Find out why 1M+ professionals read Superhuman AI daily.

In 2 years you will be working for AI

Or an AI will be working for you

Here's how you can future-proof yourself:

1. Join the Superhuman AI newsletter – read by 1M+ people at top companies

2. Master AI tools, tutorials, and news in just 3 minutes a day

3. Become 10X more productive using AI

Join 1,000,000+ pros at companies like Google, Meta, and Amazon that are using AI to get ahead.

Sign up and start learning AI

Today:

• Llama 4 Boasts Massive Context

• OpenAI Considers Buying Altman, Ive Startup

• Runway Launches Gen-4 Turbo Model

• Google Adds Multimodal Search Feature

• ElevenLabs Launches MCP Voice Server

Llama 4 Meta's "NEARLY INFINITE" Context Window... (also Reasoning?)

Meta has released Llama 4, a new family of AI models aimed at reclaiming dominance in open-source AI. The standout is Llama 4 Scout, offering an unmatched 10 million-token context window, far surpassing rivals. 

Other versions—Maverick and Behemoth—excel in multimodal tasks and reasoning, with strong performance-to-cost ratios. Despite “open-source” claims, the models have restrictive licenses, especially in the EU. More details are expected at Llama Con on April 29.

WATCH THE VIDEO ON YOUTUBE

OpenAI Considers Buying Altman, Ive Startup

OpenAI has discussed acquiring a startup co-created by Sam Altman and ex-Apple designer Jony Ive to build a new AI device. The gadget, possibly a screenless phone or smart home assistant, aims to deliver sci-fi-style voice interaction. The project, called io Products, could cost over $500 million. This move would deepen OpenAI’s push into consumer hardware and could position it in direct competition with Apple and other tech giants.

Why it matters

1. Hardware Shift: Signals OpenAI’s serious move beyond software into physical devices.
   

2. User Experience: Could redefine how people interact with AI—more naturally, without screens.
   

3. Big Tech Competition: Raises stakes in the AI assistant race, potentially challenging Apple, Google, and Meta.

Runway Launches Gen-4 Turbo Model

Runway has launched Gen-4 Turbo, a faster, cheaper, and more efficient AI video model that creates 10-second videos in just 30 seconds. It’s five times quicker and costs less than half as much to run compared to Gen-4. Users report better color consistency, motion detail, and responsiveness. The release has sparked excitement, with creators hoping for future upgrades like generating starting frames and expanding video length.

Why it matters

1. Creative Tools Boost: Makes AI video creation more accessible and affordable.
   

2. Speed Leap: Drastically shortens production time for content creators and studios.
   

3. Quality Upgrade: Improves realism and control, bringing AI video closer to professional standards.

Google Adds Multimodal Search Feature

Google is expanding AI Mode, now adding multimodal search that lets users snap or upload images and ask questions about them. This feature combines Google Lens with Gemini AI to understand entire scenes, including object details and relationships. Users get rich, contextual answers with links and follow-up options. Early testers praised its fast, helpful results for complex questions. AI Mode is now available to more users via Google Labs in the U.S.

Why it matters

1. Multimodal Leap: Combines text and image understanding, making AI more intuitive and powerful.
   

2. Richer Search: Enhances how people explore and learn visually, beyond traditional keyword searches.
   

3. AI for Everyday Use: Brings cutting-edge AI into daily mobile tasks, boosting mainstream adoption.

🧠RESEARCH

Multi-SWE-bench: A Multilingual Benchmark for Issue Resolving

Multi-SWE-bench is a new dataset to test AI on fixing code issues across seven programming languages—not just Python. It includes over 1,600 expert-reviewed examples and supports future reinforcement learning efforts. The project is open-source to help grow community contributions and push AI toward broader problem-solving skills.

MegaMath: Pushing the Limits of Open Math Corpora

The MegaMath project introduces a massive, open dataset built to train AI in math reasoning. It combines cleaned web content, math-related code, and AI-generated examples. With 371 billion tokens, it’s the largest high-quality math dataset available, aiming to improve how language models understand and solve complex math problems.

Agentic Knowledgeable Self-awareness

KnowSelf is a new method that helps AI agents act more like humans by using knowledge only when it's truly needed. Instead of blindly applying all information, agents learn to judge situations and decide wisely. This leads to better performance with fewer resources across planning tasks.

🛠️TOP TOOLS

Mindgrasp AI Free - Transform class lectures and readings into study tools with AI powered summaries, detailed notes, flashcards, quizzes, and an interactive AI-tutor trained on your course material.

ColorMagic - Generate color palettes with AI

HuxliAI - Designed to enhance and streamline academic work for students, leveraging advanced AI technologies to provide comprehensive support.

Amper Music AI - Generate a wide array of music types.

Conker AI - Designed to streamline the process of creating quizzes and assessments for teachers, particularly in the K-12 education sector.

📲SOCIAL MEDIA

🗞️MORE NEWS

• ElevenLabs launched its MCP server, letting apps like Claude use its AI voice tools. It enables tasks like AI phone calls and helps with coding. Developers see it as a step toward unified AI systems.

• IBM has acquired AI consultancy Hakkoda to boost its data and AI services for sectors like healthcare and finance. The move supports IBM’s growing focus on AI, following similar recent acquisitions.

• Google’s Gemini Live now supports screensharing on Pixel 9 and Galaxy S25 phones, with more devices coming soon for paid users. The feature lets users ask AI questions about what’s on-screen or in view.

• A Meta executive denied claims that its new Llama 4 AI models were trained on test sets to boost benchmark scores. The rumor followed mixed model performance and differences between public and benchmarked versions.

• Amazon’s updated AI video model, Nova Reel 1.1, can now create videos up to two minutes long with consistent style and shot control. It’s available on AWS, though questions remain about how its training data was sourced.

• The White House has ordered U.S. agencies to name chief AI officers and create AI strategies, removing Biden-era rules aimed at oversight. The new policy promotes broader, faster adoption of AI across government operations.