01tech5 sources2dOpenAI Says GPT-5.6 Sol Found PHP Flaw, With More Discoveries ComingOpenAI said GPT-5.6 Sol helped identify a PHP security flaw and described it as the first of several discoveries still to be announced. The disclosure adds specific model attribution to OpenAI’s broader account of model-assisted PHP vulnerability research and remediation efforts.@gdb · +4 sources+
02tech2 sources2dTesla Highlights FSD Adoption, Next-Gen Roadster and Semi Production in Q3 UpdateTesla said FSD Supervised was active on more than 75% of new vehicle deliveries in North America, unveiled its next-generation Roadster, reported record deliveries in Canada, Australia, Japan and Taiwan, and entered volume production of the Tesla Semi.@elonmusk · +1 source+
03tech3 sources2dElectronic Arts to Cut 400 Jobs, Shut Down BioWare and Cancel Next Mass EffectElectronic Arts plans to eliminate about 400 jobs, shut down BioWare and cancel the next Mass Effect game as part of a restructuring of its video-game business.@andrewcurran_ · +2 sources+
04tech3 sources2dMeta Ordered to Pay $25 Billion in New Mexico Privacy LawsuitMeta was ordered to pay $25 billion after New Mexico alleged the company deceived users about its data-privacy practices. CNBC reported that the state’s initial award was $219.5 billion before the penalty was reduced.@JonathanVanian · +2 sources+
05tech3 sources2dAI Research Startup Flapping Airplanes Raises $1 Billion at $6 Billion ValuationFlapping Airplanes, a year-old AI research startup without a commercial product, raised $1 billion at a $6 billion valuation in a round led by Index Ventures and Lux Capital. Sequoia Capital, Alphabet’s GV and Kleiner Perkins also participated.@nmasc_ · 2.5× normal · +2 sources+
06tech2 sources · 6 posts2dRecursive’s Flint Claims Best-in-Class Agentic Coding Performance and Local Inference SpeedRecursive’s open-source Flint coding model claims top performance among sub-50B mixture-of-experts models on SWE-bench Pro v2, Terminal-Bench 2.1, and SWE-bench Verified. Richard Socher said it rivals similarly capable dense models while running substantially faster locally, reaching more than 125 tokens per second on an M4 Pro MacBook Pro and 150 tokens per second on a 64GB NVIDIA RTX Spark. Recursive said Flint was also a real-world test of its Eureka Machine automated research system, which handled ideation, implementation, validation, and selection of subsequent experiments throughout development.@_rockt · 5.3× normal · +1 source+
07tech2 sources · 3 posts2dSundar Pichai Defends Google’s Aggressive AI Capex Despite Negative Free Cash FlowIn an interview about Gemini 4 Argon and Google’s AI compute constraints, CEO Sundar Pichai said Google was willing to let free cash flow turn negative to fund higher AI capital spending, citing early signals and confidence in the opportunity.@alexeheath · 2.3× normal · +1 source+
08tech2 sources2dNVIDIA Introduces DSX PODs for AI Data Centers, Promising Deployment in 14 DaysNVIDIA introduced DSX POD at GTC Berlin as an integrated, tested AI-factory building block that can be deployed at prepared sites in as little as 14 days, adding a rapid-deployment capability to its modular data-center infrastructure.@nvidia · 2.2× normal · +1 source+
09tech1 source2dSatya Nadella Outlines Microsoft’s Frontier Ecosystem StrategyIn his annual letter, Microsoft CEO Satya Nadella said the company is building a frontier ecosystem to translate technological progress into real-world outcomes and economic opportunity.@satyanadella+
10tech1 source2dOpenAI Brings Frontier Models and Codex to Google Cloud Model GardenOpenAI said its frontier models and Codex agentic workflows will be available through Google Cloud’s Model Garden, giving enterprises access to OpenAI technology within their existing cloud environment.@OpenAI · 2.3× normal+
11tech2 sources2dMeta Plans Non-Voting Share Class to Fund AI InvestmentMeta plans to create a new non-voting share class that would let it raise capital for AI investment without reducing Mark Zuckerberg’s voting control.@ft-companies · +1 source+
