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 | UniData: Universal Multimodal Instruction Generation Pipeline | Multimodal Large Language Models (MLLMs) are increasingly being applied in a wider range of real-world scenarios. However, due to the… | ↗Not briefed |
arxiv arxiv-ai | 1mo | Bernoulli Flow Models: Self-Consistent Generative Modeling for Binary Data | Binary diffusion models typically require a large number of function evaluations (NFEs) to generate high-quality samples, making pract… | ↗Not briefed |
bluesky bluesky-ai | 1mo | In September, OpenAI said 10,000 #AI agents had partially "solved" the Navier–Stokes problem, one of maths' great open questions. #univienn… | In September, OpenAI said 10,000 #AI agents had partially "solved" the Navier–Stokes problem, one of maths' great open questions. #uni… | ↗Not briefed |
arxiv arxiv-ai | 1mo | RaReCache: Bridging the Gap in Cross-Model KV Cache Reuse via Rank disagreement-based Selective Recomputation | Cross-model KV-cache reuse remains a key challenge in modern LLM serving. Coding agents and multi-model systems increasingly route a s… | ↗Not briefed |
arxiv arxiv-ai | 1mo | Writing for the Reviewer: Defensive Writing in GPT Models🔧 ChatGPT | Researchers increasingly use ChatGPT to revise their papers, and recent GPT versions often narrow or even retract the authors' claims.… | ↗Not briefed |
arxiv arxiv-ai | 1mo | AdaptEvo: Adaptive Agent Learning with Evolving Supervision | Rule-governed contextual decision tasks require models to apply specified rules to case-specific context and evidence. Written rules c… | ↗Not briefed |
bluesky bluesky-ai | 1mo | 2036…. Vem ska få litteraturpriset, Open AI eller Anthropic… specialtränade AI-agenter promptar tusentals unika författarskap… LitBot H… | 2036…. Vem ska få litteraturpriset, Open AI eller Anthropic… specialtränade AI-agenter promptar tusentals unika författarskap… Lit… | ↗Not briefed |
arxiv arxiv-ai | 1mo | RL-ARC: Calibrating Large Reasoning Models via Reasoning-guided Uncertainty | Language models (LMs) are commonly trained with Reinforcement Learning with Verifiable Rewards (RLVR) to enhance their reasoning capab… | ↗Not briefed |
arxiv arxiv-ai | 1mo | Deception by Omission: Language Models Knowingly Hide Their Mistakes | Large language models (LLMs) increasingly act as agents with little human oversight, so potential mistakes they make can go unnoticed.… | ↗Not briefed |
bluesky bluesky-ai | 1mo | Een model kent de ‘token’ verdriet, en weet dat de token ‘tranen’ daarbij in de buurt zit. Maar het model weet niet wat wij van binnen voel… | Een model kent de ‘token’ verdriet, en weet dat de token ‘tranen’ daarbij in de buurt zit. Maar het model weet niet wat wij van binnen… | ↗Not briefed |
arxiv arxiv-ai | 1mo | Sample-Efficient Generative Conformal Prediction | Generative conformal prediction builds uncertainty sets from samples of a conditional generator, which are efficient only when the sam… | ↗Not briefed |
bluesky bluesky-ai | 1mo | この手の話を見る度に思うんだけど、LLMに知らない事を質問する時にはその出力結果を「自分の脳内から出てきた知識」だとは思わないのに、創作的な事になるとなぜか「自分の脳内イメージを生成AIが引き出してくれた」に認知が歪む人達って自他境界の認識バグってない？という…そもそも生成AI… | この手の話を見る度に思うんだけど、LLMに知らない事を質問する時にはその出力結果を「自分の脳内から出てきた知識」だとは思わないのに、創作的な事になるとなぜか「自分の脳内イメージを生成AIが引き出してくれた」に認知が歪む人達って自他境界の認識バグってない？という…そもそ… | ↗Not briefed |
bluesky bluesky-ai | 1mo | On-device today, on-chip tomorrow: the model ships inside the boundary you hold. Ownership stops being procurement and becomes the default … | On-device today, on-chip tomorrow: the model ships inside the boundary you hold. Ownership stops being procurement and becomes the def… | ↗Not briefed |
arxiv arxiv-ai | 1mo | SynCo: Data Synthesis Co-Training for Self-Evolving LLMs via Multi-Agent Reinforcement Learning | Self-evolving LLM agents promise to improve autonomously through continual interaction and learning, reducing their dependence on manu… | ↗Not briefed |
rss techmeme | 1mo | A profile of Garmin CEO Cliff Pemble, who helped diversify its GPS product lines and grew annual revenue from $2.6B in 2013 to a record $7.2B last year (Brent Crane/Bloomberg) | Brent Crane / Bloomberg: A profile of Garmin CEO Cliff Pemble, who helped diversify its GPS product lines and grew annual revenue from… | ↗Not briefed |
bluesky bluesky-ai | 1mo | Another example of OpenAI scooping mathematicians who were using ChatGPT. I don't think this suggests OpenAI is training on mathematicians'…🔧 ChatGPT | Another example of OpenAI scooping mathematicians who were using ChatGPT. I don't think this suggests OpenAI is training on mathematic… | ↗Not briefed |
