Delphi-2M AI Predicts Diseases Years in Advance Using Health Records

PLUS: OpenAI tests ChatGPT alpha models with agent capabilities. Unitree's G1 robot shows rapid recovery skills. Google reveals 90% developer AI adoption. OpenAI recruits talent from Tesla's Optimus project.

In today’s agenda:

1️⃣ Delphi-2M AI system forecasts disease risks across 1,200+ conditions using UK Biobank data

2️⃣ OpenAI briefly released experimental ChatGPT alpha models with agent capabilities

3️⃣ Unitree's G1 humanoid robot demonstrates rapid recovery capabilities at $16,000 price point

  • Google's DORA survey shows 90% of developers now use AI tools daily despite 30% having trust issues

  • OpenAI recruits robotics experts from Tesla's Optimus project for humanoid development

MAIN AI UPDATES / 25th September 2025

🏥 AI predicts health risks years ahead 🏥
A new open-source AI model published on Nature can forecasts diseases like weather prediction using massive health datasets.

Researchers have developed Delphi-2M, a revolutionary GPT-based health prediction model trained on 400,000 UK Biobank records and validated on 1.9 million Danish registry individuals. This breakthrough system can forecast disease risks across more than 1,200 conditions, creating detailed health timelines that predict years into the future. The AI assesses probabilities for developing major conditions including cancer, diabetes, heart disease, and respiratory disorders, representing a paradigm shift toward preventive healthcare through predictive analytics. This advancement could transform medical practice by enabling early interventions and personalized health management strategies.

🤖 OpenAI tests ChatGPT alpha models with agent capabilities 🤖
Experimental models briefly appeared in ChatGPT selector before being pulled back.

Some ChatGPT users spotted a mysterious "alpha models" section that activated agent mode with variants labeled "Agent with truncation" and "Agent with prompt expansion." The models appeared to test different system prompt setups and underlying architectures for tool-augmented AI workflows. OpenAI quickly rolled back the accidental release, making detailed comparisons challenging, but the brief glimpse suggests the company is actively developing more sophisticated agentic capabilities for power users.

Unitree's G1 robot shows rapid recovery skills
A $16K humanoid with rapid recovery, dexterous hands, and cloud-shared skills.

Chinese robotics company Unitree has showcased impressive capabilities with its G1 humanoid robot, featuring an innovative 'anti-gravity' mode that enables rapid recovery after being knocked down. Priced at just $16,000, the G1 stands 92cm tall, weighs 35kg, and includes dexterous hands for complex manipulation tasks. The robot connects to Unitree's G1 Cloud platform, creating a collaborative ecosystem where robots can upload and share learned skills.

INTERESTING TO KNOW

🤖 Google reveals 90% developer AI adoption 🤖

Google Cloud's comprehensive DORA survey of nearly 5,000 tech workers reveals that 90% of developers now use AI tools daily, spending approximately two hours with AI assistants. Despite widespread adoption, significant trust gaps persist with 30% trusting AI outputs only minimally. However, productivity gains are substantial - 80% report faster work completion and 59% observe improved code quality. The study also highlights enhanced security capabilities through quicker vulnerability detection, demonstrating AI's dual role in accelerating development while raising questions about reliability and oversight in critical software development processes.

🦾 OpenAI expands into humanoid robotics sector 🦾

OpenAI has been strategically scaling its robotics ambitions through targeted team expansion, recruiting experts in mechanical design, sensing, and real-world mobility software. Recent job postings explicitly reference 'general-purpose robotics' objectives, while the company has secured trademarks for 'user-programmable humanoid robots'. Key hires include Stanford roboticist Chengshu Li and multiple engineers from Tesla's Optimus project, signaling serious intent beyond research.

📩 Have questions or feedback? Just reply to this email , we’d love to hear from you!

🔗 Stay connected: