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:

• GPT-5 Catches Them All: Outpaces Grok 4.20 & Gemini 3.0

• Claude Opus Can Now End Chats in Rare Abuse Cases

• Meta Unveils DINOv3: Vision AI That Learns Without Labels

• OpenAI Staff Cash Out $6B in Shares at $500B Valuation

• ChatGPT Mobile App Hits $2B, Crushing AI Rivals

GPT-5 just caught them all (Grok 4.20 and Gemini 3.0)

Elon Musk and Sam Altman reignited tensions over app rankings. Grok 4.0 is coming, but GPT-5 already impresses, beating Pokémon Red with far fewer steps. Igor Babuschkin left xAI to launch Babushkin Ventures, focusing on AI safety.

Meanwhile, former OpenAI researcher Leopold Aschenbrenner’s new hedge fund is booming, betting on the AI boom. OpenAI’s model even won gold at the IOI.

WATCH THE VIDEO ON YOUTUBE

Claude Opus Can Now End Chats in Rare Abuse Cases

Anthropic gave Claude Opus 4 and 4.1 the ability to end chats in rare cases of repeated harmful or abusive user behavior. This experimental feature stems from research into possible AI welfare. Claude avoids conversations involving violence, abuse, or disturbing content. It only ends chats after trying to redirect multiple times. Most users won’t notice the change unless they directly ask Claude to stop.

Why This Matters

1. AI Welfare Research – Treating models as entities that may experience distress pushes the frontier of ethical AI development.
   

2. Alignment and Safety – Shows how models can be trained to avoid harmful tasks and protect both users and themselves.
   

3. Experimental Guardrails – Introduces a low-impact safeguard for extreme edge cases, balancing user freedom with responsible AI behavior.

Meta Unveils DINOv3: Vision AI That Learns Without Labels

Meta’s DINOv3 is a new computer vision model that learns from images without needing human labels. It beats many top models in tasks like object detection and segmentation. It’s trained on 1.7 billion images, works across domains like satellites and medicine, and doesn’t need fine-tuning. Smaller, efficient versions are also available. Real-world partners like NASA and WRI are already using it for robotics, climate monitoring, and environmental restoration.

Why This Matters

1. Label-Free Learning at Scale
   DINOv3 proves that self-supervised learning (SSL) can outperform label-dependent models, unlocking massive unannotated datasets across domains like healthcare, agriculture, and space.
   

2. Universal Vision Backbone
   A single model backbone can now handle multiple vision tasks without fine-tuning, making AI deployment cheaper, faster, and more versatile—especially valuable for edge and low-resource environments.
   

3. Open Access with Real-World Impact
   Meta’s open release of models, code, and tools empowers the research and developer community, while real deployments by groups like NASA and WRI show its practical, high-impact potential today.

OpenAI Staff Cash Out $6B in Shares at $500B Valuation

OpenAI staff are set to sell $6 billion in shares to investors like SoftBank, Thrive, and Dragoneer, valuing the company at $500 billion. This employee stock sale gives insiders liquidity without going public. SoftBank is also leading a $40B funding round at a lower $300B valuation. The move reflects OpenAI’s rapid growth—projected to hit $12.7B in 2025 revenue—and growing investor confidence following the GPT-5 launch.

Why This Matters

1. Skyrocketing Valuation Signals Dominance
   At $500B, OpenAI would become the world’s most valuable startup, surpassing even SpaceX—cementing its leadership in the AI race.
   

2. Massive Capital Fuels AGI Ambitions
   The funds and investor confidence will support OpenAI's plans to spend trillions on AI infrastructure, accelerating development of more advanced models.
   

3. Talent Retention in a Competitive Market
   Letting employees cash out helps OpenAI retain top researchers amid fierce poaching efforts from rivals like Meta, which directly affects AI innovation pace.

🧠RESEARCH

We-Math 2.0: A Versatile MathBook System for Incentivizing Visual Mathematical Reasoning

We-Math 2.0 is a system that helps AI models understand and solve math problems more effectively. It uses a detailed math knowledge map, challenging practice problems, and step-by-step training. This method improves the AI’s reasoning and performance across many topics, making it smarter and more reliable at math tasks.

NextStep-1: Toward Autoregressive Image Generation with Continuous Tokens at Scale

NextStep-1 is a large AI model that creates images from text and edits images with high accuracy. Unlike past models, it uses both word-like and smooth image pieces without losing details. It performs better than other similar models and will be shared openly to support further research.

ToonComposer: Streamlining Cartoon Production with Generative Post-Keyframing

ToonComposer is an AI tool that simplifies cartoon animation by combining two steps—filling in motion between frames and adding color—into one process. Using just a few sketches and one colored frame, it produces smooth, high-quality animation. It reduces manual work while giving artists more control and flexibility in production.

🛠️TOP TOOLS

Rose AI - Cloud-based data platform that revolutionizes financial data analysis and visualization through the power of generative AI

DupDub - AI-powered content creation platform designed to streamline the production of multimedia content

Gliglish - AI-powered language learning platform designed to help users improve their speaking and listening skills through interactive conversations

ClassPoint AI - Instantly generate engaging quiz questions from slide content

MyVocal AI - AI-powered platform that offers rapid voice cloning and text-to-speech capabilities

📲SOCIAL MEDIA

🗞️MORE NEWS

• ChatGPT’s mobile app has earned $2 billion since 2023, far surpassing rivals. It averages $2.91 per install, with 690 million downloads. Its 2025 revenue is $1.35 billion—over 50 times more than Grok’s.

• Tencent’s new Hunyuan-Large-Vision AI model ranks top in China, trailing only GPT-5 and Gemini 2.5 Pro globally. Built with 389 billion parameters, it excels at visual reasoning, translation, and 3D analysis, available via Tencent Cloud.

• NVIDIA released Granary, a massive open-source speech dataset with 1 million hours of audio, and two powerful AI models—Canary and Parakeet—for fast, accurate transcription and translation across 25 European languages, boosting global speech AI development.

• GPT-5 is being updated to sound “warmer and friendlier” after user feedback said it felt too formal. OpenAI claims internal tests show the tone change doesn’t increase flattery or fake agreement.

• Cornell PhD student Jack Morris modified OpenAI’s gpt-oss-20B model into a freer, less-aligned “base” version. It skips reasoning steps, bypasses filters, and outputs uncensored text—sparking praise and debate over safety and openness.

• AI can now detect early signs of vocal fold lesions by analyzing voice recordings for subtle changes. A new study found that variability in the harmonic-to-noise ratio may help identify throat cancers before symptoms appear.

• MIT researchers used generative AI to design two novel antibiotics that kill drug-resistant bacteria like MRSA and gonorrhea. These lab-tested compounds work differently than existing drugs and could fight superbugs more effectively.

• As AI threatens office jobs, blue-collar careers like plumbing, HVAC, and auto repair are gaining popularity for their hands-on skills and job security. Gen Z is increasingly choosing trades to avoid debt and automation risks.

• AI experts returning from China warn that the U.S. power grid is too weak to support rapid AI growth. China, by contrast, has overbuilt its energy infrastructure, giving it a major edge in the AI race.