*Tons of new updates in Artificial Intelligence.* So I summarized everything from DeepSeek, OpenAI, Figure, Unitree, Google, Microsoft, BridgeP, Alibaba, Mistral, Meta, and more. Here's everything you need to know and how to make sense out of it: 1. *Chinese startup DeepSeek dropped an open-source image-generation model, Janus-Pro* - It tops DALL-E 3 and StabIe Diffusion in image quality and accuracy benchmarks. Another big move for open source, regardless of the R1 controversy. 2. *OpenAI released its latest small reasoning AI, o3-mini* - Available via ChatGPT and API, the model is 24% faster than the o1-mini and 93% cheaper than the much larger o1 model. 3. *Alibaba's Qwen team added to the recent Chinese AI model hype with two big releases.* Qwen2.5-VLM performs computer-use (like Operator) while Qwen2.5-Max MoE model beats top models on key benchmarks like Arena-Hard & LiveBench 4. *Unitree's H1 robots performed folk dance at China’s Spring Festival Gala* The robots used 3D Lidar and reinforcement learning to perform beat-synced full-body dance, including handkerchief spins and leg kicks. 4. *Google released two new Search Labs experimental features that handle phone calls on a user's behalf* - The first is an ‘Ask for Me’ for inquiring about local services. The second is a ‘Talk to a Live Representative’ for customer service calls. 5. *Microsoft added OpenAI's o1 model to Copilot as a new 'Think Deeper' mode.* Like DeepSeek's DeepThink, it handles complex questions by taking a few seconds to assess them from all angles. _Free for everyone!_ 6. *Chinese startup BridgeDP Robotics installed its robotic control system on AGIBOT's A2 humanoid, turning it into a dancer.* The system uses multi-sensor perception, motion tracking, and an open-source control algorithm to move A2's hands and legs in sync. 7. *Mistral launched Small 3, a 24B param open model that beats or matches GPT-4o mini and larger models like Llama 3.3 70B or Qwen 32B.* It is also 3x faster than larger models can run on your 32GB Macbook. 8. *Mark Zuckerberg reportedly plans to spend 65B on AI infra in 2025* Meta will deploy 1.3M GPUs with 1GW of compute power for a data center. This follows Stargate news, heating up the AI infra race in the US. That's it for this week's AI breakdown. We share the latest research every week, so follow us: https://whatsapp.com/channel/0029Va8iIT7KbYMOIWdNVu2Q Written by: Niraj Lunavat (https://www.linkedin.com/in/nirajlunavat)