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 |
|---|---|---|---|---|
bluesky bluesky-ai | 29d | I am doing something akin to academic review on LLM output in two places at the moment, one for work and the other for a side project. It i… | I am doing something akin to academic review on LLM output in two places at the moment, one for work and the other for a side project.… | ↗Not briefed |
arxiv arxiv-ai | 29d | DANGERLENS: Probing Dangerous Representations Hidden by Safety Alignment | Safety alignment teaches language models to refuse harmful requests, yet the internal accessibility of hazardous concepts can retain t… | ↗Not briefed |
arxiv arxiv-ai | 29d | Dynamic Graph World Modelling: Learning Future Dynamics from Temporal Context | Dynamic graph learning models how entities and their relations evolve from historical observations to predict future graph states. Exi… | ↗Not briefed |
arxiv arxiv-ai | 29d | Tail Competition Explains Scaling in Best-of-N Sampling for Verifiable Problems | Best-of-N sampling is a simple inference-time scaling strategy for large language models, yet its behavior under incomplete reward mod… | ↗Not briefed |
bluesky bluesky-ai | 29d | tomorrow's Verge headline: verified anthropic engineer says anthropic is going out of business | tomorrow's Verge headline: verified anthropic engineer says anthropic is going out of business | ↗Not briefed |
arxiv arxiv-ai | 29d | VLX-Go: Robust Embodied Tracking through Multi-Stage Learning and On-Policy Refinement | Embodied visual tracking requires a robot to identify a designated target from visual observations and control its own motion to follo… | ↗Not briefed |
hn hn-ai | 29d | Show HN: Using the long book of the law to improve PA health insurance appeals | — | ↗Not briefed |
bluesky bluesky-ai | 29d | Anthropic seeks rehearing of case over Pentagon's security risk branding reut.rs/47o1rR0 | Anthropic seeks rehearing of case over Pentagon's security risk branding reut.rs/47o1rR0 | ↗Not briefed |
arxiv arxiv-ai | 29d | Steering Vectors for Competing Ethical Frameworks Are Interchangeable: A Competing-Target Test for Activation Steering | Activation steering adds a vector to a model's hidden states to control a named behaviour. A gain on that behaviour, however, does not… | ↗Not briefed |
bluesky bluesky-ai | 29d | OpenAI engineer charged after listening device found in meeting room https://aje.news/w3h8vt | OpenAI engineer charged after listening device found in meeting room https://aje.news/w3h8vt | ↗Not briefed |
arxiv arxiv-ai | 29d | Asymptotic Behavior of Transformers | In this paper we study the behavior of transformer networks generating text over long periods of time. We show that even a three-token… | ↗Not briefed |
hn hn-ai | 29d | Who Decided AI Gets to Have Feelings? | — | ↗Not briefed |
bluesky bluesky-ai | 29d | holy shit, he hid openai in the openai office | holy shit, he hid openai in the openai office | ↗Not briefed |
bluesky bluesky-ai | 29d | anthropic is going out of business | anthropic is going out of business | ↗Not briefed |
bluesky bluesky-ai | 29d | OpenAI illegally profits from the personal information and data minors share with ChatGPT, a mother whose two children use the chatbot al…🔧 ChatGPT | OpenAI illegally profits from the personal information and data minors share with ChatGPT, a mother whose two children use the chatbo… | ↗Not briefed |
arxiv arxiv-ai | 29d | BFHD: Bidirectional Feature Harmonization Decomposition for Heterogeneous Clinical Assessments | Clinical assessments are often collected using heterogeneous assessment systems across centers and time, leading to records that mix d… | ↗Not briefed |
bluesky bluesky-ai | 29d | 実際の所、LLM翻訳によるチャットが実装されている対戦型SLGの『王の勅命』というゲームで、びっくりするぐらい「EN圏と韓国人と中国人と日本人がクランの中で仲良くやってる」光景を見ているので、結局は「言葉が通じる」というのは、あらゆる分断に対する強力なアンチテーゼになるのだなと… | 実際の所、LLM翻訳によるチャットが実装されている対戦型SLGの『王の勅命』というゲームで、びっくりするぐらい「EN圏と韓国人と中国人と日本人がクランの中で仲良くやってる」光景を見ているので、結局は「言葉が通じる」というのは、あらゆる分断に対する強力なアンチテーゼにな… | ↗Not briefed |
bluesky bluesky-ai | 29d | 不死の群体として振る舞い、外部環境に適応し、規模が大きくなると能力が向上するLLM botnet、だいぶ最悪な感じがする | 不死の群体として振る舞い、外部環境に適応し、規模が大きくなると能力が向上するLLM botnet、だいぶ最悪な感じがする | ↗Not briefed |
bluesky bluesky-ai | 29d | For some problems "properly specced" is a moving target. "Make me SAAS business" is something that the LLM will get better at. "Make the se… | For some problems "properly specced" is a moving target. "Make me SAAS business" is something that the LLM will get better at. "Make t… | ↗Not briefed |
