Krea AI just open-sourced Krea Realtime, a 14B autoregressive model that is 10x larger than existing open-source counterparts. Running on a single B200 GPU, it generates long-form video at 11 fps. Weights and a technical report are available now.
Friend founder Avi Schiffmann just showed up at a New York protest against his own $129 AI companion pendant, where critics shredded cardboard cutouts chanting "get real friends." The backlash followed a million-dollar subway ad blitz. Schiffmann leaned into the drama, talking to protesters and pledging never to sell out to Big Tech for surveillance.
Kohler just launched Dekoda, a $599 toilet-mounted sensor that uses optical spectroscopy and algorithms to analyze your bathroom breaks. It tracks hydration, assesses gut health, and flags potential health issues like blood in the bowl, sending encrypted data to a companion app. Why it matters: It turns an everyday routine into continuous preventive screening, though it requires an ongoing subscription.
Fish Audio just publicly launched Fish Audio S1, pitching it as the market's most expressive and natural TTS model. The platform is priced six times cheaper than ElevenLabs, boasting 20,000 active developers, $5 million in ARR, and free voice cloning to get started.
Deepseek just open-sourced an OCR system that compresses image-based text by up to 10x while retaining 97% of information. By drastically reducing vision tokens to as few as 64 per page, it lets language models process massive context without hitting compute limits.
Anthropic just launched Claude for Life Sciences to assist researchers from early drug discovery to regulatory compliance. Powered by Claude Sonnet 4.5, which outperforms human baselines on lab protocol QA, the platform adds connectors to PubMed, Benchling, and BioRender alongside modular agent skills. Why it matters: LLMs are moving beyond generic office tasks to run end-to-end scientific workflows.
Anthropic just launched Claude Code on the web, letting developers delegate coding tasks straight from the browser without touching a local terminal. Tasks run in parallel inside isolated cloud sandboxes, automatically submitting PRs across multiple GitHub repos—and there's even an iOS preview for coding on the go. Why it matters: It turns routine bug fixing and repo maintenance into hands-off background work.
Napster just launched View, a 65g holographic display designed to make AI agents feel physically present on your desk without headsets or glasses. It connects to M1 or newer Macs via USB-C to display 3D specialists for coding, writing, and screen sharing. It is available for $99 standalone or free with an annual plan.
OpenAI researchers just walked back claims that GPT-5 solved 10 decades-old Erdős math problems after experts pointed out the model merely surfaced existing papers that a database author had missed. DeepMind CEO Demis Hassabis called the sloppy communication "embarrassing," exposing the awkward gap between AI research hype and actual scientific progress.
Google just overhauled its AI Studio to streamline the developer experience, merging Gemini, Veo 3.1 video, and speech models into a single unified Playground. Developers can now generate text, video, and audio in one continuous workflow, save and reuse system instruction templates, ground outputs with Google Maps, and manage API keys by project.
Elon Musk just announced that X will replace all its heuristic recommendation algorithms with Grok within 4 to 6 weeks. The AI will process over 100M posts and videos daily to boost quality content from small accounts, and users will be able to tweak their feeds simply by talking to Grok.
Facebook just rolled out an opt-in AI feature that scans your camera roll via cloud processing to surface your best photos, assemble collages, and suggest creative edits. Meta promises your media won't train its models unless you use its editing tools. Why it matters: Sifting through hundreds of vacation shots is a chore, and Facebook wants AI to do the curation—starting in the U.S. and Canada.
Meshy just rolled out Meshy 6 Preview, introducing studio-grade mesh precision and sculpting-level details directly to AI 3D creation. The update sharpens hard surfaces, refines organic structures, and adds a workspace to convert 3D models into cinematic media alongside 1-click batch generation for up to 10 assets.
LabOS just introduced an AI co-scientist that bridges digital reasoning with physical lab benches via XR glasses. The system tracks your hands and tools to catch wrong reagents or missed steps in real time. Both software and hardware are slated to be open-sourced, with access requests now open.
OpenAI just blocked users from generating AI videos of Martin Luther King Jr. in Sora following requests from Dr. King's estate. The move comes after users generated offensive and disrespectful depictions of the civil rights icon on the newly launched app. OpenAI confirmed that authorized representatives of public figures can now opt out of Sora cameos.
