🗞 News of the day

🤖 OpenAI Strawberry

OpenAI's new AI model, OpenAI o1, demonstrates revolutionary advancements in problem-solving and safety.

• Enhanced Problem-Solving: OpenAI o1 models have been trained to improve problem-solving and reasoning, achieving performance levels comparable to PhD students. They excel particularly in math and coding, scoring 83% on reasoning tasks and reaching the 89th percentile in Codeforces coding competitions.

• Advanced Safety Measures: OpenAI has implemented a new safety training approach, significantly enhancing the model's adherence to safety guidelines. The o1-preview model scored 84/100 in jailbreaking tests, compared to GPT-4o's 22/100, reflecting stronger safety protocols and partnerships with U.S. and U.K. AI safety institutes.

• Accessibility for Developers: The o1-mini model offers a more affordable option for developers, priced at 80% less than the o1-preview, while still providing robust reasoning capabilities. Access to these models is available to ChatGPT Plus and Team users, with plans to extend to ChatGPT Free users.

OpenAI continues to refine and expand their models, aiming to make advanced reasoning tools more accessible and beneficial for a wide range of professional applications.

🔖 Today's Cool Story

🚨 Google Robots

DeepMind's recent advancements in robotic dexterity are pushing the boundaries of what AI can achieve in dynamic environments.

• AI Systems Revolutionize Dexterity: DeepMind introduces ALOHA Unleashed and DemoStart, two innovative systems designed to enhance robot performance. These systems help robots learn complex tasks by imitating human actions and translating images into actions.

• Breakthroughs with ALOHA Unleashed: ALOHA Unleashed significantly improves bi-arm manipulation tasks, allowing robots to perform intricate activities such as tying shoelaces and repairing robots. It builds on previous versions with better ergonomics and learning processes, using diffusion methods for action prediction.

• DemoStart's Simulated Learning Success: DemoStart leverages reinforcement learning to master dexterous behaviors with 100x fewer simulations than real-world examples. It achieves high success rates in both simulated and real-world tasks, showcasing the potential for sim-to-real knowledge transfer using the MuJoCo simulator.

These advancements illustrate the growing capabilities of AI in robotics, highlighting the potential for robots to assist in various complex tasks across different environments.

🚀 Try New Coding Prompts

🎨Image Prompts

Prompt #1: Futuristic soldier in a high-tech combat suit

Prompt #2: Breathtaking mountain landscape at sunrise

Prompt #3: Cyberpunk-inspired megacity

See all Midjourney prompts →