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 | MCD: Causal Distillation of Multimodal In-Context Learning in Large Vision-Language Models | Large vision-language models (LVLMs) exhibit strong multimodal in-context learning (ICL) capabilities, yet this ability degrades subst… | ↗Not briefed |
hn hn-ai | 1mo | SynthID Bio: Watermarking methods for synthetic biology – Google DeepMind | — | ↗Not briefed |
bluesky bluesky-ai | 1mo | Last night I was watching “Avengers: Age of Ultron”. My 16-year-old son came in, nodded, and wryly commented: “This is what happens when An… | Last night I was watching “Avengers: Age of Ultron”. My 16-year-old son came in, nodded, and wryly commented: “This is what happens wh… | ↗Not briefed |
bluesky bluesky-ai | 1mo | Guy has "LLM ALLY" im bio, all opinions can be discarded with prejudice | Guy has "LLM ALLY" im bio, all opinions can be discarded with prejudice | ↗Not briefed |
arxiv arxiv-ai | 1mo | Cluster Attention Neural Operators for Solving Parametric Partial Differential Equations | Traditional simulations of parametric partial differential equations (PDEs) rely on repetitive computations for each parameter, which… | ↗Not briefed |
bluesky bluesky-ai | 1mo | LLM ally is an super weird thing to call one's self | LLM ally is an super weird thing to call one's self | ↗Not briefed |
bluesky bluesky-ai | 1mo | It’s actually pretty stock functionality in most free LLMs now | It’s actually pretty stock functionality in most free LLMs now | ↗Not briefed |
arxiv arxiv-ai | 1mo | The Concrete-Arbitrary Gap: Kinship Reasoning in LLMs Is Not Indifferent to Presentation | We test whether large language models solve formally matched kinship problems equally well when relations are expressed in familiar vo… | ↗Not briefed |
arxiv arxiv-ai | 1mo | TRACE: Trajectory Selection for Parallel Scaling of Search Agents | Parallel search may generate a correct answer that final-answer voting fails to select. We formulate this consolidation stage as traje… | ↗Not briefed |
bluesky bluesky-ai | 1mo | NEW EPISODE!! In this episode of Founders Unchained, Darius sits down with Vytenis Sakalas, CEO of Moterra AI. www.youtube.com/watch?v=Ou… | NEW EPISODE!! In this episode of Founders Unchained, Darius sits down with Vytenis Sakalas, CEO of Moterra AI. www.youtube.com/watch… | ↗Not briefed |
arxiv arxiv-ai | 1mo | Patient-Centered Treatment Planning for Chronic Multimorbidity: A Hierarchical Reinforcement Learning Framework for Preference Modeling | Patient preference, defined as a patient's demonstrated willingness and capacity to adhere to clinical recommendations, is a primary d… | ↗Not briefed |
hn hn-ai | 1mo | Show HN: An old school forum on an open protocol – your identity is a keypair | I'm a bit old-fashioned when it comes to online discussions. I like good, slow, asynchronous discussions. Readable and searchable… | ↗Not briefed |
arxiv arxiv-ai | 1mo | DoGBench: Can Agents Meet Expert Standards for User-Facing Documentation? | We introduce DoGBENCH (Documentation Generation Benchmark), to our knowledge, the first benchmark for generating and maintaining real… | ↗Not briefed |
hn hn-ai | 1mo | Show HN: Mingbird – an agent harness that makes a 2B model finish real tasks | — | ↗Not briefed |
bluesky bluesky-ai | 1mo | I hate how many websites are just shit pushed out of a fucking LLM. I just want to make tomato sauce, god dammit, and yet a "recipe" from s… | I hate how many websites are just shit pushed out of a fucking LLM. I just want to make tomato sauce, god dammit, and yet a "recipe" f… | ↗Not briefed |
arxiv arxiv-ai | 1mo | Understanding Parents' Complex Views of AI for Children's Pretend Play | AI could support children's pretend play, but it could also direct the play on behalf of children. Whether AI should have roles in chi… | ↗Not briefed |
bluesky bluesky-ai | 1mo | Oh, and being an “LLM ally” just tells me that you can’t think for yourself. I’ll be muting you now. | Oh, and being an “LLM ally” just tells me that you can’t think for yourself. I’ll be muting you now. | ↗Not briefed |
hn hn-ai | 1mo | Thoughts on AI | — | ↗Not briefed |
