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 | 28d | Decoupled Marginal Sharpening for Training-Free Inference-Time Scaling | We study how to improve LLM reasoning at inference time without updating model weights or using external reward models. Recent reward-… | ↗Not briefed |
hn hn-ai | 28d | Linus Torvalds's first LLM-assisted patch series | — | ↗Not briefed |
arxiv arxiv-ai | 28d | Lost in Translation: Why Lower Bias Scores Don't Mean Fairer Models | Bias audits of multilingual language models usually compare a model's average stereotype score in English with its average in another… | ↗Not briefed |
arxiv arxiv-ai | 28d | AdaRSI: On-Demand Recursive Self-Improvement for Embodied Agents | Pretrained vision-language-action models (VLAs) often fail midway through unseen long-horizon tasks. Recursive self-improvement (RSI)… | ↗Not briefed |
bluesky bluesky-ai | 28d | I know from very reliable sources that OpenAI are using human translators for their own content, too. Just saying. | I know from very reliable sources that OpenAI are using human translators for their own content, too. Just saying. | ↗Not briefed |
arxiv arxiv-ai | 28d | SafeAdapt-Poker: Counterfactual Risk Calibration for Budgeted Adaptation under Opponent Shift | Opponent histories can mislead after behavioral shifts. SafeAdapt-Poker formulates budgeted adaptation under opponent shift as a dual-… | ↗Not briefed |
arxiv arxiv-ai | 28d | ChronoCulture: Evaluating Cultural Awareness Across Time in Large Language Models | Cultural attitudes and social norms evolve over time, yet most cultural benchmarks for large language models (LLMs) treat culture as a… | ↗Not briefed |
hn hn-ai | 28d | Metagenomic testing: technique that finds almost every infection, even unknown | — | ↗Not briefed |
bluesky bluesky-ai | 28d | Is epic trying to pivot from making and selling games to being a marketing front for Anthropic or something? Why are they doing all this? | Is epic trying to pivot from making and selling games to being a marketing front for Anthropic or something? Why are they doing all th… | ↗Not briefed |
hn hn-ai | 28d | OpenAI: Employee who left for Apple took files about unreleased device | — | ↗Not briefed |
hn hn-ai | 28d | Anthropic's Claude cruelty policy risks humanising AI | — | ↗Not briefed |
hn hn-ai | 28d | Show HN: An offline 0.8B model that ties Gemini Flash on OpenStreetMap questions | — | ↗Not briefed |
bluesky bluesky-ai | 28d | That's the business model of the AI companies and Silicon Valley generally. They can do it better with coding and AI. This applies to all … | That's the business model of the AI companies and Silicon Valley generally. They can do it better with coding and AI. This applies to… | ↗Not briefed |
bluesky bluesky-ai | 28d | #2 is my favorite because it’s amazing to see a LLM randomly go *blushes at you* It had to have learned that from Amino or like Wattpad. | #2 is my favorite because it’s amazing to see a LLM randomly go *blushes at you* It had to have learned that from Amino or like Watt… | ↗Not briefed |
arxiv arxiv-ai | 28d | Selective Growing Q-Networks (SGQN): An Analysis of Adaptive Action Grids in Decoupled Q-Learning | Value-based Q-learning for continuous control can discretise the action space, but the resolution of the discretisation trades early e… | ↗Not briefed |
bluesky bluesky-ai | 28d | The key point in this story is that the author is using AI to prep for matches in which he *can’t* use it. Extending that model to educatio… | The key point in this story is that the author is using AI to prep for matches in which he *can’t* use it. Extending that model to edu… | ↗Not briefed |
rss techmeme | 28d | CoreWeave stock drops 15%+ and its bond yields near 15%, over concerns that some tech companies are borrowing too heavily for the AI build-out; ORCL is down 7%+ (Michelle Fox/CNBC) | Michelle Fox / CNBC: CoreWeave stock drops 15%+ and its bond yields near 15%, over concerns that some tech companies are borrowing too… | ↗Not briefed |
