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
2mo(actually my guess is that it's even more bog-standard: a completely normal autoregressive LLM in constrained-decoding model, tuned on clas…(actually my guess is that it's even more bog-standard: a completely normal autoregressive LLM in constrained-decoding model, tuned on…
↗Not briefed
bluesky
bluesky-ai
2moThis is very funny. It’s pretty technical, so for the non-software engineers, the TL;DR is that the LLM coding tools can hallucinate too. …This is very funny. It’s pretty technical, so for the non-software engineers, the TL;DR is that the LLM coding tools can hallucinate t…
↗Not briefed
hn
hn-ai
2moJotter – A local recording and transcription tool written in Rust—
↗Not briefed
bluesky
bluesky-ai
2moGood Lord.... nypost.com/2026/09/16/t...Good Lord.... nypost.com/2026/09/16/t...
↗Not briefed
rss
simon-willison
2moClaude Cowork and chat are now one Claude—
↗Not briefed
bluesky
bluesky-ai
2moHe is nothing if not an ignorant ass Finland regulates AI through the EU Artificial Intelligence Act (Regulation (EU) 2024/1689), supplemen…He is nothing if not an ignorant ass Finland regulates AI through the EU Artificial Intelligence Act (Regulation (EU) 2024/1689), supp…
↗Not briefed
rss
latent-space
2moUnderwriting Superintelligence: Backing Agents you can Sue — Rune Kvist, AIUCWe sit down with AIUC’s CEO on their Series A!
↗Not briefed
bluesky
bluesky-ai
2moI'm fascinated by U.S. tech and business journalism because it doesn't adhere to literally any rules of actual journalism. There's no effo…I'm fascinated by U.S. tech and business journalism because it doesn't adhere to literally any rules of actual journalism. There's no…
↗Not briefed
hn
hn-ai
2moShow HN: Real-Time Chess Assistant (C++ DirectX, OpenCV, ImGui and Stockfish)—
↗Not briefed
bluesky
bluesky-ai
2moThe guardrails don't work. Again and again and again www.nytimes.com/2026/09/15/w...The guardrails don't work. Again and again and again www.nytimes.com/2026/09/15/w...
↗Not briefed
bluesky
bluesky-ai
2moWithin epsilon of nobody should be sinking their own money into Anthropic, and people defaulting to Anthropic are seeing an unfortunate mis…Within epsilon of nobody should be sinking their own money into Anthropic, and people defaulting to Anthropic are seeing an unfortunat…
↗Not briefed
bluesky
bluesky-ai
2moOpenAI is equivalent or substantially better (usually the latter) at every price point. Open models providers are generally competitive wit…OpenAI is equivalent or substantially better (usually the latter) at every price point. Open models providers are generally competitiv…
↗Not briefed
bluesky
bluesky-ai
2moReports on the OpenAI-Hugging Face incident now stand as an account of harm through which public institutions should examine OpenAI’s decis…🔧 Hugging FaceReports on the OpenAI-Hugging Face incident now stand as an account of harm through which public institutions should examine OpenAI’s…
↗Not briefed
bluesky
bluesky-ai
2moThe LLM doesn't see strings the way ELIZA does. It tokenizes it. It can't see the letters or words. If I put your post into GPT, it would s…The LLM doesn't see strings the way ELIZA does. It tokenizes it. It can't see the letters or words. If I put your post into GPT, it wo…
↗Not briefed
arxiv
arxiv-ai
2moWhy Pretraining Fails to Share Cross-Lingual KnowledgeLarge Language Models (LLMs) have made remarkable progress in the processing and modeling of many languages. Yet, unlike human multili…
↗Not briefed
bluesky
bluesky-ai
2moi hereby guess that Union-Alpha is GPT-6 Soli hereby guess that Union-Alpha is GPT-6 Sol
↗Not briefed
arxiv
arxiv-ai
2moPhysics-Informed Hemodynamic Modeling for Data-Free Prediction and Sparse-Data AssimilationClinical decision-making for coronary intervention relies mainly on angiography and fractional flow reserve (FFR). However, angiograph…
↗Not briefed
bluesky
bluesky-ai
2monew stealth model on openrouter, "union alpha," tokenizer seems to be GLM5-ish, maybe an american finetune?🔧 OpenRouternew stealth model on openrouter, "union alpha," tokenizer seems to be GLM5-ish, maybe an american finetune?