Cisco just unveiled Project CodeGuard, an open-source framework designed to stop AI coding agents from generating vulnerable code. It integrates across the entire coding lifecycle, flagging security risks like hardcoded secrets and missing input validation before, during, and after generation. While Cisco notes human review is still essential, it aims to keep low-hanging bugs from slipping into production.
Amazon just shared the first look at its planned nuclear plant in Washington State. The Cascade Advanced Energy Facility will use small modular reactors to generate 960 megawatts across three sections—enough for 770,000 homes while taking up only a few city blocks. Construction is expected toward the end of the decade.
The University of Maryland's RoboScout team just demoed an AI-powered disaster response squad that pairs autonomous drones with Boston Dynamics robot dogs. Drones locate survivors, while a Spot robot uses an onboard LLM to talk to victims, check vitals, and beam triage data straight to medics' phones. It could be a lifesaver when patients drastically outnumber first responders.
New York just became the first US state to ban landlords from using AI algorithms to coordinate and hike rental prices. Under the new law taking effect in 60 days, property owners relying on pricing software like RealPage will be legally treated as colluding. Why it matters: Regulators are finally drawing a hard antitrust line against algorithmic collusion in housing.
Character.ai just rolled out Scenes to all users, turning open-ended chats into structured, story-driven role-play scenarios. Instead of building a brand-new bot for every new narrative, creators can now define a setting, goal, and opening hook that any character can drop into. It gives role-players an improv stage without the tedious setup.
Pinterest just launched new feed controls that let users dial back the amount of AI-generated images they see. Pinners can now restrict generative AI content across heavy-hit categories like art, beauty, fashion, and home decor under recommendation settings. It is a much-needed move to protect the platform's core identity after months of user complaints over synthetic filler.
Bubble just launched Bubble AI Agent, letting users build production-ready apps by toggling between natural language prompts and its visual editor. The company noted that only 9% of its builders trust raw AI code for critical workflows due to security fears. Why it matters: Instead of wrestling with black-box generated code, builders get the speed of AI with pixel-level visual control.
Spotify just inked deals with Sony, Universal, Warner, and Merlin to build generative AI music products that put copyright first. The system will let artists opt in or out of AI tools and collect payouts when their tracks are used in AI creations. Why it matters: After backlash over fake AI bands taking over feeds, Spotify is betting that giving labels control is the only way forward.
Microsoft just rolled out a major wave of Copilot updates for Windows 11, bringing "Hey Copilot" voice wake words and screen-reading Copilot Vision to all supported markets. Windows Insiders are also testing Copilot Actions to automate workflows directly across local files. Why it matters: AI is graduating from a chat sidebar into the desktop's core nervous system.
Anthropic just introduced Agent Skills, letting users bundle instructions, scripts, and local files into reusable folders that Claude loads only when needed. The setup works identically across Claude apps, the developer API, and Claude Code, with support for executable scripts. It basically turns tedious prompt engineering into portable, plug-and-play micro-tools.
Anthropic just rolled out a Microsoft 365 connector and enterprise search for Claude, letting the chatbot tap directly into your daily work tools. Claude can now pull context across SharePoint, OneDrive, Outlook emails, and Teams threads to answer questions. It basically turns Claude into an internal company knowledge hub, available now for Team and Enterprise tiers.
OpenAI just leveled up Sora 2 with a new Storyboards feature on the web for Pro subscribers. The update also extends video generation limits across the board: all users can now generate up to 15-second clips on mobile and web, while Pro users on the web can push that to 25 seconds. It makes multi-shot storytelling with AI far more practical.
Google DeepMind just partnered with Commonwealth Fusion Systems (CFS) to accelerate the path to net fusion energy. The collaboration combines DeepMind's plasma simulator TORAX with reinforcement learning to run millions of virtual simulations for CFS's compact SPARC tokamak, optimizing fuel injection and magnetic controls before turning the machine on. AI will also act as a real-time pilot to safely manage intense heat. Why it matters: Mastering plasma control could dramatically shorten the timeline to abundant, zero-carbon grid power.
Meta just broke ground on a massive AI-optimized data center in El Paso, Texas, designed to scale up to 1 gigawatt of capacity. The facility incorporates a closed-loop liquid cooling system and flexible designs to support future hardware generations. It marks another major infrastructure bet as tech giants race to secure long-term power and compute for superintelligence.
