HN: Is there a credible gut health test out there? | I've been shopping around for a gut health test / biome test, for both me and my partner (adults), as well as for our young… | ↗Not briefed |
bluesky bluesky-ai | 17d | Here’s me acting as ATC, giving my AI pilot agent a hard time. “Do a 360 for spacing”, “there’s a family of coyotes on 16R, can you land on… | Here’s me acting as ATC, giving my AI pilot agent a hard time. “Do a 360 for spacing”, “there’s a family of coyotes on 16R, can you la… | ↗Not briefed |
arxiv arxiv-ai | 17d | Don't Skip the Baseline: A Controlled Comparison of Dynamic Expert Skipping in MoEs | A growing body of work proposes dynamic expert-skipping methods for Mixture-of-Experts (MoE) models that adjust the number of experts… | ↗Not briefed |
hn hn-ai | 17d | What can and can't be said about revenues at OpenAI and Anthropic | — | ↗Not briefed |
arxiv arxiv-ai | 17d | Relative Representations Enable Portable White-Box Monitoring Across Language Models | Independently trained autoregressive language models often acquire similar capabilities, but it remains unclear whether their internal… | ↗Not briefed |
arxiv arxiv-ai | 17d | Context-Aware Prompt Injection Detection via Small Language Models in RAG Pipelines: nokast-secureRAG | Local Retrieval-Augmented Generation (RAG) systems are vulnerable to indirect prompt injection and retrieval poisoning, where maliciou… | ↗Not briefed |
hn hn-ai | 17d | Finops-scan – Open-source CLI that ranks your top AWS cost waste | — | ↗Not briefed |
arxiv arxiv-ai | 17d | FALCA: Falsification-Aware Reinforcement Learning with Hierarchical Credit Assignment for Long-Horizon Deep Search Agents | Long-horizon deep search requires LLM agents to iteratively decompose complex information needs, gather evidence from external sources… | ↗Not briefed |
arxiv arxiv-ai | 17d | Confessions of a Masked MDP | In reinforcement learning, one presumes the action set to be known ahead of time, and, thus, one must know if any state-action pairs a… | ↗Not briefed |
rss marktechpost | 17d | Google Open-Sources TorchTPU, a Native PyTorch Backend for Cloud TPUs | Google and Meta open-sourced TorchTPU, a native PyTorch backend for Cloud TPUs. It runs eager by default, compiles through torch.compi… | ↗Not briefed |
hn hn-ai | 17d | Safe REA Skill – open-source reverse engineering in a double-layer sandbox | — | ↗Not briefed |
arxiv arxiv-ai | 17d | CountVideoBench: A Benchmark to Evaluate Prediction Biases on the Object Counting task by Video-Language Models | Object counting in videos remains a poorly characterized capability of multimodal large language models. Counting questions are includ… | ↗Not briefed |
arxiv arxiv-ai | 17d | Beyond the Point Mass: Acceptance-Derived Weights and Distributional Losses for Speculative Decoding | In speculative decoding, a lightweight draft model proposes tokens that the target model verifies in parallel, speeding up generation… | ↗Not briefed |
rss mit-tech-review-ai | 17d | The new economics of enterprise AI | Imagine your company just discovered electricity. At first only a few machines run on it, and nobody worries about efficiency — consum… | ↗Not briefed |
arxiv arxiv-ai | 17d | S-Agent 2.0: Spatial Reasoning with a General Coding Agent | Spatial reasoning asks a model to work out where objects are in 3D space, how they relate to one another, and how they move. Agentic V… | ↗Not briefed |
rss huggingface-blog | 17d | ALTK-Evolve: How learned memory tops the τ²-bench leaderboard on customer service tasks without a frontier model | — | ↗Not briefed |
bluesky bluesky-ai | 17d | maybe a big part of the problem with LLM prose editing is that we hand them the entire text and that's not how people read. so I'm experime… | maybe a big part of the problem with LLM prose editing is that we hand them the entire text and that's not how people read. so I'm exp… | ↗Not briefed |
arxiv arxiv-ai | 17d | RayCo4D: Ray-Conditioned Continuous Motion Model for Monocular 4D Dynamic Reconstruction | Reconstructing dynamic 3D scenes from monocular video requires inferring motion from noisy and intermittently visible observations. Lo… | ↗Not briefed |
bluesky bluesky-ai | 17d | my response next time will be up instantly delete everything and block list through llm | my response next time will be up instantly delete everything and block list through llm | ↗Not briefed |
arxiv arxiv-ai | 17d | Gated Prototype Transfer for Zero-Support Medical Vision Language Model | Realistic few-shot adaptation of medical vision-language models (VLMs) can leave low-prevalence classes without any labeled support. T… | ↗Not briefed |
bluesky bluesky-ai | 17d | 10月21日 「あかりの日」 「バック・トゥ・ザ・リサイクルの日」 ですよ✨ 2015/10/21に戻れたら何したいですか🤔 【 参加者募集中 】 ブログ・日記・イラスト・写真・小説をBlueskyアカウントで投稿・販売できる複合型コンテンツサービス ➡️ atelie… | 10月21日 「あかりの日」 「バック・トゥ・ザ・リサイクルの日」 ですよ✨ 2015/10/21に戻れたら何したいですか🤔 【 参加者募集中 】 ブログ・日記・イラスト・写真・小説をBlueskyアカウントで投稿・販売できる複合型コンテンツサービス ➡️ a… | ↗Not briefed |
bluesky bluesky-ai | 17d | Anthropic has already stolen my book and about half a million other authors' books. Its products are theft. | Anthropic has already stolen my book and about half a million other authors' books. Its products are theft. | ↗Not briefed |
hn hn-ai | 17d | Pencil versus Prompt | — | ↗Not briefed |
hn hn-ai | 17d | Debloat Any Android or Google TV over Wi-Fi and Get an Apple TV-Style Home | — | ↗Not briefed |
arxiv arxiv-ai | 17d | Characterizing Cross-Relay Generalization Failure in Machine Learning-Based Power System Protection | Machine learning-based fault detectors are often evaluated at the relay locations used for training, although local voltage and curren… | ↗Not briefed |
hn hn-ai | 17d | Show HN: Jev Cards, Can you beat a vision model at a picture quiz? | I built a quiz game around TypeSafe's new Jev model, which has vision capabilities. Jev looks at a picture and a question and pic… | ↗Not briefed |
hn hn-ai | 17d | Show HN: Penmentor – Claude marks up your handwritten math like a teacher | — | ↗Not briefed |
hn hn-ai | 17d | Launch HN: Respan (YC W24) – Open-source Rust AI gateway built around plugins🔧 Groq | Hey HN, this is Andy from Respan (<a href="https://www.respan.ai">https://www.respan.ai</a>). We are building an o… | ↗Not briefed |
rss wired-ai | 17d | An AI Agent Joined a Listserv of Top Internet Researchers. The Reply-All Chaos Is Still Unfolding | Hysteria over the agent's appearance in a 7,500-person group reveals that internet experts have outdated notions of privacy. | ↗Not briefed |
arxiv arxiv-ai | 17d | DeclineCraft: A RAG Abstention Benchmark for Jointly Measuring Combined Retrieval and Parametric Knowledge | Reliable retrieval-augmented generation (RAG) requires recognizing when neither retrieved documents nor parametric knowledge supports… | ↗Not briefed |
hn hn-ai | 17d | In-memory privacy shield to prevent context leaks in MCP tools | — | ↗Not briefed |