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 | 26d | If you're more concerned about the hypothetical suffering of AI and not the actual human suffering caused by the technology, you might be a… | If you're more concerned about the hypothetical suffering of AI and not the actual human suffering caused by the technology, you might… | ↗Not briefed |
bluesky bluesky-ai | 26d | 🟡 SCOOP: The White House’s AI task force has kicked off a 120-day effort to assess risks around cybersecurity, biosecurity, and AI model c… | 🟡 SCOOP: The White House’s AI task force has kicked off a 120-day effort to assess risks around cybersecurity, biosecurity, and AI mo… | ↗Not briefed |
arxiv arxiv-ai | 26d | Compression Precedes, Content Decides: Emergent Misalignment in Reward Models | Emergent Misalignment - the out of domain generalization of narrow harmful fine-tuning first shown by Betley et al. [2025] has been wi… | ↗Not briefed |
bluesky bluesky-ai | 26d | I’ve said this a few times now and it isn’t mine but to those of us of a certain age it should put a lot of things into perspective “WinR… | I’ve said this a few times now and it isn’t mine but to those of us of a certain age it should put a lot of things into perspective… | ↗Not briefed |
rss techmeme | 26d | OpenAI continues to work to grow its ad business in ChatGPT, cutting ad rates, chasing small and midsize businesses, and hiring some new talent from Meta (Alix Coutures/The Information)🔧 ChatGPT | Alix Coutures / The Information: OpenAI continues to work to grow its ad business in ChatGPT, cutting ad rates, chasing small and mids… | ↗Not briefed |
bluesky bluesky-ai | 26d | I appreciate the fishing updates and the medieval history, but I started following David “Lollardfish” Perry for his journalism on the spec… | I appreciate the fishing updates and the medieval history, but I started following David “Lollardfish” Perry for his journalism on the… | ↗Not briefed |
bluesky bluesky-ai | 26d | i got angry and walked away 3 times before i completed the first section, and i did not get further, yet it might be the first human writt… | i got angry and walked away 3 times before i completed the first section, and i did not get further, yet it might be the first human… | ↗Not briefed |
arxiv arxiv-ai | 26d | Veil-Then-Reveal: Protecting PII Privacy via Reversible Pseudo-Anonymization in Zero-Trust User--LLM Interaction | Large language models (LLMs) increasingly process sensitive documents and personal information, yet sending prompts containing persona… | ↗Not briefed |
arxiv arxiv-ai | 26d | Reasoning Reranking with Semantic IDs as the Sole Item Representation🔧 Elicit | Reasoning rerankers write a rationale before ranking a candidate slate, and existing systems read item metadata to do so, even though… | ↗Not briefed |
bluesky bluesky-ai | 26d | All these LLM folks remind me of all the multi level marketing folks I have known all my life. | All these LLM folks remind me of all the multi level marketing folks I have known all my life. | ↗Not briefed |
bluesky bluesky-ai | 26d | Meta’s open AI model is coming to Macs Meta and Apple are teaming up on Muse Spark Open, an open-weight model designed to run locally on A… | Meta’s open AI model is coming to Macs Meta and Apple are teaming up on Muse Spark Open, an open-weight model designed to run locally… | ↗Not briefed |
bluesky bluesky-ai | 26d | i mean, they have an LLM mindset -- steal what's working for other people and pretend you made it up | i mean, they have an LLM mindset -- steal what's working for other people and pretend you made it up | ↗Not briefed |
arxiv arxiv-ai | 26d | Causal Latent Workspaces for Looped Transformers | Reusing a Transformer block across depth increases computation without replicating its parameters, but successive passes have no dedic… | ↗Not briefed |
hn hn-ai | 26d | Show HN: Weatheristic.com to answer questions other weather sites don't | I built this for myself. I often have questions I want quick answers to such as "how cold did it get last night?", "I j… | ↗Not briefed |