12tech2 sources2dAmanda Smith Jailed After Fraudulent Cash App Account Was Opened in Her NameAmanda Smith says she spent three nights in jail and faces up to six years in prison after a scammer allegedly used her identity to open a Cash App account and commit fraud. She says account records and other data do not match her, while Cash App has declined to provide details because it considers the account not hers.@jason_koebler · 2.5× normal · +1 source+
13tech1 source2dStratechery and Stripe Form PartnershipStratechery analyst Ben Thompson and Stripe have formed a new partnership, praised by commentator Patrick Collison.@patrickc+
14tech1 source2dNate Lambert Publishes Deck on the State of Open AI ModelsNate Lambert assembled a presentation covering open-model licenses, competitiveness, adoption, distillation, and the risks and benefits of openness, with plans to record a video lecture based on it.@natolambert · 1.7× normal+
15tech5 sources · 7 posts2dSpaceX Reports $4.4 Billion Net Profit, With $2.8 Billion From Core OperationsSpaceX reported roughly $4.4 billion in net profit, including about $2.8 billion in operating profit from its core businesses and approximately $1.6 billion from interest income, other income and a tax benefit. The results came as Elon Musk confirmed SpaceX and Tesla are exploring a possible deal.@edludlow · 2.2× normal · +4 sources+
16tech1 source2dPieter Abbeel Describes Faster, Cheaper Humanoid Full-Body Control Training MethodPieter Abbeel described a method for training a single humanoid-control policy across many motions using one GPU. The approach combines a faster world model with backpropagation through time to gather more training signal than conventional policy-gradient methods.@pabbeel+
17tech1 source2dOpenAI Details Safeguards for Distribution PartnersOpenAI described how it works with distribution partners to support responsible access, uphold consistent safeguards and clarify expectations under its Usage Policies.@openai-news+
18tech1 source2dChristopher Manning Discusses Plan for Single-Researcher Frontier-Language-Model Pre-TrainingChristopher Manning expressed excitement and skepticism about @bneyshabur’s plan to make frontier-performance language-model pre-training feasible for a single researcher, noting that a $10 million academic budget could be substantial and praising @mirendil’s results.@chrmanning+
19tech1 source2dGoogle DeepMind Reports AI-Driven Discovery of Long-Running Simple ProgramsGoogle DeepMind researchers reported results on discovering simple programs that run for unusually long periods before halting, highlighting progress in AI-driven open-ended discovery across mathematics and theoretical computer science.@ZoubinGhahrama1+
20tech1 source2dGoogle DeepMind Demonstrates Gemini 4 Argon 3D Anatomy VisualizationGoogle DeepMind demonstrated Gemini 4 Argon combining three open-source datasets to create an interactive 3D visualization of human anatomy, including animated layers, isolated muscles and a zoomable beating heart.@GoogleDeepMind · 3.3× normal+
21tech1 source2dAnthropic Appoints Christopher Stubbs to Long-Term Benefit TrustAnthropic said experimental physicist and former Harvard Dean of Science Christopher Stubbs is joining its Long-Term Benefit Trust. Stubbs has also advised the U.S. government on national security and arms control.@AnthropicAI+
22tech1 source2dAI Security Leaderboard Says Four Frontier Models Clear Minimal Safeguard StandardThe AI Security Leaderboard reported that all four tested frontier models now clear its Minimal Standard for Safeguards, with no universal jailbreaks found. It said two models had previously been broken for under $300 in tests involving chemical, radiological and nuclear misuse.@farairesearch · 5.0× normal+
23tech1 source2dElon Musk Promotes Grok for Credit-Rating AssistanceElon Musk promoted Grok as a tool that can help users protect and repair their credit ratings, though the post did not provide details about how the capability works or its accuracy.@elonmusk+
