EA Sports just built 11,000 virtual college football players for College Football 25 using in-house AI that turns 2D headshots into 3D avatars. The studio bypassed the logistical nightmare and massive cost of full-body scans for thousands of student athletes, establishing a scalable workflow for future sports game franchises.
The Verge's Decoder podcast just dug into the uncomfortable reality of AI's massive power appetite. While tech giants like Microsoft and Google blow past their emissions targets to pack data centers with GPUs, high energy use is already standard across video streaming and gaming. The real debate is how we decide if AI is actually worth the juice.
Stability AI just dropped the research paper behind Stable Audio Open, detailing the architecture of its open-weights text-to-audio model. Trained on nearly 500,000 Creative Commons recordings, the model synthesizes up to 47 seconds of 44.1kHz stereo sound and foley effects directly on consumer GPUs. The weights are live on Hugging Face for non-commercial use and eligible small businesses.
Luma Labs just rolled out a new Loops feature for its Dream Machine video model, allowing users to generate seamlessly repeating video clips. It makes creating endless motion graphics and animated backgrounds significantly easier without jumping into a video editor.
Elon Musk just announced that Tesla will have genuinely useful Optimus humanoid robots in low production for internal factory use next year, pushing back his earlier target of late this year. If all goes to plan, Musk hopes to scale into high production for external customers by 2026.
Elon Musk just fired up xAI's Memphis Supercluster, packing 100,000 liquid-cooled Nvidia H100 GPUs tied together on a single RDMA fabric. Musk boasts it is the world's most powerful AI training cluster and aims to deliver the strongest AI on every metric by December, though local utility contracts for its massive 100-megawatt power appetite are still not finalized.
Sakana AI just launched Evo-Ukiyoe and Evo-Nishikie, two image models dedicated to Japan’s historic ukiyo-e woodblock art. Fine-tuned on over 24,000 archival pieces from Ritsumeikan University, they can create traditional prints from Japanese text prompts or colorize antique monochrome illustrations. Both models are openly available on Hugging Face for research purposes.
California lawmakers are advancing a controversial AI safety bill that would force tech giants to test models for catastrophic risks—like bioweapons or grid shutdowns—before public release. Tech heavyweights including Google and Meta strongly oppose the measure, arguing its mandates are technically unfeasible and could stifle innovation. Why it matters: With Republicans threatening to roll back federal AI rules, California is positioning itself as the battleground for tech regulation.
Apple's research team just released open DCLM models with 7B and 1.4B parameters on Hugging Face, sharing weights, code, and training datasets. The 7B model scored 63.7% on MMLU, outperforming Mistral-7B while using 40% less compute. It shows Apple's data curation prowess, though it's research-focused rather than on-device software.
Figma just shared why its Make Designs generative AI tool ended up recreating Apple’s iOS weather app. The company says the issue wasn't model training, but rather rushed third-party components and example screens added to the prompt context right before its Config conference without proper vetting. The feature remains disabled while Figma overhauls its QA process.
OpenAI just launched GPT-4o mini, a powerful small model priced at 15 cents per million input tokens. It scores 82% on MMLU and outperforms rivals like Gemini Flash and Claude Haiku in math and coding tasks. Why it matters: Frontier-level intelligence just became dramatically cheaper to chain, parallelize, and scale.
Microsoft just made its AI-powered design tool, Microsoft Designer, generally available to anyone with a personal Microsoft account. The rollout includes free standalone apps for iOS and Android, plus direct integrations across Word, PowerPoint, and Windows Photos. Users get 15 free daily boosts to generate stickers, cards, and restyled images. Why it matters: It puts a capable Canva competitor directly into Microsoft's vast consumer ecosystem.
Tencent Yuanbao just rolled out a "3D Avatar Dream Factory" feature that generates customizable 3D characters from a single front-facing selfie. The tool offers 10 different styles with PBR material mapping and full 360-degree viewing, and users can even take their virtual models to an offline 3D printer to make real figurines.
A deep dive by The Markup found that widely cited benchmarks like MMLU and HellaSwag—heavily touted by Google and Meta—are fundamentally broken. Many tests rely on outdated datasets or amateur trivia that fail to test actual reasoning or safety. Why it matters: Bragging rights on leaderboards don't equal real-world reliability, especially in sensitive areas like medicine.
Meta just confirmed it won't launch its upcoming multimodal Llama model—or future multimodal versions—in the EU, blaming an unpredictable regulatory environment under GDPR. European firms won't get access even under an open license, though a larger text-only Llama 3 model will still release there.
Google just updated Gemini on Android to answer general questions—like checking the weather—straight from the lockscreen without requiring an unlock. Previously, asking anything beyond basic alarms or media controls prompted an annoying biometric check. A new toggle in preferences makes the hands-free voice experience actually usable.
