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
13dIt helps to put rules in AGENTS.md. Early on an LLM (that we were required to use) was writing code with memory leaks. It didn't seem to …It helps to put rules in AGENTS.md. Early on an LLM (that we were required to use) was writing code with memory leaks. It didn't see…
↗Not briefed
bluesky
bluesky-ai
13dLightweight AI spots forest fires fast 🔥 A new model called YOLOv8-CNR detects forest fires with higher accuracy and speed than standard …Lightweight AI spots forest fires fast 🔥 A new model called YOLOv8-CNR detects forest fires with higher accuracy and speed than stan…
↗Not briefed
hn
hn-ai
13dShow HN: Hackertube for Roku TVKarpathy recently built Hackertube.<p>I watch a lot of Youtube on my TV and wanted to replace that with hackertube, so I built a roku…
↗Not briefed
bluesky
bluesky-ai
13dIt's The Guardian; they decided to let the picture desk go and replaced it with an LLMIt's The Guardian; they decided to let the picture desk go and replaced it with an LLM
↗Not briefed
hn
hn-ai
13dShow HN: Elo Ratings for Tests, ODIs, and T20Is—
↗Not briefed
arxiv
arxiv-ai
13dSafeguarding Agentic AI in Radiology Workflows: A Multimodel Evaluation of Rule-based and Large Language Model-based VerificationPurpose: To estimate how often tool-using large language model (LLM) agents record clinically relevant false or unsupported informatio…
↗Not briefed
bluesky
bluesky-ai
13dThis is why Schumer refused to help Mamdani win. Schumer's daughter went from working for Amazon to Open AI. He choose his daughters job o…This is why Schumer refused to help Mamdani win. Schumer's daughter went from working for Amazon to Open AI. He choose his daughters…
↗Not briefed
hn
hn-ai
13dWhat Should We Learn? What LLMs Taught Me: Understanding Isn't Producing—
↗Not briefed
bluesky
bluesky-ai
13dLolLol
↗Not briefed
arxiv
arxiv-ai
13dPRISM: Causal Counterfactual Minimization of Agent Tool Privileges via Information Bottleneck Role InductionThis paper shifts the object of role engineering from human users to large language model (LLM) agents, and proposes PRISM to address…
↗Not briefed
arxiv
arxiv-ai
13dWeight-Space Programming of Variational Quantum Circuits: Large Language Models Write Quantum Circuit Parameters Without an OptimizerVariational quantum algorithms (VQAs) optimize circuit angles through repeated quantum-device evaluations, often facing vanishing grad…
↗Not briefed
bluesky
bluesky-ai
13dLoad bearing is auth boundary and retry semantics, not the server. I put one internal tool behind an MCP server and spent a day on scoped t…Load bearing is auth boundary and retry semantics, not the server. I put one internal tool behind an MCP server and spent a day on sco…
↗Not briefed
arxiv
arxiv-ai
13dOstinato: Fast CPU-GPU MoE Inference for Local Agents via Resident Experts in Every Forward PassLanguage-model agents increasingly run on workstations with a few consumer GPUs, where large mixture-of-experts (MoE) models keep part…
↗Not briefed
bluesky
bluesky-ai
13dour Operation Paperclip will be relocating them all to woke FEMA camps and replacing every LLM with Clippyour Operation Paperclip will be relocating them all to woke FEMA camps and replacing every LLM with Clippy
↗Not briefed
arxiv
arxiv-ai
13dDriveFoundry: Co-Evolving Data and Model for Efficient Closed-Loop Iteration in Autonomous DrivingA driving foundation model must learn reusable decision knowledge from rare critical events and apply it under strict runtime constrai…
↗Not briefed
arxiv
arxiv-ai
13dProbing Transformer Representations Layer by Layer, Module by ModuleWhich representation should be probed in a transformer-based foundation model? Recent work shows that the standard practice of probing…
↗Not briefed
hn
hn-ai
13dGoldman Sachs Is Pulling Claude Code from Its Bankers and Traders—
↗Not briefed
bluesky
bluesky-ai
13dObviously I still have a lot to dig into about BLOOM, but it's cool to see a completely different approach to genAI! I appreciate the empir…Obviously I still have a lot to dig into about BLOOM, but it's cool to see a completely different approach to genAI! I appreciate the…