Google just launched Veo 3.1 alongside major creative upgrades to its AI filmmaking platform Flow, which has already generated over 275 million videos. The new release integrates rich generated audio directly into workflows like frame bridging and video extension, while adding precision tools like "Insert" to seamlessly place new elements with matching lighting. Veo 3.1 is available now in Flow, the Gemini app, and via Vertex AI.
Spotify just upgraded its AI DJ for Premium subscribers, letting you type in your music requests instead of having to speak out loud. The update also brings request support to its Spanish-language DJ, DJ Livi, alongside personalized prompt suggestions. Why it matters: You can finally tweak your soundtrack on crowded commutes or in quiet rooms without looking weird talking to your phone.
Anthropic just released Claude Haiku 4.5 to all users, delivering coding performance on par with Claude Sonnet 4 at one-third the cost and more than twice the speed. It even outperforms Sonnet 4 on computer-use tasks, priced at $1 per million input tokens and $5 per million output tokens.
Apple just refreshed the Vision Pro with an M5 chip and a cushioned Dual Knit Band, starting at $3,499. The new silicon delivers up to 50% faster system AI tasks, renders 10% more display pixels, and bumps battery life to 2.5 hours. Why it matters: Apple is tackling the biggest pain points—comfort and horsepower—while keeping its premium spatial computing bet moving.
Apple just announced its next-gen M5 chip, packing a dedicated Neural Accelerator inside every GPU core. That translates to over 4x peak GPU compute for AI workloads compared to M4, alongside a 30% jump in unified memory bandwidth to 153GB/s. Why it matters: Running local LLMs and diffusion models on Macs and iPads is about to get dramatically faster.
Mozilla just made Perplexity available as a built-in search option for all Firefox desktop users globally, letting people swap classic link lists for cited AI answers. You can toggle it directly from the unified address bar or set it as default, with mobile rollout coming in the months ahead.
OpenAI is loosening restrictions on ChatGPT after previously overtightening guardrails around mental health concerns. In a few weeks, a new update will let users opt into a human-like personality reminiscent of 4o, complete with emojis and friendly banter. By December, OpenAI plans to complete age-gating to allow verified adults access to erotica under a "treat adult users like adults" approach.
Microsoft just launched ExCyTIn-Bench, an open-source benchmark testing AI agents inside simulated security operations centers with 57 live log tables. In testing, GPT-5 with high reasoning led with a 56.2% score, but dropping its reasoning depth tanked performance by nearly 19%. Why it matters: Static Q&A tests are useless for infosec—defenders now have a reliable way to see if an AI agent can actually investigate complex threats.
Oracle and NVIDIA just unveiled the OCI Zettascale10 computing cluster, a massive infrastructure play delivering up to 16 zettaflops of peak AI compute. Powered by next-gen NVIDIA GPUs and Spectrum-X Ethernet, it is designed to interconnect millions of GPUs while natively baking NVIDIA NIM and AI Enterprise directly into Oracle's cloud console and flagship database.
OpenAI just formed the Expert Council on Well-Being and AI, recruiting top researchers in psychology, psychiatry, and human-computer interaction to guide ChatGPT and Sora. The council focuses heavily on teen development and healthy boundaries, building on early work that refined parental controls and distress notifications.
Cisco just released new enterprise data showing only 18% of companies have successfully moved AI use cases into production, even as 83% race to deploy agents next year. The report warns that rushing deployments without scalable networks or security creates mounting "AI Infrastructure Debt." Moving fast without solid foundations will cost far more to fix later.
Anthropic and Salesforce just expanded their partnership to make Claude a preferred AI model for Agentforce, specifically targeting strict sectors like healthcare and finance. All traffic stays securely inside Salesforce's private cloud, while Salesforce engineers deploy Claude Code worldwide. It's a pragmatic push to put agentic AI into mission-critical corporate workflows.
Anthropic just published a wide-ranging policy roadmap exploring how governments should handle AI-driven labor disruption. Developed with leading economists, the proposals range from immediate workforce training grants and faster data center permitting to radical scenarios like taxing token generation or creating AI sovereign wealth funds.
Microsoft AI just released MAI-Image-1, its first fully in-house text-to-image model that debuted straight into LMArena's top 10. Tuned for photorealistic lighting and rapid iteration, it is rolling out to Bing Image Creator alongside DALL-E 3. Why it matters: Microsoft continues to quietly build up its own frontier stack rather than relying purely on OpenAI.
