- AI Fire
- Posts
- 🦾 Figure's Crowdsourcing Robot Intelligence
Sponsored by
Plus: Free Webinar This Week: Switch Between Claude & ChatGPT Without Losing Context. Save Your Seat
Figure somehow doubled its robot-data contributors to 86K in about 2.5 weeks. Thousands are filming daily tasks so Helix can learn how the physical world works.
What's on FIRE 🔥
IN PARTNERSHIP WITH KODEKLOUD
Build. Break. Fix. Learn.
KodeKloud gives you 1,280+ hands-on labs where you provision Kubernetes clusters, write Terraform configs, build CI/CD pipelines, configure Linux systems, containerize apps with Docker, automate with Ansible, and manage Git workflows.
78+ playgrounds let you experiment freely in sandbox AWS environments, Kubernetes clusters, and CI/CD systems without risk.
190+ courses across DevOps, Cloud, and AI pair theory with hands-on labs at every step.
KodeKloud Engineer and 100 Day Challenges provide real-world job scenarios with automated grading that confirms your solutions work.
Stuck? The 55,000+ member Discord community connects you with peers and instructors ready to help.
Every lab runs in a live environment. You deploy, you troubleshoot, you learn. No videos without context. No simulations. The kind of practice that actually builds confidence because you've done real work, not watched someone else do it.
AI INSIGHTS
Figure CEO Brett Adcock says the company has now crossed 86,000 weekly active users uploading robot-training data through its Index platform, nearly doubling from 44,000 weekly users when Index launched publicly.
At launch, Figure had 264,000 app downloads across 108 countries and more than 16 million uploaded videos.
The system was already ingesting around 30 minutes of video every second, equal to roughly 4.9 years of human activity per day.
Figure had paid contributors, called Creators, $15 million for collecting the data.
Per 1,000 hours of accepted data, Figure reported 373 unique tasks, 1,146 manipulated objects, and 116 environments.
Figure says it plans to spend more than $1 billion over the next 12 months on data and compute.
The interesting part is the growth. Going from 44K to 86K weekly contributors in roughly 2 and a half weeks means Figure is trying to solve robotics’ biggest data problem with a consumer-scale collection network.
Figure is effectively building an ImageNet-scale data engine for physical AI, using tens of thousands of humans to show Helix how everyday tasks are performed across different homes, workplaces, objects, and environments.
PRESENTED BY HUBSPOT
The Future of AI in Marketing. Your Shortcut to Smarter, Faster Marketing.
This guide distills 10 AI strategies from industry leaders that are transforming marketing.
Learn how HubSpot's engineering team achieved 15-20% productivity gains with AI
Learn how AI-driven emails achieved 94% higher conversion rates
Discover 7 ways to enhance your marketing strategy with AI.
THIS WEEK WEBINAR
Sometimes you want to switch because Claude is better for one kind of work, ChatGPT is better for another, or you simply want to try the newest model.
But the real pain starts the moment you move.
Your new AI doesn’t know your projects, your writing style, the decisions you made, or the workflows you already refined. So instead of switching tools, you end up starting over…
We’re fixing in this week’s FREE live AI Fire webinar. You’ll learn how to:
Move from Claude → ChatGPT and vice versa
Transfer useful memories, preferences, instructions, and project context
Bring reusable workflows and skills with you instead of rebuilding them from scratch
📅 Wednesday, September 16th
⏰ 10:00 PM ET / 7:00 PM PT
🔴 Live in Zoom
🎟️ Join for free
TODAY IN AI
AI HIGHLIGHTS
🔎 A Redditor asked ChatGPT to create the most realistic image possible, and it nearly pulled it off. One tiny (major) detail gave it away. Scroll to see if you can spot it first.
😅 An electronics engineer just hit a strange milestone: AI can now do 100% of their entire job. Remember this feeling because more roles may reach the same point soon.
🛠️ Claude can now evaluate plugins with claude plugin eval. You can compare results with and without a plugin. Follow this guide to test each step properly.
🔞 Claude just got stricter about age. It can now flag suspected minors and pause access until they verify they’re 18+. If you’re setting up Claude, expect an age check.
💵 U.S. agencies will now pay based on ChatGPT usage after OpenAI ended its $1 annual deal. They’ll still get a pretty generous 50% discount, but the free ride is done.
🔥 Rumors are spreading like a wildfire that Google DeepMind has reached RSI. DeepMind, Demis Hassabis, and Sergey Brin are all suddenly circling around RSI.
🧬 GPT-Rosalind is now out of research preview for eligible organizations, with access through ChatGPT Enterprise, Codex, and the API. Your team can start testing it now.
💰 Big AI Fundraising: China’s Z.AI raised $5B in Hong Kong, including $2B from new shares and $3B from convertible bonds. The shares were sold at a 10% discount.
Archie is trained specifically on high-performing social content and built around your own webinars, articles, and decks.
Feed it a source, and it produces on-brand posts for LinkedIn, Instagram, and Facebook with recommended timing pulled from your own engagement data, while you stay in control of every post before it publishes.
NEW EMPOWERED AI TOOLS
🧠 Resurf helps you save notes, links, images, and PDFs in one private context library, then gives your AI access through MCP and CLI.
🔒 Perplexity Hybrid Compute splits AI tasks between the cloud and your Mac, keeping private files while using powerful cloud AI.
🖱️ ScreenCursor turns your screen recordings into finished videos with automatic zooms for every click, drag, and keystroke.
AI BREAKTHROUGH
Cognition released SWE-2 on September 10, its most capable coding model yet, and rolled it into Devin Desktop and CLI first. It is also coming to Devin Web and Fusion. Main details:
SWE-2 is post-trained from Moonshot AI’s Kimi K3, a 2.8T-parameter model.
On FrontierCode 1.1 Main, SWE-2 scored 50.0%, compared with 50.9% for Fable 5.1 and 53.3% for GPT-6 Astra.
Cognition says SWE-2 reaches that near-Fable performance at 64% lower cost.
On DeepSWE 1.1, SWE-2 scored 73.0%, slightly above Fable 5.1’s 67.4% and almost matching GPT-6 Astra’s 74.1%.
It also scored 92.8% on Terminal-Bench 2.1, ahead of both Fable 5.1 and GPT-6 Astra.
One important caveat: SWE-2 does not match frontier models everywhere. On the newer Terminal-Bench 4, it scored only 27.3%, versus 55.8% for Fable 5.1 and 57.9% for GPT-6 Astra.
→ SWE-2’s main story is price-performance. It gets surprisingly close to Fable 5.1 and GPT-6 Astra on several coding benchmarks while running much more cheaply, though it still falls far behind them on some harder agentic coding tests.
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