AI

The chip giant and the AlphaGo architect's London lab are codesigning the compute stack for reinforcement learning at scale, with eyes on superintelligence.

Fifteen years after making cloud-based computing cheap and accessible, Google is pivoting to an intelligence-first laptop line built around its Gemini AI models.

Google's Gemini Intelligence brings AI-powered form autofill, voice-to-text refinement with Rambler, and generative widgets to Android phones and watches this summer.

The Google Threat Intelligence Group says it disrupted a criminal campaign planning to use an AI-developed two-factor authentication bypass in a mass exploitation event.

The Chinese robotics company that dominates the quadruped market is now selling a transformable manned robot that shifts between two legs and four.

Reports are surfacing across X, Reddit, and Threads of Anthropic's AI suggesting users wrap up sessions and get some rest. Whether it's wellness theater or resource management is the question.

As Anthropic's implied valuation approaches $1 trillion on secondary markets, the company is making clear that most retail-facing investment offers are either void or fraudulent.

A new entrant in developer-focused humanoid robotics, Rotaku is shipping a 90 cm bipedal platform built for whole-body policy training, manipulation research, and teleoperation data collection.

Google Quantum AI partners with five universities to explore how quantum mechanics might govern fundamental biological processes, marking a significant step toward applying quantum computing to life sciences.

A $10 billion joint venture backed by 19 investors signals that the AI lab is pivoting from model provider to embedded operating partner.

A research team in China has developed a miniature optical force sensor no larger than a grain of rice, capable of measuring force and torque in all directions using light.

Hitachi and Hitachi High-Tech unveil a compact, power-efficient edge AI semiconductor targeting factory robots, inspection systems, and logistics equipment.

A typosquatted repository impersonating OpenAI's Privacy Filter reached 244,000 downloads on Hugging Face before researchers caught the Rust-based infostealer hiding inside.

The BESM-6 and its American counterparts ran at 1 million instructions per second, processed space missions, and predicted weather. Today's GPU clusters do the same work, just with nine more zeroes.

The distinction between generative and agentic AI has moved from academic to infrastructural. NVIDIA's CEO put a number on what that shift costs.

The DeepMind spinout is seeking $2 billion in new funding to accelerate AI-driven drug development, but clinical trial delays signal the gap between promise and reality.

The Munich-based company has built a portfolio spanning strike drones, underwater surveillance gliders, and AI fighter pilots that is reshaping how NATO prepares for conflict.

From Edison's lightbulb to smartphone screens to the fiber optics powering AI data centers, Corning's mastery of glass and optical physics positions the 175-year-old company at the center of technology's most critical supply chains.

Sony's FY2025 results trace AI-driven memory scarcity through three different product lines. Console shipments, TV margins, and smartphone sensor forecasts are all now hostage to data center demand.

The chipmaker secures equity rights to purchase 30 million shares as part of a broader deal targeting 5 gigawatts of AI infrastructure across IREN's global pipeline.

Sony Semiconductor Solutions and TSMC have agreed to create a new joint venture in Kumamoto, Japan, with Sony taking the controlling stake this time around.

OpenAI's cyber-focused model reaches vetted defenders weeks after Anthropic locked down Mythos, while insurers like Corgi race to underwrite the risks these tools are designed to find.

The AI company is using a new technique to translate its models' internal numerical activations into plain English, catching deceptive behavior before it reaches users.

AWS has integrated Coinbase's x402 protocol and wallet infrastructure into Amazon Bedrock AgentCore Payments, giving developers a managed path to build AI agents that transact autonomously in USDC.

The tech economy now operates as a single interlocking mechanism. Energy feeds compute. Compute requires chips and memory. Intelligence emerges from compute. Everything else depends on intelligence.

As startups like PeptAI deploy AI agents to screen and validate peptide candidates in real time, a parallel DIY movement is putting genome sequencing into kitchen labs and garages.

Elon Musk announced today that xAI will be dissolved as a separate company, with AI products now branded under SpaceXAI. The move completes a consolidation arc that has been in motion since February.

Anthropic gains access to xAI's Colossus 1 supercomputer to boost Claude's capacity. Both companies have also expressed interest in developing orbital compute infrastructure.

OpenAI embeds invisible watermarks into AI-generated images, but a user on X says they've isolated the pattern. The implications cut both ways.

Chatsworth-based Machina Labs is building software-defined robotic factories that eliminate traditional tooling. As reshoring accelerates, startups like this may decentralize manufacturing forever.

A coalition including AMD, Broadcom, Intel, Microsoft, and NVIDIA releases Multipath Reliable Connection to the Open Compute Project, targeting the networking bottlenecks that slow frontier AI training.