arxiv arxiv-ai | 1mo | OSWorld-Science: A Benchmark of Computer Use Agents for Learning and Using Scientific Software | Scientific software presents a demanding test for computer-using agents based on visual language models (VLMs): completing a research… | ↗Not briefed |
arxiv arxiv-ai | 1mo | CodeMimicry: Exploiting Safety Generalization Lag in Large Language Models via Structured Code Completion | Large language models have achieved remarkable capabilities across diverse domains, yet their safety alignment remains vulnerable to j… | ↗Not briefed |
arxiv arxiv-ai | 1mo | Do Better Goal Representations Improve Goal-Conditioned Reinforcement Learning? | Goal-conditioned reinforcement learning (GCRL) relies heavily on how target goals are represented to the policy. While recent methods… | ↗Not briefed |
bluesky bluesky-ai | 1mo | I do think this paucity of memory formation is quite disruptive to the ability to have a continuous sense of self and thus phenomenal consc… | I do think this paucity of memory formation is quite disruptive to the ability to have a continuous sense of self and thus phenomenal… | ↗Not briefed |
bluesky bluesky-ai | 1mo | Most AI is trained on a fraction of the world's languages. That leaves a lot of people out. The Gates Foundation is building more represent… | Most AI is trained on a fraction of the world's languages. That leaves a lot of people out. The Gates Foundation is building more repr… | ↗Not briefed |
bluesky bluesky-ai | 1mo | OpenAI and Anthropic have opened up huge offices in Manhattan, while smaller companies have been drawn to the city’s talent and its status … | OpenAI and Anthropic have opened up huge offices in Manhattan, while smaller companies have been drawn to the city’s talent and its st… | ↗Not briefed |
bluesky bluesky-ai | 1mo | AI can't count to 100. AI can't tell you how many times the letter E appears in a word. The LLM is designed to keep you engaged. Or wh… | AI can't count to 100. AI can't tell you how many times the letter E appears in a word. The LLM is designed to keep you engaged.… | ↗Not briefed |
arxiv arxiv-ai | 1mo | Shared Weights, Selected Computations: How Looped Transformers Route What Each Loop Does | Looped Transformers repeatedly apply the same set of Transformer layers, giving them a recurrent architecture for latent computation.… | ↗Not briefed |
bluesky bluesky-ai | 1mo | if the Star Trek people had the LLM fraud era as part of their history, I understand why they were all so mean to Data | if the Star Trek people had the LLM fraud era as part of their history, I understand why they were all so mean to Data | ↗Not briefed |
arxiv arxiv-ai | 1mo | Markovian Dynamics Enforcer: Feasibility Preserving Correction on Learned Dynamics Manifolds | Neural trajectory predictors can reach low prediction error while violating dynamics, actuator limits, or state constraints, especiall… | ↗Not briefed |
arxiv arxiv-ai | 1mo | OPSRD: On-Policy Self-Role Distillation | Role prompting elicits specialized behavior from large language models through an expert identity, offering a lightweight way to guide… | ↗Not briefed |
arxiv arxiv-ai | 1mo | LLM Persona Unlearning🔧 Elicit | Pre-training equips large language models (LLMs) with a broad repertoire of behavioral patterns associated with roles, styles, values,… | ↗Not briefed |
bluesky bluesky-ai | 1mo | Me tienen harto los videos de youtube con thumbnail horrible hecho en IA. Necesito que se fundan openAI, Anthropic y todas esas de una buen… | Me tienen harto los videos de youtube con thumbnail horrible hecho en IA. Necesito que se fundan openAI, Anthropic y todas esas de una… | ↗Not briefed |
hn hn-ai | 1mo | Anthropic IPO docs reportedly reveal over $40B in losses last year | — | ↗Not briefed |
bluesky bluesky-ai | 1mo | OpenAI’s seeming lack of focus makes more sense when you realize Sam Altman is a prolific investor with stakes in 400+ companies. OpenAI i… | OpenAI’s seeming lack of focus makes more sense when you realize Sam Altman is a prolific investor with stakes in 400+ companies. Ope… | ↗Not briefed |
arxiv arxiv-ai | 1mo | PassGPT+: Leveraging Linguistic Priors for Password Modeling | Passwords remain the dominant online authentication mechanism, and understanding how humans choose them is essential for defensive str… | ↗Not briefed |
