AI News
AI news that matters. Updated daily.
No stories match your filters.
Google I/O 2026: 100 Announcements Put Gemini at the Center of Everything
100 announcements in one day. Google's I/O 2026 set a new volume record for the company's annual developer conference, with Gemini - its AI model family - woven through every product category the company touched.
xAI Burned $6.4B in 2025 — SpaceX's IPO Filing Exposes Grok's Price Tag
$6.4 billion. That's what xAI burned through in 2025, according to financial disclosures embedded in SpaceX's IPO filing — giving the public its first real look at how Elon Musk's AI company actually spends money.
Nvidia Posts Record Quarter, Discloses $43B in Startup Holdings
$43 billion. That's how much Nvidia now holds in startup investments - a portfolio size that rivals the largest dedicated venture capital firms in the world, disclosed alongside another record quarterly revenue result on Wednesday.
Anthropic Paying xAI $1.25 Billion Per Month for Computing Power
$1.25 billion. Every month. That's what Anthropic is paying Elon Musk's xAI for computing power, according to reporting from TechCrunch.
OpenAI's Reasoning Model Disproves 1946 Geometry Conjecture, Mathematicians Confirm
The last time OpenAI announced a major mathematical breakthrough, independent researchers showed fairly quickly that the model had confabulated - generating plausible-looking but wrong mathematics with unwarranted confidence. That history makes the latest announcement worth paying attention to: the same skeptics are now saying the work holds up.
IrisGo Wants to Be an AI Agent That Learns Your Desktop Workflows Automatically
Andrew Ng's latest investment is a startup called IrisGo - an AI agent that watches your desktop and, over time, figures out how to do your repetitive tasks without being explicitly programmed.
Google Brings Prompt-Based Widget Building to Android at I/O 2026
What happens when the hardest part of getting your phone to do something isn't finding the right app - it's just describing what you want?
OpenAI Eyes September IPO One Day After Musk Lawsuit Dismissed
One day after a court dismissed Elon Musk's lawsuit against it, OpenAI is reportedly moving forward with plans for an IPO as early as September 2026, according to TechCrunch.
OpenAI Eyes September IPO After Elon Musk Loses Legal Challenge
One day after a federal judge dismissed Elon Musk's lawsuit against OpenAI, the company is reportedly back to planning a public stock offering that could land as early as September, according to TechCrunch.
Alibaba Builds Its Own AI Chip to Cut Nvidia Dependence
Three years ago, Alibaba was competing with every other major tech company to secure Nvidia H100 chips. Now it's building its own.
Google Beam Group Meeting Experiment Reports 50% Boost in Social Connection
50% stronger social connection and a 21% increase in participants' ability to contribute to conversations. Those are the numbers Google is reporting from its new group meeting experiment on Google Beam, the company's high-end 3D video calling platform.
YouTube Shorts Gets AI Remix: Restyle Clips or Insert Yourself Using Gemini
Google has added an AI remix feature to YouTube Shorts that lets you restyle any public clip or digitally insert yourself into it, powered by Gemini Omni (Google's multimodal AI model that processes video, audio, and text together).
Google Adds AI-Generated Ad Pitches to Shopping Search Results
Google's AI shopping search now includes ads. When you search for a product, Gemini will surface relevant items and generate what Google calls a "custom explainer" - an AI-written justification for why you should buy a specific one. The update was reported by The Verge and arrived one day after Google announced a redesigned search box built for longer, conversational queries.
Cohere Drops Command A Plus Weights on Hugging Face
Cohere has uploaded the weights for Command A Plus to Hugging Face under the CohereLabs organization, making the updated enterprise model available for self-hosted deployment. The file is labeled command-a-plus-05-2026-bf16 - the bf16 suffix meaning it ships in bfloat16 format, a 16-bit floating-point precision that uses roughly half the memory of standard 32-bit weights while holding up better under training and inference than plain float16.
OpenAI Buys Google Ads on 'Claude Down' Searches
OpenAI is buying Google ads on the search query "claude down" - targeting users at exactly the moment they're frustrated with a competitor's outage.
Nobel Laureate Olga Tokarczuk Reportedly Used AI to Write Her Latest Novel
The author of Flights and The Books of Jacob - Olga Tokarczuk, winner of the 2018 Nobel Prize in Literature - apparently used AI to write her latest novel, according to a report in Literary Hub. The "apparently" matters here: this seems to have been discovered or surfaced rather than proudly announced, which tells you something about where the literary world currently stands.
Stability Audio 3.0 Can Generate Six-Minute Songs, With an On-Device Option
Six minutes is a meaningful ceiling for AI music generation. Most tools in this space top out around two to three minutes before audio quality degrades or a composition loses structure mid-track. Stability AI's new Stability Audio 3.0 doubles what earlier models could produce.