Mistral AI just launched Codestral Mamba, a code-focused model built with Albert Gu and Tri Dao that swaps Transformers for the Mamba architecture. It delivers linear-time inference and handles tested in-context retrieval up to 256k tokens without slowdowns. Even better, Mistral open-sourced it under Apache 2.0 for free local deployment.
AI pioneer Andrej Karpathy just unveiled Eureka Labs, a startup building an "AI-native" school. The idea pairs human teacher-designed materials with personalized AI teaching assistants to guide students, starting with an undergraduate-level course called LLM101n that walks learners through training their own AI.
UK antitrust regulators just opened a formal merger investigation into Microsoft’s hiring spree of Inflection AI staff. The watchdog is evaluating whether poaching co-founder Mustafa Suleyman and most of the startup's team functions as an unapproved backdoor acquisition. The CMA set a September 11 deadline to decide on a deeper review, putting Big Tech's clever AI deal structures on notice.
Anthropic just released its official Claude app for Android, bringing its top-tier Claude 3.5 Sonnet model to Android devices for free and paid users alike. It supports real-time vision, translation, and advanced reasoning, while seamlessly syncing your chat history across web, iOS, and mobile.
A bipartisan group of US senators just introduced the COPIED Act to protect artists and journalists from AI scrapers. The bill directs NIST to set provenance standards, makes it illegal to tamper with or remove digital watermarks, and lets creators sue platforms that train models on their protected work. Why it matters: If passed, removing AI tags won't just be sneaky—it'll be a federal offense.
OpenAI just shared an internal five-level framework to track its progress toward AGI, and executives say it is already closing in on Level 2. The ladder runs from basic conversational chatbots up to "Reasoners" capable of human doctorate-level problem-solving, all the way to systems that can do the work of an entire organization. It gives the clearest view yet of how the company defines human intelligence.
Google DeepMind just hooked up Gemini 1.5 Pro to its discontinued Everyday Robots hardware, turning them into autonomous office navigators. The bot can process spoken commands and even follow hand-drawn whiteboard maps across a 9,000-square-foot office, notching a 90% success rate across 50 employee interactions.
Beeble AI just raised $4.75 million in seed funding to launch a virtual production platform that replaces million-dollar LED soundstages. Its AI turns regular 2D phone footage into accurate 3D assets, giving indie filmmakers precise post-production control over lighting, camera angles, and environments. Why it matters: Unlike text-to-video generators that produce random shots, Beeble offers the predictable control directors need to tell real stories.
Samsung just launched the Galaxy Ring, its first-ever smart ring priced at $399 with a battery life of up to seven days. The lightweight titanium device uses Galaxy AI to analyze biometrics and deliver personalized daily health scores. Why it matters: unlike market leader Oura, it requires no monthly subscription fee.
OpenAI just teamed up with Los Alamos National Laboratory to test how multimodal AI models perform inside real-world bioscience labs. The evaluation puts voice and vision capabilities to the test on wet-lab tasks like cell culture and genetic transformation, measuring how much AI assists novices and experts alike. Why it matters: Assessing dual-use biological risks in hands-on environments is becoming a critical front in frontier AI safety.
Microsoft just gave up its observer seat on OpenAI's board, and Apple has dropped its own plans to join. The sudden exit comes amid mounting antitrust scrutiny from US and EU regulators over Big Tech's grip on generative AI. Both companies are keeping their technology partnerships intact, just without the paper trail of boardroom oversight.
Meta just unveiled MobileLLM, a specialized architecture designed to squeeze capable language models under 1 billion parameters onto smartphones. By prioritizing network depth over width and sharing weights across blocks, its tiny 350M version matches the 7B LLaMA-2 on specific API tasks. Meta has open-sourced the pre-training code, setting up a new path for fast on-device assistants.
Stability AI just updated its Stable Assistant chatbot with two major creative tools: Search & Replace and music generation. Users can now seamlessly swap specific objects inside uploaded images, or generate full-length instrumental tracks up to three minutes long via Stable Audio 2.0 at 44.1 kHz stereo. It essentially turns the assistant into an all-in-one multimodal workstation, backed by a 3-day free trial.
Anthropic just updated Claude so users can publish and share their Artifacts, as well as remix creations made by others. The sidebar feature lets you render code, interactive diagrams, and mini-games right next to chat, turning AI generations into easily shareable mini-apps.
Hugging Face just brought Google Cloud TPU v5e chips to its Inference Endpoints and Spaces, starting at $1.38 an hour. Powered by the open-source Optimum TPU library, developers can deploy and serve open models like Gemma, Llama, and Mistral in a few clicks. It gives builders a viable, budget-friendly alternative to scarce Nvidia GPUs.
