AINews Portal

News

50 items (filtered)

The raw AI news stream — everything we ingest across RSS, Hacker News, arXiv, GitHub, Bluesky, YouTube and more, newest first.

SourceAgeTitleSummaryStatus
bluesky
bluesky-ai
2moSo the fun thing about that Anthropic "Pacing the Frontier" post, is that the proposed regulatory body (METR) is both financially and liter…So the fun thing about that Anthropic "Pacing the Frontier" post, is that the proposed regulatory body (METR) is both financially and…
↗Not briefed
hn
hn-ai
2moMore Tree-sitter, more neocaml, more elisp—
↗Not briefed
bluesky
bluesky-ai
2moNo, we cannot trust METR to meaningfully check the progress of Anthropic’s AI technology in an unbiased way, because METR is often quite li…No, we cannot trust METR to meaningfully check the progress of Anthropic’s AI technology in an unbiased way, because METR is often qui…
↗Not briefed
bluesky
bluesky-ai
2morob miles here has a great point about ai anthropomporphism drama which is that it's approximately as correct to say an ai model can "think…rob miles here has a great point about ai anthropomporphism drama which is that it's approximately as correct to say an ai model can "…
↗Not briefed
bluesky
bluesky-ai
2moAnd this www.verysane.ai/p/is-metr-a-...And this www.verysane.ai/p/is-metr-a-...
↗Not briefed
arxiv
arxiv-ai
2moTACTICS: Taxonomy-Aware Intelligent Corpus Sampling for Machine TranslationLarge-scale machine-translation (MT) systems are typically evaluated on random samples from a corpus whose distributional composition…
↗Not briefed
rss
techmeme
2moZuckerberg says "Meta delayed shipping Muse for several months to focus on safety and security" and didn't call on other AI labs to do the same before acting (Tom Giles/Bloomberg)Tom Giles / Bloomberg: Zuckerberg says “Meta delayed shipping Muse for several months to focus on safety and security” and didn't call…
↗Not briefed
hn
hn-ai
2moThe proof passed. Can you see why it works?—
↗Not briefed
bluesky
bluesky-ai
2moThe argument I heard is *not* the Terminator/War Games/machines turn on us model. Rather I heard it as more “insane people (eg mass shooter…The argument I heard is *not* the Terminator/War Games/machines turn on us model. Rather I heard it as more “insane people (eg mass sh…
↗Not briefed
hn
hn-ai
2moShow HN: A chess analyzer that runs in the browserI play chess casually and usually struggle to figure out why Lichess analysis say my move is a blunder. Almost all the analysis tools…
↗Not briefed
bluesky
bluesky-ai
2moA real idea that someone should fund, or just volunteer and do: Governmental pdf freeing via LLM tools. From personal experience, there is…A real idea that someone should fund, or just volunteer and do: Governmental pdf freeing via LLM tools. From personal experience, the…
↗Not briefed
hn
hn-ai
2moA better way of blocking macOS updates—
↗Not briefed
arxiv
arxiv-ai
2moEDCT-Bench: Uncovering Faithfulness Gaps in VLMs via Explanation-Driven Counterfactual TestingVision-Language Models (VLMs) can produce Natural Language Explanations (NLEs) that sound plausible yet remain inconsistent with the v…
↗Not briefed
rss
bing-news-ai
2moChinese hacking firm’s AI ambitions reveal new model for private intelligence: ReportA cache of internal records offers an unusual look at a Chinese cybersecurity company seeking to combine hacking, data brokerage and a…
↗Not briefed
arxiv
arxiv-ai
2moMaximum Strong Independent Sets in Hypergraphs: Reductions, Bounds, and Greedy CertificatesWe study the maximum strong independent set problem in a finite hypergraph: find the largest vertex set that intersects every hyperedg…
↗Not briefed
hn
hn-ai
2moAsk HN: eBPF verifier packet tracking boundaries under volumetric socket drops?How do you handle eBPF verifier pkt_ptr tracking limits under volumetric socket drops? Bypassing userspace L7 bottlenecks via L4 trans…