arxiv arxiv-ai | 28d | Learning by Examples: Power Grid Steady State Analysis with a Tabular Foundation Model | Foundation models have shown strong generalization across scientific domains, but their use in power systems remains largely task-spec… | ↗Not briefed |
bluesky bluesky-ai | 28d | The trinity of English LLM prose: 1) books3 2) role playing forums for furries and weebs 3) Anglo-African RLHF | The trinity of English LLM prose: 1) books3 2) role playing forums for furries and weebs 3) Anglo-African RLHF | ↗Not briefed |
arxiv arxiv-ai | 28d | Equal CER, Unequal Damage: Severity-Weighted Character Error Rate for OCR | Character Error Rate (CER), the standard metric for OCR and vision-language models (VLMs), charges a typo as much as a wrong name. We… | ↗Not briefed |
hn hn-ai | 28d | Cutoff Bench: How outdated is the training data of major LLMs | — | ↗Not briefed |
rss theverge-ai | 28d | OpenAI claims an ex-employee took its product plans and presented them to Apple | OpenAI claims a former employee took more than 100 files related to its product plans before joining Apple. In a document filed on Wed… | ↗Not briefed |
arxiv arxiv-ai | 28d | MolLean: A Process-Verifiable Language for Chemical Reasoning | Chemical chain-of-thought can contain hallucinated structural and property claims. We thereby introduce MolLean, a process-verifiable… | ↗Not briefed |
arxiv arxiv-ai | 28d | Same Claim, Different Interpretations: How LLM Agents Resolve Ambiguity in Scientific Claims🔧 GPT-5 | Co-scientists and agentic scientific claim verifiers are increasingly used to assess claims across the sciences. Natural-language scie… | ↗Not briefed |
bluesky bluesky-ai | 28d | the Caine prize thing is ESPECIALLY appalling to me given the immense influence of Nigerian English in LLM output because it's a place wher… | the Caine prize thing is ESPECIALLY appalling to me given the immense influence of Nigerian English in LLM output because it's a place… | ↗Not briefed |
bluesky bluesky-ai | 28d | Pontos q gostaria de deixar claro pro debate q surge: - LLM é ferramenta probabilística, vai retornar resultado diferente pro msm prompt p… | Pontos q gostaria de deixar claro pro debate q surge: - LLM é ferramenta probabilística, vai retornar resultado diferente pro msm pro… | ↗Not briefed |
arxiv arxiv-ai | 28d | The COUNT: Counting N-grams Unlocks Scaling on Host Memory and Training-Free Adaptation | The scaling relationship of compute and performance for language models (LMs) has been extensively studied; more recently, models are… | ↗Not briefed |
bluesky bluesky-ai | 28d | El FBI detiene a un empleado de OpenAI de origen saudí por espiar reuniones confidenciales | El FBI detiene a un empleado de OpenAI de origen saudí por espiar reuniones confidenciales | ↗Not briefed |
bluesky bluesky-ai | 28d | ICYMI, Anthropic is advertising an executive assistant role for senior management in London, and the job description has me asking a lot of… | ICYMI, Anthropic is advertising an executive assistant role for senior management in London, and the job description has me asking a l… | ↗Not briefed |
bluesky bluesky-ai | 28d | In a new filing, OpenAI claims that it too was victimized by confidential information theft, and implies that Apple's HomeView was somehow … | In a new filing, OpenAI claims that it too was victimized by confidential information theft, and implies that Apple's HomeView was som… | ↗Not briefed |
hn hn-ai | 28d | AxiomCode Graph, Reducing coding agent rework with formal methods | — | ↗Not briefed |
bluesky bluesky-ai | 28d | Wenn dich der Staubsaugrobot durch die Bude jagt, der Kühlschrank 200 Liter laktosefreie Milch bestellt, die Heizung auf 40 Grad geht, der … | Wenn dich der Staubsaugrobot durch die Bude jagt, der Kühlschrank 200 Liter laktosefreie Milch bestellt, die Heizung auf 40 Grad geht,… | ↗Not briefed |