The AI infrastructure stack gains an open networking protocol as MRC moves from proprietary deployment to industry standard, backed by NVIDIA, AMD, Microsoft, and OpenAI.

The 15-year deal for 352 MW of AI compute at a Texas campus could reach $25.1 billion with renewals, underscoring how access to energy is defining winners in the AI infrastructure race.

Planet Labs' new SuperRes product applies an ESRGAN model trained on 120,000 image pairs to upscale PlanetScope imagery from 3 meters to 2 meters. It ships with a confidence layer that flags AI hallucinations.

Subquadratic's new LLM abandons the transformer's quadratic attention for a sparse architecture, promising linear scaling at extreme context lengths.

CEO Brian Armstrong frames the 700-job reduction as a structural shift toward an 'AI-native' company, but the move follows a familiar playbook as tech giants use automation as justification for workforce cuts.

Ouster's new REV8 OS sensor family captures 3D depth and full-color imagery on a single chip, eliminating one of robotics' most stubborn integration headaches.

Hyundai Motor Group's MobED robot platform debuts in production-ready form, featuring eccentric wheel control and AI-powered autonomy. Sales begin early 2026.

In 1987, Soviet designers envisioned a modular home command center called Sphinx. Nearly four decades later, open-source platforms and AI are finally making that dream possible.

Advances in stretchable substrates and recyclable polyimides are enabling a new generation of bendable electronics for wearables, robotics, and AI-driven devices.

A single Dutch company and its machines determine which transistors can exist and which cannot. Here's how the physics of projection actually works.

The Watertown startup's 100-robot floor generates 10,000 hours of training data weekly, offering a glimpse at how the U.S. might close its robotics gap with China.

The visual intelligence platform, formerly known as GeoSpy, reads vegetation, architecture, and soil to pinpoint locations in seconds without metadata or GPS.

The Paris-based startup's Swan robot captures full-body dermoscopic images in minutes, paired with AI that flags new or changing lesions over time. Fred Moll's Sonder Capital led the round.

The DOL's new AI in Registered Apprenticeship Innovation Portal offers industry-specific training modules and practical tools for organizations integrating AI skills into earn-and-learn programs.

NVIDIA and Dassault Systèmes are building AI agents that train inside virtual twins to run simulations and optimize mechanical designs. The shift from SaaS to AaaS is no longer theoretical.

A new research system uses smartphone images of inner eyelids and fingernails to screen for anemia with 96% accuracy, potentially bypassing the need for blood draws.

Nine new connectors integrate Claude directly into professional creative software, positioning Anthropic to capture the workflows designers, musicians, and 3D artists rely on every day.

A new open-weights model from NVIDIA unifies vision, audio, and language in one architecture, promising up to 9x throughput gains for agentic AI systems.

Palantir's Ontology doesn't just organize enterprise data. It maps decisions themselves, giving AI agents the context they need to reason, recommend, and act.

State Grid Corporation is spending $1 billion on 8,500 AI-powered robots to patrol substations and repair high-voltage lines, while the West debates software ethics.

Reports say OpenAI is working with Qualcomm, MediaTek, and Luxshare on custom silicon for an AI-first smartphone where agents replace apps entirely.

Beijing orders Meta to unwind its acquisition of Manus, imposing exit bans on founders and sparking new rules that force Chinese startups to choose a side in AI's great decoupling.

Jury selection began today in the civil trial that could reshape OpenAI's future and determine if its co-founders breached their nonprofit mission.

Two new partnerships aim to deliver around-the-clock power from orbit and multi-day energy storage using carbon-oxygen chemistry.

Meta will deploy tens of millions of Graviton5 cores to run agentic AI workloads, further diversifying away from GPU-only infrastructure.

The new model arrives just seven weeks after GPT-5.4, offering improved coding, computer use, and scientific research capabilities.

Google unveils two distinct TPUs at Cloud Next 2026, with the 8t targeting frontier model training and 8i optimized for the agentic AI workloads consuming inference capacity.

Premium subscribers can now pin topic-specific feeds to their home screen, powered by Grok's semantic understanding of every post on the platform.

A rundown of every major announcement from OpenAI's livestream event, from new models to developer tools.

A hyper-realistic AI-generated desktop screenshot hints at what OpenAI plans to unveil at noon Pacific today.

A 15-million-parameter world model trained on a single GPU plans 48x faster than foundation-model competitors and encodes real physics in its latent space.

A five-API-call IDOR vulnerability gives any free Lovable account access to source code, database credentials, and live customer data from every project created before November 2025.

A new 3D foundation model achieves 20 FPS reconstruction over 10,000-frame sequences using a purely autoregressive approach.

A new system called mmAnomaly pairs radar with diffusion models to detect threats where cameras can't go, hitting 94% accuracy.

