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
1moScore Centering Stabilizes Off-policy Reinforcement LearningReinforcement learning (RL) of large language models is notoriously sensitive to small differences between training and inference engi…
↗Not briefed
arxiv
arxiv-ai
1moAn Empirical Study of Harness Design for Coding AgentsCoding harnesses shape how autonomous coding agents translate model capabilities into long-horizon software-engineering performance, y…
↗Not briefed
bluesky
bluesky-ai
1moI don't agree with this critique at all. It would be trivial for an LLM model to replicate the collective and distributed compute that cry…🔧 ReplicateI don't agree with this critique at all. It would be trivial for an LLM model to replicate the collective and distributed compute tha…
↗Not briefed
bluesky
bluesky-ai
1moAs a cyber security researcher I'm smart enough to know that it doesn't take a datacenter full of GPUs to run semantically guided cyber wea…As a cyber security researcher I'm smart enough to know that it doesn't take a datacenter full of GPUs to run semantically guided cybe…
↗Not briefed
bluesky
bluesky-ai
1mothe retro gaming space is currently 50% "AI is an abomination that destroys everything it touches and I will NEVER forgive anyone who uses …the retro gaming space is currently 50% "AI is an abomination that destroys everything it touches and I will NEVER forgive anyone who…
↗Not briefed
bluesky
bluesky-ai
1moAnthropic 100% deserves plenty of criticism, sometimes for not really following its core researchers' principles when they're difficult and…Anthropic 100% deserves plenty of criticism, sometimes for not really following its core researchers' principles when they're difficul…
↗Not briefed
bluesky
bluesky-ai
1modefinitely important to remember that core people at Anthropic are mostly people who worked at OpenAI & left because they found (very speci…definitely important to remember that core people at Anthropic are mostly people who worked at OpenAI & left because they found (very…
↗Not briefed
arxiv
arxiv-ai
1moJEPA-Anything: Learning Predictive Models across Different WorldsWorld modeling enables intelligence to anticipate consequences, guide interventions, and learn from interaction. Yet predictive models…
↗Not briefed
arxiv
arxiv-ai
1moPosteriorBench: From Point Estimates to Posterior Matching in Evaluating Generative Inverse SolversGenerative models are increasingly used to solve scientific inverse problems, but existing evaluations still focus primarily on whethe…
↗Not briefed
hn
hn-ai
1moShow HN: The browser tab as a back end—
↗Not briefed
arxiv
arxiv-ai
1moRetireOPD: Self-Retiring On-Policy Distillation for Agentic Reinforcement LearningMulti-turn agents trained with reinforcement learning (RL) receive a single scalar reward per trajectory, which motivates self on-poli…
↗Not briefed
bluesky
bluesky-ai
1moAmid calls for urgent AI action from Congress, House heads home to campaignAmid calls for urgent AI action from Congress, House heads home to campaign
↗Not briefed
arxiv
arxiv-ai
1moHarm Laundering in GPT Models: Evidence That Gender Discrimination Is Transformed Rather Than Reduced Across Safety-Trained Generations🔧 GPT-5Safety evaluations for large language models rely on surface-form classifiers that report declining harm scores across model generatio…
↗Not briefed
bluesky
bluesky-ai
1moAll that being said: I am still not an LLM even with all my age confusion and em dashes.All that being said: I am still not an LLM even with all my age confusion and em dashes.
↗Not briefed
hn
hn-ai
1moPixel phones getting Harry Potter ad disguised as theme pack—
↗Not briefed
arxiv
arxiv-ai
1moGeoAAC: Geometry-Based Adaptive Action Chunking from Denoising Trajectories in VLA PoliciesAction chunking is widely used for action generation and execution in Vision-Language-Action (VLA) policies, yet existing approaches c…
↗Not briefed
bluesky
bluesky-ai
1mo* OPENAI, NVIDIA CEOS TO ATTEND TRUMP-XI STATE DINNER: ABC* OPENAI, NVIDIA CEOS TO ATTEND TRUMP-XI STATE DINNER: ABC
↗Not briefed
hn
hn-ai
1moShow HN: I wrote a custom assembler for CHIP-8 in C++I'm writing a custom assembler for CHIP-8 and just finished adding support for the DB/.byte directive so I can embed sprites…
↗Not briefed
bluesky
bluesky-ai
1mo*OPENAI, NVIDIA CEOS TO ATTEND TRUMP-XI STATE DINNER: ABC*OPENAI, NVIDIA CEOS TO ATTEND TRUMP-XI STATE DINNER: ABC
↗Not briefed
bluesky
bluesky-ai
1moI'd forgotten that these people think climate change is "overplayed" because it won't cause the instant extinction of humanity and $640 bil…I'd forgotten that these people think climate change is "overplayed" because it won't cause the instant extinction of humanity and $64…
↗Not briefed
bluesky
bluesky-ai
1moIt’s like there were movies that foretold this www.independent.co.uk/tech/securit...It’s like there were movies that foretold this www.independent.co.uk/tech/securit...
