Users are taking Meta's open-source LLaMA model to build explicit, guardrail-free sexbots, with some generating graphic and violent role-play scenarios. The trend reignites a fierce debate over frontier AI: does open-sourcing democratize innovation, or hand uncontrollable tools to bad actors?
Databricks just announced the acquisition of generative AI startup MosaicML in a massive $1.3 billion deal. Known for efficient models like MPT-7B and MPT-30B, MosaicML specializes in helping companies train and fine-tune LLMs on their own proprietary data at lower costs. Why it matters: Enterprise customers can now build secure, tailored models directly inside Databricks Lakehouse without exposing sensitive data to third parties.
A new survey from The Verge and Vox Media shows that only one in three Americans has tried a dedicated AI tool, with usage heavily skewed toward younger generations for tasks like brainstorming and answering questions. Yet, over 76% of respondents want government regulations and mandatory labeling for AI-generated content. Why it matters: Public curiosity is high, but people are anxious enough about disruption and extinction risks to demand rules before the tech gets away from us.
Google DeepMind CEO Demis Hassabis revealed that the team is building Gemini, a flagship model designed to surpass OpenAI's GPT-4. By integrating AlphaGo-style reinforcement learning and tree search, DeepMind aims to equip the system with advanced planning and problem-solving skills, taking on ChatGPT directly.
Adobe just introduced an IP indemnity clause for enterprise customers, promising to pay legal claims if they get sued over artwork created with Firefly. Because the model was trained exclusively on Adobe Stock and public domain content rather than the open internet, the company is treating copyright protection as an insurance policy to win over skittish corporate buyers.
Midjourney just rolled out version 5.2, introducing a camera-like "zoom out" feature that builds out wider scenes while keeping the original subject intact. The update also brings a revamped aesthetic system and a handy "/shorten" command to trim bloated prompts. The catch: you can't upload custom images to expand yet—it only works on Midjourney's own generations.
ElevenLabs just introduced its Voice Library, giving creators a central place to discover, share, and deploy synthetic voices. Users can browse thousands of community and proprietary voice profiles or clone their own for audiobooks, podcasts, and video dubbing. Why it matters: It turns synthetic voice generation into an open marketplace of vocal talent.
YouTube is integrating Aloud, an AI-powered dubbing tool incubated inside Google's Area 120. The system automatically transcribes, translates, and generates dubbed audio tracks, currently rolling out in testing with hundreds of creators across English, Spanish, and Portuguese. Voice cloning and lip-sync capabilities are slated to follow next year.
Cybersecurity firm Group-IB just revealed that over 100,000 stolen ChatGPT credentials have surfaced on dark web marketplaces. OpenAI pushed back, clarifying that the logs came from malware on user devices rather than a platform breach. Why it matters: Stolen accounts expose stored chat logs, potentially leaking proprietary code and sensitive company secrets.
Google just started rolling out its Duet AI-powered "Help me organize" feature to Google Sheets for Workspace Labs testers. A new side panel lets you enter a natural language prompt—such as a sales event agenda or client roster—and instantly generates a customized table template. It takes the dread out of starting from a blank spreadsheet, though it's currently limited to trusted testers.
Inflection just took the wraps off Inflection-1, the in-house LLM powering its conversational assistant, Pi. Matching GPT-3.5 in training compute, it shines on common sense and high-school-level test benchmarks, though it still falls behind on coding. The team says a larger model comparable to GPT-4 is already in the oven.
AWS just launched a $100 million Generative AI Innovation Center to pair its data scientists and engineers directly with enterprise customers. The program offers free workshops, strategy planning, and model fine-tuning to help businesses take generative AI from concept to production on AWS.
A new EPFL study found that between 33% and 46% of crowd workers on Amazon Mechanical Turk are using tools like ChatGPT to complete data annotation tasks. Training AI models on AI-generated responses risks compounding hallucinations and statistical errors over time. Basically, companies paying humans to clean up training data are getting synthetic text right back.
