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
arxiv
arxiv-ai
27dRetrieved but Not Read: Understanding Why Search-Augmented LLMs Fail to Use Available EvidenceWeb search allows LLMs to access up-to-date information, but it remains unclear why they sometimes produce outdated answers even when…
↗Not briefed
hn
hn-ai
27dKarpathy on LLM Writing Style—
↗Not briefed
hn
hn-ai
27dShow HN: Codecity, a 3D map of any Git repository—
↗Not briefed
hn
hn-ai
27dShow HN: GPX Route Planner and GPX, Fit and TCX Editing ToolsInitially built Maps to GPX functionality as I could not find a decent option. Later implemented GPX Route Planner to create GPX files…
↗Not briefed
hn
hn-ai
27dBoxr: A rootless OCI container engine written in Rust—
↗Not briefed
hn
hn-ai
27dShow HN: server-noonien – conflict-free agent memory across machines, no server—
↗Not briefed
arxiv
arxiv-ai
27dFIRM: Fisher-Informed Retention Merging for Federated LLM UnlearningFederated LLM unlearning addresses targeted knowledge removal when forget and retain data remain distributed across private clients an…
↗Not briefed
hn
hn-ai
27dShow HN: Rails-preflight, a static check of what blocks your next Rails upgrade—
↗Not briefed
arxiv
arxiv-ai
27dInternational Multi-Centre Validation of Unsupervised Domain Adaptation for Precise Discrimination between Normal and Abnormal Retinal DevelopmentArrested retinal development can cause childhood visual impairment, but grading optical coherence tomography scans is difficult to sta…
↗Not briefed
hn
hn-ai
27dKavach: A booking engine built to survive a Tatkal-style stampede—
↗Not briefed
hn
hn-ai
27dShow HN: Custom Firmware for Insta360 Wave—
↗Not briefed
rss
techmeme
27dSources and docs: while Bo Hines helped craft the crypto-friendly Genius Act, he had a 33% stake in consulting firm Nxum, which received millions from Tether (Sam Kessler/Wall Street Journal)Sam Kessler / Wall Street Journal: Sources and docs: while Bo Hines helped craft the crypto-friendly Genius Act, he had a 33% stake in…
↗Not briefed
bluesky
bluesky-ai
27d"Fable-class model for cents per task" is pretty punchy"Fable-class model for cents per task" is pretty punchy
↗Not briefed
arxiv
arxiv-ai
27dEncoded but Exposed: Language Embeddings can Reveal More Than They Should in Recommender SystemsLanguage embeddings are increasingly used in recommenders to represent item content, but their privacy risks remain poorly understood.…
↗Not briefed
arxiv
arxiv-ai
27dRobust Covariance Estimation for Entry-wise Contaminated and Partially Observed Data via Expectation Maximization AlgorithmRobust covariance estimation under entry-wise contamination and missing data is challenging in various signal processing applications.…
↗Not briefed
hn
hn-ai
27dCPython 3.15 (JIT) Performance Showdown—
↗Not briefed
bluesky
bluesky-ai
27dLLM commoditization is an interesting thesis, I’ve been hearing it more and more and in the abstract I get it. It’s hard for me to fully wr…LLM commoditization is an interesting thesis, I’ve been hearing it more and more and in the abstract I get it. It’s hard for me to ful…
↗Not briefed
bluesky
bluesky-ai
27dNew form of AI psychosis droppedNew form of AI psychosis dropped
↗Not briefed
arxiv
arxiv-ai
27dHuman Interventions as Labels: Localizing, Steering, and (Sometimes) Editing the Computations Behind LLM Agent FailuresWhen an LLM agent misbehaves, a human steps in, and that intervention is usually discarded or replayed as training data. We treat it i…
↗Not briefed
rss
techmeme
27dSource: the US FTC issues ~50-page civil investigative demands to OpenAI and Anthropic over safety incidents, safeguarding practices, and consumer protection (Maria Curi/Axios)Maria Curi / Axios: Source: the US FTC issues ~50-page civil investigative demands to OpenAI and Anthropic over safety incidents, safe…
↗Not briefed
bluesky
bluesky-ai
27dIsn't this just a new way of surveilling us? "Oops, it self-replicated" Sure it fucking did.Isn't this just a new way of surveilling us? "Oops, it self-replicated" Sure it fucking did.
