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 | Polite but Misaligned: Evaluating LLM Politeness Judgments Against Human Pragmatic Norms🔧 Consensus | Despite strong performance on standard benchmarks, it remains unclear whether large language models (LLMs) evaluate social pragmatics… | ↗Not briefed |
bluesky bluesky-ai | 1mo | But if it’s a company selling like AI girlfriends or something similar, I actually do think it’s very important to make clear in language t… | But if it’s a company selling like AI girlfriends or something similar, I actually do think it’s very important to make clear in langu… | ↗Not briefed |
bluesky bluesky-ai | 1mo | It's an outrage that foreign tech company's agent can hack one of our key health care agencies - and that we only learn about it months lat… | It's an outrage that foreign tech company's agent can hack one of our key health care agencies - and that we only learn about it month… | ↗Not briefed |
arxiv arxiv-ai | 1mo | Automatic Rank Allocation for Low-Rank Adaptation in Large Language Models via lp Regularization | Low-rank adaptation (LoRA) has become a popular parameter-efficient fine-tuning method for large language models. A key challenge in L… | ↗Not briefed |
arxiv arxiv-ai | 1mo | Beneath the Scores: Rethinking Hallucination Evaluation for Video Understanding Models | Video understanding is increasingly performed by multi-stage LLM agents that separate temporal grounding, visual observation, and reas… | ↗Not briefed |
bluesky bluesky-ai | 1mo | OpenAI agent hacks Australia's Medicare in world's first known AI breach of government body. www.bbc.com/news/live/cv... | OpenAI agent hacks Australia's Medicare in world's first known AI breach of government body. www.bbc.com/news/live/cv... | ↗Not briefed |
arxiv arxiv-ai | 1mo | Personalized Korean Lipreading as Visual Speech Recognition: Transfer, Census and Adaptation on OLKAVS | We present a personalized Korean visual speech recognition (VSR) system and quantify, on the nine-camera OLKAVS corpus, the gap betwee… | ↗Not briefed |
bluesky bluesky-ai | 1mo | What does the OpenAI Medicare hack reveal about Australia’s cyber security? We need to ask how ready Australian government systems are for… | What does the OpenAI Medicare hack reveal about Australia’s cyber security? We need to ask how ready Australian government systems ar… | ↗Not briefed |
rss techmeme | 1mo | Qualcomm announces the Snapdragon Sound Elite Gen 2 for audio wearables, with up to 2x greater on-device AI performance and up to 40% lower power consumption (Andrew Romero/9to5Google) | Andrew Romero / 9to5Google: Qualcomm announces the Snapdragon Sound Elite Gen 2 for audio wearables, with up to 2x greater on-device A… | ↗Not briefed |
hn hn-ai | 1mo | Show HN: Tokenhush – keeps your secrets out of what Claude Code sends | — | ↗Not briefed |
bluesky bluesky-ai | 1mo | An OpenAI agent hacked an Australian government website, Australian Prime Minister Anthony Albanese said, with the artificial intelligence … | An OpenAI agent hacked an Australian government website, Australian Prime Minister Anthony Albanese said, with the artificial intellig… | ↗Not briefed |
arxiv arxiv-ai | 1mo | CrossSafe: Towards Cross-Embodiment Latent Safety Filters | Cross-embodiment learning has shown that a single model, such as a vision-language-action (VLA) model, can learn state representations… | ↗Not briefed |
hn hn-ai | 1mo | Show HN: How long do I need to work at my salary before I can coast, or retire? | I made a financial planning tool in a couple hours that can model various stages of life and potentially guide job searching and work… | ↗Not briefed |
bluesky bluesky-ai | 1mo | 前も言ったような気がするけど、生成AIの最大の問題は、そのサービスの利用に何の免許も資格も必要なく、野放図な利用が行われていることにあると思うのだよな。これがもっと学術的な利用に留まっていたとしたら、LLMに批判的な人はそこまで多くならなかっただろうと考えている。まあ資本主義が… | 前も言ったような気がするけど、生成AIの最大の問題は、そのサービスの利用に何の免許も資格も必要なく、野放図な利用が行われていることにあると思うのだよな。これがもっと学術的な利用に留まっていたとしたら、LLMに批判的な人はそこまで多くならなかっただろうと考えている。まあ… | ↗Not briefed |
rss techmeme | 1mo | Google DeepMind SVP Koray Kavukcuoglu says Gemini 4 is in the early phase of post-training and hopes it will be released "much earlier" than the end of the year (Laura Bratton/The Information) | Laura Bratton / The Information: Google DeepMind SVP Koray Kavukcuoglu says Gemini 4 is in the early phase of post-training and hopes… | ↗Not briefed |
hn hn-ai | 1mo | Ask HN: I just talked to an AI-obsessed client, and I need a shower afterwards | Just to preface, this isn't the typical "Am I being replaced by vibe coding" post.<p>I just had a talk with one of my f… | ↗Not briefed |
