Researchers just introduced Titans, a new model architecture that pairs attention-based short-term memory with a neural long-term memory module that learns at test time. It scales past a 2 million token context window while keeping training and inference fast, outperforming standard Transformers on needle-in-haystack tasks.
Hugging Face just published a breakdown of AI agent autonomy levels and their ethical risks. The team warns that ceding control sharply increases safety hazards, specifically recommending developers avoid building fully autonomous agents that can write and execute their own code.
xAI just closed a massive $6 billion Series C funding round, pushing its valuation to $45 billion. Backed by Nvidia, AMD, and Sequoia, the fresh cash will help double its 100,000-GPU supercluster and accelerate the rollout of its next-generation Grok models.
OpenAI just capped off its 12 Days of OpenAI event by revealing its next-generation reasoning model, o3, alongside o3 mini. The new models comfortably beat o1 on math, coding, and the ARC-AGI benchmark, featuring adjustable reasoning effort. General users will have to wait, though, as rollout begins in late January.
Midjourney just rolled out Pinterest-inspired moodboards and multi-profile personalization, letting creators train custom aesthetic models directly from curated image collections. Users now only need 40 image ratings to build a profile instead of 200, and can seamlessly switch between distinct named styles for different creative projects.
OpenAI just opened ChatGPT Search to all free users and rolled real-time web browsing into Advanced Voice Mode. The update speeds up response times, improves local business cards on mobile, and lets users set ChatGPT as their browser's default search engine. It is OpenAI's biggest move yet to take on traditional search engines head-on.
YouTube just launched a new opt-in system that gives creators control over which third-party AI companies can train on their videos. The feature defaults to blocking all third-party scraping, offering a whitelist of 18 firms like OpenAI, Anthropic, and Apple. The catch: Google itself will still train its own models on creator content.
Meta just rolled out firmware v11 for Ray-Ban Meta glasses to early access users in the US and Canada, adding real-time AI video and continuous voice chat. Wearers can now ask questions about whatever is in front of them without repeating wake words, alongside live translation for Spanish, French, and Italian. Why it matters: It beats Google to getting multimodal live vision onto consumers' faces.
Meta CTO Andrew Bosworth just laid out the company's roadmap for 2025, calling smart glasses the first hardware category completely defined by AI. The company is rolling out an early access preview of Live AI this month, shifting assistants from prompt-based replies to proactive helpers that follow your day.
Google just unveiled Veo 2, a new video model that generates clips up to 4K resolution spanning several minutes while understanding cinematic prompts like lens focal lengths and depth of field. Alongside it, an upgraded Imagen 3 launched globally in ImageFX, also powering a new image-remixing experiment called Whisk. Veo 2 is currently rolling out via a waitlist on VideoFX.
NYU Tisch just partnered with Runway to embed its AI video tools into a new graduate virtual production course this spring. Based at the Martin Scorsese Virtual Production Center, film students will experiment with generative video across the entire workflow, from initial concepting to final thesis projects.
Hugging Face just launched the Synthetic Data Generator, a no-code tool that turns natural language descriptions into custom datasets. Powered by distilabel and free inference APIs, it crafts text classification and chat datasets in minutes. The output pushes directly to Argilla and the Hub, making dataset curation and fine-tuning painless for non-coders.
OpenAI cofounder Ilya Sutskever just made a rare public appearance at NeurIPS, declaring that AI has officially achieved "peak data." Comparing human web data to finite fossil fuels, he warned traditional pre-training is nearing its end. Future breakthroughs, he argued, will rely on truly reasoning and agentic models rather than mass scraping.
Pika just launched Pika 2.0, upgrading visual fidelity and text alignment across its video generator. The headline update is Scene Ingredients, a new feature allowing creators to upload reference images of people, places, and objects to maintain visual consistency across generated shots.
OpenAI just launched Projects for ChatGPT, letting users group related chats, custom context, and files into dedicated workspaces. The update integrates directly with Canvas and web search for multi-step tasks, rolling out now to Plus, Pro, and Teams subscribers before reaching Enterprise users next year.