↗Not briefed
bluesky
bluesky-ai
27dIntroducing our new columnist, professor Hannah Fry. Read on for Hannah's take on the "existential crisis" engulfing maths, and come back e…Introducing our new columnist, professor Hannah Fry. Read on for Hannah's take on the "existential crisis" engulfing maths, and come b…
↗Not briefed
bluesky
bluesky-ai
27dYou get that the LLM doesn’t exist when it’s not loaded into memory and processing a query, right? It’s a fixed set of weights, it’s just a…You get that the LLM doesn’t exist when it’s not loaded into memory and processing a query, right? It’s a fixed set of weights, it’s j…
↗Not briefed
bluesky
bluesky-ai
27dUltramarine-1-flash — a GLM-5.3-Flash fine tune that outperforms Fable at 1/60th the cost open weights, open harness, open training data,…Ultramarine-1-flash — a GLM-5.3-Flash fine tune that outperforms Fable at 1/60th the cost open weights, open harness, open training…
↗Not briefed
hn
hn-ai
27dShow HN: Official MCP ServersWe maintain an updated list of official MCP servers repo: 390+ well-known and widely used projects, all from the companies behind the…
↗Not briefed
bluesky
bluesky-ai
27dNapster was poor people stealing. Anthropic is rich people stealing. Thank you for joining this Bluesky talk.Napster was poor people stealing. Anthropic is rich people stealing. Thank you for joining this Bluesky talk.
↗Not briefed
bluesky
bluesky-ai
27dOh I mean, individuals have and will get in trouble. Whether AI companies will, another matter. (To be clear: I’m very anti-AI in general.…Oh I mean, individuals have and will get in trouble. Whether AI companies will, another matter. (To be clear: I’m very anti-AI in gen…
↗Not briefed
arxiv
arxiv-ai
27dShiftQAT: Quantization-Aware Training for the Numerical Contract of a Fixed-Function Edge CNN AcceleratorFixed-function CNN accelerators make edge inference fast and energy-efficient, but they execute only networks that obey a precise nume…
↗Not briefed
bluesky
bluesky-ai
27dso google and meta are funding llm tokens for linux development so i ask what, exactly, is the mission of linux as an operating system?so google and meta are funding llm tokens for linux development so i ask what, exactly, is the mission of linux as an operating syst…
↗Not briefed
bluesky
bluesky-ai
27dThe ex-OpenAI employee allegedly wrapped a USB device in toilet paper and inserted it into a television located in an "executive conference…The ex-OpenAI employee allegedly wrapped a USB device in toilet paper and inserted it into a television located in an "executive confe…
↗Not briefed
rss
techmeme
27dAI industry extinction warnings are genuine rather than corporate plots for regulatory capture or hype, and they go as far back as the mid-2000s and the 2010s (Scott Alexander/Astral Codex Ten)Scott Alexander / Astral Codex Ten: AI industry extinction warnings are genuine rather than corporate plots for regulatory capture or…
↗Not briefed
bluesky
bluesky-ai
27dWarum schaut man nicht kurz in die Wikipedia, fragt eine Suchmaschine oder meinetwegen ein LLM, bevor man sich mit einem bildungsfernen Kom…Warum schaut man nicht kurz in die Wikipedia, fragt eine Suchmaschine oder meinetwegen ein LLM, bevor man sich mit einem bildungsferne…
↗Not briefed
bluesky
bluesky-ai
27dIt's difficult because a lot of health care systems are pushing health care providers to use LLM tools because they want to increase speed …It's difficult because a lot of health care systems are pushing health care providers to use LLM tools because they want to increase s…
↗Not briefed
arxiv
arxiv-ai
27dAbstracts or full text? Evaluating input scope and computational efficiency in biomedical lay summarizationBiomedical research articles are mainly written for experts and often contain terms that are difficult for the general public to under…
