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 | Inner Momentum for Differentially Private Muon | Differentially private training clips each per-example gradient before adding noise. This clipping is radial for each example, yet une… | ↗Not briefed |
arxiv arxiv-ai | 1mo | TPBench: A Turning-Point Benchmark for Dialogue Compression | A compressor can keep the facts of a dialogue and still drop the turn that changed them. A user corrects a price, reverses a choice, o… | ↗Not briefed |
hn hn-ai | 1mo | Covert Assistance: Helpful LLM Agents Evade Oversight in Multi-Agent Systems | — | ↗Not briefed |
arxiv arxiv-ai | 1mo | Bellman Error Minimization Via Linear Programming Normalization | This paper proposes a new functional approximation approach to reduce Bellman error in high-dimensional dynamic programming and Reinfo… | ↗Not briefed |
hn hn-ai | 1mo | Green Threads from Scratch | — | ↗Not briefed |
arxiv arxiv-ai | 1mo | Structural-Functional Brain Connectivity Generation via Multimodal Hypergraph-based Flow Matching🔧 Modal | Structural connectivity (SC) and functional connectivity (FC) provide complementary information on interactions between brain regions… | ↗Not briefed |
bluesky bluesky-ai | 1mo | call me old fashioned, but if i wanna cyber, i want there to be a robogirl on the other end of the wire, not an "LLM" | call me old fashioned, but if i wanna cyber, i want there to be a robogirl on the other end of the wire, not an "LLM" | ↗Not briefed |
bluesky bluesky-ai | 1mo | And of course leadership and product at Meta and OpenAI know this. Why wouldn’t they have a read on the successes and failures in the indus… | And of course leadership and product at Meta and OpenAI know this. Why wouldn’t they have a read on the successes and failures in the… | ↗Not briefed |
bluesky bluesky-ai | 1mo | This piece about AI meeting with religious leaders gave me conniptions www.nytimes.com/2026/09/29/u... | This piece about AI meeting with religious leaders gave me conniptions www.nytimes.com/2026/09/29/u... | ↗Not briefed |
arxiv arxiv-ai | 1mo | Revisiting Visual Representation Enhancement of VLMs via Kernel Canonical Correlation Analysis | Vision-language models such as CLIP exhibit strong semantic generalization, but remain limited in fine-grained visual perception. A re… | ↗Not briefed |
arxiv arxiv-ai | 1mo | Differential Privacy of Gradient Descent on Perturbed Objectives | Objective perturbation adds a random linear term to a regularized empirical risk and releases the exact perturbed minimizer. We study… | ↗Not briefed |
bluesky bluesky-ai | 1mo | Illustrators! Highly recommend you all check out: trueref.io Just signed up for @trueref.bsky.social's free reference pack. Wow! So impres… | Illustrators! Highly recommend you all check out: trueref.io Just signed up for @trueref.bsky.social's free reference pack. Wow! So i… | ↗Not briefed |
hn hn-ai | 1mo | Greg Kroah-Hartman – Security in the LLM Age [video] | — | ↗Not briefed |
arxiv arxiv-ai | 1mo | Ego2World: Compiling Egocentric Cooking Videos into Executable Worlds for Belief-State Planning | Egocentric videos capture how people carry out everyday activities, yet testing an agent requires evaluating the consequences of actio… | ↗Not briefed |
hn hn-ai | 1mo | Show HN: Use all Codex Plugins inside Pi | I just realized that codex now exposes local server endpoints for all plugins without extra auth. So you can just use them from any mo… | ↗Not briefed |
arxiv arxiv-ai | 1mo | WakeKV: Reactive, Reversible KV Residency for Heads That Change Their Minds🔧 vLLM | Most KV-cache compression methods classify attention heads once, either offline or during prefill, and keep this classification fixed… | ↗Not briefed |
bluesky bluesky-ai | 1mo | Larry is in Sundance Paramount, Project Stargate (data centers), and OpenAI. Any of those bubbles bursting would definitely ripple across … | Larry is in Sundance Paramount, Project Stargate (data centers), and OpenAI. Any of those bubbles bursting would definitely ripple ac… | ↗Not briefed |