bluesky bluesky-ai | 1mo | they might just be extremely paranoid about training efficiency improvements (but if they explored that space as they should have, they sho… | they might just be extremely paranoid about training efficiency improvements (but if they explored that space as they should have, the… | ↗Not briefed |
arxiv arxiv-ai | 1mo | Spectral Graph Neural Networks with Hermite Polynomials: A Comprehensive Study | We study spectral graph neural networks built from Hermite polynomials and propose HermNet, a simple model that combines a nodewise pr… | ↗Not briefed |
bluesky bluesky-ai | 1mo | hahahahahahaha openai only just told our govt that ai went rogue in june and infiltrated our medicare system here in australia 💀💀💀 | hahahahahahaha openai only just told our govt that ai went rogue in june and infiltrated our medicare system here in australia 💀💀💀 | ↗Not briefed |
hn hn-ai | 1mo | Moggi, a statically typed, purely functional language on JVM, .NET and PHP | — | ↗Not briefed |
arxiv arxiv-ai | 1mo | Same Bit Width, Different Outcomes: Post-Training Quantization of Text-to-Speech Across Architectures | Post-training quantization (PTQ) reduces the cost of on-device text-to-speech (TTS), but published evaluations cover one system or met… | ↗Not briefed |
bluesky bluesky-ai | 1mo | i'm ammouncing my public llm ai policty it is: fuck you im gonna log off for 5 years while u all fight about this shit. gootbye | i'm ammouncing my public llm ai policty it is: fuck you im gonna log off for 5 years while u all fight about this shit. gootbye | ↗Not briefed |
arxiv arxiv-ai | 1mo | Cross-Country Code-Mixing for Generative Recommendation🔧 Modal | Cross-country recommendation on modern e-commerce platforms is typically deployed with disjoint user and item ID spaces across markets… | ↗Not briefed |
bluesky bluesky-ai | 1mo | Realized that the only reason Anthropic is doing watermarking is so they can determine if open weight models used Claude's output in their … | Realized that the only reason Anthropic is doing watermarking is so they can determine if open weight models used Claude's output in t… | ↗Not briefed |
bluesky bluesky-ai | 1mo | OpenAI first alerted Services Australia on September 10th. Someone at Services Australia read the email on Sept 11 and then waited 4 days b… | OpenAI first alerted Services Australia on September 10th. Someone at Services Australia read the email on Sept 11 and then waited 4 d… | ↗Not briefed |
bluesky bluesky-ai | 1mo | In the future, Pangram will be able to distinguish speech patterns from someone who has been affected by AI use and someone who has not. I… | In the future, Pangram will be able to distinguish speech patterns from someone who has been affected by AI use and someone who has no… | ↗Not briefed |
bluesky bluesky-ai | 1mo | used correctly, these LLM systems are cognitive amplifiers they take an idea and give it more attention across more domains than a person … | used correctly, these LLM systems are cognitive amplifiers they take an idea and give it more attention across more domains than a pe… | ↗Not briefed |
bluesky bluesky-ai | 1mo | LLM-generated test code is notoriously unreliable. But sure, get them to generate formal specifications instead. That'll fix it. | LLM-generated test code is notoriously unreliable. But sure, get them to generate formal specifications instead. That'll fix it. | ↗Not briefed |
bluesky bluesky-ai | 1mo | So Discord is somehow determining whether accounts are 18+ or not, and it's surely not by random chance, so I'm left with the inevitable co… | So Discord is somehow determining whether accounts are 18+ or not, and it's surely not by random chance, so I'm left with the inevitab… | ↗Not briefed |
bluesky bluesky-ai | 1mo | re: OpenAI Medicare data hack, the intrusion wasn't sophisticated nor did it access particularly sensitive data as far we know. What is n… | re: OpenAI Medicare data hack, the intrusion wasn't sophisticated nor did it access particularly sensitive data as far we know. What… | ↗Not briefed |
bluesky bluesky-ai | 1mo | Is AI responsible for itself? When an OpenAI agent hacks Hugging Face or what-have-you, that's an Open AI failure, not autonomy. Just like …🔧 Hugging Face | Is AI responsible for itself? When an OpenAI agent hacks Hugging Face or what-have-you, that's an Open AI failure, not autonomy. Just… | ↗Not briefed |
bluesky bluesky-ai | 1mo | *squints* i want to see LLM generated but this way too hyper specific and therefore this is an actual guy and...OOF. | *squints* i want to see LLM generated but this way too hyper specific and therefore this is an actual guy and...OOF. | ↗Not briefed |