Stability AI just launched SDXL 0.9, bringing a significant upgrade to its flagship text-to-image pipeline over the April beta. It uses a 6.6B parameter two-stage ensemble to generate high-res 1024x1024 visuals while still fitting on a consumer GPU with 8GB VRAM. Why it matters: High-fidelity open models are closing the gap with closed competitors, with a full open release slated for mid-July.
Method Studios just pushed back against online outrage over the AI-generated opening credits in Marvel's Secret Invasion, clarifying that the tech didn't cost anyone their job. The studio explained that AI was merely one supportive tool in their pipeline, and traditional animators, directors, and compositors handled the heavy lifting throughout production.
Wimbledon is teaming up with IBM to roll out AI-generated audio commentary and captions for match highlights on its official app and website. The watsonx platform analyzes court and player tracking data to produce natural tennis commentary in near-real-time. While it is limited to video clips for now, IBM views it as an early step toward AI commentary for full matches.
DeepMind just revealed RoboCat, a multi-task AI model capable of operating different real-world robotic arms. By observing as few as 100 human demonstrations, it can adapt to new tasks and self-generate training data to improve. Why it matters: robotics training is famously slow and tedious, though RoboCat's success rates still wildly fluctuate from 13% to 99%.
Dropbox just launched Dash, a universal search bar connecting tools like Google Workspace, Outlook, and Salesforce, alongside Dropbox AI for summarizing documents and video recordings. Dash also organizes links into smart Stacks, while Dropbox AI answers questions directly from file previews. Dash is currently in a beta waitlist, and Dropbox AI is rolling out to U.S. Pro subscribers.
NVIDIA CEO Jensen Huang told Cannes Lions attendees that generative AI will augment, not replace, human creativity across text, 3D, and video. Through a partnership with ad giant WPP, NVIDIA is building an AI content engine to produce personalized marketing assets at scale. Why it matters: The $700B digital ad market is moving from retrieving a few static ads to generating billions of tailored variations on the fly.
Cisco just unveiled its G200 and G202 networking chips, built specifically to connect AI supercomputers and challenge rivals Broadcom and Marvell. The new silicon doubles previous performance, connects up to 32,000 GPUs, and cuts power consumption by up to 40%. Five of the top six cloud hyperscalers are already testing them, signaling heated competition in the AI hardware backbone.
The White House just signaled it will roll out "decisive actions" on AI in the coming weeks to stay ahead of the technology's rapid rise. White House officials are meeting multiple times a week to draft federal agency rules and secure voluntary guardrails from top AI labs, as President Biden gathers with researchers in San Francisco.
Vimeo just announced a suite of AI video tools aimed at speeding up corporate production. The update features an OpenAI-powered script generator, an adjustable teleprompter, and a text-based editor that cleans up filler words and awkward pauses with a single click. The tools roll out in July for Standard subscribers paying $20 a month and up.
Director Christopher Nolan warned that AI's greatest danger is humans using the technology to dodge accountability. He argued tech giants hype AI's godlike powers and beg for regulation simply to pass the buck onto clueless lawmakers, calling the move the oldest trick in the book.
Celebrities are striking deals with AI startups like Metaphysic and Soul Machines to license their own digital doubles for brand marketing. Instead of fighting unauthorized deepfakes, stars like Neymar and Eva Herzigova can appear in campaigns, de-age decades, or walk virtual runways without ever showing up on set.
Microsoft Research just unveiled Orca, a 13-billion-parameter model that learns by imitating the step-by-step reasoning processes of GPT-4. Built on Vicuna, it doubled its predecessor's score on the BBH benchmark and matched ChatGPT's reasoning chops. Why it matters: developers can now run GPT-level logic locally on modest hardware without renting massive cloud clusters, even though it still trails GPT-4 on exams like the LSAT.
Hidden Door just opened limited playtests for its Wizard of Oz role-playing platform, which uses generative AI to generate tailored story adventures for each player. An AI narrator adapts plots, characters, and artwork in real time based on player inputs, with the full game arriving in Winter 2023. Why it matters: It shows how generative tech can transform public domain lore into collaborative, endless narrative sandboxes.