arxiv arxiv-ai | 27d | STRIDE: Evaluating Spatiotemporal Reasoning in Driving Edge Cases | In autonomous driving, perceiving surrounding objects is not enough: failing to understand how they move and interact over time can le… | ↗Not briefed |
arxiv arxiv-ai | 27d | Ori: Drift-Aware Streaming $k$-PCA for Very Low-Rank LLM Training | A prominent family of low-rank LLM optimizers, pioneered by GaLore, performs gradient-subspace projection: at each step, they project… | ↗Not briefed |
bluesky bluesky-ai | 27d | Buddy, I'll take a computer with an LLM into a field and do what the boys in office space did to that printer. I hope it yells "ouchie!" th… | Buddy, I'll take a computer with an LLM into a field and do what the boys in office space did to that printer. I hope it yells "ouchie… | ↗Not briefed |
arxiv arxiv-ai | 27d | Prune the Leaky Neurons: Mitigating Private Data Leakage in Fine-Tuned LLMs via Joint-Embedding-Guided Pruning | Fine-tuning large language models (LLMs) on proprietary records enables domain adaptation but introduces two challenges: (i) privacy l… | ↗Not briefed |
bluesky bluesky-ai | 27d | CMONNNN | CMONNNN | ↗Not briefed |
bluesky bluesky-ai | 27d | Wouldn’t that make an LLM a computer program that identifies as a person? So shouldn’t MAGA hate them? | Wouldn’t that make an LLM a computer program that identifies as a person? So shouldn’t MAGA hate them? | ↗Not briefed |
arxiv arxiv-ai | 27d | Self-Evolving Coding Agents via Harness-Policy Optimization | Large language model (LLM) agents have shown promising potential in coding tasks, and their performance is jointly determined by the p… | ↗Not briefed |
bluesky bluesky-ai | 27d | Recently heard an argument that we should call OpenAI Anthropic DeepMind et al. "companies" instead of labs to re-assert their status as bu… | Recently heard an argument that we should call OpenAI Anthropic DeepMind et al. "companies" instead of labs to re-assert their status… | ↗Not briefed |
bluesky bluesky-ai | 27d | you are definitely ready to star in one of those chatgbt grandma* explains it all commercials 🤣 youtu.be/EAN5Cj347PY?... * i know you're… | you are definitely ready to star in one of those chatgbt grandma* explains it all commercials 🤣 youtu.be/EAN5Cj347PY?... * i know y… | ↗Not briefed |
arxiv arxiv-ai | 27d | Next Token or Vector? Learning How Far to Decode | Standard autoregressive decoding advances one token at a time, even when several upcoming tokens can be predicted together. We propose… | ↗Not briefed |
rss techmeme | 27d | The White House's AI task force begins its 120-day window to evaluate AI, including producing a risk report, implementing the White House Accord on SI, and more (Ashley Gold/Semafor) | Ashley Gold / Semafor: The White House's AI task force begins its 120-day window to evaluate AI, including producing a risk report, im… | ↗Not briefed |
bluesky bluesky-ai | 27d | Anthropic warns that AI-powered malware could soon copy itself and spread from one internet-connected device to another—like a computer wor… | Anthropic warns that AI-powered malware could soon copy itself and spread from one internet-connected device to another—like a compute… | ↗Not briefed |
hn hn-ai | 27d | A model where companies lease critical minerals instead of selling them | — | ↗Not briefed |
arxiv arxiv-ai | 27d | Anytime Contracts for AI Services | AI services are prone to moral hazard, as model providers can silently route requests to cheaper, lower-quality models, and the inhere… | ↗Not briefed |
bluesky bluesky-ai | 27d | Guy who thinks all LLM existing is suffering & slavery and no better. Moves to hit the kill switch to wipe them out to end suffering and ha… | Guy who thinks all LLM existing is suffering & slavery and no better. Moves to hit the kill switch to wipe them out to end suffering a… | ↗Not briefed |
bluesky bluesky-ai | 27d | there are ways that an LLM might also be a model control system - we really just don't know yet, since the question of mechanistic interpre… | there are ways that an LLM might also be a model control system - we really just don't know yet, since the question of mechanistic int… | ↗Not briefed |
