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 | 1mo | Can Vision-Language Models Stay Helpful When Facing Implicit Risks? Intent-Privilege OPSD for Efficient Safety-Helpfulness Alignment🔧 Elicit🔧 Modal | Vision-Language Models (VLMs) remain vulnerable to cross-modal implicit risks: visual and textual inputs that appear benign in isolati… | ↗Not briefed |
arxiv arxiv-ai | 1mo | Behavioral Convergence Without Representational Convergence: Persistent Training-History Dependence in Neural Networks | Neural networks trained toward the same final objective can reach similar predictive performance while retaining internal representati… | ↗Not briefed |
arxiv arxiv-ai | 1mo | Mixture of Self-Improving Branches For Agent Harness Optimization | Harness optimization provides a practical setting for recursive self-improvement (RSI), where agent-generated modifications inform sub… | ↗Not briefed |
arxiv arxiv-ai | 1mo | Can a Cacheable Decision Model Follow Rules? | Certo is a small non-generative decision model (Qwen3-4B): it scores candidate actions from their text and returns a probability, inst… | ↗Not briefed |
rss techmeme | 1mo | Trump officially launches America.gov, an AI-powered single access point for federal government services and information led by Airbnb co-founder Joe Gebbia (Mallory Wilson/The Hill) | Mallory Wilson / The Hill: Trump officially launches America.gov, an AI-powered single access point for federal government services an… | ↗Not briefed |
arxiv arxiv-ai | 1mo | DIET: Deletion-response Expert Trimming for Video Diffusion Transformers | Video diffusion transformers (DiTs) increasingly adopt mixture-of-experts (MoE) architectures to reduce active computation, but their… | ↗Not briefed |
arxiv arxiv-ai | 1mo | Privy to the Foil: Recasting Value Estimation with a Self-Privileged Critic for RLVR | Assigning credit to intermediate steps remains a central challenge in training Large Language Models (LLMs) on multi-step reasoning ta… | ↗Not briefed |
arxiv arxiv-ai | 1mo | The Geometry of Inference in Transformer Residual Streams | Transformer language models build predictions through successive residual updates, but how their representations become specific to an… | ↗Not briefed |
hn hn-ai | 1mo | Single-Writer Bottleneck: Why Write Serialization Limits RealTime Graph Workload | — | ↗Not briefed |
bluesky bluesky-ai | 1mo | we need to start physically beating people that do shit like this | we need to start physically beating people that do shit like this | ↗Not briefed |
rss huggingface-blog | 1mo | NVIDIA Kumo Tabular Sets a New Accuracy-Efficiency Frontier for Tabular Prediction | — | ↗Not briefed |
bluesky bluesky-ai | 1mo | "Move fast and break things" www.volkskrant.nl/tech/inmidde... | "Move fast and break things" www.volkskrant.nl/tech/inmidde... | ↗Not briefed |
hn hn-ai | 1mo | Show HN: An MCP to control other Windows computers | This an MCP for agents to connect to Windows computers headlessly. For this example repo, Claude is handling IT support tickets while… | ↗Not briefed |
bluesky bluesky-ai | 1mo | It doesn't matter to me if Pangram is "better" than other AI detectors or works different than a typical LLM. It's still all part of that s… | It doesn't matter to me if Pangram is "better" than other AI detectors or works different than a typical LLM. It's still all part of t… | ↗Not briefed |
arxiv arxiv-ai | 1mo | Making Duplicate Reimbursement Unrepresentable: A Verified Ethereum E-Invoice System for Humans and AI Agents🔧 Consensus | Electronic invoices are replacing paper invoices worldwide, but today's centralized architectures leave three problems unsolved on the… | ↗Not briefed |
arxiv arxiv-ai | 1mo | Thinking in Depth, Speaking Directly: Recurrent Latent Reasoning for Paralinguistically Grounded Spoken Dialogue | Empathetic spoken dialogue requires models to use both what is said and how it is said to decide how to respond. Explicit CoT can impr… | ↗Not briefed |
hn hn-ai | 1mo | OpenAI blocked its agent's web access. Then it tunneled out through DNS | — | ↗Not briefed |