↗Not briefed
bluesky
bluesky-ai
1moIn the exact same convo he also admitted that Claud fucking sucked at coding and he hates AI in general, so it's even more insane to me tha…In the exact same convo he also admitted that Claud fucking sucked at coding and he hates AI in general, so it's even more insane to m…
↗Not briefed
arxiv
arxiv-ai
1moSemantic Action Graph: A Shared Representation for Agent Grounding and Human Interpretation of Sports HighlightsGenerative agents are increasingly used to select and narrate video highlights, but they typically operate over unstructured or frame-…
↗Not briefed
bluesky
bluesky-ai
1mo“rumored” 🙃“rumored” 🙃
↗Not briefed
hn
hn-ai
1moThe Life Sciences Verification Program—
↗Not briefed
arxiv
arxiv-ai
1moCalibrated RF-Fingerprinting Under Interference With Heterogeneous Transmission ProtocolsRadio Frequency(RF)-Fingerprinting is a spectrum monitoring technique that identifies specific transmitters based on hardware impairme…
↗Not briefed
arxiv
arxiv-ai
1moAgile-WAM: An Agile Tactile World Action Model for Contact-Rich Robot ControlWorld Action Models (WAMs) advance beyond conventional visuomotor policies by jointly predicting future world states and robot actions…
↗Not briefed
hn
hn-ai
1moShow HN: Windows-style Excel ribbon shortcuts and formatting macros for MacI worked in finance for 5+ years and became extremely proficient with Excel on Windows. When I left to build a start-up, I switched to…
↗Not briefed
arxiv
arxiv-ai
1moPrediction-Powered Smoothing and Validation for Disaggregated AI EvaluationEvaluating an AI system requires disaggregated assessment, as performance varies across domains such as benchmark task types or conver…
↗Not briefed
hn
hn-ai
1moOpenWispr, a free, open-source push-to-talk AI dictation app for Android—
↗Not briefed
arxiv
arxiv-ai
1moOPTED: On-Policy Fine-Tuning for End-to-End Driving using a Render-Free TeacherAs scaling pre-training data alone yields diminishing returns, post-training is becoming increasingly important across physical AI dom…
↗Not briefed
arxiv
arxiv-ai
1moRAFT: A Stateful Retrieval-Augmented Framework for Troubleshooting AgentsEffective troubleshooting agents in enterprise customer support depend on retrieving actionable guidance from similar historical cases…
↗Not briefed
arxiv
arxiv-ai
1moLarge Language Models as Falsifiers for Cyber-Physical SystemsFalsification searches for counterexamples to formal specifications in cyber-physical systems (CPS). With specifications written in Si…
↗Not briefed
arxiv
arxiv-ai
1modQwen3.5: Hybrid-Attention Diffusion Language ModelsAdapting a pretrained autoregressive (AR) model is a cost-efficient route to a diffusion language model (DLM). While nearly all such a…
↗Not briefed
hn
hn-ai
1moPython Packaging Council Election Results—
↗Not briefed
hn
hn-ai
1moPSF Board Election Results—
↗Not briefed
bluesky
bluesky-ai
1mo‘Trust is built on accountability, but the AI companies — from hyperscalers like Google to model builders like OpenAI and Anthropic — have …‘Trust is built on accountability, but the AI companies — from hyperscalers like Google to model builders like OpenAI and Anthropic —…
↗Not briefed
hn
hn-ai
1moBenchmark Jev vs. Gemini Flash and Claude Fable on Code Review—
↗Not briefed
hn
hn-ai
1moWorkout by Muscle – Claude Solution—
↗Not briefed
arxiv
arxiv-ai
1moMILER: Semantic Mid-Level Representation for Sim-to-Real Reinforcement Learning in Unstructured Autonomous DrivingReinforcement learning constitutes a promising approach owing to its potential for superhuman performance and self-learned policies. H…
↗Not briefed
bluesky
bluesky-ai
1mo*OPENAI, NVIDIA CEOS TO ATTEND TRUMP-XI STATE DINNER: ABC yes of course they are*OPENAI, NVIDIA CEOS TO ATTEND TRUMP-XI STATE DINNER: ABC yes of course they are
↗Not briefed
bluesky
bluesky-ai
1moAn agent couldn't retrieve real county earnings figures, so it made numbers up and presented them as real data. Nvidia's CEO says the answe…An agent couldn't retrieve real county earnings figures, so it made numbers up and presented them as real data. Nvidia's CEO says the…
↗Not briefed
arxiv
arxiv-ai
1moVideo DeltaNet: A Video-Native Hybrid Attention for Livestream Video GenerationVideo diffusion models repeatedly process long spatiotemporal token sequences during denoising, making attention a major computational…
↗Not briefed
hn
hn-ai
1moProjects Redesigned: From Folder to Conversation—
↗Not briefed
bluesky
bluesky-ai
1moeven vibecoders can't be bothered to tell an llm to write documentation for themeven vibecoders can't be bothered to tell an llm to write documentation for them
↗Not briefed
bluesky
bluesky-ai
1moReally wish "AI" hadn't become the word for everything. An LLM that can't reliably count the e's in "seventeen" and a purpose-built syste…Really wish "AI" hadn't become the word for everything. An LLM that can't reliably count the e's in "seventeen" and a purpose-built…
↗Not briefed
hn
hn-ai
1moCompiler-style optimization for drawing via Skia—
↗Not briefed
hn
hn-ai
1moHack: Coding at No Cost in Codex, Claude Code, Hermes, VSCode—
↗Not briefed
hn
hn-ai
1moI built a multi platform torrent video streamer—
↗Not briefed
hn
hn-ai
1moAn ahead-of-time JavaScript compiler—
↗Not briefed