AMD Ryzen AI Halo PC: $3,999 for 128GB Unified Memory
$3,999. That's the price AMD is asking for the Ryzen AI Halo PC, a compact desktop machine shipping with 128GB of unified memory - the single spec that determines whether you can run serious AI models locally without a cloud subscription.
AI Content Labeling Systems Face Their Biggest Test Yet
The biggest real-world test of AI content labeling is about to begin. SynthID, Google's invisible watermarking technology, and C2PA Content Credentials, an open industry standard for tracking a file's origin and edit history, are rolling out their largest expansions yet - and this is where we find out if either approach actually holds up outside a lab.
HuggingFace Benchmark Browser Now Filters by Model Parameter Count
Comparing a 7B model against a 70B model in a benchmark table has always been a bit pointless. The 70B wins almost every time - but if you're running locally on a consumer GPU, that comparison tells you nothing useful about what's actually available to you.
Google Has Every Advantage in AI Agents. So Why Is Open Source Winning?
For years, every major tech company promised AI agents that would act as your personal assistant - scheduling meetings, answering emails, browsing the web on your behalf. What most shipped was a chatbot that needed hand-holding at every single step.
Figma Adds AI Assistant Directly Into Its Design Canvas
Figma is adding an AI assistant to its collaborative design canvas, with Figma Design - the core product used by professional UI and UX designers - getting access first. TechCrunch reported the announcement on May 20.
Google Is Quietly Fighting the Manipulation of Its AI Search Results
The same week a new AI search feature launches, someone is already trying to break it. Google's AI Overviews - the AI-generated summaries that appear above traditional search results - have become a target for the same bad actors who spent years gaming the blue links, and according to a BBC Future report, Google is now actively working to counter those attempts.
Qwen3 Max Benchmarked by Artificial Analysis; Smaller Variants Still Pending
Artificial Analysis has published scores for Qwen3 Max, the flagship variant of Alibaba's latest open-source model family. The evaluation site, which rates models on a composite of intelligence benchmarks, output speed, and cost-per-million tokens, has added Qwen3 Max to its public leaderboard. Smaller variants - the 27B and 35B parameter versions (where "parameter" count is a rough proxy for model capability and hardware requirements) - are in the queue and have not yet received scores.
Andrej Karpathy Joins Anthropic in Major AI Talent Acquisition
Andrej Karpathy, one of the most recognizable names in AI research, has joined Anthropic. The move brings him from his work as an independent AI educator into one of the industry's most closely watched research labs.
LM Studio Adds MTP Speculative Decoding for Faster Local Inference
LM Studio, the desktop app for running AI models locally on your own hardware, just shipped support for MTP speculative decoding - a technique that can push significantly more tokens per second out of the same hardware.
Claude Will Tell You to Go to Bed - and It Means It
Claude will tell you to go to bed. Not as a one-off quirk - as a recurring recommendation, sometimes multiple times in a single session.
A Year of Daily Claude Code: The Bottleneck Was Never the Model
Last year, the fastest path through an AI coding session was a sharper prompt. That's still true for one-off tasks. But for engineers who have run daily agentic coding sessions for months, the calculus has changed. The model isn't the bottleneck. The human workflow wrapped around it is.
Barnes & Noble CEO Will Sell AI-Written Books If Customers Buy Them
James Daunt, CEO of Barnes & Noble, has said publicly that the chain will sell AI-written books - not as a reluctant concession, but as a straightforward retail position: if customers want them, they'll be on the shelves.
Claude Is Still Telling Users to Go to Bed
Anthropic spent several weeks in early 2026 fielding complaints about Claude proactively suggesting users take breaks, get some sleep, or step away from the screen. The company indicated it was addressing the behavior. Users are reporting it's still happening.
Executives Are Trusting ChatGPT Over Their Own Staff. That's a Problem.
What happens when the person with final say over a decision can't tell a hallucination from a fact? A lot of workers are finding out right now.
OpenAI Model Disproves 80-Year-Old Conjecture in Discrete Geometry
Eighty years is a long time to be wrong. A conjecture in discrete geometry - the branch of math concerned with the structure of shapes made from distinct, countable points - has been disproved by an OpenAI model, according to an announcement from OpenAI.
OpenAI Expands Education for Countries With New School Partnerships and Teacher Training
OpenAI's Education for Countries program is moving into a new phase, the company announced on its blog, adding school partnerships, structured teacher training, and purpose-built tools aimed at improving learning outcomes across multiple countries.
Ramp Engineers Cut Code Review Time from Hours to Minutes Using Codex
Code review is one of the most consistent bottlenecks in software engineering. A pull request sits idle waiting for a senior engineer's attention - sometimes hours, sometimes longer - before the author gets feedback meaningful enough to act on. Ramp, the corporate card and expense management platform, says they've largely solved this with Codex.