24tech1 source2dNVIDIA and Partners Expand Global Compute Supply CapacityNVIDIA said it is working with Amkor, IBIDEN, KYEC, Samsung Mechanics, SPIL and Unimicron to expand global supply capacity for components, advanced packaging and testing amid extraordinary demand for compute.@nvidia+
25tech3 sources · 5 posts3dSpaceX Reports Strong Growth and Eyes $1 Trillion Revenue by 2028SpaceX reported $14.6 billion in third-quarter 2026 revenue, up 161% year over year, and said demand for its compute capacity exceeds supply. Elon Musk also said $1 trillion in annual revenue by 2028 is no longer impossible, alongside plans to bring roughly 10 gigawatts of additional compute online in 2027.@elonmusk · 4.5× normal · +2 sources+
26tech1 source2dElevenLabs Opens Limited Alpha of Real-Time Voice TranslationElevenLabs opened a limited alpha of Eleven Realtime Translation, a product that provides live voice translation during calls, following the launch of its Eleven Realtime voice-agent platform.@ElevenLabs · 7.0× normal+
27tech1 source2dOpenAI Packages Codex Skill to Check Whether Websites Are AffectedOpenAI packaged a Codex skill that guides users through checking whether their websites are affected, directing Codex to a linked software repository.@OpenAI · 3.7× normal+
28tech1 source2dChris Manning Compares Smaller Model With Thinky Machines’ InklingChris Manning praised an unspecified model for performing nearly as well as Thinky Machines’ Inkling despite being more than 10 times smaller, while saying he would wait before buying a Mac Studio.@chrmanning · 2.6× normal+
29tech1 source · 4 posts2dCohere Rerank 5 Pro Claims Top Performance on Major Retrieval BenchmarksCohere says Rerank 5 Pro outperformed every model it tested on the nanoMTEB, ViDoRe V3 and BRIGHT public benchmarks, while Rerank 5 Fast delivers nearly three times Pro’s throughput and roughly three times faster response times. Cohere says Fast also outperforms Voyage AI Rerank 3 Lite and edges out Qwen3-Reranker-8B.@cohere · 4 posts+
30tech1 source2dU.S. Marines Train With Small Drones Aboard USS Portland in Arabian SeaU.S. Marines trained with small unmanned aerial systems aboard the USS Portland in the Arabian Sea as CENTCOM’s innovation battle lab tests emerging technologies for deployed service members.@CENTCOM+
31tech1 source2dThomas Wolf Shows 4-Bit lm_head Quantization Nearly Matches bf16Thomas Wolf found that quantizing a language model’s lm_head to 4 bits produced performance within 0.7 points of a bf16 version, challenging the conventional practice of keeping the component in high precision.@Thom_Wolf · 4.5× normal+
32tech1 source2dWIRED’s Will Knight Praises Flint for Fast, Reliable CodingWill Knight said he used Flint to build multiple programs and found the software fast, capable and reliable compared with similarly sized models, Claude and Codex.@RichardSocher+
33tech1 source2dMicrosoft Reports First Headcount Decline in a DecadeMicrosoft’s annual report says its workforce fell to approximately 223,000 employees as of June 30, down about 5,000 from 228,000 the prior year—the company’s first reported headcount reduction in 10 years.@tomwarren+
34tech1 source2dOpenAI and AT&T Partner to Expand Access to In-House Legal KnowledgeAT&T is working with OpenAI through its LegalEdge team to make legal knowledge easier to access, streamline intake and scale legal judgment across its in-house legal operations.@openai-news+
35tech1 source2dNVIDIA Introduces OpenShell Reference Architecture for Securing AI AgentsJensen Huang introduced NVIDIA’s OpenShell Reference Architecture, designed to help businesses define boundaries for AI agents and stop them when they operate outside those limits.@nvidia · 2.4× normal+
36tech1 source2dCohere Introduces Context Focus to Reduce LLM Context UsageCohere introduced Context Focus, a retrieval feature that returns precise passages with relevance scores instead of entire documents. The company says it can reduce token consumption by up to 41% while maintaining accuracy.@cohere+