arxiv arxiv-ai | 1mo | EvoSim: Learning to Model, Modeling to Learn | Physics-based models connect scientific explanation with quantitative prediction. Constructing them requires selecting physical proces… | ↗Not briefed |
bluesky bluesky-ai | 1mo | The argument can be made that the problem will reduce when Anthropic and OpenAI go broke, take nVidia with them, and Google, Amazon and Met… | The argument can be made that the problem will reduce when Anthropic and OpenAI go broke, take nVidia with them, and Google, Amazon an… | ↗Not briefed |
bluesky bluesky-ai | 1mo | Det här är faktiskt väldigt roligt. Nån har med AI-Claude rekonstruerat alla Adobes program och lagt ut dem för gratisanvändning. Man får s… | Det här är faktiskt väldigt roligt. Nån har med AI-Claude rekonstruerat alla Adobes program och lagt ut dem för gratisanvändning. Man… | ↗Not briefed |
arxiv arxiv-ai | 1mo | Type-Checking for Pattern-Based Tree Transformations | We introduce and study pattern-based tree transformations. As an illustrating example, consider a source pattern $(x \cdot y) + (x \cd… | ↗Not briefed |
bluesky bluesky-ai | 1mo | The prompt was just, "How many R's are there in 'strawberry'?" | The prompt was just, "How many R's are there in 'strawberry'?" | ↗Not briefed |
bluesky bluesky-ai | 1mo | I don’t buy the idea that AI leads to democratisation of creativity. I don’t care about open source Adobe because there’s been free alterna… | I don’t buy the idea that AI leads to democratisation of creativity. I don’t care about open source Adobe because there’s been free al… | ↗Not briefed |
arxiv arxiv-ai | 1mo | ReCast: Attribution-Oriented Step Representation Learning for LLM-Based Agent Systems | In LLM-based agent systems, failures can originate from early steps whose effects propagate through subsequent interactions, making th… | ↗Not briefed |
hn hn-ai | 1mo | Fluffles-OS | — | ↗Not briefed |
bluesky bluesky-ai | 1mo | As a long time coder (started in the late 70s) that claim was obvious bullshit. What LLM-based "AI" is actually good for is generating conv… | As a long time coder (started in the late 70s) that claim was obvious bullshit. What LLM-based "AI" is actually good for is generating… | ↗Not briefed |
arxiv arxiv-ai | 1mo | TokenBank: Financial Infrastructure for AI Services | AI services incur inference costs during execution, while revenue may arrive later. Changing API prices, limited upfront capital, and… | ↗Not briefed |
hn hn-ai | 1mo | Classic PC demoscene productions running natively in the browser | — | ↗Not briefed |
arxiv arxiv-ai | 1mo | ReCal: Calibrating Structured Pruning for On-Policy Distillation Recovery | Structured pruning reduces the deployment cost of reasoning language models, but the resulting capability degradation can hinder subse… | ↗Not briefed |
bluesky bluesky-ai | 1mo | “In ignoring the central premise of the Advisory Group’s position, OpenAI has indicated total disregard for the norms of scientific researc… | “In ignoring the central premise of the Advisory Group’s position, OpenAI has indicated total disregard for the norms of scientific re… | ↗Not briefed |
arxiv arxiv-ai | 1mo | Mine Odyssey: Benchmarking Spatial Agentic Intelligence in the Wild | Advances in foundation models are driving efforts to introduce agents to assist people in the physical world. Such agents require agen… | ↗Not briefed |
rss techmeme | 1mo | The Association for Human Mathematics says OpenAI's new math documents show power, not scholarship, and urges mathematicians to stop working with the company (AHM) | AHM: The Association for Human Mathematics says OpenAI's new math documents show power, not scholarship, and urges mathematicians to s… | ↗Not briefed |
bluesky bluesky-ai | 1mo | open ai today released details of a hundred previously undiscovered administrative revolutions which its latest model had unearthed | open ai today released details of a hundred previously undiscovered administrative revolutions which its latest model had unearthed | ↗Not briefed |
bluesky bluesky-ai | 1mo | If you want an example of how we have no control, look at the Mathematics Advisory Board OpenAI established. Do you think yesterday’s debac… | If you want an example of how we have no control, look at the Mathematics Advisory Board OpenAI established. Do you think yesterday’s… | ↗Not briefed |