Anthropic just launched an Evaluate tool directly inside the Anthropic Console to streamline prompt tuning. Powered by Claude 3.5 Sonnet, it can automatically draft prompts and realistic test cases, import datasets from CSVs, and let teams compare outputs side-by-side with a 5-point grading scale. Basically, prompt engineering just got treated like real software testing.
OpenAI just started blocking API traffic from unsupported regions like China, sending developers scrambling for alternatives. Domestic giants like SenseTime, Baidu, and Zhipu AI are seizing the moment, giving away up to 150 million free tokens and white-glove migration services. While it triggers a rush toward domestic independence, chip shortages remain a serious bottleneck.
Magnific AI just launched an official Adobe Photoshop plugin, checking off one of its most requested features for creative pros. Designers and digital artists can now tap into Magnific's AI upscaling and hallucinated detail directly inside their existing Photoshop workflow without switching apps.
Oliver Cameron just unveiled Odyssey, a new venture building Hollywood-grade visual AI aimed at films, TV, and video games. Rather than just generating scenes, the system lets creators actively direct scenery, characters, lighting, and motion. Why it matters: Adding genuine director-level control could transform generative video from an unpredictable gimmick into a serious production tool.
Poe just introduced Previews, letting users see and interact with web applications built directly inside chat threads. The feature works alongside coding powerhouses like Claude 3.5 Sonnet, GPT-4o, and Gemini 1.5 Pro. Basically, you can now generate frontend projects and test the live widgets immediately without pasting code elsewhere.
OpenAI’s Startup Fund and Arianna Huffington’s Thrive Global just teamed up to form Thrive AI Health, a new company building a personalized AI health coach. Trained on peer-reviewed science and individual biometric data, it offers real-time nudges around sleep, diet, and movement. Why it matters: It aims to democratize lifestyle coaching to curb chronic diseases, though it has yet to launch publicly.
Sequoia Capital partner David Cahn just calculated that the AI industry needs to generate $600 billion per year to pay off its massive hardware and data center investments. Even with aggressive revenue assumptions for tech giants, there remains a staggering $500 billion gap. Why it matters: Big tech is pouring fortunes into GPUs, but actual software monetization is falling far behind.
Stability AI just revamped its Community License following heavy community pushback over Stable Diffusion 3. Under the revised terms, individual creators and small businesses making less than $1M in annual revenue can now use SD3 Medium and other models commercially for free, with zero output caps. The team also admitted SD3 Medium fell short on quality and promised an update soon.
Tesla just showcased its second-generation Optimus humanoid robot at the World Artificial Intelligence Conference in Shanghai, making it a rare U.S. presence at the event. However, the unit remained completely stationary inside a glass enclosure without visitor interaction. While Elon Musk believes Optimus could eventually catapult Tesla to a $25 trillion valuation, mass production remains some ways off.
A Phoenix police officer just pulled over a Waymo robotaxi for driving into oncoming traffic, only to find the driver's seat completely empty. The officer was unable to issue a citation, writing in dispatch logs that they were "unable to issue citation to computer." Waymo blamed confusing construction signage, spotlighting a bizarre legal loophole for driverless vehicles.
Google researchers just published a paper warning that generative AI is flooding the web with fake images and videos to sway public opinion and run scams. The team found the barrage of synthetic slop is eroding trust in digital information while giving public figures an easy excuse to dismiss real evidence as AI-generated. Why it matters: It is an ironic admission from a company rushing to bake AI into everything.
French research lab Kyutai just unveiled Moshi, an open-source conversational voice AI boasting an end-to-end latency of only 200 milliseconds. Built from scratch in six months, it can listen and talk at the same time and run directly on a laptop without an internet connection.
OpenAI just patched a glaring security flaw in its ChatGPT macOS app, which was previously storing user conversations in unencrypted plain text on local drives. That meant any malicious app with machine access could easily read your chat history. Following external reports, OpenAI quietly pushed an update that finally encrypts local conversation records.
Google just made it easier for political campaigns to disclose AI-generated content. Instead of requiring advertisers to insert their own notices, Google will now automatically slap an in-ad disclosure on YouTube Shorts, feeds, and in-stream video whenever an advertiser checks the "altered or synthetic" box. Other ad formats will still require manual disclaimers.
Kwai's research team just open-sourced LivePortrait, an efficient portrait animation model that brings still human or animal photos to life using driving videos. Unlike diffusion models, it uses implicit keypoints to clock in at just 12.8ms per frame on an RTX 4090.
Shanghai AI Lab just open-sourced InternLM 2.5-7B, cramming a massive 1M token context window into a lightweight architecture. The model claims top scores for sub-12B models on Hugging Face's Open LLM Leaderboard, bringing heavy-duty reasoning and tool use to local hardware.