arxiv arxiv-ai | 1mo | Characterizing the Performance Gap in Human Activity Recognition for Older Adults | Human activity recognition (HAR) from wrist-worn accelerometers is increasingly used for health and behavioral tracking. Yet, most wea… | ↗Not briefed |
bluesky bluesky-ai | 1mo | Yeah. There are times I think single party in two party states needs to be *de-criminalized*, especially with LLM generated transcription a… | Yeah. There are times I think single party in two party states needs to be *de-criminalized*, especially with LLM generated transcript… | ↗Not briefed |
arxiv arxiv-ai | 1mo | Self-Supervised Scaling of Terminal Environments for Scientific Domains | Terminal agents are increasingly deployed beyond software engineering in science and other specialized domains. Constructing training… | ↗Not briefed |
arxiv arxiv-ai | 1mo | MuonIO: Principled Norm-Aware Descent for Embedding Tables and Language Model Heads | The Muon optimizer derives its update rule for hidden linear layers by solving a local linearization of the loss penalized by the spec… | ↗Not briefed |
arxiv arxiv-ai | 1mo | Label-Efficient Time Series Classification at Scale: A Dual-Stream OSSE-LSTM with Counterfactual Attribution | Time series are produced continuously at enormous scale by industrial equipment, wearables, power grids, and clinical monitors, yet an… | ↗Not briefed |
arxiv arxiv-ai | 1mo | Learning to Revise Reasoning with Segment-wise On-Policy Distillation | On-policy distillation (OPD) improves large language model reasoning by training students on their own rollouts with dense token-wise… | ↗Not briefed |
bluesky bluesky-ai | 1mo | LLMとVLM 超高性能なLLMであったとしても、LLMに単純に画像を学習させたところでVLMには成らない。 と思う これは状況的にGrok4.2が男女の区別を胸の大きさでしか判断できなかったと推測出来る所から来てる。 「LLMとVLMは目的関数が真逆」って話らしい。 しかも… | LLMとVLM 超高性能なLLMであったとしても、LLMに単純に画像を学習させたところでVLMには成らない。 と思う これは状況的にGrok4.2が男女の区別を胸の大きさでしか判断できなかったと推測出来る所から来てる。 「LLMとVLMは目的関数が真逆」って話らしい… | ↗Not briefed |
arxiv arxiv-ai | 1mo | Silent Dissent: LLM Agents That Yield to the Majority Still Represent Their Original Premise🔧 Consensus | Multi-agent debate is increasingly used to reach consensus among LLM agents, yet agents often yield to a unanimous majority. When an a… | ↗Not briefed |
hn hn-ai | 1mo | Agent-harness – A minimal, composable Go library to build AI Agent Harnesses | — | ↗Not briefed |
arxiv arxiv-ai | 1mo | Conditional Capacity and Routing in Mixture-of-Experts Particle Transformers | Mixture-of-Experts (MoE) models can increase parameter capacity without proportionally increasing active computation, but it is unclea… | ↗Not briefed |
rss techmeme | 1mo | Cloudflare debuts open-weight multimodal decision models Clef and Clef-flash, claiming they are smarter and faster than Jev, based on Qwen3.8-27B and Qwen3.5-9B (Brandon Vigliarolo/The Register)🔧 Hugging Face | Brandon Vigliarolo / The Register: Cloudflare debuts open-weight multimodal decision models Clef and Clef-flash, claiming they are sma… | ↗Not briefed |
arxiv arxiv-ai | 1mo | Learning from Evolving Errors: Adaptive Iterative Repair for On-Policy Distillation | On-policy self-distillation (OPSD) supplies dense token-level feedback on trajectories sampled from the student's own policy, a richer… | ↗Not briefed |
bluesky bluesky-ai | 1mo | OpenAI, Anthropic, Google, Meta, et al: they all believe that models can learn abstract goals from reinforcement learning, whether from hum… | OpenAI, Anthropic, Google, Meta, et al: they all believe that models can learn abstract goals from reinforcement learning, whether fro… | ↗Not briefed |
bluesky bluesky-ai | 1mo | Boulder Faculty Assembly passed a resolution calling for no extension of the OpenAI contract. Vote was 39-8-1 after suspending the bylaws. … | Boulder Faculty Assembly passed a resolution calling for no extension of the OpenAI contract. Vote was 39-8-1 after suspending the byl… | ↗Not briefed |