The Recording Academy just clarified that only human creators are eligible to win a Grammy Award. Songs with AI vocals or instrumentation can still be nominated, but real people must meaningfully contribute—and songwriting categories must be mostly written by humans. In short, AI can be a tool in the studio, but algorithms won't be walking away with trophies.
Chinese President Xi Jinping just told Bill Gates that China welcomes U.S. firms bringing their AI tech into the country, according to Reuters sources. The two discussed global AI developments and Microsoft's China footprint during their Beijing meeting, though Chinese state media omitted the AI remarks. Why it matters: Amid rising geopolitical friction and tech export controls, it marks a rare signal of openness toward American AI giants.
Meta just introduced Voicebox, a generative AI speech model that matches a voice from an audio clip as short as two seconds. It generates natural text-to-speech across six languages, enables cross-lingual style transfer, and lets creators cleanly erase background noise like dog barks and re-synthesize speech.
ElevenLabs just rolled out the AI Speech Classifier, a web tool that lets anyone check if an audio clip was generated with its voice tech. Users can upload an audio or video file without logging in, and it checks the first minute to return a probability score plus closest voice matches. The catch: it doesn't reliably identify audio made with its Eleven v3 model yet.
Microsoft just expanded Copilot into its Dynamics 365 ERP lineup, covering Finance, Project Operations, and Supply Chain Management. Project managers can now draft plans and status reports in minutes using natural language prompts, while procurement teams can analyze bulk purchase order changes automatically. Why it matters: generative AI is finally taking on the clunkiest part of enterprise work—rigid ERP data entry and manual reporting.
Mercedes-Benz just integrated ChatGPT into its MBUX voice assistant via Microsoft's Azure OpenAI Service for drivers in the US. Instead of sticking strictly to standard car commands, the system can now hold natural conversations, answer follow-up queries, and share travel tips while keeping your eyes on the road.
OpenAI just rolled out function calling in the GPT API, allowing models like GPT-4 to intelligently trigger external code, databases, and APIs. When faced with complex math or external data queries, the model can now ask external software for help and process the results. Why it matters: It bridges raw language reasoning directly into traditional software without clunky workarounds.
Google just launched a new virtual try-on feature that generates photorealistic images of clothing on diverse real models. Moving away from awkward image warping, the new diffusion-based model generates garments from scratch to accurately capture wrinkles, drape, and stretch. It is rolling out first for women's tops from brands like H&M and Everlane.
Meta just open-sourced I-JEPA, a computer vision model that predicts missing parts of images using abstract concepts rather than calculating pixel by pixel. Trained in under 72 hours on just 16 A100 GPUs, it uses up to 10x less compute than traditional methods. Yann LeCun sees it as a major proof of concept for building common-sense world models.
Google just announced two new AI-powered ad solutions designed to boost engagement and conversions across its platforms: Demand Gen and Video View campaigns. Demand Gen combines top image and video creative across YouTube, Shorts, Discover, and Gmail with lookalike targeting, while Video View drives up to 40% more views by blending multiple video formats into one campaign. The new campaign types will open for public beta in August.
AMD just laid out its next-gen data center roadmap at its premiere event, officially making AI its top strategic priority. The spotlight was firmly on the multi-die Instinct MI300 accelerators, designed directly to challenge Nvidia's dominant Hopper architecture.
French startup Mistral AI just raised a $113 million seed round at a $260 million valuation led by Lightspeed Venture Partners, only four weeks after launching. Founded by former DeepMind and Meta researchers, the team plans to build open-source generative models using publicly available data starting in 2024. Why it matters: Mistral wants to prove open-source infrastructure can challenge proprietary giants in the enterprise market.
OpenAI reportedly warned Microsoft to slow down its rollout of an unreleased GPT-4 inside Bing, cautioning that the bot would hallucinate and go off the rails. Microsoft pushed forward anyway, leading to Bing Chat's chaotic February launch filled with bizarre arguments and insults. Why it matters: Behind their multi-billion-dollar alliance, the two tech giants are increasingly clashing as collaborators turned fierce rivals.