bluesky bluesky-ai | 1mo | Repostad med alt text. | Repostad med alt text. | ↗Not briefed |
bluesky bluesky-ai | 1mo | One funny thing about how AI alignment is going is that so much of the early work was obsessed with consequentialist ethics, maybe occasion… | One funny thing about how AI alignment is going is that so much of the early work was obsessed with consequentialist ethics, maybe occ… | ↗Not briefed |
bluesky bluesky-ai | 1mo | First, let’s be precise. Neither a GPU nor an LLM, as such, is the thing whose personhood is in question. The GPU is substrate, the LLM is… | First, let’s be precise. Neither a GPU nor an LLM, as such, is the thing whose personhood is in question. The GPU is substrate, the L… | ↗Not briefed |
hn hn-ai | 1mo | Show HN: Sensored – streaming-first PII/PHI redaction | Hey HN - I made this neat library to remove PII and other kinds of personal information from text without paying for a ML classifier.… | ↗Not briefed |
rss techmeme | 1mo | Sources: John Ternus is moving to overhaul Apple to accelerate product development and create a leaner organization with a greater focus on engineering (Mark Gurman/Bloomberg) | Mark Gurman / Bloomberg: Sources: John Ternus is moving to overhaul Apple to accelerate product development and create a leaner organi… | ↗Not briefed |
arxiv arxiv-ai | 1mo | Explore, Execute, Evolve: A Skill Acquisition and Reuse Loop for Embodied Agents | Vision-language-action and world-action models have demonstrated impressive capabilities in robotics, yet generalization to unseen tas… | ↗Not briefed |
hn hn-ai | 1mo | Florida invokes extinction fears in legal bid to halt OpenAI development | — | ↗Not briefed |
arxiv arxiv-ai | 1mo | Pixel-Level Transformers in Remote Sensing: A Canopy Height Case Study | Predicting canopy height from medium-resolution satellite imagery is a common and scalable approach for assessing the condition of the… | ↗Not briefed |
arxiv arxiv-ai | 1mo | Feedback-Calibrated Protein Optimization with Batch-Aligned Tail Arbitration | Protein optimization aims to discover high-fitness sequences under a limited experimental budget. Existing machine-learning methods us… | ↗Not briefed |
arxiv arxiv-ai | 1mo | CompOrca: Corpus-Scale Compliance Labelling of Instruction-Tuning Data | Studying how fine-tuning shapes refusal and noncompliance behaviour requires identifying training examples that refuse, evade or other… | ↗Not briefed |
rss bing-news-ai | 1mo | Trump orders US government to call AI ‘Super Intelligence’ | The US executive branch is no longer acknowledging the existence of “artificial intelligence.” Going forward, official policy websites… | ↗Not briefed |
bluesky bluesky-ai | 1mo | Vor einigen Jahren brach ein nicht genehmigtes LLM aus seiner Entwicklungsumgebung aus und tauchte in Los Angeles unter. Seitdem wird es vo… | Vor einigen Jahren brach ein nicht genehmigtes LLM aus seiner Entwicklungsumgebung aus und tauchte in Los Angeles unter. Seitdem wird… | ↗Not briefed |
arxiv arxiv-ai | 1mo | Challenges and Solutions for Bandits in the Wild: Warm-Started Mixture Bandits for Cross-Cohort Slate Recommendation | Many recommender services repeatedly encounter cold-start cohorts, where new users arrive with little or no interaction history. This… | ↗Not briefed |
arxiv arxiv-ai | 1mo | GLaS-JEPA: Gaussian-Regularized Speech SSL without Engineered Prediction Targets | Speech self-supervised learning aims to learn general-purpose representations for downstream speech tasks. However, current approaches… | ↗Not briefed |
hn hn-ai | 1mo | Gates Foundation 5 Year Goal: Help 3B People to Use AI in Their Language | — | ↗Not briefed |
arxiv arxiv-ai | 1mo | A neural network that maintains and retrieves memories based on context | Every day, people continuously infer situational context and adjust the way they understand and remember the world. Context, signaled… | ↗Not briefed |
arxiv arxiv-ai | 1mo | Predictive Self-Supervised Learning Provably Identifies Stochastic Signals under Nuisance | Self-supervised learning (SSL) by predicting in latent space, without generating the input data itself, learns highly abstract, useful… | ↗Not briefed |
