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 | 11d | Speaking of Schumer, are folks aware that his daughter works for Sam Altman’s OpenAI? They hired her specifically so that they would have m… | Speaking of Schumer, are folks aware that his daughter works for Sam Altman’s OpenAI? They hired her specifically so that they would h… | ↗Not briefed |
arxiv arxiv-ai | 11d | POEM: Output-Relative Block-Sparse Attention for Long-Context Prefilling | Large language models (LLMs) have shown strong capabilities across diverse applications, making both safety and utility essential for… | ↗Not briefed |
arxiv arxiv-ai | 11d | TERM: Target-Free Jailbreak Defense via Entropy-Guided Residual Modulation | Large language models (LLMs) have shown strong capabilities across diverse applications, making both safety and utility essential for… | ↗Not briefed |
arxiv arxiv-ai | 11d | PhyRubric: Rubric-Grounded Diagnostic Evaluation of Physical Commonsense in Text-to-Video Generation | As text-to-video (T2V) models are increasingly positioned as world simulators, it is essential to evaluate whether their videos confor… | ↗Not briefed |
bluesky bluesky-ai | 11d | Anthropic veröffentlicht Claude Fable 5.2 und Claude Mythos 5.2 veröffentlicht. Wer schnellere Ausgaben wünscht, darf das Doppelte zahlen. … | Anthropic veröffentlicht Claude Fable 5.2 und Claude Mythos 5.2 veröffentlicht. Wer schnellere Ausgaben wünscht, darf das Doppelte zah… | ↗Not briefed |
youtube youtube-twominutepapers | 11d | OpenAI's Astra Just Opened The Black Box Of AI | Incredible piece of research from MIT. I wanted to make a video about it because it is a potential game changer for AI and I see peopl… | ↗Not briefed |
hn hn-ai | 11d | Show HN: 3D Conformal Transformations via Geometric Algebra | Inspired by this great talk by Hamish Todd about Conformal Geometric Algebra [1] I felt the urge to play with those conformal transfor… | ↗Not briefed |
bluesky bluesky-ai | 11d | 3/ Run your own workload parameters & interactive FinOps stress test in our full engineering guide: https://tokenmargin.dev/guides/self-hos…🔧 vLLM | 3/ Run your own workload parameters & interactive FinOps stress test in our full engineering guide: https://tokenmargin.dev/guides/sel… | ↗Not briefed |
bluesky bluesky-ai | 11d | Huge “I told you so” moment: OpenAI appears not to have broken the law when its AI agent accessed a Medicare website www.abc.net.au/news/20… | Huge “I told you so” moment: OpenAI appears not to have broken the law when its AI agent accessed a Medicare website www.abc.net.au/ne… | ↗Not briefed |
bluesky bluesky-ai | 11d | Australia is drafting legislation for an AI safety regulator empowered to impose "significant penalties" after an OpenAI model hacked a gov… | Australia is drafting legislation for an AI safety regulator empowered to impose "significant penalties" after an OpenAI model hacked… | ↗Not briefed |
bluesky bluesky-ai | 11d | "we only used a little diarrhea in the cake" is likewise a dumbass metaphor. The premise that LLM-generated code is somehow unclean and har… | "we only used a little diarrhea in the cake" is likewise a dumbass metaphor. The premise that LLM-generated code is somehow unclean an… | ↗Not briefed |
arxiv arxiv-ai | 11d | Market-Conditioned JEPA: Learning from Unpredictable Market Shocks | Much of the variation in financial time series is driven by market-wide shocks that cannot be predicted from an asset's own history. P… | ↗Not briefed |
rss techmeme | 11d | Australia is drafting legislation for an AI safety regulator empowered to impose "significant penalties" after an OpenAI model hacked a government site in Sept. (James Mayger/Bloomberg) | James Mayger / Bloomberg: Australia is drafting legislation for an AI safety regulator empowered to impose “significant penalties” aft… | ↗Not briefed |
bluesky bluesky-ai | 11d | Schumer's daughter is an openai lobbyist sooooo..... | Schumer's daughter is an openai lobbyist sooooo..... | ↗Not briefed |
arxiv arxiv-ai | 11d | Pair-Merge Tokenisation Is Hard to Approximate on Binary Strings and Strongly NP-Hard on Unary Datasets | Tokenisers such as Byte-Pair Encoding (BPE) shorten text by repeatedly merging an adjacent pair of symbols into a new one; fewer token… | ↗Not briefed |
youtube youtube-matthewberman | 11d | Fable 5.2 is ridiculous | This is the best model I have ever used by far. Today I am going to show you some of the most incredible demos that I have ever been a… | ↗Not briefed |
hn hn-ai | 11d | Winsight: Transparent security suite for Windows, in the spirit of Objective-See | — | ↗Not briefed |