↗Not briefed
rss
techcrunch-ai
2moWe don’t need AI regulation — leave safety to us, Nvidia’s Jensen Huang saysAI isn't some kind of new form of "alien mind," according to Jensen Huang. It's just hardware and software, so safety can be engineere…
↗Not briefed
bluesky
bluesky-ai
2mogonna download something off github and ask an LLM to do a review in the voice of pynchon or burroughsgonna download something off github and ask an LLM to do a review in the voice of pynchon or burroughs
↗Not briefed
bluesky
bluesky-ai
2mo[JP] LLM強化学習の罠「マシュー効果」とは?難問を攻略するNever Give Up(NGU)アプローチ [EN] The Trap of LLM Reinforcement Learning: What is the "Matthew Effect"? Introduc…[JP] LLM強化学習の罠「マシュー効果」とは?難問を攻略するNever Give Up(NGU)アプローチ [EN] The Trap of LLM Reinforcement Learning: What is the "Matthew Effect"? Int…
↗Not briefed
bluesky
bluesky-ai
2moyou ever be playing music, and someone asks you what you just did, and now not only can you not explain it, while trying to think about how…you ever be playing music, and someone asks you what you just did, and now not only can you not explain it, while trying to think abou…
↗Not briefed
bluesky
bluesky-ai
2moOpenAI & Anthropic both promised to create the tools that destroy humanity unless someone stops them. Corporations don't have an obligation…OpenAI & Anthropic both promised to create the tools that destroy humanity unless someone stops them. Corporations don't have an oblig…
↗Not briefed
bluesky
bluesky-ai
2mo"Whatever our goals are, we’ve operationalized them primarily through proving theorems. Almost all papers or PhD theses have a main theorem…"Whatever our goals are, we’ve operationalized them primarily through proving theorems. Almost all papers or PhD theses have a main th…
↗Not briefed
bluesky
bluesky-ai
2moAt least 2% of OpenAI and 5% of Anthropic Source: I made it up but it’s probably trueAt least 2% of OpenAI and 5% of Anthropic Source: I made it up but it’s probably true
↗Not briefed
hn
hn-ai
2moDivide, Consult, Conquer: Capability Laundering Through Aligned LLMs—
↗Not briefed
bluesky
bluesky-ai
2moRepubs are doing better polling on AI???!!!!!! You have got to be fucking kidding me! Oh, wait! Schumer's daughter works for OpenAI. A…Repubs are doing better polling on AI???!!!!!! You have got to be fucking kidding me! Oh, wait! Schumer's daughter works for OpenA…
↗Not briefed
arxiv
arxiv-ai
2moASPIRE: Asynchronous Batched Self-Speculative Decoding for Long-Context LLM InferenceLong-context LLM inference is bottlenecked by attention, whose repeated KV-cache reads make decoding memory-bound. Self-speculative de…
↗Not briefed
bluesky
bluesky-ai
2moAI doesn't need nukes or killer robots to destroy humanity. All it has to do is replace every aspect that makes us human with its own codin…🔧 ChatGPTAI doesn't need nukes or killer robots to destroy humanity. All it has to do is replace every aspect that makes us human with its own…
↗Not briefed
arxiv
arxiv-ai
2moOn the Identifiability of Mixed Ordinal and Exponential Family Causal DAGs under Linear Parametric ModelsThe problem of identifiability in linear parametric models (LPMs) whose nodes follow either an ordered logit model or a regular one-pa…
↗Not briefed
bluesky
bluesky-ai
2moif this is just an excuse to curtail their runaway cash-burning, this is just straight-up collusion techcrunch.com/2026/09/15/o...if this is just an excuse to curtail their runaway cash-burning, this is just straight-up collusion techcrunch.com/2026/09/15/o...
↗Not briefed
bluesky
bluesky-ai
2mo‘Tech oligarchs agree calling AI “too dangerous” way cooler than just admitting it doesn’t work’ www.thebeaverton.com/2026/09/tech...‘Tech oligarchs agree calling AI “too dangerous” way cooler than just admitting it doesn’t work’ www.thebeaverton.com/2026/09/tech...
