xAI just released the base model weights and architecture of its Grok-1 language model under the permissive Apache 2.0 license. This is the raw pre-training checkpoint from October 2023, trained from scratch as a Mixture-of-Experts system without fine-tuning for specific tasks like dialogue. Developers can now access instructions directly on GitHub.
Apptronik just inked a commercial deal with Mercedes-Benz to trial its Apollo bipedal robots on auto assembly lines. Starting at a factory in Hungary, the 160-pound bots will inspect and deliver parts to human workers. The pitch is simple: carmakers can plug humanoids into existing facilities to tackle grueling, repetitive tasks without having to redesign the plant.
Reddit just disclosed that the FTC is inquiring into its partnerships that license user posts to Google and others for AI training. Ahead of its upcoming IPO, Reddit had pitched $203 million in expected licensing revenue to investors, but US regulators are now questioning the privacy and fairness of selling user-generated content.
Apple just acquired Canadian startup DarwinAI to bolster its artificial intelligence division ahead of iOS 18. DarwinAI specializes in making AI models smaller and more efficient, a key technical piece as Apple pushes to run generative AI features directly on user devices for privacy.
Figure just integrated OpenAI's vision and speech models into its Figure 01 humanoid robot, enabling it to hold full conversations while executing physical tasks. The robot can process visual context, follow verbal commands, and even pauses naturally while speaking. Why it matters: Bringing ChatGPT into a physical frame turns generative AI into an embodied agent, accelerating the race toward real-world AGI.
Anthropic just released Claude 3 Haiku, its fastest and most affordable model to date. Built for enterprise workloads, it processes 21K tokens per second—roughly 30 pages—making it three times faster than peers while slashing document processing costs in half. Why it matters: developers can now analyze 400 legal cases or 2,500 images for just one dollar.
Adobe's Firefly image generator just ran into the exact same diversity overcorrection that tripped up Google Gemini. In tests by Semafor, prompts for historical figures generated Black Nazi soldiers and Black Vikings due to underlying safety guardrails. Adobe clarified that Firefly is built for creative ideation rather than factual history, but the blunder proves that prompt-injection guardrails breaking historical accuracy is an industry-wide headache, not just a Google problem.
OpenAI just confirmed that its buzzy text-to-video generator Sora will be publicly available later this year, potentially in "a few months." CTO Mira Murati revealed plans to eventually add audio and video editing features, while working to push operating costs down close to DALL-E levels.
The European Parliament just approved the EU AI Act, establishing the world's first comprehensive rulebook for artificial intelligence with an overwhelming 523 votes in favor. The landmark legislation classifies AI systems by risk level and bans "unacceptable" hazards outright. It is set to take effect in May, with enforcement phased in starting in 2025.
Google DeepMind just introduced SIMA, a generalist AI agent capable of navigating 3D virtual environments and video games using only screen pixels and natural language instructions. Trained across titles like No Man’s Sky and Teardown, SIMA executes over 600 basic skills via standard keyboard and mouse controls. It marks a shift from chasing high scores to building versatile virtual assistants.
OpenAI's web team just accidentally leaked teaser text for GPT-4.5 Turbo on major search engines ahead of an official announcement. The indexed snippets revealed a 256k token context window—double that of GPT-4 Turbo—and a knowledge cutoff of June 2024, hinting at a potential mid-year rollout. While the link currently leads to a 404 page, the move shows OpenAI is eager to counter Anthropic's Claude 3.
Deepgram just launched Aura, a real-time text-to-speech API designed to help conversational AI agents speak without awkward pauses. It renders realistic voices in under half a second—often under 0.3 seconds—at $0.015 per 1,000 characters, making natural phone-based customer service bots much cheaper and faster.
Google just started restricting Gemini from answering questions about upcoming elections worldwide, rolling out the block across the U.S., India, and other key voting markets. Ask about a specific candidate or party, and the bot will bluntly redirect you to traditional Google Search. Facing election-year heat and regulatory crackdowns, Google is choosing total silence over potential hallucinations.
Researchers just showed that large language models can predict whether neuroscience experiments will succeed better than human experts, scoring 81% accuracy against 64%. Using just 7B-parameter models fine-tuned on scientific literature, the system helps researchers quickly pinpoint which studies are actually worth pursuing.
Inflection just rolled out Inflection-2.5, claiming it rivals GPT-4 while using only 40% of the training compute. The upgrade powers its personal assistant Pi, joining Claude 3 Opus and Gemini 1.5 in a sudden wave of contenders breaking OpenAI's performance barrier.