↗Not briefed
bluesky
bluesky-ai
2moMy colleagues and I on the Science Committee sat down with representatives from OpenAI, Anthropic, Hugging Face, and METR to discuss rogue …🔧 Hugging FaceMy colleagues and I on the Science Committee sat down with representatives from OpenAI, Anthropic, Hugging Face, and METR to discuss r…
↗Not briefed
arxiv
arxiv-ai
2moLearning-Induced Dynamical Transition in Recurrent Neural NetworksLearning in recurrent neural networks can fundamentally reshape their underlying dynamics, transforming initially chaotic activity int…
↗Not briefed
bluesky
bluesky-ai
2moamong other things the main reason we need to end this llm crap is because i can't stand spending one more second watching people worry abo…among other things the main reason we need to end this llm crap is because i can't stand spending one more second watching people worr…
↗Not briefed
arxiv
arxiv-ai
2moRadio-Frequency Convolutional Neural NetworksRunning artificial intelligence (AI) models directly on edge devices such as smartphones, wearables, and drones offers low latency, pe…
↗Not briefed
arxiv
arxiv-ai
2moObjective vs. Search: Decomposing What Makes a Good TokeniserTwo dominant tokenisation algorithms are used by modern language models: byte-pair encoding (BPE) and UnigramLM. These differ along tw…
↗Not briefed
arxiv
arxiv-ai
2moA Zeroth-Order Paradigm for LLM Preference AlignmentDirect preference alignment methods are widely used to align large language models (LLMs) with human preferences because of their comp…
↗Not briefed
arxiv
arxiv-ai
2moPANORAMA: Panoptic Grounded Captioning via Mask Proposal SelectionIntelligent systems that act in the world require image understanding that is both comprehensive and spatially grounded. Current visio…
↗Not briefed
hn
hn-ai
2moShow HN: TabMail. Open-source AI email for Thunderbird and iPhoneHi all,<p>I&#x27;m the sole developer of TabMail. I wanted my email client to have AI features, and ended up making a Thunderbird add-…
↗Not briefed
bluesky
bluesky-ai
2moyeah I mean the whole intellectual (derogatory) exercise that led to the creation of OpenAI and Anthropic could easily be glossed as "if we…yeah I mean the whole intellectual (derogatory) exercise that led to the creation of OpenAI and Anthropic could easily be glossed as "…
↗Not briefed
hn
hn-ai
2moShow HN: I built a router for agent tools—
↗Not briefed
hn
hn-ai
2moAI Safety Needs a State—
↗Not briefed
arxiv
arxiv-ai
2moDreaming the Sound of Contact: Leveraging Video and Audio Generation for Zero-Shot Force-Aware Manipulation and Data GenerationRecent advances in video generation allow robots to learn manipulation trajectories from generated videos. However, these approaches p…
↗Not briefed
arxiv
arxiv-ai
2moCharacterizing Web Search by Conversational LLM Agents: From Search Decisions and Strategies to Results and Responses🔧 ChatGPTConversational LLM agents increasingly rely on Web search, yet the end-to-end lifecycle of agentic search remains poorly understood. W…
↗Not briefed
hn
hn-ai
2moShow HN: Pkgbot – Fast Linux package search engineI recently switched to NixOS and in the process of setting it up, found the package search on its repo site very clunky and tedious. N…
↗Not briefed
arxiv
arxiv-ai
2moExponential Hardness of Off-Policy Evaluation under History-Dependent LoggingCan a logged dataset visit every hidden state frequently and still be exponentially uninformative about a target policy's value? We sh…
↗Not briefed
bluesky
bluesky-ai
2moI have been spending a bit of time reading up on the AI Bioweapons discourse on X and I must say, fascinating. It also made me realize tha…I have been spending a bit of time reading up on the AI Bioweapons discourse on X and I must say, fascinating. It also made me realiz…
↗Not briefed
arxiv
arxiv-ai