37tech1 source2dVivix, Siemens, Snowflake and NVIDIA Develop Digital Twin for Glass FurnacesVivix is working with Siemens, Snowflake and NVIDIA on an executable digital twin for float-glass furnaces that aims to reduce energy use by 3% to 5% and extend furnace life.@snowflake-blog+
38tech1 source2dGPT-Rosalind Identifies Possible Clue to TSC2-Linked Tuberous Sclerosis SeverityGPT-Rosalind found a possible explanation for why TSC2 mutations may cause more severe tuberous sclerosis than TSC1 mutations in an analysis of a mouse-brain dataset, focusing on differences in cerebellar neurons and mTOR regulation.@OpenAI+
39tech2 sources2dThe New York Times Explains How It Verifies A.I.-Generated ImagesA New York Times reporter and editors discussed the challenges of determining whether A.I.-generated images are authentic while reporting the news.@nyt-business · +1 source+
40tech1 source2dGitHub Moves 99.99999% of Repositories to Azure for Agent-Scale InfrastructureGitHub said it moved a copy of 99.99999% of its repositories to Azure and outlined per-reference quorums and object storage to migrate the remainder and scale repository reads independently.@github-blog+
41tech1 source2dWebsite Estimates When AI Could Deprecate Your JobLucas Beyer (bl16) shared a website created by his brother that estimates when a user’s job could be deprecated and explains how AI might affect it, reflecting concerns about automation and career disruption.@giffmana · 4.3× normal+
42tech1 source2dYorkshire Ambulance Service Uses Microsoft Copilot to Reduce Administrative WorkYorkshire Ambulance Service is using Microsoft Copilot to reduce administrative work, improve data use and support a better experience for medical personnel and patients, representing an early NHS England AI deployment.@satyanadella+
43tech1 source2dNVIDIA Says European Companies Are Working With It on AINVIDIA said L’Oréal, Revolut, Spotify, Merck, Vodafone, Zara, Porsche, Nestlé and HSBC are among European companies working with the chipmaker on AI.@nvidia+
44tech1 source2dXbox Cloud Gaming to Open to Everyone in November With Pay-As-You-Go PricingMicrosoft plans to open Xbox Cloud Gaming to anyone in November and will offer pay-as-you-go access starting at $3.99 for 10 hours of cloud gaming.@tomwarren · 7.4× normal+
45tech1 source2dDust Builds Multi-Model AI Platform for Enterprise Agent ControlDust co-founder and CTO Stanislas Polu outlined the startup’s multi-model AI platform, which helps companies manage and maintain control over agentic traces.@snowflake-blog+
46tech1 source2dMistral AI and NVIDIA Collaborate on AI for Industrial EngineeringMistral AI announced a collaboration with NVIDIA to accelerate industrial engineering through faster simulation, open physics AI benchmarks and datasets, and evaluations for agentic AI in computer-aided engineering and electronic design automation.@MistralAI · 5.8× normal+
47tech1 source2dGPT-Rosalind Identifies Potential Cholesterol-Transport Clue to Protect NeuronsGPT-Rosalind surfaced a drug-discovery lead involving cholesterol transport and neuronal protection in Perry syndrome. The finding came from mouse-brain data involving DCTN1, a gene linked to the rare neurodegenerative disease.@OpenAI · 2.7× normal+
48tech1 source2dNVIDIA Says Co-Design Accelerated AI Performance Millionfold in 10 YearsNVIDIA said its extreme hardware-software co-design approach helped accelerate AI performance by one million times over the past decade, with further gains planned.@nvidia+
49tech1 source2dSundar Pichai Says Google Trades Cloud Revenue for Frontier AI ComputeSundar Pichai said Google Cloud sometimes forgoes short-term revenue by reallocating compute capacity to support frontier AI research and development.@kantrowitz+
50tech1 source2dAI21 Labs Says Its Post-Training Work on NVIDIA Nemotron 3.5 Super VL Was Recognized at GTC BerlinAI21 Labs said its Forward-Deployed Researchers post-trained NVIDIA’s recently released Nemotron 3.5 Super VL with Alice, and that the work was recognized during Jensen Huang’s keynote at NVIDIA GTC Berlin 2026.@AI21Labs+