Google just rolled out two new AI-powered features to help users track trending content. Discover now shows expandable topic summaries with source links in the U.S., South Korea, and India, while mobile Search is adding a "What's new" feed for sports teams and players rolling out soon in the U.S.
California Governor Gavin Newsom just signed AB 1043, requiring device operating systems and app stores to prompt users for their age during setup. Taking effect in 2027, the law avoids ID uploads and won backing from Google and Meta, though Apple has yet to endorse it.
Google Meet just launched an AI-powered makeup filter featuring 12 different styles to save you from having to get camera-ready. The clever part: it stays locked onto your face even when you sip coffee, so the illusion won't break mid-call. It is off by default and rolling out across mobile and web.
Google just gave NotebookLM’s Video Overviews a massive visual upgrade powered by its Gemini image model, "Nano Banana." Users can now generate illustrated explainer videos in six distinct visual styles—including anime, watercolor, and whiteboard. The update also adds a bite-sized "Brief" summary format alongside custom focus prompts. It rolls out to Pro users this week before reaching everyone.
Andrej Karpathy just dropped nanochat, an 8,000-line codebase that packs an entire end-to-end ChatGPT training and inference pipeline into a single repo. Running a single script on an 8xH100 node costs about $100 and yields a working, chat-ready model in just four hours, covering everything from tokenizer training to GRPO reinforcement learning. Why it matters: It turns the entire modern LLM post-training playbook into a readable, easily forkable playground for developers.
Governor Newsom just signed a sweeping package of legislation targeting emerging tech risks for kids. The bills force "companion chatbot" platforms to detect self-harm cues, bar them from posing as medical pros, and let victims of deepfake porn sue facilitators for up to $250,000. Crucially, AI creators can no longer dodge liability by claiming the model acted autonomously.
Slack is testing a complete overhaul of Slackbot, turning the simple notification bot into a proactive workspace AI assistant. It can parse unread messages across channels, find files via conversational prompts, coordinate Google and Outlook calendars, and even spin up project plans in Canvas. It is currently in testing with 70,000 Salesforce staff ahead of a rollout later this year.
Anduril and Meta just unveiled EagleEye, an AI-powered mixed-reality helmet that turns soldiers into real-time command centers. The system overlays live sensor streams and threat tags directly into a heads-up display, letting warfighters command drones via voice or gestures. Backed by $159M in Army contracts, it looks to replace Microsoft's stalled military gear.
NVIDIA just started shipping DGX Spark, a compact desktop AI supercomputer powered by its Grace Blackwell GB10 superchip. It packs 1 petaflop of compute and 128GB of unified memory, allowing developers to run inference on 200B parameter models right on their desks. Jensen Huang even recreated history by hand-delivering one of the first units to Elon Musk at SpaceX.
OpenAI just inked a massive multi-year deal with Broadcom to deploy 10 gigawatts of OpenAI-designed AI accelerators and Ethernet networking systems. The goal is to bake frontier model lessons straight into silicon, with rack deployments kicking off in late 2026 and wrapping up by 2029.
Meta just expanded its AI translation tools for Reels, adding voice dubbing and lip syncing in Hindi and Portuguese alongside English and Spanish. The tool clones the creator's voice and mimics their mouth movements in the translated language. It's completely free for public Instagram accounts and Facebook creators with over 1,000 followers.
OpenAI just introduced a new evaluation framework to measure and reduce political bias in ChatGPT across 100 topics and five distinct behavioral axes. An analysis of real production traffic found that less than 0.01% of all responses display signs of political bias. Why it matters: Moving past basic multiple-choice benchmarks is crucial for keeping AI objective under real-world pressure.
DC Comics president Jim Lee just drew a hard line against generative AI, pledging at New York Comic Con that the publisher will "not now, not ever" support AI-generated art or storytelling. Lee argued that fans instinctively reject what feels fake, emphasizing that AI merely aggregates data without feeling. The firm stance follows past backlash where DC had to pull variant covers suspected of AI use.
The Browser Company just launched Dia for all Mac users, bringing its AI-first browser out of private beta. It centers on a tab-based chat interface where users can @-mention open pages, run built-in 'Skills' for tasks like fact-checking, and build study guides from notes. Core features are free on Apple silicon Macs, with an unlimited Pro tier priced at $20 a month.
Anthropic just revealed that poisoning an AI doesn't require millions of malicious files—just around 250. Across models ranging from 600M to 13B parameters, attack success relied on a fixed document count rather than dataset percentage, easily planting backdoors that trigger gibberish outputs. Why it matters: Hijacking frontier models during pretraining is vastly cheaper and easier than the industry previously assumed.
