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 | A Tilted Bowl Is Not a Slippery Slope: Compressing Looped Models | Looped models reason by applying the same block of weights many times, so compressing that block saves memory traffic on every loop. C… | ↗Not briefed |
bluesky bluesky-ai | 1mo | I'm sure that you can say to an LLM 'I like X, Y and Z bands - can you recommend some more that I might like?' and it'll come up with somet… | I'm sure that you can say to an LLM 'I like X, Y and Z bands - can you recommend some more that I might like?' and it'll come up with… | ↗Not briefed |
arxiv arxiv-ai | 1mo | ReTaCo: Residual-Target Control for On-Policy Distillation | On-policy distillation (OPD) trains a student on its own generated prefixes with token-level teacher feedback, but transmitting or sto… | ↗Not briefed |
rss marktechpost | 1mo | Perplexity Introduces Photon: A Rust-Based Retrieval Engine That Cuts p99 Latency From 800 ms to 65 ms | Perplexity has released Photon, an in-house retrieval and ranking engine written in Rust. It replaces an open-source engine Perplexity… | ↗Not briefed |
bluesky bluesky-ai | 1mo | Also now true for other forms of reasoning. Jev is really not an reasoning LLM and can't use tools, but still can guess validity of logical… | Also now true for other forms of reasoning. Jev is really not an reasoning LLM and can't use tools, but still can guess validity of lo… | ↗Not briefed |
arxiv arxiv-ai | 1mo | RW-Flow: One-Step Generation on Compact Manifolds via Wasserstein Gradient Flows | Manifold-valued data, and consequently the distributions they induce, are prevalent across many domains, ranging from the locations of… | ↗Not briefed |
arxiv arxiv-ai | 1mo | A Time-Aware Bag-of-Receptive-Fields for Interpretable Irregular Time Series Classification | Irregular time series, characterized by non-uniform sampling intervals, missing observations, and variable lengths, are ubiquitous in… | ↗Not briefed |
arxiv arxiv-ai | 1mo | Concept Subspaces Compute Beyond the Logit Lens: A Weights-Only Test for Locating Representations Upstream of Readout | A concept subspace's effect on model behavior does not establish how it relates to the output readout. We introduce a two-sided geomet… | ↗Not briefed |
bluesky bluesky-ai | 1mo | "농이 지나치시네요. 이차원 기술 연구부가 대단한건 알지만 그렇게까지 과장할 필요는 없어요. 로라가 정말 그렇다면 도대체 그런 고위험 작전을 어떻게 수행한다는 거죠? 그 정도 연산력으로는 로컬 LLM도 힘들텐데." "인공지능 말인가요? 그럴 필요가… | "농이 지나치시네요. 이차원 기술 연구부가 대단한건 알지만 그렇게까지 과장할 필요는 없어요. 로라가 정말 그렇다면 도대체 그런 고위험 작전을 어떻게 수행한다는 거죠? 그 정도 연산력으로는 로컬 LLM도 힘들텐데." "인공지능 말인가요? 그… | ↗Not briefed |
arxiv arxiv-ai | 1mo | Jacobian Rank Collapse in Decision-Focused Learning | Decision-focused learning (DFL) trains predictors through downstream objectives, but a different loss need not provide an independent… | ↗Not briefed |
arxiv arxiv-ai | 1mo | Effective Does Not Mean Useful: Conditional Functional Substitutability for Redundancy and Scaling in Transformers | Modern neural networks scale predictably, yet the mechanisms behind these regularities remain unclear. Neural redundancy is typically… | ↗Not briefed |
bluesky bluesky-ai | 1mo | J'ai plus Spotify, maintenant j'ai Claude qui demande à ChatGPT de faire un Top 50 alternatif avec que des musiques qui viennent de Suno et…🔧 ChatGPT🔧 Suno | J'ai plus Spotify, maintenant j'ai Claude qui demande à ChatGPT de faire un Top 50 alternatif avec que des musiques qui viennent de Su… | ↗Not briefed |
arxiv arxiv-ai | 1mo | From Benchmarks to Production: Transferring Time Series Anomaly Detection Methods for Electricity Production Monitoring | Accurate forecasting of electricity production is essential for maintaining the operational efficiency and strategic planning of energ… | ↗Not briefed |
bluesky bluesky-ai | 1mo | What an odd thing to say. Let me fix it: If you're not a fool you are dispatching deterministic jobs to tools available to the LLM. Not do… | What an odd thing to say. Let me fix it: If you're not a fool you are dispatching deterministic jobs to tools available to the LLM. N… | ↗Not briefed |
bluesky bluesky-ai | 1mo | RE: https://infosec.exchange/@david_chisnall/117358990093685067 100%. The whole "terrorists can use an LLM to build a bomb" is a dumb movi… | RE: https://infosec.exchange/@david_chisnall/117358990093685067 100%. The whole "terrorists can use an LLM to build a bomb" is a dumb… | ↗Not briefed |