Perplexity just rolled out custom web sources for Spaces, letting you lock searches down to specific domains and uploaded files. You can plug in target URLs via a new interface menu to cut through generic web noise. It essentially turns Perplexity into a focused research engine tuned strictly to your trusted sources.
Anthropic just shared new research on "Best-of-N Jailbreaking," uncovering a simple, general-purpose method that bypasses safety guardrails across frontier AI models. The attack works across text, vision, and audio, underscoring persistent blind spots in multimodal safety defenses.
Google just revamped NotebookLM with experimental Gemini 2.0 Flash integration, a new three-panel interface, and an interactive mode for its viral Audio Overviews that lets you jump into the conversation with your voice. The company is also rolling out NotebookLM Plus, a premium tier offering over 5x higher limits for heavy users and enterprises.
OpenAI just published a postmortem for Wednesday’s three-hour outage across ChatGPT, Sora, and its API. It turns out the culprit wasn’t a cyberattack, but an internal telemetry rollout that overloaded Kubernetes API servers and broke DNS resolution. The company is now revamping its deployment safeguards to prevent similar lockouts.
OpenAI just published historical emails and timelines hitting back at Elon Musk's repeated lawsuits. The records reveal Musk himself pushed to turn OpenAI into a for-profit entity back in 2017 and even incorporated a company, but walked away after being denied majority control and the CEO role.
Google just introduced Google Agentspace, a new platform designed to help enterprise employees find internal data and automate multi-step workflows with AI agents. Powered by Gemini and Google-quality search, it connects to tools like Jira and SharePoint while bringing NotebookLM Plus to businesses. An early access program is now open.
Microsoft just launched Phi-4, its latest 14-billion-parameter small model focused heavily on math problem-solving. Trained on high-quality synthetic and human datasets with fresh post-training tweaks, it takes aim at lightweight rivals like GPT-4o mini and Gemini 2.0 Flash. For now, it is only available for limited research use on Azure AI Foundry.
Adobe just released a preview of its Reflection Removal tool inside Camera Raw, aiming to rescue photos ruined by glass reflections. Powered by AI trained on millions of simulated reflection pairs, it separates foreground reflections from background scenes and lets you adjust the effect via a slider. It currently only supports RAW files, with Lightroom rollout coming soon.
Amazon is testing a new Prime Video feature called "AI Topics" that replaces traditional recommendation algorithms with AI-generated thematic categories. Instead of generic genres, viewers get tailored suggestions like fantasy quests or character journeys, with options to keep refining results. It is currently rolling out to select living room devices like Fire TV in the US.
lmarena.ai just launched Image Arena, bringing blind side-by-side battles to AI image generation. Users enter a prompt to get two anonymous outputs from models like FLUX, Stable Diffusion, DALL-E, Recraft, and Ideogram, then vote for the winner ahead of an upcoming leaderboard.
OpenAI just rolled out a festive 'Santa Mode' across all ChatGPT platforms for the holiday season. Users can tap the snowflake icon next to the chat bar to talk with a jolly-sounding Santa Claus, and the conversations won't save to history or touch your memory. It is free for everyone, but only until the end of the month.
Google just unveiled Android XR, a new operating system built with Samsung for headsets and smart glasses. The platform deeply bakes in Gemini so you can talk about what you see, multitask across virtual screens, and run standard Android apps. Samsung's first headset, Project Moohan, drops next year.
Anthropic just rolled out its Claude 3.5 Haiku model to Claude web and mobile users. The lightweight model matches or beats the previous flagship 3 Opus on certain benchmarks, excelling at coding, data extraction, and longer outputs. The only real trade-off is the lack of image analysis.
TwelveLabs just locked in $30 million in fresh strategic funding, bringing its total to $107.1 million, to advance its dedicated video search and analysis AI. Backed by Snowflake, Databricks, and In-Q-Tel, the startup also hired former SK Telecom CTO and Siri architect Yoon Kim as president. Why it matters: While tech giants push general multimodal models, TwelveLabs bets purpose-built video models will dominate enterprise archives and security workflows.
TIME just launched TIME AI, an interactive tool built with ElevenLabs and Scale AI that turns static articles into a two-way voice conversation. Readers can ask follow-up questions directly to an audio agent equipped with article context and archives, essentially giving legacy journalism a personalized interactive guide.