Anthropic just introduced plugins for Claude Code, letting developers bundle custom slash commands, specialized subagents, MCP servers, and hooks into one-command installs. Anyone can host a curated marketplace via a simple Git repo, and plugins can be toggled on and off to keep prompts clean. It effectively turns the CLI coding assistant into an extensible developer ecosystem.
Figure AI just introduced Figure 03, its third-generation humanoid robot engineered for the home and mass manufacturing. Built to power its proprietary Helix AI, the robot features tactile fingertips sensitive to three grams of pressure, washable fabric coverings, and 2 kW wireless inductive charging. Why it matters: It shifts humanoids from experimental lab hardware toward consumer-ready scale.
Google just launched Gemini CLI extensions, letting terminal developers connect directly to tools like Figma, Stripe, Postman, and Google Cloud without context switching. Built on top of MCP, each extension bundles tools and a built-in playbook that guides the AI agent. It basically turns your terminal into a customized AI command center.
Google just expanded its virtual try-on tool to footwear, letting shoppers see how sneakers or heels look on them using a full-length photo. The feature rolls out today in the U.S. and heads to Australia, Canada, and Japan in the coming weeks. Why it matters: Online shoe shopping is finally cutting down on the guesswork.
Qualcomm just announced an agreement to acquire open-source hardware pioneer Arduino for an undisclosed sum. Alongside the buyout, the two introduced the Arduino UNO Q, powered by Qualcomm's Dragonwing platform, plus a new App Lab development environment. Arduino will keep its independent brand and open spirit, effectively giving 33M+ makers direct access to Qualcomm's edge AI silicon.
Google just expanded AI Mode in Search to over 35 new languages and 40+ countries, bringing total availability to more than 200 regions worldwide. Powered by custom Gemini models, the feature handles complex queries and multimodal inputs. Google also noted that users are already asking questions nearly three times longer than traditional searches.
Google just launched the Gemini 2.5 Computer Use model in public preview, letting developers build agents that navigate web and mobile interfaces by clicking, scrolling, and filling out forms. Built on Gemini 2.5 Pro, it tops browser benchmarks with lower latency and step-by-step safety controls. Why it matters: AI can now tackle messy real-world web chores that lack structured APIs.
Google just rolled out Opal, its no-code AI mini-app builder, to 15 new countries including Canada, India, and Japan. The update also brings visual step-by-step debugging with real-time error tracking, faster app startup, and parallel execution for complex workflows. Why it matters: Google is making it significantly easier for non-coders to assemble and troubleshoot production-ready AI workflows.
Secret Weapons just launched EbSynth 2 as a browser-based tool that lets creators restyle an entire video by editing a single keyframe. Instead of relying on generative models trained on external data, it uses a texture-synthesis algorithm to propagate changes without AI hallucinations. The new version introduces timeline navigation, layer support, and prompt server-side data deletion.
Anthropic just announced plans to open an office in Bengaluru in early 2026, marking its second hub in the Asia-Pacific region after Tokyo. India currently ranks as Claude's second-largest user base behind the U.S., driven heavily by software development and IT services. The company is also upgrading Claude's support for Hindi and nearly a dozen Indic languages.
Anthropic just hired former Stripe CTO Rahul Patil as its new Chief Technology Officer to lead product, compute, and infrastructure engineering. Co-founder and previous CTO Sam McCandlish is moving to Chief Architect to focus entirely on large-scale model pretraining and RL systems. With over 300,000 enterprise customers to support, Anthropic is clearly leveling up its production engineering muscle.
IBM just partnered with Anthropic to integrate Claude across its enterprise software portfolio, kicking off with a brand-new AI-first IDE. Internal trials across 6,000 IBM engineers showed an average 45% productivity jump without sacrificing code governance. The duo is also teaming up on open Model Context Protocol (MCP) standards to help enterprises deploy agents safely.
Taylor Swift fans are turning detective after an online Google scavenger hunt for her upcoming album revealed promotional videos that look noticeably AI-generated. The backlash is growing as speculation swirls that Google quietly used its Veo 3 video tech for the stunt. Google has yet to comment, but fans are far from impressed.