rss techmeme | 1mo | Outmarket, which uses AI to help insurance agencies and brokers automate their tedious paperwork, raised a $34.5M Series B, a source says at a $335M valuation (Marina Temkin/TechCrunch) | Marina Temkin / TechCrunch: Outmarket, which uses AI to help insurance agencies and brokers automate their tedious paperwork, raised a… | ↗Not briefed |
arxiv arxiv-ai | 1mo | Client and Training Data Selection for Computationally Efficient Synchronized Federated Learning | Federated learning (FL) is a promising paradigm of machine learning, which preserves user privacy by enabling learning without sharing… | ↗Not briefed |
rss bing-news-ai | 1mo | Trump Signs ‘Super Intelligence’ Order: What Actually Changes for AI | President Trump signed an executive order directing federal agencies to use “Super Intelligence” instead of AI. Here’s what changes an… | ↗Not briefed |
rss bing-news-ai | 1mo | Trump Signs ‘Super Intelligence’ Order: What Actually Changes for AI | President Trump signed an executive order directing federal agencies to use “Super Intelligence” instead of AI. Here’s what changes an… | ↗Not briefed |
bluesky bluesky-ai | 1mo | AI safety discourse is hilarious My brother in Christ you spent hours jail breaking a multimillion dollar model to become a shit version … | AI safety discourse is hilarious My brother in Christ you spent hours jail breaking a multimillion dollar model to become a shit ver… | ↗Not briefed |
arxiv arxiv-ai | 1mo | Trust the Critic More | Standard language model RL algorithms credit every token of a long rollout with the same advantage determined by the terminal reward.… | ↗Not briefed |
bluesky bluesky-ai | 1mo | With all this talk of "rogue AIs" (that do exactly what they're made to do) and the recent revelation that Meta's Muse is basically a malic… | With all this talk of "rogue AIs" (that do exactly what they're made to do) and the recent revelation that Meta's Muse is basically a… | ↗Not briefed |
bluesky bluesky-ai | 1mo | i have very little bad feelings about LLM themselves. but encountering groupchat-itis sufferers, cultists, and other people in a now very m… | i have very little bad feelings about LLM themselves. but encountering groupchat-itis sufferers, cultists, and other people in a now v… | ↗Not briefed |
arxiv arxiv-ai | 1mo | Right Answer, Wrong Mechanism: Detecting Pernicious Divergence in Causal Interventions | Causal interventions such as activation patching and distributed alignment search (DAS) are the main tool for making mechanistic claim… | ↗Not briefed |
arxiv arxiv-ai | 1mo | A differentiability framework for zigzag persistent homology via linear interpolation | Persistent homology can be differentiated and incorporated into learning pipelines, but no analogous framework exists for zigzag persi… | ↗Not briefed |
bluesky bluesky-ai | 1mo | Something I haven't seen people talk about in regards to LLM-use is the almost complete collapse of people giving feedback on the tools the… | Something I haven't seen people talk about in regards to LLM-use is the almost complete collapse of people giving feedback on the tool… | ↗Not briefed |
bluesky bluesky-ai | 1mo | Baffling that people really think there are no one avoiding these LLM bullshit | Baffling that people really think there are no one avoiding these LLM bullshit | ↗Not briefed |
rss bing-news-ai | 1mo | White House accord on Super Intelligence includes humiliating United States typo | "Maybe we need some super intelligence running typo checks on these docs," Another person jeered. A fourth person added that Trump sho… | ↗Not briefed |
bluesky bluesky-ai | 1mo | Oh look, having AI run on your system could be killing your ram modules. (Consumer RAM isn't designed to run this sort of code constantly,… | Oh look, having AI run on your system could be killing your ram modules. (Consumer RAM isn't designed to run this sort of code consta… | ↗Not briefed |
hn hn-ai | 1mo | Practical Secrets Extraction Against Black-Box LLMs | — | ↗Not briefed |
arxiv arxiv-ai | 1mo | 4MT-VLM: How Coarse Is a VLMs Cognitive Map?🔧 GPT-5 | An agent that moves must recognise a place from a viewpoint it has never seen. We introduce 4MT-VLM, a dataset of procedurally generat… | ↗Not briefed |