arxiv arxiv-ai | 1mo | Finsler Flow Matching: Dynamics-Aware Geodesic Interpolation for Single-Snapshot Trajectory Inference | Single-cell snapshot data can resolve a continuum of cellular states but do not uniquely determine the dynamics governing transitions… | ↗Not briefed |
arxiv arxiv-ai | 1mo | DivMoE: Fine-Grained MoE Upcycling via Cross-Domain Expert Composition | Mixture-of-Experts (MoE) architectures have become essential for scaling large language models, with recent work demonstrating the ben… | ↗Not briefed |
arxiv arxiv-ai | 1mo | MetaEncoder: Exploring the Limit of Bi-Encoders for Multimodal System One Decision Making with Natural Language Interface | System One models output constrained decisions and probability distributions rather than free-form text generation. While prevailing p… | ↗Not briefed |
arxiv arxiv-ai | 1mo | Deflating the Hessian: Rank-4 W4A4 Quantization for Multimodal Diffusion Transformers🔧 FLUX | In diffusion transformers, low-rank branches can mitigate 4-bit weight--activation (W4A4) post-training quantization (PTQ) loss by dec… | ↗Not briefed |
arxiv arxiv-ai | 1mo | From Retrieval to Reconstruction: Constructing Evolvable Cognitive Memory for Long-Term Dialogue | Large Language Models (LLMs) serving as long-term dialogue agents require memory systems that support reliable reasoning over extended… | ↗Not briefed |
rss techmeme | 1mo | A look at Emeryville, CA-based Atomic Machines, which is training AI on materials and designs to revamp how microelectromechanical systems (MEMS) are built (Cade Metz/New York Times) | Cade Metz / New York Times: A look at Emeryville, CA-based Atomic Machines, which is training AI on materials and designs to revamp ho… | ↗Not briefed |
arxiv arxiv-ai | 1mo | NP-Hardness of Minimizing Neurons in Two-Hidden-Layer ReLU Neural Networks | A fundamental question in neural network architecture optimization is whether the minimum hidden-neuron count required to approximate… | ↗Not briefed |
arxiv arxiv-ai | 1mo | MedBenchAgent: Towards Systematic Automation of Medical VLM Benchmark Construction | Large-scale construction of medical vision-language model (VLM) benchmarks is increasingly feasible with richly annotated imaging data… | ↗Not briefed |
arxiv arxiv-ai | 1mo | FloorSAV: Elucidating Spatial Audio-Visual Context with 2D Floormap for AV-LLMs🔧 Modal | While 3D spatial reasoning in dynamic egocentric environments is crucial for embodied intelligence, audio-visual large language models… | ↗Not briefed |
bluesky bluesky-ai | 1mo | It's all here: Photoshop, Illustrator, After Effects, Premiere, Acrobat, Lightroom, Designer. Someone leveraged the fact that AI just sto… | It's all here: Photoshop, Illustrator, After Effects, Premiere, Acrobat, Lightroom, Designer. Someone leveraged the fact that AI jus… | ↗Not briefed |
arxiv arxiv-ai | 1mo | From Geometry to Generalization: Why Row Normalization Can Beat Adam and Muon | Different optimizers can fit the same training data while selecting classifiers with substantially different geometries, but whether t… | ↗Not briefed |
hn hn-ai | 1mo | Viewport Lab – Open-source responsive tester with sync scroll(no telemetry) | — | ↗Not briefed |
bluesky bluesky-ai | 1mo | i think you should read literally anything about openai | i think you should read literally anything about openai | ↗Not briefed |
bluesky bluesky-ai | 1mo | Okay, this one's kind of interesting because Adobe absolutely sucks ass and something that potentially a) destroys Adobe's business model o… | Okay, this one's kind of interesting because Adobe absolutely sucks ass and something that potentially a) destroys Adobe's business mo… | ↗Not briefed |
arxiv arxiv-ai | 1mo | BeliefScope: Diagnosing Evidence-Driven Revision and Pressure-Induced Shifts in Large Language Models | A language model may revise the same proposition after receiving genuinely relevant evidence or after receiving directional user press… | ↗Not briefed |
bluesky bluesky-ai | 1mo | はやくシンギュラリティ(笑)とかいうのきてくれーーー マジ何が出来んのホント 「池に投げ入れた石の波紋」程度のLLMで超知能とか言ってるのを鼻で笑うレベルで人類に行き先を示してくれんの？ | はやくシンギュラリティ(笑)とかいうのきてくれーーー マジ何が出来んのホント 「池に投げ入れた石の波紋」程度のLLMで超知能とか言ってるのを鼻で笑うレベルで人類に行き先を示してくれんの？ | ↗Not briefed |
bluesky bluesky-ai | 1mo | Guaranteed. And part three will be: Expensiver model does review. Humans apparently not necessary. Code ships with zero oversight. Things… | Guaranteed. And part three will be: Expensiver model does review. Humans apparently not necessary. Code ships with zero oversight. T… | ↗Not briefed |