This week in AI research: ‘o3’ and reasoning version of ‘Gemini 2.0’ redefine AI capabilities

Long the stuff of sci-fi, humanoid robots are moving from prototypes toward practical use in factories and labs. AI is shifting from massive models toward smarter reasoning — and now Google is getting in the AI reasoning game while OpenAI is teasing the next iteration of its reasoning model. Meanwhile, biotech efforts like comprehensive epigenetic mapping and improved medical imaging AI are setting the stage for breakthroughs in healthcare.

Additionally, Google DeepMind’s Project Astra hints at a future where universal AI assistants could aid users in real-time across devices. And Google’s new “reasoning” AI model, highlighted by a recent TechCrunch report, suggests that more powerful, context-aware tools are on the horizon.

---

What’s new this week?

Humanoid robots nearing real-world deployment

• Scaling up: Chinese companies are mass-producing humanoid robots, approaching 1,000 units. Near-term applications could include lab assistants and factory workers. On December 19, the company Pudu Robotics published a video (below) highlighting the capabilities of the PUDU D9 robot, highlighting how it could tackler, for instance, janitorial tasks. (Sources: Newsweek) and Interesting Engineering)
• U.S. push: U.S. players like Tesla are also advancing, targeting market-ready robots by 2026.
• Why it matters: Expect increased automation in research and manufacturing, potentially improving efficiency and safety by handling repetitive or dangerous tasks. Also anticipate the gradual uptick in adoption of AI agents that can perform basic tasks like data entry and evolve over time to handle more sophisticated duties.

AI reasoning continues to evolve

• Smarter models: OpenAI’s new “o3” model prioritizes enhanced reasoning capabilities over sheer size. The company touted continued progress in R&D and STEM–related tasks such as coding and math. (Source: The Information)
• Google’s new reasoning AI model: Google has launched its own “reasoning” AI model, demonstrating improved logical processing and self-checking steps to provide more reliable results. The model is available for free to test via https://aistudio.google.com/ (sign-up required). (Source: Tech Crunch)
• Universal Assistants (Project Astra): Google DeepMind’s Project Astra prototype hints at a universal AI assistant that can operate across devices, use visual context, and interact with tools like Maps and Lens. This could herald the emergence of more integrated, context-aware AI that can support research and daily workflows, potentially giving users an alternative to the chatbot interface that has defined many genAI systems. The video below is a demo Google released from about seven months ago showing Project Astra in action. (Source: Google Deepmind)