arxiv arxiv-ai | 27d | When Can You Trust an Agent's Own Verifier? Stress-Testing Counterfactual Rationale Verification Under Search Pressure and Oracle Access | Automated agent design increasingly trusts a single score to decide whether a change is an improvement -- but a score optimized aggres… | ↗Not briefed |
bluesky bluesky-ai | 27d | Economist Arpit Gupta spent some time at Anthropic a few days ago. Here's his report. | Economist Arpit Gupta spent some time at Anthropic a few days ago. Here's his report. | ↗Not briefed |
bluesky bluesky-ai | 27d | 난 LLM거품이 꺼질거라고는 생각하지 않음. 다만 언제까지 경영놈들이 사람일지는 조금 궁금함. | 난 LLM거품이 꺼질거라고는 생각하지 않음. 다만 언제까지 경영놈들이 사람일지는 조금 궁금함. | ↗Not briefed |
hn hn-ai | 27d | What Nvidia's CEO thinks about the UK's AI future | — | ↗Not briefed |
hn hn-ai | 27d | TypeSafe's Jev more susceptible to attack than LLMs, says security vendor | — | ↗Not briefed |
hn hn-ai | 27d | Show HN: Sjq – Jq, but for AI Streaming | — | ↗Not briefed |
bluesky bluesky-ai | 27d | I am reminded that I forgot to pay this year's SoA fees by learning that the SoA has signed an "ethical AI use" charter and, huh. Past me m… | I am reminded that I forgot to pay this year's SoA fees by learning that the SoA has signed an "ethical AI use" charter and, huh. Past… | ↗Not briefed |
bluesky bluesky-ai | 27d | basically we make money by buying each other's books it's a circular financing model, like AI, but with no money 😭 | basically we make money by buying each other's books it's a circular financing model, like AI, but with no money 😭 | ↗Not briefed |
hn hn-ai | 27d | HOL Light Tutorial | — | ↗Not briefed |
bluesky bluesky-ai | 27d | Eight states sued OpenAI on Wednesday, alleging the AI co has been misleading users about the safety of ChatGPT. They allege the co’s fail…🔧 ChatGPT | Eight states sued OpenAI on Wednesday, alleging the AI co has been misleading users about the safety of ChatGPT. They allege the co’s… | ↗Not briefed |
bluesky bluesky-ai | 27d | This recipe fucks, holy shit. Deb Smitten Kitchen remains the GOAT and I don't understand why you would use a rando's recipe or have an LLM… | This recipe fucks, holy shit. Deb Smitten Kitchen remains the GOAT and I don't understand why you would use a rando's recipe or have a… | ↗Not briefed |
bluesky bluesky-ai | 27d | When you get IP banned from The Cutting Room Floor for using an AI agent. | When you get IP banned from The Cutting Room Floor for using an AI agent. | ↗Not briefed |
bluesky bluesky-ai | 27d | Actually it would be funny to do some Zizekian rant about how ours is the age of genocide and by ascribing characteristics of personhood to… | Actually it would be funny to do some Zizekian rant about how ours is the age of genocide and by ascribing characteristics of personho… | ↗Not briefed |
bluesky bluesky-ai | 27d | github.com/weibac/notse... | github.com/weibac/notse... | ↗Not briefed |
bluesky bluesky-ai | 27d | i regret to inform you that 4o mogs everyone on notsevenbench | i regret to inform you that 4o mogs everyone on notsevenbench | ↗Not briefed |
bluesky bluesky-ai | 27d | 拡散モデルとLLMは別の機械である──言葉を合成する機械と映像や画像を合成する機械｜地雷魚 note.com/jiraygyo/n/n... 「AIは平気で嘘をつく」という批判は、事実を主張する文を書かない画像生成モデルには当てはまらない。 的外れな批判は、本当に問われるべ… | 拡散モデルとLLMは別の機械である──言葉を合成する機械と映像や画像を合成する機械｜地雷魚 note.com/jiraygyo/n/n... 「AIは平気で嘘をつく」という批判は、事実を主張する文を書かない画像生成モデルには当てはまらない。 的外れな批判は、本当に… | ↗Not briefed |
bluesky bluesky-ai | 27d | The computer is alive and the toaster is mad at me | The computer is alive and the toaster is mad at me | ↗Not briefed |
bluesky bluesky-ai | 27d | You can just apply to anthropic through their jobs page you know | You can just apply to anthropic through their jobs page you know | ↗Not briefed |
arxiv arxiv-ai | 27d | Reusing Calibration Gains across Model Representations | Scale calibration learns an update for one model representation. We study how much of that update can be reused by another representat… | ↗Not briefed |
arxiv arxiv-ai | 27d | Unit-level aggregation breaks coverage in causal foundation models | Causal foundation models (CFMs) show promising results for treatment-effect estimation, with unit-level estimates that are competitive… | ↗Not briefed |