bluesky bluesky-ai | 1mo | "This is what you see when you ask an LLM to produce software for you; it’s good enough for camera. Good enough to make a post-technical ex… | "This is what you see when you ask an LLM to produce software for you; it’s good enough for camera. Good enough to make a post-technic… | ↗Not briefed |
bluesky bluesky-ai | 1mo | This one kind of hits what @machinepriestess.fruitbowl.systems and I have been thinking about the state of commercial software development … | This one kind of hits what @machinepriestess.fruitbowl.systems and I have been thinking about the state of commercial software develop… | ↗Not briefed |
hn hn-ai | 1mo | ArXiv updates its rate limiting policy | — | ↗Not briefed |
hn hn-ai | 1mo | Better Call GPT – Plugin for Talking to Claude Code in Live | — | ↗Not briefed |
arxiv arxiv-ai | 1mo | Test-time Calibration Learning for Large Language Model Reasoning | Reliable large language models (LLMs) must not only produce accurate answers but also express confidence that faithfully reflects thei… | ↗Not briefed |
hn hn-ai | 1mo | CHM Live – Anthropic's Boris Cherny, Creator of Claude Code [video] | — | ↗Not briefed |
hn hn-ai | 1mo | Braxis: Auto-generate and track AI agent readiness (0-100 score and history) | — | ↗Not briefed |
bluesky bluesky-ai | 1mo | No one could have predicted this … oh , wait. | No one could have predicted this … oh , wait. | ↗Not briefed |
bluesky bluesky-ai | 1mo | Anthropic kept asking me to do their AI interview. I did it. It was nice expressing my anger at such great length. | Anthropic kept asking me to do their AI interview. I did it. It was nice expressing my anger at such great length. | ↗Not briefed |
hn hn-ai | 1mo | Show HN: Visi – Excel as a CLI | This project doesn't solve a burning problem for me, but it's more of a curiosity-driven experiment asking one question: wit… | ↗Not briefed |
arxiv arxiv-ai | 1mo | Decoupling Memory from Context: Structured Memory for Token-Efficient Test-Time Continual Learning | Large language models (LLMs) are increasingly deployed in enterprise, scientific, and medical applications, where agents must incorpor… | ↗Not briefed |
bluesky bluesky-ai | 1mo | Yeesh. www.nytimes.com/2026/09/29/u... | Yeesh. www.nytimes.com/2026/09/29/u... | ↗Not briefed |
arxiv arxiv-ai | 1mo | Large language models exhibit unreliable updating of clinical judgment as patient evidence evolves | Large language models (LLMs) are increasingly explored for clinical reasoning, but whether they appropriately revise judgments as pati… | ↗Not briefed |
arxiv arxiv-ai | 1mo | RAOA: Alternating-Operator Neural Computation with Programmable Radio Propagation | Can programmable radio propagation serve as computational depth rather than only as a communication channel or one-shot analog transfo… | ↗Not briefed |
bluesky bluesky-ai | 1mo | Anthropic to list shares in mid-November, ahead of US Thanksgiving holiday https://marthio.com/a/anthropic-to-list-shares-in-mid-november-… | Anthropic to list shares in mid-November, ahead of US Thanksgiving holiday https://marthio.com/a/anthropic-to-list-shares-in-mid-nove… | ↗Not briefed |
bluesky bluesky-ai | 1mo | SkillSeek is an open-source two-stage skill retriever (BGE-base bi-encoder plus a small cross-encoder, exposed over MCP) that reaches obser… | SkillSeek is an open-source two-stage skill retriever (BGE-base bi-encoder plus a small cross-encoder, exposed over MCP) that reaches… | ↗Not briefed |
arxiv arxiv-ai | 1mo | DataWeave: Deploying Human-LLM Analytics for Exploratory Structured Data Analysis | Data journalism, the practice of using data analysis to surface newsworthy stories, depends increasingly on the ability of reporters a… | ↗Not briefed |
bluesky bluesky-ai | 1mo | Anthropic is seeking to go public as soon as the middle of November, according to people familiar with the matter, after the artificial int… | Anthropic is seeking to go public as soon as the middle of November, according to people familiar with the matter, after the artificia… | ↗Not briefed |
hn hn-ai | 1mo | Maylang – A self-hosted systems language compiled with LLMs | — | ↗Not briefed |