↗Not briefed
arxiv
arxiv-ai
2moBeyond the Previous Layer: Residual Predictive Structure in Sparse MoE RoutingSparse mixture-of-experts models route each token through a sequence of expert selections. We ask whether the immediately preceding se…
↗Not briefed
hn
hn-ai
2moShow HN: Plurnk (Yet *Another* AI Harness)I am desperate for feedback on my homebrewed harness, &quot;Plurnk.&quot;<p>I&#x27;ve been dogfooding it on my RTX5070Ti with only 16G…
↗Not briefed
bluesky
bluesky-ai
2moa JEV is an llm that has taken a vow of silencea JEV is an llm that has taken a vow of silence
↗Not briefed
rss
techmeme
2moDemocrats and Republicans proposed many bills to address major AI threats over the past two years, but they remain stymied by partisan disputes and lobbying (New York Times)New York Times: Democrats and Republicans proposed many bills to address major AI threats over the past two years, but they remain sty…
↗Not briefed
bluesky
bluesky-ai
2moEvery GPT model is "the one" up until it comes out and Altman eventually admits that it is actually shit and the NEXT one will be the one …🔧 GPT-5Every GPT model is "the one" up until it comes out and Altman eventually admits that it is actually shit and the NEXT one will be the…
↗Not briefed
bluesky
bluesky-ai
2moas far as i can tell, it’s purely an RL feat seems like they 1. take a fully pretrained LLM (decoder-only??) 2. slap a new 255-element c…as far as i can tell, it’s purely an RL feat seems like they 1. take a fully pretrained LLM (decoder-only??) 2. slap a new 255-elem…
↗Not briefed
bluesky
bluesky-ai
2mowww.itmedia.co.jp/news/article... 「クローズなgAI使うのやめて」 (アンソロピックがユーザーの情報全公開)www.itmedia.co.jp/news/article... 「クローズなgAI使うのやめて」 (アンソロピックがユーザーの情報全公開)
↗Not briefed
arxiv
arxiv-ai
2moLocating Hidden Failures Makes Long-Horizon Agents More ReliableAs AI agents take on long, autonomous tasks, we increasingly oversee rather than perform the work, yet we still judge them almost enti…
↗Not briefed
hn
hn-ai
2moProbably the best laptop with Linux support and Local LLM AI [video]—
↗Not briefed
rss
theverge-ai
2moVoters mostly don’t like AI and data centers, but neither party seems to have an edgePoll data released Tuesday by the New York Times and Siena University confirms what we've already been seeing, and what politicians ar…
↗Not briefed
arxiv
arxiv-ai
2moSymmetry without a manifold: intrinsic dimension on orbitsThe standard geometric derivation of neural scaling exponents takes the intrinsic dimension of a data manifold as its input. On modula…
↗Not briefed
bluesky
bluesky-ai
2moPrompt "Make an image of how I interact with you" When OpenAI is a little too honest with a prompt.Prompt "Make an image of how I interact with you" When OpenAI is a little too honest with a prompt.
↗Not briefed
bluesky
bluesky-ai
2mothe timing on this is something. Anthropic prepares Claude Money for personal financethe timing on this is something. Anthropic prepares Claude Money for personal finance
↗Not briefed
bluesky
bluesky-ai
2mo"If you were some sort of cynic, you might think the models were running out of steam and the vendors were running out of money.""If you were some sort of cynic, you might think the models were running out of steam and the vendors were running out of money."
↗Not briefed
arxiv
arxiv-ai
2moCollaborative Memory for Multi-Agent VLM SystemsVision-language model (VLM) agents combine specialized perception, tools, and reasoning to address complex visual tasks. In multi-agen…
↗Not briefed
rss
techmeme
2moQ&A with Matt Sheehan, a senior fellow at the Carnegie Endowment for International Peace, on US-China AI race, distillation, US chip export controls, and more (New York Times)New York Times: Q&A with Matt Sheehan, a senior fellow at the Carnegie Endowment for International Peace, on US-China AI race, distill…
↗Not briefed
hn
hn-ai
2moNobody Needs a Human Router—
↗Not briefed
arxiv
arxiv-ai
2moEdgeReMIND: A Scalable, Top-Ranked Memorization Baseline for Temporal Multi-Relational Link PredictionTemporal link prediction on the Temporal Graph Benchmark 2.0 (TGB 2.0) faces a scalability ceiling: on the benchmark's three largest d…
↗Not briefed
bluesky
bluesky-ai
2mo“Do you publish a paper that lists “GPT-6 Astra” or “Claude Fable” as the author—but then let the AI profusely thank you in the acknowledgm…“Do you publish a paper that lists “GPT-6 Astra” or “Claude Fable” as the author—but then let the AI profusely thank you in the acknow…
↗Not briefed
hn
hn-ai
2moShow HN: Agenttik – work on multiple projects in parallel with AI agentsHi! I&#x27;ve been reading HN daily for almost 16 years. Really glad this community exist. This is my first Show HN, so I hope you fin…
↗Not briefed