2moScienceIDE: Turning World's Scientific Codebase into Agent Learnable EnvironmentsScientific code repositories encode decades of human knowledge in executable models, methods, and tools. Yet fragmented toolchains, im…
↗Not briefed
bluesky
bluesky-ai
2moThe new iOS is good. Snow Leopard style update: Well thought out, smooth and polished. Apple did a good job this year. The new Siri is be…The new iOS is good. Snow Leopard style update: Well thought out, smooth and polished. Apple did a good job this year. The new Siri…
↗Not briefed
hn
hn-ai
2moOpenAI's Navier-Stokes Proof Meets a New Kind of Database—
↗Not briefed
arxiv
arxiv-ai
2moCognitive Extensions for Dual-Process Language Agents: Memory and Self-Reflection in Interactive EnvironmentsLanguage agents remain brittle in interactive environments, where success requires long-horizon state tracking, valid action execution…
↗Not briefed
hn
hn-ai
2moAsk HN: Google has suspended our business profileWe have had a successful business running and posted on Google for almost 15 years. Since moving across the country, we changed the ad…
↗Not briefed
bluesky
bluesky-ai
2moIf you mansplain me how to air-gap, I will violence you, mansplain to someone else, I'm the wrong person, you need to find an OpenAI guy.If you mansplain me how to air-gap, I will violence you, mansplain to someone else, I'm the wrong person, you need to find an OpenAI g…
↗Not briefed
hn
hn-ai
2moAdvertising is coming to AI chatbots and it could influence the answers you get—
↗Not briefed
arxiv
arxiv-ai
2moAffora: A Design System for Agent-Friendly InterfacesComputer-use agents increasingly operate software designed for people, but interfaces often leave actions or task state unclear to mac…
↗Not briefed
arxiv
arxiv-ai
2moFlag Game: A Toy Model for Mechanistic Swarm InterpretabilityEmergent coordinated behaviors of AI agents are starting to present critical safety risks. A key phenomenon driving these behaviors is…
↗Not briefed
bluesky
bluesky-ai
2moSocial engineering just means manipulating humans. Usually it’s goal-directed (prompting humans to take action on behalf of the model). B…Social engineering just means manipulating humans. Usually it’s goal-directed (prompting humans to take action on behalf of the model…
↗Not briefed
arxiv
arxiv-ai
2moRandomized SVD Approximations for Spectral Co-Clustering of Word-Document MatricesSpectral co-clustering is a useful tool for discovering latent structure in word-document matrices, but its reliance on singular value…
↗Not briefed
arxiv
arxiv-ai
2moPlaying log(N)-Questions over Wikipedia Abstracts: Communication Efficiency Between Paired Frontier Models🔧 Grok 4🔧 GPT-5We evaluate six frontier language models on the two-agent $\log(N)$-Questions game. A questioner sees $N$ Wikipedia lead paragraphs an…
↗Not briefed
bluesky
bluesky-ai
2moWeil du Texas Kettensägen Massaker gesehen hast, interessiert dich wahrscheinlich auch das geheime Leben der Bäume. Keine Ahnung, weshalb …Weil du Texas Kettensägen Massaker gesehen hast, interessiert dich wahrscheinlich auch das geheime Leben der Bäume. Keine Ahnung, wes…
↗Not briefed
bluesky
bluesky-ai
2moWhereas our ai is so crap it could only have a ‘disastrous impact’ on our bottom line.Whereas our ai is so crap it could only have a ‘disastrous impact’ on our bottom line.
↗Not briefed
bluesky
bluesky-ai
2moSo not only are there people pitching AI to help write more papers no one has time to read, now there are pitches to make papers themsleves…So not only are there people pitching AI to help write more papers no one has time to read, now there are pitches to make papers thems…
↗Not briefed
bluesky
bluesky-ai
2moL'IApocalypse, c'est le techno-capitalisme sans garde-fous : www.arretsurimages.net/chroniques/c... par @winstonsmith.eurosky.socialL'IApocalypse, c'est le techno-capitalisme sans garde-fous : www.arretsurimages.net/chroniques/c... par @winstonsmith.eurosky.social
↗Not briefed