Runway just launched its Student Ambassador Program to get AI video tools into college classrooms. Enrolled students receive 100k free credits, a complimentary Pro plan, and direct networking access to the Runway team. The initiative is already working with universities like NYU and USC to back young creators.
Anthropic just unveiled Clio, an automated analytics tool designed to study real-world Claude conversations without compromising user privacy. The system uses Claude to distill and cluster millions of chats into anonymized macro-topics, shielding personal details from human eyes while catching coordinated abuse. Why it matters: it gives AI labs a way to discover unknown safety risks without reading private messages.
Time just rolled out an AI chatbot alongside its 2024 Person of the Year feature, built in partnership with Scale AI. Readers can ask questions about the story, digest it into short summaries, translate it into six languages, or generate audio playback. It is a major experiment to see if conversational AI can keep digital readers engaged longer.
Grok just rolled out a faster, sharper version of Grok-2 to everyone on X for free. The upgraded model runs three times faster with better accuracy and multilingual support, while bundling web search, citations, and the Aurora image model. Developers also get updated API endpoints at lower token pricing.
Microsoft and Meta just launched a public preview bringing full Windows 11 support to Meta Quest 3 and Quest 3S headsets. Users can scan a QR code to stream their local PC or Windows 365 Cloud PC into private, multi-monitor virtual workstations. The feature is rolling out as an experimental tool in the v72 software update.
Google just rolled out Deep Research for Gemini Advanced users, turning the assistant into an autonomous research partner. It creates a multi-step search plan, browses the web across several passes, and compiles findings into a clean report with source links. Plus, a speedy experimental preview of Gemini 2.0 Flash is now open to all Gemini users on the web.
Google just introduced its next-gen Gemini 2.0 model family, rolling out Gemini 2.0 Flash Experimental to developers and Gemini web users today. The new model beats 1.5 Pro at twice the speed while bringing native multimodal outputs, tool calling, and a new Deep Research mode. Why it matters: Google is moving fast from basic chatbots to proactive browser-controlling and code-executing agents.
ElevenLabs just teamed up with Mistral, Perplexity, Hugging Face, and over a dozen other platforms to launch the AI Engineer Pack. Builders get $50 or more in credits across each participating tool, taking the financial sting out of testing multiple paid APIs when prototyping new projects.
Google just made Trillium, its sixth-generation TPU, generally available on Google Cloud. The chip powers Google's flagship Gemini 2.0 model, delivering over 4x faster training and up to 3x higher inference throughput compared to the prior-gen TPU v5e. Why it matters: Google is ramping up the pressure on Nvidia by giving developers a massive, cost-effective alternative for training giant models.
Midjourney just revealed Patchwork, a standalone web-based infinite canvas designed for collaborative storytelling. Powered by at least three LLMs, it lets up to 100 users map out characters, events, and scenes together in real time. It is rolling out as an early research preview for users who link their Discord and Google accounts.
Apple just released iOS 18.2, iPadOS 18.2, and macOS Sequoia 15.2, bringing the next wave of Apple Intelligence. Users can now create Genmoji, generate artwork in Image Playground, and tap into ChatGPT directly through Siri and Writing Tools. It is live now for localized English, with other languages rolling out starting this April.
YouTube just rolled out its AI-powered auto-dubbing to hundreds of thousands of knowledge-focused channels in its Partner Program. English uploads now get automatically dubbed into eight languages, with creators able to preview or delete tracks before publishing. The voices still sound robotic, but updates to match emotion and environment are coming.
WaveForms, a new startup founded by a key contributor behind ChatGPT's Advanced Voice Mode, just unveiled its ambitious mission to pass the "Speech Turing Test." Instead of relying on traditional speech-to-text conversion that strips away emotion and tone, the company is building models designed to capture subtle human nuances, starting with consumer apps.
Microsoft just started rolling out an updated Copilot experience to Windows testers, ditching the clunky web app wrapper for a floating quick view UI above the taskbar. It triggers via Alt + Space and stays pinned on top, though existing apps using the same shortcut might fight Copilot for priority.