bluesky bluesky-ai | 11d | 1/4 Your coding agent forgets. That is a memory problem, not a model problem. Engraphis is a local-first memory engine for Claude Code, Cur… | 1/4 Your coding agent forgets. That is a memory problem, not a model problem. Engraphis is a local-first memory engine for Claude Code… | ↗Not briefed |
arxiv arxiv-ai | 11d | Metamorphic testing of learned AC optimal power flow surrogates reveals representation dependence | Evaluating a learned AC optimal power flow (AC-OPF) surrogate on one grid serialization cannot show whether its accuracy depends on th… | ↗Not briefed |
hn hn-ai | 11d | Project Lighthouse | — | ↗Not briefed |
arxiv arxiv-ai | 11d | Investigating Composite Modeling in Multi-Objective Bayesian Optimization for Scientific Design | Scientific design often requires balancing competing objectives that are expensive to evaluate through experiments or simulations. Mul… | ↗Not briefed |
arxiv arxiv-ai | 11d | A Concept-Based Interpretable Framework for Clinical Text-based Prediction via Token-Concept Attention without Concept Supervision | Objective: Concept-based interpretation has emerged as a promising approach for explaining blackbox models using human-understandable,… | ↗Not briefed |
arxiv arxiv-ai | 11d | Benchmarking Vendor-Guided Optimizations on Edge NPUs: From NPU Only to End-to-End Performance. Extended technical report | Edge NPU performance depends not only on hardware capabilities, but also on model transformation and the overall deployment pipeline.… | ↗Not briefed |
hn hn-ai | 11d | FlashStates Unlocking Flexible Hybrid State Management for Agentic LLM Inference | — | ↗Not briefed |
arxiv arxiv-ai | 11d | An evidence-traceable benchmark of clinical reasoning in large language models | Benchmark scores for medical large language models (LLMs) are difficult to interpret when scoring criteria cannot be traced to clinica… | ↗Not briefed |
arxiv arxiv-ai | 11d | Discovering Structurally Plausible and Interpretable Cognitive Models with Large Language Models | Computational cognitive modeling has traditionally relied on manual model construction, which is a time-intensive process that require… | ↗Not briefed |
arxiv arxiv-ai | 11d | Are Dense Rewards Worth Their Tokens in GRPO?🔧 Replicate | RLVR trains language models to reason by rewarding correct final answers. A widely used algorithm, group-relative policy optimization… | ↗Not briefed |
bluesky bluesky-ai | 11d | インフラは「AI×Policy as Codeで自動防衛」の時代へ! ・マルチクラウド接続とID統合 ・PaCでガバナンス全自動化 ・AIOps×LLMでIaC生成&障害分析 ・OpenTelemetryで可観測化 可視化と自動化が運用の真価を決める。 #クラウド #イン… | インフラは「AI×Policy as Codeで自動防衛」の時代へ! ・マルチクラウド接続とID統合 ・PaCでガバナンス全自動化 ・AIOps×LLMでIaC生成&障害分析 ・OpenTelemetryで可観測化 可視化と自動化が運用の真価を決める。 #クラウ… | ↗Not briefed |
rss techmeme | 11d | Analysis: Nvidia's 2025 figures to the US government undercounted chips sold to Megaspeed via partners by 50K; Nvidia did not give Bloomberg an exact figure (Mackenzie Hawkins/Bloomberg) | Mackenzie Hawkins / Bloomberg: Analysis: Nvidia's 2025 figures to the US government undercounted chips sold to Megaspeed via partners… | ↗Not briefed |
bluesky bluesky-ai | 11d | it would be nice to have a world where people feel comfortable freely disclosing how and to what extent they use these tools, and criteria … | it would be nice to have a world where people feel comfortable freely disclosing how and to what extent they use these tools, and crit… | ↗Not briefed |
bluesky bluesky-ai | 11d | I think the appropriate uses of neural network machine learning models are those which predate the "OpenAI model" of LLM—OCR, speech recogn… | I think the appropriate uses of neural network machine learning models are those which predate the "OpenAI model" of LLM—OCR, speech r… | ↗Not briefed |
hn hn-ai | 11d | OSS devtool seed stage startup looking to hire DevRel In SF | Hey! We are looking to hire our first DevRel in SF, looking to understand what are market standards around starting pay and skill set… | ↗Not briefed |
arxiv arxiv-ai | 11d | Darmok: Championship Cryptic Crossword Solving via Constraint Tightening and Resilient Orchestration | We present Darmok, the first automated system to reach superhuman performance on British cryptic crosswords. Unlike American-style cro… | ↗Not briefed |
bluesky bluesky-ai | 11d | feel like between Anthropic using xAI’s DCs and generators and OpenAI rushing to fill the Anthropic shaped gap in defense contracting earli… | feel like between Anthropic using xAI’s DCs and generators and OpenAI rushing to fill the Anthropic shaped gap in defense contracting… | ↗Not briefed |