Top YouTube creator MrBeast just voiced serious concerns about AI's impact on content creators, calling it "scary times" as AI video threatens millions of livelihoods. The warning follows OpenAI's viral launch of its Sora 2 mobile app. If the world's highest-earning creator is publicly questioning the future of video making, smaller creators have every reason to be terrified.
Spotify just launched an official integration inside ChatGPT, allowing both Free and Premium users to summon tailored music and podcast recommendations directly in chat. Premium subscribers can even turn complex prompts into fresh playlists, and Spotify promises none of its audio will be used to train OpenAI's models.
OpenAI just made its Codex AI coding agent generally available, rolling out a direct Slack integration, enterprise admin tools, and a TypeScript SDK. Engineering teams can now delegate code tasks by tagging @Codex in Slack threads or embed the agent directly into custom CI/CD pipelines. Codex daily usage has already surged over 10x since August.
OpenAI just rolled out interactive third-party apps directly inside ChatGPT alongside a preview of its open-source Apps SDK built on MCP. Users can now summon services like Spotify, Canva, and Figma mid-chat to view interactive playlists, slides, and listings, giving developers a direct path to reach over 800 million users. The feature is available today outside the EU, UK, and Switzerland.
OpenAI just launched AgentKit, a full-stack toolkit designed to help developers build, deploy, and optimize AI agents. It packs a visual canvas called Agent Builder, an embeddable ChatKit UI, and expanded evaluation tools with reinforcement fine-tuning (RFT). Why it matters: Teams can now drag and drop multi-agent workflows and test them in hours instead of months.
OpenAI and Allied for Startups just released the Hacktivate AI report, outlining 20 policy and industry proposals ahead of the EU’s upcoming Apply AI Strategy. Ideas range from individual learning accounts to a shared GovAI Hub and regulatory simplification. OpenAI’s workplace data shows adoption remains heavily uneven across sectors.
Anthropic just teamed up with Deloitte to roll out Claude to more than 470,000 employees globally, marking its largest enterprise deployment to date. The firm will also establish a Claude Center of Excellence and certify 15,000 specialists to build compliance-focused AI solutions for regulated industries like healthcare and finance. Why it matters: Anthropic is proving that enterprise safety might be the ultimate winning ticket for corporate adoption.
Google DeepMind just unveiled CodeMender, an autonomous AI agent powered by Gemini Deep Think models that diagnoses and patches software vulnerabilities. Over six months, it has already upstreamed 72 security fixes to open-source projects, though human researchers still review every patch before submission.
Anthropic just released Petri, an open-source auditing tool that deploys automated agents to grill AI models in multi-turn conversations and flag dangerous behaviors like deception, self-preservation, or unauthorized whistleblowing. In pilot tests across 14 frontier models, Claude Sonnet 4.5 edged out GPT-5 as the lowest-risk system. Why it matters: as models get smarter, manual red-teaming can't scale—now researchers can stress-test frontier systems in minutes.
AMD and OpenAI just announced a massive deal to deploy 6 gigawatts of AMD GPUs for next-gen AI infrastructure. The rollout kicks off in the second half of 2026 with an initial 1-gigawatt deployment of Instinct MI450 chips. AMD is also granting OpenAI warrants for up to 160 million shares tied to delivery milestones.
OpenAI just introduced new Cameo restrictions in Sora, letting users set custom rules on how others can generate videos with their likeness—like blocking political commentary or specific words. The update also improves watermark visibility, tunes safety moderation, and fixes an annoying bug that tied Sora account deletion directly to ChatGPT.
OpenAI's new Sora iOS app just climbed to the top of the App Store, but users are already flooding it with clips of Pokémon, SpongeBob, and Ronald McDonald. Legal experts warn the startup is walking straight into a copyright minefield. OpenAI says it's honoring takedown requests and rolling out character-level controls, but critics argue making creators opt out won't fly in court.
IBM just open-sourced Granite 4, a new language model family combining Transformer attention with Mamba-2 state space architecture. The models range in size from 3 billion to 32 billion parameters. In benchmarks, the 7B hybrid model consumed just 15GB of RAM—one-sixth of its predecessor—dramatically lowering long-prompt inference costs.
Uber just acquired Segments.ai to beef up its Uber AI Solutions division. The startup specializes in multi-sensor and lidar annotation for autonomous vehicles and robotics, and its entire team will join Uber. Why it matters: Uber is aggressively turning its internal data and autonomy infrastructure into a commercial labeling powerhouse for outside AI builders.
