- AI Fire
- Posts
- 🦾 Physical Prompting for Robots is Here!
🦾 Physical Prompting for Robots is Here!
WOW, 100% Free university from Claude Academy!!!
Wendy
August 20, 2026
In partnership with
Robots may need way less training than we thought. Generalist’s GEN-1.5 learned tasks from 3-12 second demos, then jumped to 83% success with only 10 updates.
What's on FIRE 🔥
IN PARTNERSHIP WITH AGORAPULSE
If your LinkedIn reach has cratered this year, it's not you, it's the platform. Reach gave way to relevance, and likes now matter less than saves, sends and real comments.
Our free field guide covers what changed and how to adapt, no algorithm-cracking hype.
AI INSIGHTS
Generalist introduced GEN-1.5, a robot model that can learn a new physical task from a single 3–12 second demonstration and attempt it almost immediately.
Generalist calls this “physical prompting.” Instead of retraining the robot from scratch, a short demonstration is placed inside GEN-1.5’s 30-second context window, and it uses that example to guide its next action. Main findings:
Across 10 simple tasks, one demonstration produced 59% average success.
With 5 minutes of extra data and only 10 weight updates, success increased to 83%.
GEN-1.5 can imitate some demonstrations performed by a human hand.
It can learn from simulated demonstrations and transfer the behavior to a real robot.
The model can combine two physical prompts to handle a new task.
It can also improvise with tools it did not see during training.
There is also an important distinction between its 2 learning modes. The one-shot demo is in-context learning. The few-shot version actually updates the model, making it closer to continual learning.
🎁 Today's Trivia - Vote, Learn & Win!
Get a 3-month membership at AI Fire Academy (500+ AI Workflows, AI Tutorials, AI Case Studies) just by answering the poll.
What pushed GEN-1.5 to 83% success? |
WE WANT TO COLLABORATE WITH YOU
AI is moving fast, and some of the most valuable lessons come from people who are actually building, testing, and applying it.
We’re opening guest expert collaborations for upcoming AI Fire webinars.
If you have a practical AI workflow, real-world case study, unique perspective, or experience that can help others use AI better, we’d love to hear from you.
You bring the expertise. We’ll help shape the session, host the conversation, and introduce your ideas to the AI Fire community.
As an AI Fire guest expert, you’ll have the opportunity to:
🎤 Share your expertise with an AI-focused community
🌎 Reach professionals, builders, and AI enthusiasts across AI Fire channels
📣 Introduce your work, project, or resources to a new audience
🎥 Create a recorded session you can continue sharing afterward
Think your topic could be a great fit?
TODAY IN AI
AI HIGHLIGHTS
🤓 Claude Academy is now live with free courses & tutorials for everyone, from AI beginners to daily Claude users. Save this when you want a structured way to learn.
😂 OK, can we actually cool data centers with our pee? This Liquid Death ad joked about it, but the idea isn’t fully ridiculous for experts. Don’t try it yourself, though.
🌀 Grok is randomly sending gibberish to some users, with reports centered on Grok Lite. xAI calls it a rare and temporary glitch, so start a fresh chat if yours breaks.
💸 Replit launched Free Mode for paid users, so you can handle simpler tasks or everyday work with GPT-5.6 Luna for no credit cost. Bigger jobs can still jump to Sol.
💬 ChatGPT can now literally draft replies, and text people for you through Apple Messages plug-in. And please keep approval on before anything gets sent as you.
💻 Meta AI launched this Mac app with system-wide dictation, screen awareness, and business integrations. I’d test this before adding another productivity app.
💰 Big AI Data Boom: Micro1’s AI training data business jumped from a $100M to $500M gross annual run rate in just 8 months, showing how fast demand for high-quality AI data is growing.
NEW EMPOWERED AI TOOLS
📄 Meibel Document Intelligence lets you build AI agents that understand your documents, preserve tables and structure, and score every answer for confidence. Sign up free and get $25 in credits now!
⚡ Ramp Router gives developers one API to access and auto switch between top AI models based on cost, speed, or task difficulty.
🎮 Meta Pocket lets you vibe-code interactive games with AI prompts, then share, remix, and play creations in a social feed.
🤖 OpenBot is an AI coworker platform that works with any agent system, featuring generative UI, computer use, and human handoffs.
AI BREAKTHROUGH
Anthropic says Claude can now run parts of protein-design research largely on its own, showing how general-purpose AI could assist early drug discovery. Main findings:
Anthropic tested Mythos Preview and Claude Opus 4.8 on autonomous protein-design campaigns.
Claude produced working molecules for 14 of 15 targets.
Reported binding success rates reached 22–35%, above Anthropic’s cited industry baseline of roughly 10–15%.
Twist Bioscience and Adaptyv Bio handled the real-world lab synthesis and testing.
Claude generated candidates, evaluated results, revised ideas, and continued the research loop with minimal human direction.
In a separate experiment, Claude Opus 5 analyzed raw instrument files and estimated a sample at 96.4% purity in 19 minutes, while the lab’s normal report took 4 days.
Anthropic’s broader point is that Claude is moving beyond answering biology questions toward running multi-step scientific workflows. Its latest model evaluations also show stronger performance on biological design and prediction tasks.
Blu Dot surpasses 2,000% ROAS with self-serve CTV ads
Home furniture brand Blu Dot blew up on CTV with help from Roku Ads Manager. Here’s how:
After a test campaign reached 211,000 households and achieved 1,010% ROAS, the brand went all in to promote its annual sales event. It removed age and income constraints to expand reach and shifted budget to custom audiences and retargeting, where intent was strongest.
The results speak for themselves. As Blu Dot increased their investment by 10x, ROAS jumped to 2,308% and more page-view conversions surpassed 50,000.
“For CTV campaigns, Roku has been a top performer,” said Claire Folkestad, Paid Media Strategist, Blu Dot. “Comping to our other platforms, we have seen really strong ROAS… and highly efficient CPMs, lower than any other CTV partner we've worked with.”
Using Roku Ads Manager, the campaign moved from a pilot to a permanent performance engine for the brand.
We read your emails, comments, and poll replies daily
How would you rate today’s newsletter?Your feedback helps us create the best newsletter possible |
Hit reply and say Hello – we'd love to hear from you!
Like what you're reading? Forward it to friends, and they can sign up here.
Cheers,
The AI Fire Team
Reply