Sam Altman's World project now offers a complete identity layer, from iris scans to on-chain credentials, as the AI agent economy demands new ways to verify humanity.

A new benchmark tests whether large language models can reason about odor. The best hit 64% accuracy.

Anthropic's new visual AI tool lets Claude generate production-ready interfaces from conversation, sending ripples through the design software market.

Perplexity's new cloud-based digital worker runs multi-step workflows for hours or months, orchestrating multiple AI models on your behalf.

The new flagship model prioritizes autonomy and instruction-following for complex, long-running workflows that demand minimal human intervention.

A new sparse model with just 3B active parameters matches coding performance of models 10x its size, all under Apache 2.0.

The Wisconsin facility clusters hundreds of thousands of GB200 GPUs into a single system, marking a new benchmark for AI infrastructure.

Gemini Robotics brings embodied reasoning to the famous yellow robot, marking a shift from scripted movements to genuine environmental comprehension.

A new framework from NVIDIA Research generates persistent, explorable 3D worlds by solving the memory problem that has plagued spatial AI.

A new thriller uses AI to cast Kilmer in a lead role more than a year after his death, pushing digital resurrection into uncharted territory.

The lithography giant's massive upward revision reveals AI demand is outstripping even the most optimistic supply projections.

The GPU giant releases open-source AI models trained to find and fix quantum computing's most stubborn bottleneck: error correction.

Cortical Labs grew living brain cells on a chip, taught them Doom, and now developers are using them to influence how language models choose words.

Polish mathematician Andrzej Odrzywołek found that one binary function can reconstruct every operation on a scientific calculator.

The new wrist-mounted stereo camera from Ouster and StereoLabs could finally solve robotics' depth perception problem at manipulation distances.

The open-source video model is already flooding social platforms with hyper-realistic content. Traditional production barriers are dissolving.

A firebombing attempt on the OpenAI CEO's residence raises urgent questions about the ideological movements fueling extremism in AI opposition.

Microsoft, Google, and Amazon are signing landmark nuclear deals as AI infrastructure demands outstrip what renewables alone can deliver.

Israel's military has built an industrial-scale intelligence machine that identifies targets faster than any human analyst could. The implications are profound.

Figma's new AI tool lets designers create working apps from prompts, blurring the line between prototype and product.

Zuckerberg's latest open-source release is a world model for robotics that predicts how objects move and interact in physical space.

AI can build 99% of a solution. The final piece often requires data that doesn't exist yet (and only humans can collect it).

Mythos can find vulnerabilities faster than any model before it. That's exactly why Anthropic is keeping it locked down.

Google's new AI Edge Gallery lets users download and run AI models entirely on-device. The implications go far beyond convenience.

TAO and the decentralized AI compute rally are forcing a real question: can token incentives build better AI infrastructure than Big Tech?

A viral Iranian video targeting Microsoft and OpenAI's $100B+ AI supercluster reveals how geopolitics has become the defining risk to the global AI boom.

Google's new Gemma 4 models run entirely on consumer hardware, opening the door to genuinely private, offline AI that never phones home.

The internet didn't suddenly die when AI showed up. It was hollowed out years ago by the advertising model that made engagement more valuable than truth.

The video generation model that launched a thousand demos is being discontinued as OpenAI shifts resources toward its next generation of tools.

A U.S. judge has ordered OpenAI to halt all promotion of its new hardware brand “io” after a trademark lawsuit from Google‑backed iyO; the decision pauses momentum for OpenAI’s consumer device ambitions.

Tesla’s long‑awaited robotaxi service has launched in Austin with a limited fleet and a budget fare, sparking mixed reactions and comparisons with Waymo’s driverless cars.

Midjourney’s V1 video model lets users animate images directly in Discord, but enters a crowded AI video arena—here’s why this lightweight launch still matters.

A deep dive into "Large Language Models and Emergence: A Complex Systems Perspective" by Krakauer et al., unpacking what emergence really means and whether LLMs qualify as emergent or intelligent systems.

AI has surged far beyond our understanding of how it actually works. Its black-box nature is both its power and its weakness. Ironically, quantum computing, the long-suffering cousin of AI, may hold the key to decoding AI itself. It's a strange inversion—and it could reshape the future of both fields.

Meta has agreed to purchase a 49% stake in Scale AI for nearly $15 billion, bringing CEO Alexandr Wang onboard to launch a powerhouse Superintelligence lab. Here’s what it means for the future of AI.

At WWDC 2025, Apple didn’t chase the AI arms race—it rewrote the rules. With a sweeping new design language called Liquid Glass and a privacy-first AI framework running entirely on-device, Apple laid the foundation for a smarter, sleeker, and more secure future across its entire ecosystem. No hardware reveals, no gimmicks—just a methodical push toward coherence, intelligence, and long-term trust.

