New AI video model takes over the internet

Welcome back, Superhuman. You may or may not remember that we covered a new video generation model coming out of China a few days ago. That model is now taking over the internet. We may finally have a video generation model that matches Hollywood-grade productions.

Today: AI video takes a big leap, how to create your own talking head videos with AI, and get the latest prompts and trending social posts.

TODAY IN AI

1. AI-generated videos cross the uncanny valley as Seedance 2.0 gains traction: The new AI video model from TikTok maker ByteDance is taking over social media with a wave of hyper-realistic videos. While previous models experienced distortions like floating objects and unnatural movements, the new model from ByteDance looks very close to the real thing. Some examples: Lord of the Rings reimagined, a war combat scene, and a martial arts fight.

2. OpenAI debuts Codex-Spark, a speedier coding model powered by a new chip: OpenAI released GPT-5.3-Codex-Spark, a smaller, faster version of its agentic coding tool that delivers over 1,000 tokens per second. The model is the first to run on chips from Cerebras, an Nvidia rival, marking the first milestone in a $10B+ multi-year partnership between the two companies. It's available now as a research preview for ChatGPT Pro users. Watch GPT-5.3-Codex-Spark in action here.

3. Gemini 3 Deep Think gets a major upgrade: Google released a significant update to Gemini 3 Deep Think, its specialized reasoning mode, expanding it beyond math and coding into science, engineering, and research. The upgrade sets new benchmarks on key industry tests, while also achieving gold medal-level results on the 2025 International Physics and Chemistry Olympiads. The updated model can also now turn hand-drawn sketches into 3D-printable files. Available now for Google AI Ultra subscribers. Request early access or watch the improved model in action here.

PRESENTED BY ARTISAN

Meet Ava, the AI BDR

Ava monitors buying intent and launches multi-channel outreach in minutes. Leads land in your inbox on autopilot while Ava handles the grunt work.

She's able to:

• Prospect from 300M+ contacts

• Enrich leads from public data

• Run continuous tests to improve conversion

• Find leads via live intent signals (fundraises, job posts, website visitor de-anonymization, etc.)

• Import HubSpot lists/Salesforce views in one click

• Execute multi-channel outreach + manage deliverability

Trusted by teams at Quora, SumUp, and SaaStr.

Book a demo

FROM THE FRONTIER

AI’s next benchmark? Mastering a simple children's video game

Puzzling over Pokémon. AI models can ace the bar exam and solve PhD-level math problems, but there's still one challenge that stumps them: Pokémon. What started out as a fun experiment has gotten competitive — with Google, OpenAI, and Anthropic now battling it out to master the classic Game Boy game.

Why Pokémon? AI labs have often used games like Pong, chess, and Go to benchmark intelligence. But Pokémon is a next-level challenge. It requires sustained planning, resource management, and error recovery over hundreds of hours, not just a single match. Anthropic's David Hershey says Pokémon gives researchers "a great way to just see how a model is doing and to evaluate it in a quantitative way" — the exact capability that matters most as AI moves toward real-world agents.

The competition is live. On Twitch, you can watch Gemini 3 Pro, GPT 5.2, and Claude Opus 4.6 battle through different Pokémon games in real time. Gemini was first to finish, completing Pokémon Blue in roughly 406 hours. GPT has also beaten the game, and both models have now moved on to sequels. Claude is still working through Pokémon Red with 500+ hours logged and 170,000+ steps taken.

The bottom line. AI models can outperform doctors, lawyers, and PhD students on standardized tests, but they still struggle with many tasks that come easy to humans. It's a reminder that raw intelligence and real-world reasoning are two very different things.

THE AI ACADEMY

How to create your own talking head videos with AI

• Go to Gemini, upload your photo, and use the prompt below to create a professional headshot.

Sample Prompt: Use the attached selfie to create a high-resolution professional headshot that keeps my facial features exactly the same. Add studio lighting, a soft neutral background, and formal professional attire suitable for a confident, approachable LinkedIn profile.

• Then go to Wan AI , sign up, and click New Project.

• Change media type from Video → Avatar.

• Upload your headshot.

• Record 10–15 seconds of clear audio (or type your script and choose a voice).

• Click Generate to create your talking avatar.

• Add more short clips if needed, then click Export to download your final video.

PRESENTED BY YOU. COM

You. com Founders Predict an AI Winter Is Coming in 2026

Richard Socher and Bryan McCann are among the most-cited AI researchers in the world. They just released 35 predictions for 2026, including:

• Capital floods back to fundamental research

• "Reward engineering" becomes an in-demand job

• Traditional coding will be gone by December

Read all 35 predictions.

IN THE KNOW

What’s trending on socials and headlines today

🍝 Spaghetti Test: The Will Smith spaghetti test is a way to measure the realism of AI video models. Seedance 2.0 might’ve just broken it (5M views).

🎮️ Mission Control: This prompt creates a mission control for your OpenClaw, letting you monitor every single task it does for you.

👀 AI Renaissance: Since December, AI has experienced a “step function” in what’s possible — according to a viral post from OpenAI’s President (2M views).

💻️ Technical Co-Founder: This prompt converts Claude from an assistant into a technical co-founder to help you build tools, websites, and apps.

PRODUCTIVITY

5 New & Trending AI Tools

• 🗣️ Wispr Flow*: One mic for every app. Wispr Flow lets you dictate into Slack, Gmail, Notion, ChatGPT, and anywhere else you type. Clean text, instantly.

• 🐠 heyfish: Create hyper-realistic AI digital human videos in minutes.

• 📺️ Dokie: Create smart professional slides in minutes with AI.

• 🤝 WhiskAI: Blend subject, scene, and style images to create 4K artwork with AI.

• ✅ Good Assistant: Your productivity companion that helps you stay focused and consistently progressing toward your most important goals.

^{* indicates a promoted tool, if any}

PROMPT STATION

30-Day Goal Execution Prompt

Prompt: Act as a high-performance strategist and execution coach.
My goal: {Insert your goal}
Create a realistic but challenging 30-day action plan.

Structure it like this:

1. Define Success
Rewrite the goal clearly and measurably.
Define what success looks like after 30 days.
List 3 key metrics to track.

2. Weekly Plan (4 Weeks)
Break it into:
Week 1: Foundation
Week 2: Momentum
Week 3: Stretch
Week 4: Results
For each week include:
Clear objective
Specific actions
Milestone by end of week
Common mistakes to avoid

3. Daily Execution
Create a simple daily structure:
1 main priority task
1 growth/discomfort task
1 habit to maintain
1 reflection question

4. Accountability
Weekly review format
Simple scorecard
What to do if I fall behind

Be direct and practical. No vague advice. Give concrete actions only.

Friday Fun

Whenever you’re ready to take the next step

• Check out our Top 125 AI Tools

• Get better at prompting with our Top 1,000 Prompts

• Learn how to use AI at work and get certified with our free course

Grow customers & revenue: Join companies like Amazon, HubSpot, and Salesforce. Showcase your product to our 1M+ readers and 2M+ followers on socials. Get in touch.

Until next time — Zain, Theodore, & the Superhuman AI team