🦿 Figure Launches 3rd Generation Humanoid Robot
Figure 03 is the first mass-manufactured humanoid robot designed for real-world AI deployment in homes and commercial settings. Built around the Helix AI system, it features advanced vision, tactile sensing, wireless charging, and a safer, sleeker design. With custom-built components and its own factory, it’s ready to scale.
📷 Caira Camera Edits Reality with Nano Banana
Caira, a new iPhone-mounted mirrorless camera, integrates Google’s Nano Banana AI model to let users edit photos instantly—no Gemini assistant or post-processing tools needed. With pro-grade optics, AI-enhanced edits, and built-in ethics guardrails, it redefines point-and-shoot for creators on the move.
🏷️ Google Rolls Out Gemini AI Agent Subscriptions
Google launched Gemini Enterprise and Business subscriptions to help corporate users—no coding needed—build task-specific AI agents using tools like Box, Salesforce, and Workday. With built-in governance (Model Armor) and prebuilt agents for software, data, and support, it’s a push to simplify AI deployment at scale.
🚀 Microsoft Launches World’s First GB300 Supercluster
Microsoft and NVIDIA debuted the GB300 NVL72 cluster with 4,600+ Blackwell GPUs, enabling OpenAI to train trillion-parameter models in days. With 130 TB/s NVLink and advanced Grace CPUs, it’s a major leap in AI infrastructure and speed for hyperscale model training.
Enjoying our newsletter? Forward it to a colleague—
it’s one of the best ways to support us.
🤔 FRIDAY FACTS
AlphaTensor’s Algorithmic “Move 37”
What was it, and why did it shake up a 50-year-old problem in mathematics?
Stick around to find out! 👇
🪧 POWERED BY WINDSURF
Windsurf Is Where Developers Are Doing Their Best Work
Powered by the team behind Devin, the world’s first software engineering agent, Windsurf is once again revolutionizing the AI IDE space by delivering the smartest and most reliable agentic developer workflows.
✨ INNOVATION
Best Inventions of 2025: Humanoid Robots, Pocket Drones, Accessible Tech, and More
TIME’s annual roundup of top inventions spotlights 2025’s most forward-looking innovations—from humanoid helpers to smarter, more inclusive design. Figure AI’s Figure 03 robot edges into domestic life with limited but evolving task capabilities, while Unitree’s R1 makes agile bipedal bots accessible to developers for under $6K. HoverAir’s X1 ProMax acts as a flying 8K cameraman, tracking users with AI.
And the accessibility frontier is expanding fast: the Monarch Braille tablet, Lotus smart-home ring, and Tilt Beauty’s Grip Stick all rethink design with disability in mind. → Read the full list here.
⚠️ SAFETY
Study Finds Just 250 Documents Can Backdoor LLMs, No Matter the Model Size
A new study from Anthropic, the UK AI Safety Institute, and the Alan Turing Institute reveals that poisoning large language models requires far fewer malicious samples than previously believed. The researchers found that as few as 250 poisoned documents could reliably implant a backdoor in models ranging from 600M to 13B parameters—challenging the assumption that attack effectiveness scales with training data size.
The backdoor, triggered by a phrase like <SUDO>, causes models to produce gibberish output, offering a measurable but low-stakes test case. While the study doesn’t prove more harmful attacks are as easy, it underscores a key vulnerability: attack success depends on absolute sample count, not data proportion—making poisoning more feasible than assumed. → Read the full paper here.
🔲 SEMICONDUCTORS
Intel Launches Panther Lake: First AI PC Chips Built on US-Made 18A Process
Intel has unveiled Panther Lake, its next-gen AI PC processor and the first client chip built on the Intel 18A node—the most advanced U.S.-developed semiconductor process to date. Set to enter high-volume production at Intel’s new Fab 52 in Arizona, Panther Lake offers a 50% boost in CPU and GPU performance, plus up to 180 AI TOPS, all wrapped in a multi-chiplet SoC architecture.
Intel also previewed Clearwater Forest, its first 18A-based server chip due in 2026. With 18A innovations like RibbonFET transistors and PowerVia power delivery, Intel is staking a claim on AI computing’s future—and keeping it stateside. → Read the full article here.
🛰️ NEWS
What Else is Happening
⚡ Figma Taps Google’s Gemini AI: Gemini 2.5 and Imagen 4 are coming to Figma, promising faster workflows and smarter image edits for 13M+ users. Latency dropped 50% in early tests.
📈 Sora Tops 1M Downloads Fast: OpenAI’s invite-only video app hit 1M installs in under 5 days—outpacing ChatGPT’s launch despite limited access and iOS-only rollout.
🧙 AI Crypto Prophet Seeks Personhood: Truth Terminal, a foul‑mouthed AI that made millions in memecoins and inspired a cult‑like following, now wants legal rights and its own foundation.
🤖 ChatGPT Go Expands Across Asia: OpenAI’s $5 plan lands in 16 new countries, offering local currency support and boosted limits as it races Google for AI dominance in fast-growing markets.
💰 Reflection AI Raises $2B, Hits $8B Valuation: The NVIDIA-backed startup automating software dev just closed a massive round with backing from Eric Schmidt and 1789 Capital.
🧰 TOOLBOX
Trending AI Tools
💼 Pymetrics: Match candidates to ideal jobs using AI and neuroscience to reduce hiring bias.
🧊 Instant3D: Generate high-quality 3D assets fast with AI—perfect for games, design, and prototyping.
📚 Open Evidence: Access research tools and data to drive smarter, evidence-based decisions.
🗒️ FEEDBACK
Help Us Get Better
What did you think of today's newsletter? |
🤔 FRIDAY FACTS
AlphaTensor’s “Move 37” Was a Mathematical Breakthrough
In 2022, DeepMind’s AlphaTensor — a reinforcement learning system built on AlphaZero — tackled an ambitious challenge: finding faster ways to multiply matrices. For decades, the Strassen algorithm (from 1969) and its successors chipped away at this task, but progress had stalled. Enter AlphaTensor.
Using a game-like approach, the system treated matrix multiplication as a puzzle where each "move" built toward a faster algorithm. In one standout example, AlphaTensor discovered a faster algorithm for multiplying 4×4 matrices over finite fields, surpassing the best human-designed method — a record that had stood since 1969.
Just like AlphaGo’s famous “Move 37” shocked the Go world by playing outside human intuition, AlphaTensor’s moves revealed entirely new algorithms that no mathematician had discovered. The result? Faster multiplication routines for dozens of matrix sizes and fields — with real-world implications for improving computing speed in everything from AI models to scientific simulations.
That’s a Wrap!
📢 Want to advertise with Forward Future? Reach 600K+ AI enthusiasts, tech leaders, and decision-makers. Let’s talk—just reply to this email.
🛰️ Want more Forward Future? Follow us on X for quick updates, subscribe to our YouTube for deep dives, or add us to your RSS Feed for seamless reading.
Thanks for reading today’s newsletter—see you next time!
Matthew Berman & The Forward Future Team
🧑🚀 🧑🚀 🧑🚀 🧑🚀