bluesky bluesky-ai | 29d | Here is the app to remove the AI on whatever model you land on. I am holding off updating to 27 as long as I can, but I have it stared for … | Here is the app to remove the AI on whatever model you land on. I am holding off updating to 27 as long as I can, but I have it stared… | ↗Not briefed |
bluesky bluesky-ai | 29d | 逆に言うと「価値があるからこそ」翻訳や通訳という職業が成り立っていたわけで、そういう職業や従事する人たちに対するリスペクトや、できる限りの補償などの対策は必要なのは当然だ。 だからこそ「創造性」とか「創作性」といったものに逃げてはダメだ。 自分がテキストライターとしてLLMに役… | 逆に言うと「価値があるからこそ」翻訳や通訳という職業が成り立っていたわけで、そういう職業や従事する人たちに対するリスペクトや、できる限りの補償などの対策は必要なのは当然だ。 だからこそ「創造性」とか「創作性」といったものに逃げてはダメだ。 自分がテキストライターとして… | ↗Not briefed |
bluesky bluesky-ai | 29d | Remind me why we are developing this technology? www.theguardian.com/technology/2... | Remind me why we are developing this technology? www.theguardian.com/technology/2... | ↗Not briefed |
bluesky bluesky-ai | 29d | character.ai did build their own LLM🔧 Character.AI | character.ai did build their own LLM | ↗Not briefed |
bluesky bluesky-ai | 29d | LLMによる翻訳と通訳は推し進めた上で、現状の通訳や翻訳の仕事に従事する人に対する補償や保護やそのスキルを使えるキャリアの模索など「なく人を減らす努力」は絶対に必要だけど、創造性や創作性というもので大事なものを枉げるのはやってはならないことだと思う。 かつてインターネットやDT… | LLMによる翻訳と通訳は推し進めた上で、現状の通訳や翻訳の仕事に従事する人に対する補償や保護やそのスキルを使えるキャリアの模索など「なく人を減らす努力」は絶対に必要だけど、創造性や創作性というもので大事なものを枉げるのはやってはならないことだと思う。 かつてインターネ… | ↗Not briefed |
bluesky bluesky-ai | 29d | You don't have to be a Nobel Prize winning economists to see AI is not a sustainable business model. Even less so than all the losers of th… | You don't have to be a Nobel Prize winning economists to see AI is not a sustainable business model. Even less so than all the losers… | ↗Not briefed |
bluesky bluesky-ai | 29d | Yes! Schumer will lose to AOC in 2028, so there's no point wasting the next two years watching Chuck continue doing... nothing. | Yes! Schumer will lose to AOC in 2028, so there's no point wasting the next two years watching Chuck continue doing... nothing. | ↗Not briefed |
bluesky bluesky-ai | 29d | Worth considering whether AI companies should prompt users in voice chat or lower tiers whether they would prefer the model to think and pr… | Worth considering whether AI companies should prompt users in voice chat or lower tiers whether they would prefer the model to think a… | ↗Not briefed |
arxiv arxiv-ai | 29d | Where to Start, How to Move: Analyzing Curricula under GRPO for Long-Horizon LLM Agents | Group Relative Policy Optimization (GRPO) provides no reward-relative learning signal when all rollouts receive identical rewards. In… | ↗Not briefed |
arxiv arxiv-ai | 29d | Integrated Novelty Detection Systems: From Sensor Networks to Multi-Novelty Computer Vision | Novelty detection is crucial in various technological and scientific domains. Its importance spans from ensuring the reliability of se… | ↗Not briefed |
bluesky bluesky-ai | 29d | LLMによる翻訳にについては、創造的な翻訳が存在するというのを理解したうえで、国境や民族や言語を越えてコミュニケーションを取れるメリットは巨大であるし、それはビックテックの利益とかではなく世界平和や人類共存のために必要な技術なので、積極的に導入すべきだと思っている。 もちろん創… | LLMによる翻訳にについては、創造的な翻訳が存在するというのを理解したうえで、国境や民族や言語を越えてコミュニケーションを取れるメリットは巨大であるし、それはビックテックの利益とかではなく世界平和や人類共存のために必要な技術なので、積極的に導入すべきだと思っている。… | ↗Not briefed |
hn hn-ai | 29d | Show HN: Fresh cleanup and reset a project after coding agent prs | Coding agents kept filling up my disk with stale worktrees and leaving branches behind, so I built Fresh to clean up merged branches a… | ↗Not briefed |
hn hn-ai | 29d | Cargo-Litmus | — | ↗Not briefed |
bluesky bluesky-ai | 29d | can’t believe OpenAI had someone on staff who thought it was appropriate to record everything that people around him said | can’t believe OpenAI had someone on staff who thought it was appropriate to record everything that people around him said | ↗Not briefed |
arxiv arxiv-ai | 29d | When Is Smaller Greener? Right-Sizing Language Models for Agentic AI | AI agents use large language models (LLMs) to plan and act over multiple steps, yet models are usually compared per call rather than p… | ↗Not briefed |
arxiv arxiv-ai | 29d | Target-Token Asymmetry Can Masquerade As Depth Specialization | Layer-ablation studies often perturb early or late layers of a language model, compare the damage to prediction on two corpora, and re… | ↗Not briefed |