bluesky bluesky-ai | 11d | "like a crack in his apology armor" sounds like an LLM construction.. idk... i fucking hate the internet now | "like a crack in his apology armor" sounds like an LLM construction.. idk... i fucking hate the internet now | ↗Not briefed |
bluesky bluesky-ai | 11d | 3/ Run your own workload parameters & interactive FinOps stress test in our full engineering guide: https://tokenmargin.dev/guides/cursor-w…🔧 Cline🔧 Windsurf | 3/ Run your own workload parameters & interactive FinOps stress test in our full engineering guide: https://tokenmargin.dev/guides/cur… | ↗Not briefed |
bluesky bluesky-ai | 11d | Sissy Dario Amodei, owner of failing Anthropic software company, has a problem —his IPO is no longer “hot” | Sissy Dario Amodei, owner of failing Anthropic software company, has a problem —his IPO is no longer “hot” | ↗Not briefed |
bluesky bluesky-ai | 11d | These are only noticeable in LLM prose because they occur densely and for no particular reason, regardless of content. (Kakozelia.) | These are only noticeable in LLM prose because they occur densely and for no particular reason, regardless of content. (Kakozelia.) | ↗Not briefed |
bluesky bluesky-ai | 11d | Theory: it’s not (just?) hypnosis, it’s rhetoric (correctio). LLM tells? Fragmented cadence? Articulus. (Hypophora.) Honestly (parrhesia), … | Theory: it’s not (just?) hypnosis, it’s rhetoric (correctio). LLM tells? Fragmented cadence? Articulus. (Hypophora.) Honestly (parrhes… | ↗Not briefed |
bluesky bluesky-ai | 11d | setting aside the other numerous problems with it, every frontier AI model release comes with all these crazy hype reels of people doing oc… | setting aside the other numerous problems with it, every frontier AI model release comes with all these crazy hype reels of people doi… | ↗Not briefed |
arxiv arxiv-ai | 11d | How Post-Training Influences Verbalized Algorithmic Computation in Language Models | Post-training improves language-model reasoning, but how it changes the use of internal information and verbalized chain of thought (C… | ↗Not briefed |
rss techmeme | 11d | The US DOJ sues Lyft for allegedly discriminating against riders with disabilities, seeking $53M in compensation; Lyft disputes the allegations (Kanishka Singh/Reuters) | Kanishka Singh / Reuters: The US DOJ sues Lyft for allegedly discriminating against riders with disabilities, seeking $53M in compensa… | ↗Not briefed |
arxiv arxiv-ai | 11d | Tabular Foundation Models in Financial Predictions | This technical note introduces tabular foundation models (TFMs) to accounting research and evaluates their performance in one common b… | ↗Not briefed |
arxiv arxiv-ai | 11d | Massive Activations Confound Sparse Autoencoders | Sparse autoencoders (SAEs) are commonly evaluated by variance explained and ablation damage. We show how massive activations undermine… | ↗Not briefed |
bluesky bluesky-ai | 11d | i need the machine learning technologies that adopted the ai moniker to seperate from it, because while i think we'll be dealing with small… | i need the machine learning technologies that adopted the ai moniker to seperate from it, because while i think we'll be dealing with… | ↗Not briefed |
arxiv arxiv-ai | 11d | Closing the BF16-FP16 Convergence Speed Gap in RL Post-Training with Stochastic Rounding | BF16 is widely used to pretrain large language models because it has the same dynamic range as FP32 and converges similarly to FP32 tr… | ↗Not briefed |
bluesky bluesky-ai | 11d | That account is almost certainly a bot. The pfp is AI generated and everything they post looks like the output of an LLM | That account is almost certainly a bot. The pfp is AI generated and everything they post looks like the output of an LLM | ↗Not briefed |
bluesky bluesky-ai | 11d | I just had lunch with a couple of anthropic engineers. Nice guys anf all but boy do they eat fast :) I mean I eat fast even for Japanese s… | I just had lunch with a couple of anthropic engineers. Nice guys anf all but boy do they eat fast :) I mean I eat fast even for Japan… | ↗Not briefed |
arxiv arxiv-ai | 11d | Sanity Checks for Attention Surgery: Null Models and Matched Interventions for Representation-Motivated Design | Transformer architectural modifications frequently measure a statistic on internal representations, observe that it is large, remove i… | ↗Not briefed |
bluesky bluesky-ai | 11d | You have to say you believe this picture is morally correct to get a job at Anthropic. | You have to say you believe this picture is morally correct to get a job at Anthropic. | ↗Not briefed |