arxiv arxiv-ai | 1mo | What Streaming Anomaly Detection Finds (and Misses) in Industrial Time Series | EDF relies on continuous monitoring of its power plants to detect anomalies as soon as they occur. Given the absence of a universally… | ↗Not briefed |
hn hn-ai | 1mo | I ran a 397B-parameter model on 20 GPUs with 16 GB each | — | ↗Not briefed |
arxiv arxiv-ai | 1mo | RAIM: Robust Aggregation of Inexpensive Models for Hallucination Detection | Automatic evaluation of faithfulness increasingly relies on a large language model acting as a judge, yet the most reliable judges are… | ↗Not briefed |
arxiv arxiv-ai | 1mo | Fyan: A Human--AI Harness with Semantic Auditing for Document-Level Formalization | We present FYAN, a human--AI harness for document-level mathematical formalization. Rather than treating theorems in isolation, FYAN c… | ↗Not briefed |
arxiv arxiv-ai | 1mo | Emergent Multi-View Geometry Through Self-Distillation | Over a century ago, Henri Poincaré argued that a motionless observer cannot acquire the notion of space. Yet, most visual representati… | ↗Not briefed |
arxiv arxiv-ai | 1mo | Argument Structure Prediction in Online Conversations: A Comparative Study of Modeling Paradigms and Task Architectures | Argument structure prediction (ASP) constructs complete argument structures from discourse by identifying argumentative units and thei… | ↗Not briefed |
rss techmeme | 1mo | Goldman Sachs: investors have provided ~$500B in financing to AI-linked groups in 2026 so far, as hyperscaler debt issuance spreads to the euro, CAD, and others (Financial Times) | Financial Times: Goldman Sachs: investors have provided ~$500B in financing to AI-linked groups in 2026 so far, as hyperscaler debt is… | ↗Not briefed |
arxiv arxiv-ai | 1mo | QATFactory: A Versatile, Deployment-Aligned Framework for Quantization-aware Training and Distillation of LLMs🔧 vLLM | Large language model (LLM) inference is increasingly moving toward lower precision to realize the throughput of hardware accelerators,… | ↗Not briefed |
arxiv arxiv-ai | 1mo | Learning Process Rewards via Reasoning State Propagation | Process reward models (PRMs) have demonstrated notable effectiveness in test-time scaling and reinforcement learning by providing fine… | ↗Not briefed |
bluesky bluesky-ai | 1mo | Here’s what I wrote about the incident; a critique I heard often was that OpenAI had no way of knowing this kind of thing would happen and … | Here’s what I wrote about the incident; a critique I heard often was that OpenAI had no way of knowing this kind of thing would happen… | ↗Not briefed |
bluesky bluesky-ai | 1mo | ... but I feel even this would be unpopular due to the probabilistic nature of those tools. I worry we'll take the alternative route of hav… | ... but I feel even this would be unpopular due to the probabilistic nature of those tools. I worry we'll take the alternative route o… | ↗Not briefed |
hn hn-ai | 1mo | Ulua: Pure Rust Implementation of the Luau Language with JIT and WebAssembly | — | ↗Not briefed |
arxiv arxiv-ai | 1mo | In a Streaming World, Should You Stand Still? A Comprehensive Benchmark of Anomaly Detection in Streams | Time series anomaly detection (TSAD) is increasingly deployed in streaming settings, where data arrive sequentially and may exhibit no… | ↗Not briefed |
arxiv arxiv-ai | 1mo | Minimax Additive Regression under Unknown Dependent Designs | We study additive regression under a potentially non-product random design on $[0,1]^d$, allowing the dimension $d$ to grow with the s… | ↗Not briefed |
hn hn-ai | 1mo | Show HN: Built agent orchestration system, Director->{workers } | I have recently started working on an agent orchestrator with stricht budgeting + TUI, check it here: <a href="https://githu… | ↗Not briefed |
hn hn-ai | 1mo | Show HN: I made Jev to talk | Simply by asking Jev multiple yes/no question about what would be best next word, it's possible to get some reasonable text… | ↗Not briefed |
hn hn-ai | 1mo | Show HN: Melete – an open-source alternative to OpenAI's Dots | — | ↗Not briefed |
bluesky bluesky-ai | 1mo | deepseek APIでやろうとしていることを試したが、秒で2ドルが飛んだわ ローカルLLMしか勝たん | deepseek APIでやろうとしていることを試したが、秒で2ドルが飛んだわ ローカルLLMしか勝たん | ↗Not briefed |
arxiv arxiv-ai | 1mo | UniAE-MoE: A Unified Audio Encoder via Mixture of Experts | Large Audio Language Models (LALMs) rely on effective audio encoders for multi-task performance. We introduce UniAE-MoE, a unified aud… | ↗Not briefed |