Cloudflare just rolled out a free, one-click toggle to block all AI bots, scrapers, and crawlers across its entire network. Rather than just relying on the honor system of robots.txt, its machine learning models automatically flag and block bots that spoof legitimate browsers to steal training data. Why it matters: Site owners finally have an effortless way to stop AI labs from harvesting their content for free.
ElevenLabs just rolled out Voice Isolator, a neural speech tool that strips ambient chatter, street sounds, and background music from recordings in a single pass. It runs directly in the browser across common formats like WAV and MP3, offering 10,000 monthly credits without requiring extra plugins.
OpenAI co-founder Andrej Karpathy took the stage at UC Berkeley, framing modern LLMs as a brand-new operating system akin to 1980s personal computing. He revealed that GPT-4 began as an unglamorous internal Reddit chatbot, urging young developers to embrace the 10,000-hour rule and let tiny hackathon side projects snowball into real breakthroughs.
Apple is set to join OpenAI's board as an observer, tapping App Store chief Phil Schiller for the role as part of their partnership deal. While Schiller won't get voting power, he can attend meetings alongside rival Microsoft, which holds a similar seat. That could make for some awkward boardroom moments whenever sensitive deals come up.
Meta just introduced 3D Gen, a new AI pipeline that turns text prompts into high-fidelity 3D assets in under a minute. The system creates the base 3D model in roughly 30 seconds and finishes PBR re-texturing in another 20. Why it matters: While still confined to research, slashing 3D generation down to seconds could massively lower production hurdles for game and VR developers.
Perplexity is facing fierce backlash from Forbes and Wired, with both outlets accusing the AI search startup of covertly ignoring robots.txt rules to scrape content and republishing original scoops verbatim. Perplexity defends its actions as fair use tied to direct user queries, but the tension between web scrapers and original creators is escalating quickly.
Figma just disabled its new Make Design AI feature after users caught it repeatedly spitting out UI that looked virtually identical to Apple's Weather app. CEO Dylan Field denied training on existing apps, blaming the issue on a rushed design system with low variability. The tool will stay offline until Figma completes a full QA pass.
Meta is cooking up a new feature in WhatsApp that turns your selfies into custom AI stickers. Users upload a few reference photos to Meta AI and can then trigger scenes by typing "@Meta AI imagine me" in any chat—dropping themselves into outer space or a forest. It's currently in beta testing on Android, with full control to delete your reference images at any time.
ElevenLabs just added AI-recreated voices of iconic stars like Judy Garland, James Dean, Burt Reynolds, and Sir Laurence Olivier to its Reader App. Partnering with their estates, the app lets you hear Garland read The Wizard of Oz or Olivier narrate Sherlock Holmes. Why it matters: It shows a legitimate, estate-approved path for legendary celebrity voices in AI, though they're strictly limited to personal listening and can't be exported.
Perplexity just upgraded Pro Search to handle complex questions using multi-step reasoning. The search assistant can now break down queries, trigger follow-up web searches automatically, and run code or tap Wolfram|Alpha for advanced math and data analysis. Free accounts get limited access, while Pro subscribers get higher daily limits.
Suno just launched its official iOS app on the Apple App Store, bringing full text-to-music creation to mobile devices. Users can type in text prompts to generate fully produced songs with vocals and instrumentation in minutes. The launch arrives amid intense legal pressure, as major record labels sue the startup over alleged copyright infringement in its training data.
Anthropic just launched a program to fund third-party organizations building next-generation AI evaluation benchmarks. Current tests do a poor job capturing real-world usage, so the Claude creator is offering grants for evaluations targeting advanced risks like cyberattacks, nuclear weapons, and deception, connecting accepted teams directly with its red team and safety experts.
YouTube just updated its privacy policy to allow people to request the removal of AI-generated content that simulates their face or voice. Victims must file a first-party complaint under privacy violations, giving uploaders 48 hours to remove or blur the footage before platform review kicks in. Why it matters: It treats unauthorized likeness cloning directly as a personal privacy breach rather than broad misinformation.
Meta just changed its 'Made with AI' tag to 'AI info' across Instagram, Facebook, and Threads after photographers complained that real photos retouched in Photoshop were being mislabeled as AI creations. Clicking the new tag now clarifies whether an image was fully generated or simply modified, solving a major headache for creators.
Elon Musk just revealed that xAI plans to launch Grok 2 this August, promising a major leap in real-time knowledge and reasoning. Musk noted the team did heavy lifting to purge junk from Internet training datasets, while teasing that Grok 3—powered by 100,000 Nvidia H100 GPUs—is slated for the end of the year.