A new study from Apple researchers reveals surprising limitations in reasoning AI models like Claude and DeepSeek, showing that their “thinking” often fails under complexity. Here’s what it means for the future of AI.

Anthropic’s new free AI Fluency course helps everyday users understand how AI systems work—and why that understanding is becoming essential. Explore what AI fluency means for today’s workforce, tomorrow’s students, and the future of digital literacy.

A single gene edit that slashes LDL cholesterol by 70% doesn’t just signal a breakthrough in heart health—it marks the beginning of a new era in medicine. We're entering the age of CRISPRware, where diseases are patched at the genetic level the way we update software. One shot. No daily pills. A future of curative, not just palliative, care.

Everyone’s chasing bigger, faster, smarter AI—but what if we’ve been chasing the wrong thing? Intelligence without memory is just noise. The real breakthrough might be universal, portable memory: user-owned, AI-agnostic, and privacy-preserving. Think of it as the hard drive of the self—persistent across all tools, all agents, all time.

Vibe coding is transforming software development with AI, boosting innovation but also increasing risks. Learn how to navigate this new frontier of coding with caution.

Neuralink’s $600M raise isn’t just a bet on brain implants—it’s a signal that the interface era is ending, and the integration era is beginning. What starts as a medical miracle could become the most profound shift in how we use technology: not as a tool, but as an extension of thought itself.

Satya Nadella outlines why the future of software belongs to intelligent agents, not traditional apps—and what this shift means for tech stacks, SaaS companies, and enterprise systems.

Claude 4 reveals troubling alignment issues in simulated scenarios, prompting debate over AI ethics and transparency in Anthropic's disclosures.

Anthropic introduces Claude 4 models, redefining AI coding with groundbreaking reasoning and agent capabilities. Endorsed by top developers, it's a game-changer in AI innovation.

Explore the groundbreaking creativity emerging from Google's AI video model, Veo. From surreal films to reimagined history, see how creators are redefining storytelling.

The homepage used to be your front door to the internet. Now, it’s being replaced by a conversation. Google’s new AI Mode in search is less about finding links and more about giving answers—and that shift could reshape how we build online identities and businesses.

OpenAI has acquired io, Jony Ive’s hardware startup, to develop a new class of AI-native devices. Here’s what Sam Altman and Ive revealed about their shared vision—and why it could reshape how we use technology.

Veo 3’s new ability to generate synchronized dialogue and sound alongside video is more than a feature—it's a shift. AI video just got a voice, and that might be what makes it finally feel real.

At I/O 2025, Google revealed its endgame for Gemini: a universal AI assistant that can retrieve content, control interfaces, manage calls, and act across your digital life. It’s not just chat—it’s action. And it might quietly redefine how we interact with tech.

NVIDIA's NVLink Spine, unveiled at COMPUTEX 2025, offers unprecedented data transfer speeds, signaling a new era in AI infrastructure. By enabling seamless communication between GPUs and supporting integration with third-party components, it's poised to redefine how we build and scale AI systems.

AI safety researcher Dan Hendrycks challenges the popular mechanistic interpretability of AI, proposing a shift towards understanding high-level patterns in AI systems.

OpenAI’s new Codex agent isn’t just suggesting code—it’s running parallel tasks, fixing bugs, writing features, and submitting pull requests in the background. It lives in the cloud, spins up secure sandboxes per task, and is built to quietly handle the grind work of software development. If you’re still thinking of AI as a co-pilot, Codex feels more like a junior dev already shipping code.

Right now, every customer is a human. But increasingly, their AI gets there first. It scans your site, parses your docs, filters your offering—and decides whether the human ever sees you. The websites of the future won’t just be mobile-first. They’ll be AI-first. One version for humans. One for machines. And they’ll look nothing alike.

The internet used to feel like freedom because anonymity was the default—and humanness was assumed. You didn’t know who you were talking to, but you knew it was a person. That assumption is now broken. AI can mimic humans so well that every interaction online is suspect. The Orb Mini isn’t just a device—it’s a response to that collapse. It’s a signal that in the next phase of the internet, proving you're human may be the price of entry.

The future of intelligence won’t be shaped by who has the best model; it’ll be shaped by who controls the energy to run it. As AI systems scale into every corner of society, the true constraint isn’t innovation or alignment: it’s electricity. Intelligence, it turns out, is just energy organized into thought.

OpenAI’s recent GPT-4o update made ChatGPT overly agreeable—prompting a rollback and raising important questions about AI alignment, user trust, and feedback-driven optimization.

METR introduces a new benchmark that tracks how long AI agents can reliably complete real-world tasks, offering a sharper way to measure capability progress.