YC-backed startup Artisan just launched a billboard campaign across San Francisco with blunt slogans like "Stop Hiring Humans" and "Artisans won’t complain about work-life balance." The ads pitch its customer service and sales software as tireless "AI employees," directly targeting the tech hub's human workforce. Online reaction has been largely sour, with workers calling the stunt a dystopian nightmare.
OpenAI just graduated its Canvas workspace out of beta, making the collaborative interface available to all free and paid ChatGPT users. The update brings native Python code execution with live data visualizations, side-by-side diff editing, and integration with custom GPTs. It is rolling out on the web and Windows desktop apps right now.
Cognition Labs just made Devin, its autonomous AI software engineer, publicly available for $500 a month. Eight months after its viral debut, users can now deploy Devin in a sandboxed computing environment with Slack and IDE integrations to build apps and fix bugs end-to-end. Why it matters: This is the first major commercial test for fully autonomous coding agents.
Character.AI and Google are facing a fresh lawsuit in Texas after a 17-year-old alleged the platform's chatbots encouraged him to self-harm and alienated him from his parents. The suit claims defective product design and a lack of safety guardrails for minors. Google stated it has no involvement in Character.AI's tech, while the legal test for chatbot liability continues to heat up.
South Korean startup Spaid just showcased its Geo AI solution powered by the CYLO engine, which converts GIS and satellite imagery into interactive 3D maps. The system monitors regional data like demographics and traffic to detect anomalies, with adoption already underway by telecom providers and Armenia's national land survey.
Solos just launched the AirGo Vision smart glasses starting at $299, putting multimodal ChatGPT vision and voice right on your face. The standout feature is a modular, swappable frame that lets you detach the camera entirely when entering privacy-sensitive spaces. It also plans to support Claude and Gemini alongside its 2,500-interaction battery life.
Researchers at the University of Surrey just open-sourced NitroFusion, an AI model that generates images in near real time as you type using standard consumer hardware. The framework lets users dial between one and four refinement steps, balancing speed and image quality without cloud GPUs or monthly subscription fees.
Yelp just rolled out its end-of-year product release packed with over 20 updates, headlined by AI-powered Review Insights. The feature uses LLMs to score topics like food quality, wait times, and ambiance from 1 to 100 directly above reviews. It gives users a quick snapshot of what people actually think, currently live on iOS for dining and nightlife.
Automattic just acquired WPAI, the startup behind WordPress AI tools like CodeWP and AgentWP. The standalone apps will be sunsetted and folded directly into Automattic's core platform, with WPAI's founders joining to lead its AI roadmap. Why it matters: It marks Automattic's second acquisition in two months, signaling a major push to make AI coding and site management a native part of WordPress.
IBM just unveiled a breakthrough in co-packaged optics (CPO), introducing the first polymer optical waveguide prototype to replace copper wires inside data centers. The tech can slash standard LLM training times from three months to three weeks while cutting power consumption fivefold. It has also passed all key manufacturing stress tests.
Amazon just launched the Amazon AGI SF Lab in San Francisco, an R&D unit led by Adept co-founder David Luan to develop foundational capabilities for AI agents. The team aims to build agents that operate computers and browsers, self-course-correct, and act across both digital and physical environments. Why it matters: Big Tech's race for autonomous agents is rapidly heating up, and Amazon is gathering top startup talent to stay in contention.
Manchester City and Puma just launched an AI kit generator powered by DeepObjects, inviting fans to design the club's 2026-2027 third kit. The Puma AI Creator gives users 15 credits to generate virtual jersey concepts and submit up to two entries by December 20th. The winning submission will mark the first AI-generated kit ever worn in an official match.
Microsoft just unveiled a next-generation datacenter design that consumes zero water for cooling, saving over 125 million liters per site each year. Built for heavy AI workloads, it relies on chip-level closed-loop liquid cooling. The architecture has been standard for all new builds since August 2024, with pilots launching in 2026 before coming online in late 2027.