↗Not briefed
bluesky
bluesky-ai
27dWas a bit skeptical about the LLM session at this conference, but the take home message seems to be: we checked if an LLM could do task X a…Was a bit skeptical about the LLM session at this conference, but the take home message seems to be: we checked if an LLM could do tas…
↗Not briefed
arxiv
arxiv-ai
27dMARS: A Momentum-Aware Adaptive Rating System for Personalized AssessmentAdaptive assessment systems personalize learning by estimating learner proficiency and recommending suitable questions. However, many…
↗Not briefed
bluesky
bluesky-ai
27dA fellow member of the #HouseofLords expressed astonishment that I don't use so-called AI for research or speeches. I am astonished anyone…A fellow member of the #HouseofLords expressed astonishment that I don't use so-called AI for research or speeches. I am astonished a…
↗Not briefed
hn
hn-ai
27dClaude Fable 5.5 traces have been spotted in Azure configs—
↗Not briefed
arxiv
arxiv-ai
27dRouter Drift Control: Mitigating Off-Policy Degradation in Mixture-of-Experts Reinforcement LearningReinforcement learning (RL) can become off-policy due to training-inference mismatch, mini-batch updates, and asynchronous training.Th…
↗Not briefed
arxiv
arxiv-ai
27dData Strategies for Large Language Models: A Unified Perspective on the Training LifecycleData strategy plays a central role in training large language models (LLMs). It governs the selection, weighting, and scheduling of tr…
↗Not briefed
bluesky
bluesky-ai
27di saw someone use an llm to convert usd to thb. were fucked as a species.i saw someone use an llm to convert usd to thb. were fucked as a species.
↗Not briefed
bluesky
bluesky-ai
27dBesenkompott hat völlig recht. Ganz ehrlich: Welcher normale Mensch interessiert sich schon für so abwegige Themen wie Luftfahrt, Windenerg…Besenkompott hat völlig recht. Ganz ehrlich: Welcher normale Mensch interessiert sich schon für so abwegige Themen wie Luftfahrt, Wind…
↗Not briefed
bluesky
bluesky-ai
27dOpenAI, desenvolvedora do ChatGPT, notificou a Polícia Federal sobre interações de ao menos dois usuários que manifestaram intenção de atac…🔧 ChatGPTOpenAI, desenvolvedora do ChatGPT, notificou a Polícia Federal sobre interações de ao menos dois usuários que manifestaram intenção de…
↗Not briefed
arxiv
arxiv-ai
27dNo Single Tokenizer Feature Reliably Predicts Downstream Language Model PerformanceTokenization is a critical component of the language modeling pipeline, but it is poorly understood. Tokenizers can impact downstream…
↗Not briefed
hn
hn-ai
27dShow HN: Pelago – your Claude Code and Codex history as a pixel-art world—
↗Not briefed
hn
hn-ai
27dRun: AI Now Open Source—
↗Not briefed
rss
techmeme
27dRamp raised $1.85B led by Dragoneer at a $60B valuation, up from $44B after raising $750M in June, and hits $2B in annualized revenue, up from $1.5B in June (Rebecca Torrence/Bloomberg)Rebecca Torrence / Bloomberg: Ramp raised $1.85B led by Dragoneer at a $60B valuation, up from $44B after raising $750M in June, and h…
↗Not briefed
bluesky
bluesky-ai
27dAnthropic got an AI model to run across 15 cheap TV boxes. Its goal: to explore how AI could someday spread between devices like a computer…Anthropic got an AI model to run across 15 cheap TV boxes. Its goal: to explore how AI could someday spread between devices like a com…
↗Not briefed
hn
hn-ai
27dHandwriting Animation for the Web [Multilingual]—
↗Not briefed
rss
theverge-ai
27dYour secrets aren’t safe with this cute robot toy — they’re on Chinese servers ready to be hackedEiliko. | Photo by Sean Hollister / The Verge There's a new digital pet in town - a cute little marshmallow of a robot that can chan…
↗Not briefed