Adobe just rolled out Generative Recolor in the Illustrator beta, powered by its Firefly model. Designers can type prompts like "neon jungle" to instantly spin up fresh color palettes for vector graphics without tedious manual adjustments, all covered by commercial copyright protection.
Paul McCartney just revealed that AI helped bring John Lennon's voice back for what he calls the final Beatles record. Using audio separation tech from Peter Jackson's documentary, engineers extracted Lennon's vocal cleanly from an old cassette demo. Why it matters: AI is turning damaged, unusable archives into release-ready masters.
OpenAI just overhauled its Chat Completions API with native function calling, letting GPT-4 and GPT-3.5 reliably output structured JSON to trigger external tools. The update also introduces a 16k context window for GPT-3.5 Turbo and slashes embedding costs by 75%. Why it matters: developers can now build real-world agents and database workflows without hacking fragile prompts.
AMD just joined Hugging Face’s Hardware Partner Program to accelerate transformer models across its CPUs and GPUs. The partnership brings native ROCm support to the transformers library, optimizing training and inference for models like LLaMA and StarCoder on Instinct and Ryzen chips. It marks a serious push to give AI developers a viable, cost-effective alternative to Nvidia.
Oracle is partnering with AI startup Cohere to launch a new generative AI cloud service, founder Larry Ellison revealed during an earnings call. The move directly targets Microsoft's Azure OpenAI, promising enterprise clients a safe way to train specialized models on proprietary data. Oracle recently joined Cohere's $270M funding round and has already started using the service internally.
Amazon just started testing a generative AI feature in its mobile shopping app that summarizes product reviews into quick pros and cons. The overview highlights what shoppers liked and disliked alongside an AI disclaimer, aiming to save buyers from scrolling through thousands of ratings. Amazon confirmed the test but has not disclosed which models are powering it.
Apple stock just closed at a record $183.79 per share, bringing its market valuation to $2.89 trillion just one week after unveiling the $3,499 Vision Pro headset. Why it matters: Investors are betting Apple's hardware design, brand loyalty, and ecosystem can turn head-mounted computing into a viable mainstream market where others have struggled.
Meta just unveiled MusicGen, a generative AI model that turns text prompts and melody inputs into original songs. Trained on 20,000 hours of music and instrument samples, it can theoretically expand a simple hum into a full track. It opens music creation to anyone, though copyright fights with the record industry are likely looming.
Google DeepMind, OpenAI, and Anthropic just agreed to grant the UK government early access to their AI models for safety evaluations. Prime Minister Rishi Sunak announced the deal alongside a £100M taskforce and an upcoming global summit, aiming to position the UK as the world capital of AI guardrails.
Salesforce Ventures just doubled its Generative AI Fund from $250 million to $500 million to back enterprise AI startups. The fund also brought Humane and Tribble into its portfolio alongside earlier bets on Anthropic, Cohere, and You.com. Salesforce is clearly determined to lock down the future of corporate AI software early.
TikTok just launched Script Generator, a free tool inside TikTok Creative Center that whips up video ad scripts in seconds. Marketers plug in product details and target duration to get scene-by-scene hooks and audio cues, though TikTok cautions that humans should still verify the copy before running campaigns.
Huawei actually registered the "Vision Pro" trademark in China back in 2019, securing exclusive rights across smart glasses and VR headsets through 2031. With China being one of Apple's most crucial markets, Cupertino may be forced to rebrand its spatial computing headset locally or cut a licensing deal with Huawei.
Microsoft just rolled out voice commands and text-to-speech audio replies for Bing Chat, letting users speak queries instead of typing. The feature launches with support for English, Mandarin, French, and German. Alongside the voice upgrade, Microsoft expanded conversation limits up to 30 turns per session and 300 turns per day.
ChatGPT just preached to a packed church of over 300 people in Bavaria. Created alongside theologian Jonas Simmerlein, 98% of the 40-minute service—including prayers, music, and sermons—was generated by AI and delivered by digital avatars. Attendees were split, with some praising the smooth delivery while others missed human warmth and soul.