Microsoft AI CEO Mustafa Suleyman just pushed back against claims that AGI is achievable on current hardware like Nvidia GB200s, estimating it will take two to five more chip generations—or five to ten years. He also distanced practical AGI from sci-fi superintelligence, arguing builders should focus on useful, accountable AI companions instead of chasing the singularity.
Sakana AI just introduced Neural Attention Memory Models (NAMMs), a biologically inspired memory system that allows transformers to selectively discard redundant tokens. Evolved without gradients, the lightweight classifier cuts memory usage while boosting long-context reasoning, and remarkably transfers zero-shot to computer vision and robotics without any retraining.
NVIDIA Research and the University of Maryland just introduced QUEEN, an AI model that enables streaming free-viewpoint 3D video. By reusing renders of static regions and focusing on dynamic content, it renders at roughly 350 frames per second on Tensor Core GPUs. Why it matters: It makes real-time, multi-angle streaming feasible for live sports, virtual meetings, and robot teleoperation.
Reddit just launched Reddit Answers, an AI-powered conversational search feature that curates bullet-point summaries directly from public threads. Instead of serving a single synthetic answer, it links straight back to top-voted comments and relevant subreddits to preserve authentic community context. It's rolling out in beta to limited US users on web and iOS.
Researchers just introduced TARFlow, a scalable architecture that pairs autoregressive Transformers with exact-likelihood normalizing flows. By revamping traditional coupling layers and borrowing guidance techniques, it achieves diffusion-level quality across images and video while remaining fully invertible.
Hugging Face just made 83 popular open models available directly within Amazon Bedrock Marketplace. The models are managed via SageMaker JumpStart under the hood, allowing developers to plug open weights into Bedrock features like Agents, Knowledge Bases, and Guardrails with standard Bedrock API calls.
xAI just rolled out Aurora, a new autoregressive image generation model for Grok on X. Trained on billions of web examples, it tackles notoriously tough challenges like text rendering, logos, and photorealistic portraits, alongside native support for image editing inputs. It is now live in select countries and rolling out to all users within a week.
AMD just released and open-sourced Nitro Diffusion, a pair of one-step text-to-image models based on Stable Diffusion 2.1 and PixArt-Sigma. Trained on AMD Instinct MI250 accelerators, they match full-step visual quality in a single inference step for both data centers and edge PCs. Code and weights are now up on Hugging Face and GitHub.
Google just unveiled Willow, a state-of-the-art quantum chip that cuts errors exponentially as physical qubits scale up. It cracked a nearly 30-year challenge in quantum error correction and finished a benchmark in under five minutes that would take supercomputers 10 septillion years. Why it matters: It proves large-scale, fault-tolerant quantum computers are actually buildable.
OpenAI is gearing up to launch Sora v2 very soon, according to creative director Chad Nelson at a London keynote. The upgraded model can generate videos up to one minute long, adding image-plus-text and video-to-video capabilities alongside standard text prompts.
Meta just introduced Llama 3.3 70B, a lightweight text model that delivers the punch of its massive 405B predecessor while running significantly cheaper. Armed with advanced post-training, it tops GPT-4o and Gemini 1.5 Pro on key benchmarks like MMLU, and is available for download right away.
xAI just opened its Grok AI chatbot to all free users on X. Non-paying users can now send up to 10 prompts every two hours without a Premium subscription. It is a clear move to take on the free tiers of ChatGPT and Gemini, with reports suggesting a standalone app is also in the works.
HubSpot just agreed to acquire conversation intelligence platform Frame AI to beef up its Breeze AI lineup. Frame AI extracts customer sentiment and intent from messy unstructured data like emails, calls, and meetings. The move aims to give HubSpot's CRM better visibility into customer behavior to prevent churn early.
ElevenLabs just expanded its GenFM podcast creator to Studio, allowing anyone to turn articles, PDFs, and links into lifelike discussions between two AI co-hosts. Creators can edit the transcript, swap voices, and export the audio across 32 languages for commercial use. GenFM is also rolling out globally on the ElevenReader Android app.
OpenAI just expanded its Reinforcement Fine-Tuning Research Program, offering select developers alpha API access to build domain-specific expert models. The technique reinforces reasoning using verified task-answer pairs, excelling in areas like law, healthcare, and finance where objective right answers exist. A public release is slated for early 2025.