↗Not briefed
bluesky
bluesky-ai
13dWhat would it look like to train an LLM with the ethics baked in? This is an interesting read: www.technologyreview.com/2022/07/12/1... A…What would it look like to train an LLM with the ethics baked in? This is an interesting read: www.technologyreview.com/2022/07/12/1.…
↗Not briefed
arxiv
arxiv-ai
13dData Manipulation Attacks Against Machine Learning Integrated Cyber-Physical SystemsThere is a well-established body of work demonstrating that individual machine learning (ML) algorithms can be manipulated or deceived…
↗Not briefed
bluesky
bluesky-ai
13dI also find it funny that folks are running the post through AI detectors and claiming the results actually prove anything. My dude (or bo…I also find it funny that folks are running the post through AI detectors and claiming the results actually prove anything. My dude (…
↗Not briefed
bluesky
bluesky-ai
13d"The brainstorming took the form of prompting an LLM to generate text with criticism of the draft’s substance and tone at multiple stages o…"The brainstorming took the form of prompting an LLM to generate text with criticism of the draft’s substance and tone at multiple sta…
↗Not briefed
bluesky
bluesky-ai
13d"In preparing this essay, I used an LLM for copy-editing and for brainstorming. The entire text of the essay was typed by my hand directly,…"In preparing this essay, I used an LLM for copy-editing and for brainstorming. The entire text of the essay was typed by my hand dire…
↗Not briefed
hn
hn-ai
13dRethinking Virtual Machine Monitors for Better Security in Clouds [pdf]—
↗Not briefed
bluesky
bluesky-ai
13dSimCity 3000 is coming along well! We're getting there! It does keep finding new pieces to do, but it's solidly progressing. DeepSeek v4.1 …SimCity 3000 is coming along well! We're getting there! It does keep finding new pieces to do, but it's solidly progressing. DeepSeek…
↗Not briefed
bluesky
bluesky-ai
13d#1: Verbosity The description has a shitload of technical explanations for what the hack is doing that just don't make sense to have. Here…#1: Verbosity The description has a shitload of technical explanations for what the hack is doing that just don't make sense to have.…
↗Not briefed
bluesky
bluesky-ai
13d“Ram Altman” Is Now a Linux Script That Emails Sam Altman Daily to Report How Much Unused RAM Is in Your PC“Ram Altman” Is Now a Linux Script That Emails Sam Altman Daily to Report How Much Unused RAM Is in Your PC
↗Not briefed
hn
hn-ai
13dThe efficiency trap: why optimising everything can backfire, badly—
↗Not briefed
arxiv
arxiv-ai
13dA Stability Analysis of AdamW: Unstable Equilibria and Non-ConvergenceAdamW is ubiquitous in deep learning, yet its behavior remains poorly understood. We analyze its dynamics through the lens of dynamica…
↗Not briefed
arxiv
arxiv-ai
13dBoundary Mass in Feature Space as a Label-Budget Diagnostic for RepresentationsLow-label transfer is often decided by a practical question: after a representation looks useful, how many labels are still needed to…
↗Not briefed
hn
hn-ai
13dShow HN: Email-like interface to review, decide, and annotate agent workThere is some recent discourse about what a &quot;better&quot; interface for humans managing agents.<p><pre><code> - https:&#x2F;&#x2…
↗Not briefed
bluesky
bluesky-ai
13dContract drafting/negotiation is a big part of my job. I've gotten ok at spotting when I'm opposite an LLM. The drafting is often solid but…Contract drafting/negotiation is a big part of my job. I've gotten ok at spotting when I'm opposite an LLM. The drafting is often soli…
↗Not briefed
bluesky
bluesky-ai
13dwhat if jeffrey epstein faked his death and then assumed the identity of "dario amodei" and started anthropicwhat if jeffrey epstein faked his death and then assumed the identity of "dario amodei" and started anthropic
↗Not briefed
hn
hn-ai
13dShow HN: Search Everywhere at Once. InstantlyI&#x27;ve made AntiFind, to allow you to search everywhere at once, vastly increasing the efficency of LLM&#x27;s and your own searche…