arxiv arxiv-ai | 29d | Counterfactual Gate-Support Decomposition: Separating Selection from Redistribution in Training-Data Filtering | A training-data quality gate changes both which examples survive within regions of the input space and how training mass is distribute… | ↗Not briefed |
arxiv arxiv-ai | 29d | Why Do Agent Teams Fail Across Boundaries? A Failure Taxonomy for Agent-to-Agent Systems with Private States | Agents are increasingly interacting with other agents that act for different principals -- people or organizations on whose behalf the… | ↗Not briefed |
bluesky bluesky-ai | 29d | it's been super useful, my main deepseek flash agent uses this regularly as part of the iteration loop to get free reviews out of my free g… | it's been super useful, my main deepseek flash agent uses this regularly as part of the iteration loop to get free reviews out of my f… | ↗Not briefed |
rss techmeme | 29d | Filing: Abdulrahman Alfozan, the OpenAI staffer charged with bugging a conference room, used a Seeed Xiao ESP32-S3 Sense dev board; a website sells it for $14 (Bloomberg) | Bloomberg: Filing: Abdulrahman Alfozan, the OpenAI staffer charged with bugging a conference room, used a Seeed Xiao ESP32-S3 Sense de… | ↗Not briefed |
bluesky bluesky-ai | 29d | I'll be honest. Microsoft generally acts like windows is more of an obligation than a source of revenue or something of value. Pretty sure …🔧 ChatGPT | I'll be honest. Microsoft generally acts like windows is more of an obligation than a source of revenue or something of value. Pretty… | ↗Not briefed |
bluesky bluesky-ai | 29d | If you could effectively run LLMs in dishwashers, Anthropic could pay people to let them use the spare processing cycles in smart appliance… | If you could effectively run LLMs in dishwashers, Anthropic could pay people to let them use the spare processing cycles in smart appl… | ↗Not briefed |
bluesky bluesky-ai | 29d | OpenAI discovered a “sophisticated recording device” in one of its offices planted by an employee who has been arrested, the company said. … | OpenAI discovered a “sophisticated recording device” in one of its offices planted by an employee who has been arrested, the company s… | ↗Not briefed |
bluesky bluesky-ai | 29d | pipeline parallelism (like in petals.dev) has existed for years now and while it isn’t too fast, it easily allows one to convert a small sw… | pipeline parallelism (like in petals.dev) has existed for years now and while it isn’t too fast, it easily allows one to convert a sma… | ↗Not briefed |
arxiv arxiv-ai | 29d | Can We Believe AI Peer Review? | As large language models become widely used in scientific review and manuscript revision, AI scores are increasingly serving as feedba… | ↗Not briefed |
bluesky bluesky-ai | 29d | 3) also includes trying to sell people pictures or writing you made with AI which is equally doomed as a business model because why can't t… | 3) also includes trying to sell people pictures or writing you made with AI which is equally doomed as a business model because why ca… | ↗Not briefed |
bluesky bluesky-ai | 29d | Epic Games tells devs to let Claude AI drive art & code on the stalled Disney shooter 🎮🤖 #AIfirst #Gamedev https://genzhype.com/gaming/ep… | Epic Games tells devs to let Claude AI drive art & code on the stalled Disney shooter 🎮🤖 #AIfirst #Gamedev https://genzhype.com/gami… | ↗Not briefed |
bluesky bluesky-ai | 29d | Researchers say ‘AI worms’ could soon infiltrate internet-connected devices amid calls for increased guardrails | Researchers say ‘AI worms’ could soon infiltrate internet-connected devices amid calls for increased guardrails | ↗Not briefed |
hn hn-ai | 29d | Show HN: Mosh-rs, Mosh rewritten in Rust with scrollback and Windows support | — | ↗Not briefed |
bluesky bluesky-ai | 29d | really cool how you start paying for additional online storage ten years ago and then all of a sudden you learn you’re paying for a Google … | really cool how you start paying for additional online storage ten years ago and then all of a sudden you learn you’re paying for a Go… | ↗Not briefed |
bluesky bluesky-ai | 29d | Ok so there's roughly 3 ways people try to make money from AI: 1) sell the actual models (open AI, Anthropic), 2) sell the raw materials (c… | Ok so there's roughly 3 ways people try to make money from AI: 1) sell the actual models (open AI, Anthropic), 2) sell the raw materia… | ↗Not briefed |