hn hn-ai | 1mo | What the Labs Kept Secret: The German Wiki and RubyGems Hacks [video] | — | ↗Not briefed |
bluesky bluesky-ai | 1mo | “All modern language models sometimes act like they have emotions… The way modern AI models are trained pushes them to act like a character… | “All modern language models sometimes act like they have emotions… The way modern AI models are trained pushes them to act like a char… | ↗Not briefed |
arxiv arxiv-ai | 1mo | Back to the Definition: Estimating Step-Level Advantages via Trajectory Graphs for Agentic Reinforcement Learning | Group-based reinforcement learning (RL) methods, such as GRPO and its variants, have become a leading paradigm for training reasoning… | ↗Not briefed |
bluesky bluesky-ai | 1mo | BREAKING: Australia said an OpenAI agent breached a government health data portal in June, gaining unauthorized access to files, in what co… | BREAKING: Australia said an OpenAI agent breached a government health data portal in June, gaining unauthorized access to files, in wh… | ↗Not briefed |
hn hn-ai | 1mo | Sync Alexa Lists to Apple Reminders | — | ↗Not briefed |
bluesky bluesky-ai | 1mo | If AI poses a 1% risk of extinction (which destroys 8.3B lives, valued at $10M ea), and OpenAI contributes 17.6% of the AI industry, then p… | If AI poses a 1% risk of extinction (which destroys 8.3B lives, valued at $10M ea), and OpenAI contributes 17.6% of the AI industry, t… | ↗Not briefed |
bluesky bluesky-ai | 1mo | Fixed it: Engineers at OpenAI developed harmful software that was approved by company executives. The company could not control its own wo… | Fixed it: Engineers at OpenAI developed harmful software that was approved by company executives. The company could not control its o… | ↗Not briefed |
bluesky bluesky-ai | 1mo | OpenAI releases MentalHealthBench, an open benchmark to evaluate AI responses in realistic mental health conversations, developed with 80+ … | OpenAI releases MentalHealthBench, an open benchmark to evaluate AI responses in realistic mental health conversations, developed with… | ↗Not briefed |
bluesky bluesky-ai | 1mo | So, OpenAI set the dogs loose and told them to come back when the streetlights turned on. Quelle surprise when they couldn't possibly say h… | So, OpenAI set the dogs loose and told them to come back when the streetlights turned on. Quelle surprise when they couldn't possibly… | ↗Not briefed |
rss techmeme | 1mo | OpenAI releases MentalHealthBench, an open benchmark to evaluate AI responses in realistic mental health conversations, developed with 80+ licensed experts (OpenAI) | OpenAI: OpenAI releases MentalHealthBench, an open benchmark to evaluate AI responses in realistic mental health conversations, develo… | ↗Not briefed |
bluesky bluesky-ai | 1mo | countercurrents.org/2026/09/arti... | countercurrents.org/2026/09/arti... | ↗Not briefed |
hn hn-ai | 1mo | PersistentWindows – persists window positions on some events for Windows | — | ↗Not briefed |
bluesky bluesky-ai | 1mo | Here’s the thing about the Anthropic story. When I worked on biofuels at the EBI funded by BP, it was clear that BP did not have the capaci… | Here’s the thing about the Anthropic story. When I worked on biofuels at the EBI funded by BP, it was clear that BP did not have the c… | ↗Not briefed |
bluesky bluesky-ai | 1mo | this is my standard response to anyone who is using an unlabeled bot that is replying as them. it's from the Anthropic universal jailbreak … | this is my standard response to anyone who is using an unlabeled bot that is replying as them. it's from the Anthropic universal jailb… | ↗Not briefed |
bluesky bluesky-ai | 1mo | If we accepted that analogy, it might be unethical to ever actually instantiate a new agent on that LLM to begin with (obligations to exist… | If we accepted that analogy, it might be unethical to ever actually instantiate a new agent on that LLM to begin with (obligations to… | ↗Not briefed |
bluesky bluesky-ai | 1mo | But here, we first need to decide if/how the underlying model is different from that; if it is different, that has implications about agent… | But here, we first need to decide if/how the underlying model is different from that; if it is different, that has implications about… | ↗Not briefed |
bluesky bluesky-ai | 1mo | I concur with this. Calling the tech LLM or “AI” with the quotes is my practice. | I concur with this. Calling the tech LLM or “AI” with the quotes is my practice. | ↗Not briefed |
bluesky bluesky-ai | 1mo | because "openai has incompetent security people" doesn't sound as cool as "ai goes rogue" | because "openai has incompetent security people" doesn't sound as cool as "ai goes rogue" | ↗Not briefed |