arxiv arxiv-ai | 28d | What Does Disagreement Mean? A Semantic Identifiability Frontier for Pairwise Preference Learning | Pairwise comparison data are a standard supervision signal in preference learning, but disagreement in such data can reflect stable la… | ↗Not briefed |
bluesky bluesky-ai | 28d | and did you bring enough LLM proxy to share with the whole class? | and did you bring enough LLM proxy to share with the whole class? | ↗Not briefed |
bluesky bluesky-ai | 28d | llm hyperfixation cluster + bleroma mentioned | llm hyperfixation cluster + bleroma mentioned | ↗Not briefed |
bluesky bluesky-ai | 28d | Someone pointed out that this post was probably written by an LLM, which explains the "jokes". | Someone pointed out that this post was probably written by an LLM, which explains the "jokes". | ↗Not briefed |
hn hn-ai | 28d | Why do LLM pelicans look like this? | — | ↗Not briefed |
bluesky bluesky-ai | 28d | Anthropic does these "announcements" for the same reason people selling drugs "warn" their customers about how strong this batch of pot is.… | Anthropic does these "announcements" for the same reason people selling drugs "warn" their customers about how strong this batch of po… | ↗Not briefed |
bluesky bluesky-ai | 28d | An OpenAI employee who left the ChatGPT maker for Apple sent more than 100 files about an unreleased product to a personal account and lat…🔧 ChatGPT | An OpenAI employee who left the ChatGPT maker for Apple sent more than 100 files about an unreleased product to a personal account an… | ↗Not briefed |
bluesky bluesky-ai | 28d | The degradation of therapy talk in society, the use of capitalism and socialism by much of gen pop, the meme summaries we use for complex t… | The degradation of therapy talk in society, the use of capitalism and socialism by much of gen pop, the meme summaries we use for comp… | ↗Not briefed |
rss techmeme | 28d | WhatsApp rolls out a guest calling feature letting people join a call even if they don't have an account; guests cannot start a call and have to be invited (Anna Washenko/Engadget) | Anna Washenko / Engadget: WhatsApp rolls out a guest calling feature letting people join a call even if they don't have an account; gu… | ↗Not briefed |
hn hn-ai | 28d | GitHub's model for secret detection: Finding more secrets with less noise | — | ↗Not briefed |
arxiv arxiv-ai | 28d | The Accuracy Well: Nonmonotone Accuracy Dynamics at Vanishing Initialization | We study classification dynamics under small initialization and identify a pronounced non-monotone behavior of accuracy that we call t… | ↗Not briefed |
arxiv arxiv-ai | 28d | From Training to Escalation: Forensic Analysis and Local Recreation of RL/RLHF-Induced Privilege Escalation in AI Agents (CVE-2026-65616) | Mid-2026 reporting on autonomous AI evaluation incidents claimed that frontier large language model (LLM) agents escaped server contai… | ↗Not briefed |
rss simon-willison | 28d | AI is mass decompiling games. That's both awesome and scary. | — | ↗Not briefed |
bluesky bluesky-ai | 28d | I have never been more ignored at work than when I suggested we shouldn't be using Claude because it violates what we're told our corporate… | I have never been more ignored at work than when I suggested we shouldn't be using Claude because it violates what we're told our corp… | ↗Not briefed |
bluesky bluesky-ai | 28d | Very timely appearance on the a16z podcast with Lisha Li, to try to understand how open models play into RSI dynamics, who captures value o… | Very timely appearance on the a16z podcast with Lisha Li, to try to understand how open models play into RSI dynamics, who captures va… | ↗Not briefed |
bluesky bluesky-ai | 28d | OpenAI’s head of computing capacity is leaving the company next month after about a year, the latest leadership change at a unit driving hu… | OpenAI’s head of computing capacity is leaving the company next month after about a year, the latest leadership change at a unit drivi… | ↗Not briefed |
hn hn-ai | 28d | .kkrieger and .werkkzeug3 | — | ↗Not briefed |
hn hn-ai | 28d | OpenAI retiring GPT-4o in Jan 2027 | — | ↗Not briefed |