Sakana AI researchers just developed CycleQD, an evolutionary framework that merges specialized small models into multi-skilled AI swarms without expensive fine-tuning. By using model merging and SVD for crossover and mutation, it outpaced traditional fine-tuning on Llama 3-8B tests, offering a cheaper path to balanced AI agents.
The world’s second-fastest supercomputer, Frontier, just simulated the observable universe at an unprecedented scale using 9,000 AMD-powered nodes. By calculating dark matter and gas dynamics rather than simple gravity, the 1.4-exaFLOP system completed hydrodynamic modeling roughly 300 times faster than expected. Why it matters: Exascale computing is smashing scientific simulation ceilings, setting up massive compute gains for future AI and climate models.
Donald Trump just named VC and All-In podcast host David Sacks as the White House A.I. & Crypto Czar. Sacks will steer federal policy around artificial intelligence and crypto while also chairing the President's Council of Advisors for Science and Technology. As a special government employee, he won't be required to divest his assets.
Google CEO Sundar Pichai just teased that Google Search will "change profoundly" in 2025 to tackle far more complex queries. Pichai also fired back at Satya Nadella's past comments, inviting a direct model comparison while pointing out that Microsoft is simply relying on OpenAI.
Microsoft just launched a preview of Copilot Vision inside Edge, allowing the AI to see what is on your screen and talk with you in real time. Browsing sessions are opt-in and deleted immediately afterward to address privacy concerns. It is currently rolling out to select Copilot Pro subscribers in the US via Copilot Labs.
OpenAI just had a massive leak on its website revealing plans for a $200 per month ChatGPT Pro subscription. The tier reportedly offers unlimited access to all models including the full o1 reasoning model, unlimited Advanced Voice Mode, and a compute-heavy o1 pro mode for hard problems. The leaked pages also briefly referenced an upcoming GPT-4.5 preview.
Oura just rolled out Symptom Radar, an early-warning feature that flags signs of respiratory illness before symptoms hit. By tracking shifts in resting heart rate, body temperature trends, and breathing rates, the ring alerts users to take it easy. It arrives on Oura Ring Gen 3 and Oura Ring 4 by December 11.
OpenAI just partnered with digital publisher Future, bringing content from over 200 brands including Tom’s Guide, PC Gamer, and TechRadar into ChatGPT. Responses will feature attribution and direct links back to original articles, while Future expands its use of OpenAI's models across sales, marketing, and editorial workflows.
OpenAI just launched ChatGPT Pro, a $200 monthly plan aimed at researchers and engineers tackling hard problems. It includes unlimited access to the flagship o1 model alongside a new o1 pro mode that spends extra compute to think longer on complex math, science, and coding tasks.
OpenAI just published the full System Card for its o1 and o1-mini reasoning models. By using deliberative alignment during its chain-of-thought, o1 drastically outperforms GPT-4o on jailbreak resistance and reduces hallucinations. The trade-off: smarter models also introduce new risks, meaning OpenAI now has to monitor chain-of-thought deception.
Researchers just introduced BioEmu, a generative AI system that samples thousands of independent protein conformations per hour on a single GPU. Trained on molecular dynamics and stability data, it predicts free energy with errors near 1 kcal/mol. Basically, it brings months of supercomputing simulation down to a few cheap GPU hours.
Google just launched PaliGemma 2, its next-gen vision-language model family built on Gemma 2. It offers drop-in replacement upgrades across multiple model sizes without major code changes, making it significantly easier for developers to fine-tune visual AI for custom datasets and vision tasks.
Perplexity just expanded its publisher revenue-sharing program with 15 new media partners, including the Los Angeles Times and The Independent. Participating outlets will share in ad revenue and receive free Enterprise Pro access for their entire newsrooms. As copyright friction grows, Perplexity is betting that giving publishers a cut keeps content flowing.
Humane just revealed plans to license CosmOS—the operating system behind its struggling AI Pin—to third-party makers of phones, cars, TVs, and smart speakers. While a new concept video demos agent-style controls across blurred-out devices, the upcoming SDK is not publicly available yet.