Biden just called for a ban on AI voice impersonations during his State of the Union address. While he didn't share specifics on guardrails or whether the rules would reach Hollywood, synthetic voices are clearly moving from industry labor disputes straight onto the federal agenda.
OpenAI just concluded its independent review into Sam Altman's brief ouster, officially reinstating him to the board of directors. The WilmerHale investigation found the sudden firing was sparked by a breakdown in trust, not issues with AI safety or finances. To patch up oversight, OpenAI also added three heavyweight female board members and introduced a whistleblower hotline.
Midjourney just slapped an indefinite ban on all Stability AI employees after bot-like scraping brought its service down for 24 hours. The rival was allegedly trying to harvest prompt-image pairs to train its own models. Now, Midjourney is enforcing a nuclear policy: crash our servers with aggressive automation, and your entire company gets banned.
The US Army Research Laboratory is testing OpenAI's GPT-4 Turbo and Vision models to assist with operational planning inside military video games. While GPT-4 outperformed older AI models in formulating attack and defense strategies, researchers found it also caused significantly more casualties to achieve mission objectives. Why it matters: as militaries accelerate AI adoption, models that prioritize ruthless efficiency highlight mounting ethical and safety risks.
A new bipartisan advocacy group, Americans for Responsible Innovation, just launched to push a pragmatic framework for AI governance. Instead of reinventing the wheel, the group wants regulators to adapt proven playbooks from nuclear energy, cybersecurity, and banking—such as creating an auditing oversight board and restricting cloud training access.
Microsoft just cited the humble VCR in a motion to dismiss key claims in The New York Times’ copyright lawsuit. The company argued that large language models are like video recorders or personal computers—technologies with substantial lawful uses that shouldn't be outlawed due to potential misuse. Why it matters: If accepted, this classic defense could protect the entire generative AI sector from crushing copyright liability.
Perplexity is finalizing a new funding round that will push the AI search startup to a $1 billion valuation, according to The Wall Street Journal. That nearly doubles its $520 million valuation from just two months ago, alongside topping $10 million in annual revenue. Investors are clearly eager to back a challenger to Google's search empire.
DeepMind alumni Yishu Miao and Ziyu Wang just launched Haiper, a fresh AI video platform backed by $13.8M in seed funding. Users can currently generate 2- to 4-second video clips or animate still images for free on its site. Haiper plans to build out a core model for developers and even explore open sourcing, taking directly on OpenAI's Sora.
OpenAI just published a sharp rebuttal to Elon Musk's lawsuit, releasing old emails to push back against his claims. The records show Musk once pushed to merge OpenAI into Tesla, demanded controlling equity, and explicitly agreed that building AGI didn't require open-sourcing the science. Why it matters: OpenAI is dismantling Musk's narrative before the court, showing he backed their pivot until he walked away.
Stability AI and Tripo AI just released TripoSR, an open-source model that transforms a single 2D image into a textured 3D mesh in under half a second. It runs fast even without a dedicated GPU, and both the code and weights are completely free for commercial use under an MIT license.
Sheffield doctors just deployed an AI tool that predicts kidney failure in under a minute, beating human analysts by six times. For patients with polycystic kidney disease, measuring kidney volume across 60 MRI slices used to take an hour of manual tracing; the algorithm now automates it with clinical-grade accuracy. Why it matters: ADPKD leads to irreversible kidney damage, and faster forecasting gives clinics crucial lead time for intervention before dialysis is required.
Anthropic just introduced the Claude 3 family, rolling out Haiku, Sonnet, and its flagship Opus model. Opus outperforms rival models across major reasoning and coding benchmarks, offering multimodal vision capabilities and a 200K context window. It sets a new bar for frontier models, with Opus and Sonnet live now and Haiku coming soon.
OpenAI just rolled out 'Read Aloud' to ChatGPT, letting users listen to generated answers across iOS, Android, and the web. On mobile, you simply press and hold an answer to play audio, while the web version adds an on-screen button under the output. It gives users a quick hands-free listening option on the go.
California regulators just gave Waymo permission to expand its driverless robotaxi service to highways in Los Angeles and the Bay Area at speeds up to 65 mph. The approval brushes aside objections from local transport agencies over recent safety scares, though Waymo says it has no immediate plans to send cars onto the highway right away.