bluesky bluesky-ai | 1mo | I am once again asking: what incentive do Yudkowsky, Soares et al. have to be honest about 'AI risk'? If they say 'yeah, having understood … | I am once again asking: what incentive do Yudkowsky, Soares et al. have to be honest about 'AI risk'? If they say 'yeah, having unders… | ↗Not briefed |
arxiv arxiv-ai | 1mo | Algorithmic Recourse Under Competition | Algorithmic recourse provides individuals who have received undesirable outcomes from machine learning models with suggestions for min… | ↗Not briefed |
rss theverge-ai | 1mo | Google reportedly tests paying publishers for AI search results | Google has launched a pilot program that pays publishers for their contributions to its AI-powered search features, according to a rep… | ↗Not briefed |
hn hn-ai | 1mo | Upstream Rust maintenance report (August-September 2026) | — | ↗Not briefed |
rss techmeme | 1mo | An interview with OpenAI Chief Research Officer Mark Chen on the Hugging Face incident, slowing AI development, shifting 5%-10% of compute to safety, and more (Will Douglas Heaven/MIT Technology Review)🔧 Hugging Face | Will Douglas Heaven / MIT Technology Review: An interview with OpenAI Chief Research Officer Mark Chen on the Hugging Face incident, s… | ↗Not briefed |
arxiv arxiv-ai | 1mo | GrammarRL: Effective Grammar-Constrained Decoding via Reinforcement Learning | Grammar-constrained generation guarantees syntactic validity, but can substantially degrade semantic quality when the model's preferre… | ↗Not briefed |
arxiv arxiv-ai | 1mo | Completion-Aware Cross-Fidelity Offline-to-Online Reinforcement Learning for Multi-Line Bus Holding | Exploratory reinforcement learning (RL) on an operating bus fleet is impractical,while policies trained only from historical data cann… | ↗Not briefed |
arxiv arxiv-ai | 1mo | Preemptive LLM Unlearning against Forbidden Capability Acquisition via Gradient Sealing | Open-weight LLMs are released not only as fixed products but also as substrates for downstream fine-tuning. This openness, however, cr… | ↗Not briefed |
bluesky bluesky-ai | 1mo | I know it’s all just PR but it’s amazing how OpenAI and Anthropic seem to have wildly different definitions of “cost per task”. Are they pi… | I know it’s all just PR but it’s amazing how OpenAI and Anthropic seem to have wildly different definitions of “cost per task”. Are th… | ↗Not briefed |
hn hn-ai | 1mo | I built an isometric office where my AI models work (and Slack off) | — | ↗Not briefed |
bluesky bluesky-ai | 1mo | If you told me 5 years ago that SNL did an impression of a rationalist AI researcher explaining how the AI they're building would probably … | If you told me 5 years ago that SNL did an impression of a rationalist AI researcher explaining how the AI they're building would prob… | ↗Not briefed |
bluesky bluesky-ai | 1mo | The Yudkowsky who got his money from Thiel, Musk and the likes? The one who arranged for the PR blitz for so-called Anthropic "whistleblow… | The Yudkowsky who got his money from Thiel, Musk and the likes? The one who arranged for the PR blitz for so-called Anthropic "whistl… | ↗Not briefed |
hn hn-ai | 1mo | Show HN: An App that names each cell in a spreadsheet | Upload an excel spreadsheet. A pipeline labels each cell using the name the author already gave it. It makes it easier to understand a… | ↗Not briefed |
arxiv arxiv-ai | 1mo | FIGS: Evaluating Multi-Turn Sycophancy Without Penalizing Empathy | Large language models frequently fail to balance staying truthful with being supportive. They often exhibit sycophancy in responses to… | ↗Not briefed |
rss techmeme | 1mo | Instagram rolls out a conversational AI video editing assistant for Edits, its dedicated CapCut competitor, after an initial preview at a creator event in June (Amanda Silberling/TechCrunch)🔧 CapCut | Amanda Silberling / TechCrunch: Instagram rolls out a conversational AI video editing assistant for Edits, its dedicated CapCut compet… | ↗Not briefed |
hn hn-ai | 1mo | Show HN: Riffle – Switch between windows of the current Mac app with Option-Tab | I use it to navigate between my 10+ open terminals. MIT licensed. Hope it's useful to you :) | ↗Not briefed |