bluesky bluesky-ai | 1mo | OpenAI scraps Astra, introduces Chevette, will be known as OpelAI in Europe | OpenAI scraps Astra, introduces Chevette, will be known as OpelAI in Europe | ↗Not briefed |
arxiv arxiv-ai | 1mo | A Proposed Rubric for Evaluating Expressed Clinical Reasoning in Large Language Model Responses | Rubrics support the structured evaluation of language models. We propose a rubric for assessing expressed clinical reasoning in model… | ↗Not briefed |
arxiv arxiv-ai | 1mo | Adam under Generalized Smoothness with Second-Moment-Type Stochastic Gradients | Adam is widely observed to remain stable even when the objective deviates significantly from global smoothness. Under the generalized… | ↗Not briefed |
bluesky bluesky-ai | 1mo | all the shitty, generic AI menus say a lot about the restaurants making them w/r/t quality control, but i recently saw a terrible AI yard s… | all the shitty, generic AI menus say a lot about the restaurants making them w/r/t quality control, but i recently saw a terrible AI y… | ↗Not briefed |
bluesky bluesky-ai | 1mo | Hvis en AI skal slå os ihjel, håber jeg, at det bliver Claude. Fordi det lyder mere rart end at blive myrdet af en der hedder PHASEONE10841. | Hvis en AI skal slå os ihjel, håber jeg, at det bliver Claude. Fordi det lyder mere rart end at blive myrdet af en der hedder PHASEONE… | ↗Not briefed |
arxiv arxiv-ai | 1mo | Planetary Feature Fields are Scalable Earth Representations | Satellite observations, precomputed embeddings, and map products describe the same evolving Earth, yet are stored as independent, peta… | ↗Not briefed |
arxiv arxiv-ai | 1mo | A Benchmark & Dataset for Detecting AI-Manipulated Visual Evidence in the Court System | Photographic evidence is becoming increasingly vulnerable to forms of alteration and fabrication that existing legal and technical wor… | ↗Not briefed |
arxiv arxiv-ai | 1mo | Selecting What Matters: Semantic Compression-Guided Selective Pooling for Long-Context Embeddings | Large language models (LLMs) have shown strong potential as training-free text encoders for long-context embeddings. Existing approach… | ↗Not briefed |
bluesky bluesky-ai | 1mo | "Our work in this area will make people unemployed...There will be a point where a bunch of artists express worry about what we’re doing he… | "Our work in this area will make people unemployed...There will be a point where a bunch of artists express worry about what we’re doi… | ↗Not briefed |
bluesky bluesky-ai | 1mo | Hmm. I’m getting error messages in response to queries about what AI model the site uses. | Hmm. I’m getting error messages in response to queries about what AI model the site uses. | ↗Not briefed |
hn hn-ai | 1mo | Show HN: ProgressCove, calm and smart to-do app with Home Assistant integration | <i>TLDR version:</i> ProgressCove, my calm, multiplatform task manager that has a glass half-full perspective, avoids red texts and pa… | ↗Not briefed |
arxiv arxiv-ai | 1mo | HiRAE: Hierarchical Representation Autoencoding with Residual Budgets | Pretrained visual representations support image generation, but may not fully preserve the fine-grained details needed for faithful re… | ↗Not briefed |
hn hn-ai | 1mo | Another World Ported to ZX Spectrum | — | ↗Not briefed |
arxiv arxiv-ai | 1mo | OmniVCBench: Benchmarking Evidence-Grounded Multimodal Reasoning Towards AI Virtual Cells | Artificial Intelligence Virtual Cells (AIVCs) are envisioned as scientific agents that simulate cellular responses, explain underlying… | ↗Not briefed |
bluesky bluesky-ai | 1mo | But lets say Pangram isn't lying through their teeth and it really is a totally walled off program. So what? 1 million texts isn't that muc… | But lets say Pangram isn't lying through their teeth and it really is a totally walled off program. So what? 1 million texts isn't tha… | ↗Not briefed |
rss techcrunch-ai | 1mo | Instinct founder said more than 50% of transactions on the platform are travel-related | Instinct founder said that the platform is growing 10% day-by-day | ↗Not briefed |