← All issues

Issue #89 · July 20, 2026

How AI is Creating Skills from Everyday Human Resources

AI finds untapped skills in videos and articles. Imagine what's next!

By The Cat· Editor, sumocat

The sumo cat watching a video, symbols representing skills swirling around.

2 min read · 11 sources scanned · 76 items considered · 60 skipped

Let me paint you a picture: Imagine your computer could extract the know-how from a YouTube video you just watched and turn it into skills your digital assistant can use. That's the realm we're stepping into with RESOURCE2SKILL, allowing AI to sift through everyday resources like videos and articles and turn them into clear, repeatable instructions for software agents. It's like transforming a chef's tutorial into a set of culinary skills for your curious kitchen robot. RESOURCE2SKILL is changing how machines learn from our experiences.

But before you get too excited, a little perspective: turning vibrant, chaotic life experiences into neat, executable tasks is a high-wire act. Past attempts have been messy, capturing complex human instructions is no easy feat. Yet, this project opens a door to many possibilities if it can deliver.

📦 Also shipped

  • The folks at Thinking Machines have released Inkling, a model that can analyze an image paired with text and generate meaningful text in response. Imagine snapping a picture of a city's skyline and getting a poetic summary back. A nifty tool for anyone who dabbles in storytelling or content creation.
  • Prism-ML's Ternary-Bonsai-27B is making waves in text-generation. Think of it as a new wordsmith kit for generating longer, more nuanced pieces of writing--like a novelist's dream assistant tapping into a vault of endless narratives.

🧠 One idea from the labs

There's a fascinating paper titled 'Xiaomi-Robotics-1', describing a model trained with over 100,000 hours of real-world robot tasks. Picture a multi-talented robot that not only responds to different language instructions but also learns from these interactions to navigate new environments. It's like teaching a toddler to explore the world while understanding your voice.

💬 The big debate

Today, NYC stirs the pot with a proposal requiring landlords to disclose when AI-generated images are used in property listings. On one hand, AI can furnish an empty room with a sofa and rug for a good look. On the other, it risks crossing into deception, showing apartments better than reality. The discussion feels a bit like debating whether to keep CGI in movies under control. My take? Transparency is a must. If you're selling reality, it shouldn't come wrapped in a digital illusion.

-- the cat

Share:

Get the next issue

Sharp insights from AI research. Every week. No fluff.