↗Not briefed
arxiv
arxiv-ai
13dMusic Inpainting With Similarity-Guided Latent DiffusionWe propose a music inpainting technique combining similarity matching and a diffusion-based generative model to replace large segments…
↗Not briefed
bluesky
bluesky-ai
13dWow man that's crazy that Michaelsoft and Amazon and everyone using Anthropic models was unaware of this before the gallant redditoid rode …Wow man that's crazy that Michaelsoft and Amazon and everyone using Anthropic models was unaware of this before the gallant redditoid…
↗Not briefed
bluesky
bluesky-ai
13dOne Claude fixed fingerprint sensor earlier and Mistral Le Chonk, GLM 5.3 and Deepseek teamed up on [redacted]One Claude fixed fingerprint sensor earlier and Mistral Le Chonk, GLM 5.3 and Deepseek teamed up on [redacted]
↗Not briefed
arxiv
arxiv-ai
13dK-PWM: Control-Oriented Structured Bilinear World Models under Partial ObservationWorld models learn internal simulators of environment dynamics for prediction, planning, and control, yet learning them from limited,…
↗Not briefed
arxiv
arxiv-ai
13dHydra: Towards Transferable Multi-Task Learning on Temporal GraphsReal-world evolving networks are naturally modeled as temporal graphs (TGs), where capturing temporal dynamics is essential for predic…
↗Not briefed
hn
hn-ai
13dSelf Improving Software Factory—
↗Not briefed
arxiv
arxiv-ai
13dInfCoiL: Coordinated Planner-Controller Learning for Closed-Loop Physics-Based Human-Object InteractionText- and goal-conditioned human-object interaction (HOI) provides an intuitive interface for specifying high-level intents in physics…
↗Not briefed
hn
hn-ai
13dShow HN: Claude Code status line, hooks and permissions builders (in-browser)—
↗Not briefed
bluesky
bluesky-ai
13dI know the American dream is to make money doing nothing but acting as a middleman, but if your ability to do that is entirely up to the wh…I know the American dream is to make money doing nothing but acting as a middleman, but if your ability to do that is entirely up to t…
↗Not briefed
rss
techmeme
13dExcluding Anthropic's IPO, AI companies sold or priced $300B+ of stock in 2026, with ~$100B more on tap, raising questions about how much the market can absorb (Alex Eichenstein/The Information)Alex Eichenstein / The Information: Excluding Anthropic's IPO, AI companies sold or priced $300B+ of stock in 2026, with ~$100B more o…
↗Not briefed
bluesky
bluesky-ai
13dme 🤝 the ai model samsung uses to turn a white blob into hirez moon pics “skyscrapers are supposed to be straight up and down, so uhhhh”me 🤝 the ai model samsung uses to turn a white blob into hirez moon pics “skyscrapers are supposed to be straight up and down, so uh…
↗Not briefed
arxiv
arxiv-ai
13dDetecting Common Golfing Faults Using AI and Video-Derived Biomechanical FeaturesUnderstanding how an athlete moves can improve both technique and safety, yet outside the laboratory technique is still assessed by ey…
↗Not briefed
bluesky
bluesky-ai
13dAnother annoying little fact: Jessica Schumer works for OpenAI, which is led by Sam Altman, and Alison Schumer works for Meta, which is led…Another annoying little fact: Jessica Schumer works for OpenAI, which is led by Sam Altman, and Alison Schumer works for Meta, which i…
↗Not briefed
arxiv
arxiv-ai
13dAmortized Bayesian Experimental Design with In-Context Knowledge ConditioningAmortized Bayesian experimental design (BED) enables real-time design strategies by shifting acquisition cost offline, but existing me…
↗Not briefed
bluesky
bluesky-ai
13dThere’s newer science on this in relation to LLM use! Spoiler alert: it’s badThere’s newer science on this in relation to LLM use! Spoiler alert: it’s bad
↗Not briefed
bluesky
bluesky-ai
13dOld Optane powers 2.8T AI on a CPU 💨 A homelabber reports running a 2.8-trillion-parameter Kimi K3 AI model on a CPU, using hot, disconti…Old Optane powers 2.8T AI on a CPU 💨 A homelabber reports running a 2.8-trillion-parameter Kimi K3 AI model on a CPU, using hot, dis…
↗Not briefed