Significant progress in AI and Robotics this week.
So, I summarized everything from Google, Microsoft, Alibaba, Figure, Cohere, Unitree, Field AI, Boston Dynamics, Deepseek, and more.
Here's everything you need to know and how to make sense out of it:
Google updated multiple product lines with new AI features, including:
—Pixel phones with 20+ AI features, like magic cues and live call translation
—Agentic capabilities in Search's AI Mode for AI Ultra subscribers
—Gemini for Government at $0.5/agency
Microsoft added a new Copilot function in Excel
This allows users to generate analysis, summaries, data classifications, and more by typing =COPILOT() in their cells
Microsoft has now officially caught up with Google in delivering agentic spreadsheets
Alibaba continued to ship, dropping two new products:
—Qwen-Image-Edit, an AI model for conversational editing of photos
—Qoder, an agentic coding platform that's 100% free to use; currently in public preview
Figure update: we have started making the walking of our humanoids more robust with the new Helix walking controller
As seen in this clip, it enabled our robot to walk across different surfaces without actually using the cameras to see what lies ahead
Cohere launched Command A Reasoning, a new reasoning model for enterprises
It outperforms similar privately deployable rivals, like gpt-oss and DeepSeek R1, on agentic, deep research and safety benchmarks
Field AI raised $405M in funding and introduced Field Foundation Models
These models are designed to grapple with uncertainties and the physical constraints of the real world, enabling safe robot behaviors when navigating in new environments
Boston Dynamics and Toyota Research Institute debuted Large Behavior Models (LBMs)
These policies enabled Atlas to perform multi-step tasks like folding Spot legs, rope tying, or tire manipulation, while adapting to errors in real time
Deepseek launched Deepseek V3.1, an updated hybrid reasoner
It reaches answers in less time vs. DeepSeek-R1-0528 and promises stronger performance in tool use and multi-step agent tasks
On SWE-Bench, it scored 66%, nearing SOTA closed models
South Korean company WIRobotics unveiled ALLEX, a general-purpose humanoid
Built as a modular platform, ALLEX focuses on manipulation with highly dexterous hands, with 15 degrees of freedom (DoF) across fingers and wrist
It can also lift over 30kg
Daxo Robotics demonstrated its flexible hand, where each joint is driven by multiple tendons for robust control, in a new podcast
Each hand uses 108 motors, enabling 108 actively controlled degrees of freedom and infinite passive DoF
Researchers at Stanford debuted HEAD, an autonomous navigation and reaching system
It learns navigation, locomotion, and reaching directly from human motion and vision data, allowing humanoids to move around obstacles and touch an object if needed
IBM and NASA introduced Surya, an open-source AI to predict solar flares
Trained on 9 years of high-res imagery from NASA's SDO, it predicts solar outbursts faster than ever before and beats benchmarks by 16%
We're hiring for BotQ, our new manufacturing division at @Figure_robot:
> Software Test Eng
> End of Line Lead
> Equipment Eng
> Technician
> Manufacturing Eng
I summarized everything from OpenAI, xAI, Google, Meta, Figure, Nvidia, Tencent, Unitree, Weave Robotics, Engine AI, and more.
Here's everything you need to know and how to make sense out of it:
OpenAI updated ChatGPT after GPT-5 drew backlash from users. Key changes:
—Return of GPT-4o for paid users
—Auto, Fast, and Thinking modes for GPT-5
—3K messages/week for GPT-5 Thinking with extra Thinking mini capacity
—A 'warmer' GPT-5 personality
Significant progress in AI and Robotics this week.
So, I summarized everything from OpenAI, Google, xAI, Anthropic, Figure, Unitree, OpenMind, Microsoft, Perplexity, ElevenLabs, and more.
Here's everything you need to know and how to make sense out of it:
OpenAI had a huge week, announcing:
—New flagship models: GPT-5, GPT-5 Pro, and GPT-5 Mini
—gpt-oss-120B and 20B: OAI's first open models since GPT-2
—New personalities for ChatGPT
—Deal with the U.S. govt to offer ChatGPT at $1/agency
Google DeepMind released Genie 3
It's a model capable of generating playable worlds in real-time, with consistency and visual memory up to 1 min
Major push for gaming and entertainment, as well as a new foundation for training physical AI
I summarized everything from Meta, Google, OpenAI, Figure, Microsoft, Z ai, Skild AI, Limx Dynamics, Syncere, Daxo Robotics, X-Humanoid, and more
Here's everything you need to know and how to make sense out of it:
Zuckerberg shared that Meta's new AI lab will develop "personal superintelligence" to empower individual goals, not automate work
He said “devices like glasses” will be key to computing of the future, also hinting at a shift from the "open" strategy
Google made big product announcements:
—AlphaEarth Foundations for mapping Earth in near real-time
—Video Overviews to convert files into narrated video slides
—SensorLM AI to convert sensor-captured health signals into natural language
Significant progress in AI and Robotics this week.
So, I summarized everything from Meta, xAI, Microsoft, OpenAI, Google, Alibaba, Sapient Intelligence, Unitree, and more.
Here's everything you need to know and how to make sense out of it:
Meta Reality Labs' research on a new AI-powered neuromotor interface appeared in Nature
The band-based system allows users to control devices using simple hand/finger gestures, tapping muscular signals produced during movement
Elon Musk gave a glimpse of xAI's upcoming Colossus 2 AI supercomputer
Located in Tennessee, the data center will be powered by 550K NVIDIA GB200 and GB300 GPUs, delivering AI compute equivalent of 50M units of Nvidia’s H100 in 5 years