AI chipmaker Groq just acquired analytics startup Definitive Intelligence for an undisclosed sum to scale its on-demand platform, GroqCloud. Definitive Intelligence previously helped build the cloud infrastructure, which lets developers access Groq's ultrafast LPUs on demand. Its CEO Sunny Madra will now take charge of GroqCloud, alongside a new Groq Systems division targeting government and enterprise deployments.
Motorola just teased its upcoming Moto X50 Ultra, positioning the F1-inspired device as its first true AI-powered phone to rival Samsung's Galaxy S24. The teaser hints at a reveal tied to the F1 China Grand Prix on April 21, sporting a leatherette back. Why it matters: Phone makers are rushing to make artificial intelligence the headline feature of 2024 flagships.
Elon Musk just filed a lawsuit against OpenAI and Sam Altman, claiming the startup breached its founding agreement by chasing profits instead of benefiting humanity. Musk alleges OpenAI has become a closed-source subsidiary of Microsoft and asks the court to declare GPT-4 an AGI to block commercial licensing.
Microsoft just rolled out Copilot for Finance in public preview, plugging ERP data from SAP and Dynamics 365 directly into Excel and Outlook. Analysts can now run variance checks with natural-language prompts, automate ledger reconciliations, and pull customer collection summaries inside email. Why it matters: it targets the tedious data entry that bogs down corporate finance teams.
Google and Stack Overflow just announced a partnership powered by the new OverflowAPI. Google will integrate Stack Overflow's verified developer Q&A directly into Gemini for Google Cloud and its console, citing sources so coders can double-check answers. Meanwhile, Stack Overflow plans to build AI features on Google's Vertex AI.
Microsoft just detailed DirectSR, a new Windows API built with Nvidia, AMD, and Intel to streamline AI super resolution across PC games. Instead of writing custom integrations for DLSS, FSR, and XeSS separately, developers can now support all three through a single code path. It rolls out soon via an Agility SDK preview.
Humanoid robot maker Figure AI just raised $675 million at a $2.6 billion valuation from Big Tech heavyweights, including Jeff Bezos, Nvidia, Microsoft, and OpenAI. Along with the cash, Figure signed a deal with OpenAI to develop next-generation multimodal models for its robots. Why it matters: the tech giants are officially racing to put frontier AI into physical bodies.
Apple CEO Tim Cook just told shareholders that the company will "break new ground" in generative AI later this year, right as reports confirmed Apple shuttered its electric car project to shift staff to AI. The move signals an aggressive push to catch up with rivals by baking generative intelligence directly into Siri and built-in system tools.
Lightricks just revealed LTX Studio, a platform that turns a single concept into a complete, multi-scene video with AI. It targets major pain points like maintaining consistent characters and lighting across cuts, aiming to make end-to-end AI filmmaking actually practical.
Morph Studio just partnered with Stability AI to launch an all-in-one AI video creation workflow. The platform allows creators to evolve text, images, or existing clips into stories using multi-parallel generation lines. A waitlist is now live for early access.
Apple just pulled the plug on Project Titan, its decadelong autonomous car initiative, moving most of the team over to generative AI. After years of shifting strategies, Wall Street welcomed the exit as a smart pivot back to software and disciplined spending.
Adobe just unveiled Project Music GenAI Control, a prototype that lets creators generate and customize tracks using text prompts or reference melodies. Beyond raw generation, users can fine-tune tempo, structure, and create seamless loops. It is strictly in the research phase right now with no UI or release date yet.
Ideogram just launched Ideogram 1.0, upgrading its text-to-image model with better photorealism, tighter prompt adherence, and cleaner in-image text. It also rolls out Magic Prompt to help refine your descriptions. For anyone making posters, logos, or memes, getting legible text into images just got much easier.
Klarna just revealed that its OpenAI-powered customer service bot handled 2.3 million chats in its first month, doing the equivalent work of 700 full-time reps and cutting resolution times from 11 minutes to 2. Klarna laid off 700 people back in 2022, and while the company denies a direct connection, putting an exact headcount on AI replacement is sparking plenty of backlash.
Meta CEO Mark Zuckerberg just met with LG Electronics leadership in Seoul to discuss teaming up on next-gen extended reality devices and on-device AI. Facing fresh heat from Apple's Vision Pro, Meta is looking to marry its large language models with external hardware muscle. Zuckerberg's South Korea itinerary also includes talks with Samsung and President Yoon.
